A11 Water is Wet: predicting in and estuaries.

Jessica Jane Meeuwig

Department of Biology, McGill University, Montréal

August 1998

A thesis submitted to the Faculty of Graduate Studies and Research in partial fblfillment of the requirements of the degree of Doctor of Philosophy

Q Jessica Jane Meeuwig 1998 National Library Biblioth&que nationale du Canada Acquisitions and Acquisitions et Bibliographie Services services bibliographiques 395 Wellington Sttwt 395. me Wellington OttawaûN K1A ON4 Otiawa ON KIA ON4 Canada Canada

The author has granted a non- L'auteur a accordé une licence non exclusive Licence allowing the exclusive permettant à la National Library of Canada tu Bibliothèque nationale du Canada de reproduce, loan, distribute or sel1 reproduire, prêter, distribuer ou copies of ths thesis in microform, vendre des copies de cette thèse sous paper or electronic formats. la forme de microfiche/film, de reproduction sur papier ou sur format électronique.

The author retains ownership of the L'auteur conserve la propriété du copyright in this thesis. Neither the droit d'auteur qui protège cette thèse. thesis nor substantial extracts fiom it Ni la thèse ni des extraits substantiels may be printed or otherwise de celle-ci ne doivent être imprimés reproduced without the author's ou autrement reproduits sans son permission. autorisation. nis thesis is dedicated tu Rob Peters who taught me to see patterns

" men yo~istop inventing realiy then you see things as they really are"

- Ben Okn, 1995 Astonishing the Gods Table of Contents .. . Abstract III Résumé iv Acknow ledgements v Funding Acknowledgements ix

Thesis Format and Contributions of Co-authors X Statement of originality xiii General Introduction 1 Circumventing phosphorus in management: a cornparison of 25 chlorophyll a predictions £kom land-use and phosphorus-loading models. Predicting coastal eutrophication from land-use: an empirical approach to small non-stratified estuaries. Turbid waters and clarifjmg mussels: their moderation of empirical Chknutrient relations in Prince Edward Island estuaries. Predicting coastal eutrophication in the Baltic: Vollenweider applied to Finnish estuaries. Predicting eutrophication in lakes and estuaries: a quantitative comparison of system response across a range of tidai energy, openness and salinity Generai Conclusion Appendix 1 : Lake Data Appendur 7: Raw physical and water chemistry data for PEI estuaries Appendix 3: Sumrnary data for PEI estuaries Appendix 4: List of Finnish sampling stations Appendix 5a: Growing season values for Chl by station by year for Finnish estuaries. Appendix fi: Growing season values for TP by station by year for Finnish estuaries. Appendix Sc: Growing season values for TN by station by year for Finnish estuaries. Appendix 6a: Annual pwing season averages for Ch1 Finnish estuanes Appendix 6b: Annual growing season averages for TP Finnish estuaries Appendut 6c: Annual growing season averages for TN Finnish estuaries Appendices 7: Final Chi, TP and TN values for Finnish estuaries Appendk 8: Raw data used to caicuiate watershed population densities for Finnish estuaries. Appendices 9a-9b: Mean flow rates for monitored (9a) and unmonitored (9b) in Firuiish estuaries. hppendix 10: huainonpoint source TP and TX ioads to Finnish estuaries Appendix 1 la: Annual growing season averages for Chl for sarnpling stations in US estuaries Appendix 1 lb: Annual growing season averages for TP for sampling stations in US estuaries Appendix 1 lc: Annual growing season averages for TN for sampling stations in US estuaries Appendix 12a: huaigrowing season averages for Ch1 for US estuaries Appendix 12b: Annual growing seascn averages for TP for US estuaries Appendix 12c: hua1growing season averages for TN for US estuaries Appendices 13a-13c: Mean values for US estuaries for Ch1 (13a), TP (13b) and TN (13c). Abstract

Coastal eutrophication, defined as an increase in algai biomass (as chlorophyll (Chl)) is of increasing international concem. Although coastal eutrophication wili Iikely increase as coastal populations grow, few models exist to support its management. Lake eutrophication has also long been recognized as an important environmental concem However, effective lake eutrophication management exists, supponed by regression and mas-balance models. Traditionally, these ''Vollsnweider" models hkland-use to Ch1 via total pliosphorus (TP), the nutrient considered to be limiting Chl. However, based on a data set of 63 lakes, Ch1 was more accurately predicted by models based on land-use than by those based on TP. This result provided the rationale to build Chl:land-use models for esnianes where the Chknutrient relations are unclear. Chkland-use models were developed for 15 estuaries in PEI, 19 esniaries in and 26 US esniaries. Land-use models predicted Ch1 more accurately than TP in the US estuaries and in some of the Finnish estuaries. In the Fuuiish estuaries, Ch1 was best predicted by a land-use model in estuaries dorninated by nonpoint source loading whereas Chi was most accurately predicted by the Vollenweider approach in estuaries dominated by point-source loading. In the PEI estuaries, the accuracy of the land- use model was comparable io the accuracy of the TP model. The PEI estuaries had much lower yields of Ch1 per unit nutrient than \&es suggesting differences among systems. This Ch1 deficit (expected - observed Chl) was accounted for by herbivory and turbidity, neither of which factors are exclusive to estuaries. The cornparison of Ch1 response to nutnents and land-use across lakes and estuaries demonstrated no systematic differences as a function of tidd energy, openness or salinity. The regression models based on the combined data accurately predicted Ch1 as a function of TP and pmentage of the catchment forested and mean depth. These results suggest that differences among systems are more Iikely a fiinction of pan-system properties such as herbivory rather than the presence of a lake-estuary dialectic, and that Iimnological approaches can effectively be applied to estuaries to support efforts to manage coastal eutrophication.

S.. Ill Résumé L'eutrophisation des zones côtières est une préoccupation d'une ampleur internationale, Bien que le problème d'eutrophisation des côtes tisque de s'accentuer à mesure que les populations côtières se développent, il n'existe à l'heure actuelle aucun modèle qui en permette la gestion. Par ailleurs, l'eutrophisation des lacs est aussi reconnue depuis longtemps comme constituant une problème environnemental de taille. Toutefois, dans le cas des lacs, la gestion de l'eutrophisation est rendue possible par l'application des modèles de régression et de bilans massiques. Ces modèles de type Volienweider relient l'usage du bassin versant a la chlorophylle a (chl), en passant par le phosphore total (PT), l'élément nutritif qui semble limiter la chlorophylle a. Une analyse de données provenant de 63 lacs a démontré que le type d'usage du bassin venant prédit plus justement la ch1 que ne le fait le PT. C'est sur cette observation ainsi que sur le fait que les relations chl-éléments nutritifs pour les estuaires demeurent obscures que s'est développée l'idée de construire pour les estuaires des modèles reliant la ch1 à l'usage du bassin venant. Des modèles chl-bassin versant ont donc été développés pour 15 estuaires de l'Île-du-prince-Édouard, 19 de la Finlande et 26 des États-unis. L'usage du bassin versant permet de prédire la chl avec plus de justesse que le PT dans les estuaires des États-unis et dans certains esniaires de la Finlande. Dans le cas de la Finlande, l'usage du bassin versant s'est avéré plus efficace que le PT pour prédire la ch1 la où les sources dibesde pollution dominaient sur les sources ponctuelles. Dans le cas des estuaires pollués par des sources ponctuelles, l'approche de Vollenweider s'est montrée plus juste. Dans le cas des estuaires de 1'1.-P.&, les deux types de modèles - usage du bassin venant et PT - étaient comparables en termes de justesse de prédiction de la ch1 Les estuaires de 1'1.-P.-É. montraient des rendements de chl par unités de PT plus faibles que ceux de lacs, laissant penser qu'il existe des différences entre ces deux systèmes. Néanmoins, la manque à gagner en ch1 (valeur prédite - valeur observée) des estuaires a pu être attribué à des facteurs qui sont communs aux deux systèmes: l'herbivorie et la tuhidité. Il n'y avait aucune différence entre lacs et estuaires quant à la relation chl- bassin versant et ce, pour toutes les étendues d'amplitude de marée, de temps de rétention hydraulique, ou de salinite. Il en allait de même pour la relation CM-élémentsnutritifs. Les modèles basés sur l'ensemble des données (lacs et estuaires) prédisent la chi soit à partir du PT soit à partir de la proportion du bassin versant boisé et la profondeur moyenne. Selon ces résultats, les différences entre lacs et estuaires tiendraient à des propriétés communes aux deux systèmes, telle I'herbivorie, plutôt qu'à l'existence d'une dialectique lacs-estuaires. Ainsi, les approches développées en Iuiinologie peuvent être appliquées avec succès aux estuaires afin d'appuyer les efforts de gestion de l'eutrophisation des zones côtières. I have the pnvilege of being Rob Peters' student. By example, Rob taught me to ask questions that were clearly testable. He taught me to make predictions and to evaluate the quality of my predictions. He fiequently reminded me that science was fun and, again by example, he taught me to work hard, think hard and write hard. 1haven't learned al1 of Rob's lessons but they stand as testimony to his generous mentoring. Rob ensnared me as an impressionable undergraduate when 1 attended his course in predictive ecology. His heresy was captured in two (paraphrased) statements that were sufficiently influential that I felt compelled to join his lab 6 years later: Evolution is a nice story but it isn't science Understanding is nice but, not only are we unlikely to achieve it, it is unnecessary to science. These are uncornfortable statements. In his Sceptical Essuys, Bertrand Russe11 (193535) claims that "thegreat scandals in the philosophy of science ever since the time of Hume have been causality and induction. We al1 believe in both, but Hume made it appear that our belief is a blind faith for which no rationai ground can be assigned .. . and yet, in common with everyone else, I cannot help believing there must be an answer [proving the validity of causality and induction]". Russell goes on to write that he doesn't believe anyone has yet corne up with an effective response to Hume. Rob's legacy is to heighten this tension. He makes it clear that effective ecology requires relinquishing our pursuit of the grail of understanding. Moreover, our environmental cnsis demands this. 1 also have the pnvilege of being loe Rasmussen's student. Joe's willingness to adopt me mid-way through my PhD, despite my general unwillingness to be adopted, showed bdness which 1remember with warmth. Joe has taught me to ternper my somewhat rigid position on empiricism and what some might see as a tendency towards describing the world in black and white. loe labeled me schizophrenic in my attempts to maintain my empincal position whilst musing about mechanisms. This schizophrenia is of course only a problem so long as 1 refuse to rellliquish a rather single- (bloody?) minded approach to science. 1remain unable to renounce Rob's empiricism but it is a mark of rny growth at McGill that 1 am able to appreciate (and apply) Joe's creative opportunism. Their combined gift is perhaps captured in the following rnaxim:

Always predict clearly defined ecologically relevant characteristics by whatsoever means do so most accurately.

In a recent essay on feminist empincisml, science is defuied as an activity iduenced by the values of communities. My science has certainly been highly infiuenced by the community of scientists in which 1 work. Adrian deBruyn has helped sharpen both the prose and the ideas of most of this thesis. He is also one of the last hard core ernpiricists arnong the graduate students and 1value his Company and arguments on the empincal side of the Journal Club table. Marc Trudel has for 4 years sat on the mechanist side of the Journal Club table, Both of us have moved somewhat towards each other over this time and it is a tribute to the quality of Marc's arguments that I have moved ut ail. Maite Maldonado sets a wondefil example of what a scientist should be - fiery and passionate (about iron, go figure) yet meticulous and precise. Neil Rooney Grequently reminds me of two basic truths: field work is fun (especially at the lake), and science does not have to be contact sport. Gray Stirling has prompted me frequently to think outside the confines of limnology - from Cartesian coordinates for life to patches. The richness of these 5 years has also been enormously hcreased by Andre Cimbleris, Carolyn Hall, Murray Humphries, and Renate Lehmann. My community also includes three women whose influence is strong. Joan Manden was my first mentor and it pleases me no end that she is here to see me complete another step in the process of becorning a scientist. Amanda Vincent has opened up a whole new world to me and her faith in my ideas about empiricai fisheries models is both gratifying and temfjmg. Nella Cattaneo has tau@ me about strength. 1must also thank my committee members, Yves Prairie and Neil Price. 1 suspect I didn't bug them as often as I threatened to when "inviting" them to sit on my committee.

I Nelson, L.H. 1990. Who knows: fiom Quine to a feminist cmpiricisa Philadelphia: Temple University Press. However, they were there for me when 1 did. 1particularly appreciate Yves' willingness to read a flood of Finnish(ed) manuscripts in the final months of my PD.Jaap Kale behaved as if he were on my comrnittee, taking a strong interest in my research and encouraging me to develop a more balanced and measured approach to science. I am grateful for these lessons and continue to try to leam thcm. Support and administrative staffrnake the world a happier and easier place. 1 thank Robert Lamarche, L ynda McNeil, Kathryn Peterson, Elena Roman and Carole Verdone-Smith for respectively, finding lost files, ensuring I filled in al1 my fonns, keeping me paid, facilitating lab work and providing emergency slides on ndiculously short notice. 1am grateful. This PhD would have been difficult to complete without the able and careful work of a number of research assistants. Elsie Sunderland shared 3 months of non-stop sarnpling with me in PEI with humour and endurance. Daniel Chamberland and Tanin Nayar ran thousands of chlorophyll and nutrient samples on the not-so-auto-autoanalyzer and Rosalie AlIen and Susan Innis crunched data 'tii the Pentium choked. The field work in PEI would have been impossible without the help and in-kind support of a number of people. Brenda Penak of the Bedeque Bay Environmental Management Association and Randy Angus of the Department of Fisheries and Oceans and the Southeast Environmental Association provided essential logistic support including housing, transport (at times of Trooper collapse) as well as a host of contacts and much moral support. Bruce Raymond and Clair Murphy of the PEI Department of Environment and Forestry provided valued advice as well as oppominities to introduce my research to people involved in environmental management. 1 must also thank the myriad, generally nameless, good samaritans in PEI who rescued me fiom the mud, slippery boat rarnps, ensnaring musse1 lines and temperarnentd boat and car engines. My thesis has involved much travel and I would also like to thank Lars Hâkanson, Johan Penson and Mona Petersson of the Mitute of Earth Sciences, Uppsala University and Heikki Pitkinen, Pirkko Kauppila, Petri Eckholm, Oiva Rekolainen and Ansti Heiskanen, of the Finnish Environment Institute, Helsinki, for their invitations to visit, interest, collaboration, and friendship during my visits. The fiends and colleagues 1 met

vii during these trips have done much to expand my understanding of how culture and science intertwine. And I had a lot of fûn. The remalliing challenge of these achowledgements lies in finding an innovative way to thank my parents. This is difficult in that attempted innovation rnight cause me to overlook the traditional @fisof parents to child: unconditional love, irrational belief that the child can accomplish anythng, sincere but puzzled interest in the child's research, emergency financial aid, good food and a warm bed far hma cold and lonely laboratory. In rny case, it also included a maniacal dnve tlirough a very long night lioin PEI to Montreal with f'rozen samples. They remained fiozen and my thesis was saved. I've held luctor et emergo close as a family touchstone, believing, as you, that 1 would. Fuoding Acknowledgements

1 and my research were supported by funds hma number of sources. 1 acknowledge NSERC for two years of Funding through a PGS-B scholanhip and contemporaneously for three years of research support on Rob's operating grant. I also acknowledge the support of FCAR through the Groupe de Recherche Interuniversitaire en Limnology that allowed me to spend the in . A scholanhip £?om the Fimish Center for International Mobility allowed me to complete the comparative study of Baltic and North Amencan estuaries. Intramurally, 1 am grateful for receiving the Vineberg scholarship. 1 would also like to acknowledge Don Kramer's efforts to secure hinding for the completion of my thesis following Rob's death and the generosity of the Faculty of Graduate Studies and the Department of Biology in stepping into this breach. At the same time that I acknowledge these efforts and sources of support, 1 feel it necessary to document NSERC's withdrawai of support as a result of Rob's death. NSERC apparently felt no responsibility towards the completion of Rob's research program despite having supported his research for many years. Such a view is shon- sighted as it ensures that research in progress in which resources have already been invested will remain incomplete. It is irresponsib le to those individuals, named as researchers in the grant, who had a reasonable expectation of continued huiding for their research given Rob's granting record. NSERC's decision also meant that those struggling with Rob's death on a personai level had to simultaneously re-arrange professional lives. In this NSERC shows itself hardhearted. Thesis Format and Contributions of Co-authors

Doctoral candidates at McGill University may submit a thesis based on a senes of manuscnpts that presents a coherent research programme. If this option is chosen, in accordance with Faculty regulations, the following text is included to inform the external examiner of Faculty regulations regarding the submission of a manuscnpt-based thesis.

Candidates have the option of hcluding, as part of the thesis, the text of one or more papen submitted or to be submitted for publication, or the clearly duplicated text of one or more published papen. These texts must be bound as an integral part of the thesis.

If this option is chosen, connecting texts that provide logical bridges between the different papers are mandatory. The thesis must be written in such a way that it is more than a mere collection of manuscripts; in other words, results of a series of papen must be integrated.

The thesis must still conform to al1 other requirements of the "Guidelines for Thesis Preparation". The thesis must include: a Table of Contents, an abstract in English and French, an introduction which clearly states the rationale and objectives of the study, a review of the literature, a final conclusion and summary, and a thorough bibliography or reference list.

Additional matenal must be provided where appropriate (e.g. in appendices) and in sufficient detail to allow a ciear and precise judgement to made of the importance and onginality of the research reported in the thesis.

In the case of rnanuscripts CO-authoredby the candidate and others, the candidate is required to make an explicit statement in the thesis as to who contnbuted to such work and to what extent. Supervisors must attest to the accuracy of such statements at the doctoral oral defense. Since the task of the examiners is made more difficult in these cases, it is in the candidate's interest to make perfectly clear the responsibilities of al1 the authors of the CO- authored papers. 1 have chosen to submit a manuscript-based thesis which consists of the lollowing papers. These papen will hencefonh be referred to by their roman numerais.

I MEELWIG,J.J. and Peters, R.H. 1996. Circumventing phosphorus in lake management: a comparison of chlorophyll-a predictions Erom land-use and phosphorus-loading models. Cm. J. Fish. Aquat. Sci. 53(8): 1795- 1806

II MEELWIG,J.J. n.d. Predicting estuarine eutrophication &om land-use variables. Mzr. Ecol. Prog. Ser. Accepted.

[II MEEUWG,S.S., Rasmussen, J. and Peten, R.H. n.d Thid waters and clarifjmg , mussels: their moderation of Chknutrient relations in estuaries. Mar. Ecol. Prog. Ser. . Accepted.

IV MEEWVIG,JJ., Kauppila, P., Pitkihen, H. n.d Predicting coastal eutrophication in the Baltic: Vollenweider applied to Finnish esniaries. Submitted to Can. J. Fish. Aquat. Sci.

V MEEUMG,J.J. n.d. Predicting eutrophication in lakes and esniaries: a quantitative comparison of system response across a range of tidal energy, openness and salinity. Submitted to Limnol. Oceanop In accordance with Faculty regulations, 1 must also indicate the contributions of myself and my CO-authorsto the pubiished and submitted manuscripts (indicated by roman nurnerals).

1 Rob Peters suggested that I begin my thesis by comparing TP and land-use models in lakes as regression models had already been used extensiveiy by limnologists and because there already existed Ch1:TP regressions. Thus the idea for the paper was Rob's. 1 was responsible for choosing and compiling the data, designing the analysis and writing the paper. Rob provided extensive editing help.

II The idea for this paper was mine but the paper was the result of many discussions with Rob Peten. I was responsible for the sampling design, collection of data. statistical analyses and writing. Neil Price, as a member olmy supervisory cornmittee reviewed the manuscript prior to its submission.

III This paper was the result of working in proximity to Joe Rasmussen's lab where mas-balance models abound. The idea of mass-balancing biomass was mine. However, Joe Rasmussen provided essential guidance in implementing the idea. 1 was responsible for the analyses and writing. Joe provided several critical reviews.

IV This paper was the asking pt-ice for access to the Fimish data for paper V. Heikki Pitk&en, as head of the coastal research group of the Fimish Environment Institute, wished for an in depth analysis of the Fimish coastal monitoring data and the development of predictive eutrophication models specific to Finnish estuaries. 1 and Pirkko Kauppila assembled the data set. 1 was responsible for the regression and mus- balance modeling. 1 wrote the first draft of the manuscript with the exception of parts of the Methods section detailing the monitoring data. Pirkko and Heikki both provided critical background information on the estuaries, land-use activities etc. as well as reviews.

V 1 was solely responsible for this paper.

xii Statement of Originality

This thesis represents an original contribution to the study of eutrophication in fresh and coastal waters. To my knowledge, the manuscnpts (1 - V) are the fint to:

1) predict phytoplankton biomass (as chlorophyll-a (Chl)) as a direct function of land-use in both lakes (1) and estuaries (II, IV. V) 2) predict Ch1 as a function of total nutrients in estuaries (II, IV, V) 3) apply a mas-balance approach to phytoplankton biomass to assess the affects of herbivory (III) 4) quantitatively compare responses of lakes and estuaries to total phosphoms and land-use

These results reflect a novel approach taken to eutrophication in that the traditionai focus on single nutrients was replaced with a focus on land-use. Land-use was chosen as it integraies the effects of a number of variables that likely detemine Chl suct r; TT,Ti<, iurbidity. Moreover, while nutrients may be the proximate cause of eutrophication, land- use is the ultimate cause and a variable which must be managed if eutrophication is to be addressed. The approach was also novel in that it applied regression and rnass-balance models to estuaries. These tools have successfully been used to predict and manage lake eutrophication. However, no Chl:TP or TP mass-balance analogous to the lake models have been developed for estuaries. The application of this approach to estuaries allowed me to quantitatively test the assumption that lakes and estuaries differ in their response to disturbance.

xiii It is difficult to imagine anything more pleasant than rocking gently in a boat, listening to the slap of waves against the hull. Cool crystal water demands a late aftemoon swim. It is thus little wonder that lake and coastal eutrophication are firmly entrenched in political and scientific agendas. The use of surfaces waten typifies the collision between aesthetics and economic growth: shoreline real-estate properties are the most expensive yet surface waten are the most "cost-effective" receptacle for the wastes generated by industrial society. As lakes and coastal waten tum green with algae and black with anoxia, pressure has mounted to "clean up the watef '. Govemments have responded with ambitious scientific and management programs on local, national and international scales. Interest in eutrophication is not, however, tied only to applied problems. Identifying pattems in the distribution of phytoplankton biomass has been a research theme in both limnology and oceanography since the tum of the century. This thesis continues the tradition, identi fying pattems in phytoplankton biomass in lakes and estuaries as a function of human disturbance. This chapter introduces my thesis with 1) a brief history of the science and management of eutrophication, 2) a review of current lake eutrophication research and the lessons learned, 3) a review of coastal eutrophication research and 4) a presentation of the thesis goals and hypotheses.

A Brief History of Eutrophication Eutrophication has captured the interest of limnologists since the early 1900's. Naumann (1 91 9) documented differences in phytoplankton biomass of highland mountain Iakes and lowland cultivated lakes. He related environmental factors such as temperature, light, nutrient concentration to primary producer biomass (Home and Goldrnan 1994). Adopting the tenns oligotrophic and eutrophic nom Weber's research on nutrients in (1907 in Cole 1994), Naurnann developed the fint lake typology based on trophy. Thienemann (1925) independently arrived at the same typology, dividing lakes into oligotrophic and eutrophic based on the presence or absence of benthic invertebrates and lake morphometry. Oceanographen did not use the îrophic tenninology but were also concemed with patterns of primary production. Early research focused on the differences between tropical and temperate waters (Mills 1989). Brandt (1899 as cited in Mills 1989) focused on differential rates of denitrification as the deteminant of latitudinal patterns in phytoplankton abundance. Highly influenced by agricultural chemistry, his theory was predicated on Liebig's law of the minimum which stated that: The different substances necessary to the growth of a plant, or the different articles of th& food, are ull oJrqtrul vulue; that is to Say, if one out of the whole nurnber be absent, the plant will not thrive (Liebig 1840).

Liebig's Law of the minimum has, as the paradigrnatic limiting nutrient concept, remained a cornerstone of fieshwater and marine research on primary production and eutrophication. Definitions of eutrophication are numerous (c. f. Wetzel 1975, Cole 1994, Home and Goldman 1994). Most incorporate a sense of the original Greek eutrophes or well nourished (Woo 1f 198 1) and thus nutrients are central to eutrophication. Di fferences in definition primarily reflect 1) the response variable included: none, one of biomass, primary productivity, Secchi depth or oxygen concentration, or more than one of these, and 2) whether eutrophication is a process or an effect. In this thesis, 1 define eutrophication as the increase in phytoplankton biomass, incorporating the specific response that has been most studied of al1 eutrophication effects. This choice is similar to that of Nixon (1 995) who defined eutrophication as the increase in primary productivity, also without reference to nutrients. 1 have not included nutrients explicitly as 1 intend to demonstrate that land-use predicts eutrophication and because other factors such as turbidity (Pe~ock1985) and herbivory (Officer 1982, Carpenter et al. 1985) may also determine eutrophication. Lake eutrophication is considered to be a naturd process occuning over geological time. Oligotrophic lakes progressively fil1 with both autochthonous and aIlocthonous organic material; as the lake becomes shallower, nutrient recycling, primary production and biomass increase. The lake becomes mesotrophic then eutrophic and ultimately becomes a (Cole 1994, Home and Goldman 1994). Cultural eutrophication Is defined as the acceleration of this naturai process due to increased anthropogenic nutrient loading, primarily through the addition of and agricultural fertilizers (Home and Goldman). Early work did not identify cultural eutrophication as an important environmental concem. Some lakes were considered to be naturally eutrophic thus Naumann (19 19) placed as much emphasis on the importance of geographic location (highland vs. lowland) as on human impact (pnstine vs. cultivated). Strem (1928) also attaches little importance to anthropogenic disturbance in his description of eutrophication: the natural process of the maturing of a lake is that of eutrophication ... many of the North German lakes are now in this stage of changing and it is clear that a supply of nitrogen compounds and phosphates from cultivated fields can do much to accelerate the change, even if it be not the main cause (in Welch 1935, italics added).

The lack of emphasis on human impacts extended into the 1950's: Stram's definition of eutrophication was still used in textbooks at this time (Welch 1952). However, concem about cultural eutrophication was mounting as seen in an early paper demonstrating the relationship between eutrophication and domestic drainage (Hasler 1947). Currently, there remains debate over the extent to which lakes are naturally eutrophic (Anderson 1995). While some naturally eutrophic lakes have been identi fied in catchrnents with phosphorus rich soils (Murphy et al. 1W), other lakes previously identified as naturally eutrophic had actually become eutrophic as a result of medieval land use (Klein 1993, Anderson 1995). Results such as these demonstrate the central role of land-use in determining aquatic trophy. Cultural eutrophication of lakes surfaced as an environmental issue in the early 1960's. Excessive phytoplankton biomass was associated with increased incidence of noxious and toxic algal blooms, decreased oxygen concentrations and anoxia (Hendenon- Sellen 1987). Economic impacts include loss of recreational value and aesthetics, decreased quality of potable water and loss of livestock ingesting toxic phytoplankton. As the water quality of important systems such as Lake Erie and myriad mal1 lakes declined, public pressure mounted on both govenunents and scientists to clean up the lakes. A nurnber of scientific and management initiatives were begun in the 1960's. in 1964, the Great Lakes international Joint Commission established a scientific cornmittee to evaluate water pollution in the lower St. Lawrence Great Lakes and to formulate management recommendations. In 1967, under the auspices of the US National Academy of Sciences, 600 peaons from 12 c0unb-k~attended a symposium entitled "Eutrophication: Causes, Consequences, Correctives" (National Academy of Sciences 1969). The summary of the symposium proceedings States that: Human sewage and industrial wastes are significant sources of nutrients that contribute to eutrophication of lakes. Drainage fiom farm land is also an important source, most of the nutrients coming from fmmanure ... Substances other than inorganic phosphorus and nitrogen compounds soiitribute to eutrophisation. Examples are vitmins, growth liorniones, amino aids, and trace elements. pg. 4

Phosphorus and nitrogen were identified as the most important in causing cultural eutrophication but their relative importance was not resolved. In 1972, the IJC cornmittee on pollution in the lower St. Lawrence published their report (mon. 1969). It included 19 recommendations; the first and rnost controversial recornmendation called for the total replacement of phosphate-based detergents by 1972. This recommendation provoked an immediate and strong reaction from the US detergent lobby. In 1969, industrial detergents were 30-70% sodium triphosphate by weight. Sodium triphosphate was used as a "builder" in combination with the surfactant or cleaning agent. The builders complex with calcium and magnesiurn ions to improve the e fficac y of the surfactants, essentially acting as water so fteners. In 1970, the detergent lobby did not have an alternative (Vallentyne 1974). Some scientists from academia (Lange l967), governrnent (Kerr et al. 1970) and industry (Kuentzel 1969) claimed that carbon rather than phosphonis was the key nutrient controlling lake eutrophication. In identifjmg carbon as the limiting nutrient in lake eutrophication and rejecting phosphonis abatement as an appropriate eutrophication control, these scientists assumed that cause and control were synonymous (Vallentyne 1974). However, phosphorus was targeted for management not because there was unanimity on its role in limiting phytoplankton biomass but because there was unanimity on society's ability to control it. Fint, of the three nutrients implicated in eutrophication @hosphonss, nitrogen and carbon), both nitrogen and carbon have atmosphenc pools on which phytoplankton hypothetically may draw. Access to these pools is not controllable. Altematively, phosphonis is only available via external and intemal loading, both of which can be controlled. Second, tec hnology exists to remove phosphorus fiom waste waters; targeted nitmgen removal is much more expensive. Third, as of 1969, more than 1 10' kg yil of phosphate detergents were in use in the US alone, accounting for 50% of the P in municipal waters (Vallentyne 1974). Finally, even if phosphonis is not limiting when P abatement begins, removal of P would increase N:P and C:P ratios to the point of P limitation with an eventual decrease in phytoplankton biomass (Vallentyne 1974). The importance of P in controlling eutrophication appeared clear as of 1970. Canada enacted lederal legislation requiring complete removal of P fiom detergents by 1974. In the United States, New York and indiana placed a complete ban on phosphates in detergents while others States (e.g. Connecticut, Florida and Michigan) decreased allowable detergent P levels to 8.7%. The US Federal govemment failed to enact any national legislation, choosing instead encourage jurisdictions to decrease phosphoms discharges From sewage facilities (Vallentyne 1974). This decision suggests that the US govenunent was convinced of the importance of phosphonis abaternent. That the US govemment placed the burden on municipal govemments rather than detergent manu facturers suggests that powerful lobbies were at work. Confirmation of the central role of phosphorus in causing lake eutrophication came fiom three influential studies published in the early 1970s. Using an expenmental approach, Schindler (1977) demonstrated the reliance of phytoplankton biomass on phosphoms at the ecosystem level. Lake 226 in the Experimental Lakes Area is figure-8 shaped with two basins comected by a shallow neck. The basins were separated by a vinyl curtain with carbon and nitrogen added to the southwest basin and carbon, nitrogen and phosphoms added to the northeast basin. The northeast basin rapidly developed a phytoplankton bloom while the southwest basin showed no significant effect of the treatment. Concurrently, Dillon and Rigler (1974) used a comparative, empirical approach to quanti@ the relationship between phytoplankton biomass (measured as chlorophyll-a, Chl) and total phosphoms (TP). They applied standard regression techniques to 77 Mes nom North Amenca and Japan and demonstrated that 96% of the variation in Ch1 could be accounted for by variation in TP in lakes thought to be phosphorus limited. This comparative approach was revolutionary in that it abandoned the notion that lakes are unique entities to be studied individuaily. It assumed instead that there are general patterns to be seen across lakes. Vollenweider (1975) also revolutionized the way in which lakes were studied: he abandoned the reductionist approach that compartmentalized lakes, instead modeling lakes as chemostats. He used a mass-balance mode1 to predict the in-lake concentration of TP as a function of phosphorus loading to the lake which in turn was estirnated fkom catchent land-use. Combined, these studies demonstrated that that phosphorus limits phytoplankton biomass in lakes (Schindler 1977), provide a simple mode1 to estimate the arnount of phosphorus in the system (Volienweider 1975) and allow lake managers to predict the amount of phytoplankton they can expect to see given an arnount of phosphorus (Dillon and Rigler 1974). The success of this approach in predicting eutrophication is attested to by its widespread use in lake management (OECD 1982, Hutchinson et al. 1991, Dillon et al. 1994).

Current Research in Lake Eutrophication Since these seminal studies, regional Ch1:TP relations have proliferated (n>60; Peten 1986). These studies include global (OECD 1982) and regional regression relationships, including New Zealand (Pridmore et al. 1985), Missouri (Jones and Knowlton 1993), Australia (Outridge et al. 1989) and Argentina (Quiros 1990). Many of these relationships are similar to the original Dillon and Rigler (1 974) relation. Others, such as the relation for Alberta's saline lakes, show remarkably low yields of Ch1 per unit TP (Bierhuizen and Prepas 1985). This suggests a need for a meta-analysis of lake Ch1:TP relations to idenci@ systematic differences among lakes. Research has also focused on identifjmg the role and relative importance of nitrogen in predicting eutrophication. Molot and Dillon (1991) and McCauley et al. (1 989) demonstrated that the relationship between Ch1 and TP is not linear. As TP increases, Ch1 becomes asymptotic. They interpreted this as evidence of nitrogen limitation in eutrophic systems. Prairie et al. (1989) also demonstrated the importance of nitrogen in limiting phytoplankton biomass. Covariation of TP and total nitrogen (TN) make it difficult to separate their effects on Ch1 in regression analyses. However, they were able to demonstrate systematic changes in the slopes and intercepts of the Chl:TP regression equations as trophic status increased. Research has also focused on improving estimates of the terms utilized in the mas-balance equation. There are at least 15 formulations of Vollenweider's basic mass- balance equation (Canfield and Bachrnan 198 1 ). Phosphoms retention is particularly difficult to measure directly thus researchers have developed empirical models to predict it as a function of, for example, lake area (Kirchner and Dillon 1975). Phosphorus load is also difficult and costly to mesure directly and researchen have considered the effects of land-use (Dillon and Kirchner 1975), geology (Dillon and Kirchner 1975, Duarte and Kalff 1989), sedirnent load (Rowan and Kalff 199 1), catchment size (Prairie and Kalff 1986), and drainage basin morphometry (Kirchner 1975) on P export to lakes. The result of such activities is a compilation of phosphorus export coefficients to be used in estimating phosphorus loads (Reckhow and Chapra 1983). The importance of land-use in determining the delivery of nutnents to aquatic systems is also highlighted in reviews such as that of Sharpley et al. (1994). Land-use has been rnodeled intensely, particularly with the arriva1 of geographic information systems (GIS)which allow easy and more accurate estimates of land-use. (Sumer et al. 1990). Paleolirnnology has also demonstrated the importance of land-use in affecting lake trophic status (Klein 1993, Anderson 1995, Cooper 1995). Although most lake eutrophication research has focused on bottom-up control of phytoplankton by nutrients, the trophic cascade hypothesis (Carpenter et al. 1985) has suggested an important role for top-down control of eutrophication. Quiros (1 990) compared data from 1 10 lakes and reservoirs in Argentina, demonstrating that lakes with large filter feeding zooplankton had lower levels of Ch1 at a given TP level than lakes with zooplanktivorous fish and small zooplankton. The Ch1 yield was almost an order of magnitude lower in the presence of large zooplankton suggesting a strong top-down effect. Mellina et al. (1995) present evidence for top-down control of phytoplankton by zebra mussels (Dreissenapoiymorpha) in lakes Erie and St. Clair. Mazumder (1 994) also showed a strong top-dom effect in that the Chi:TP yields in lakes with large filtering Daphnia are lower than the yields in lakes without Daphnia. Such evidence has suggested a role for biornanipulation of lakes to control eutrophication. Limnologists have clearly advanced in their abilities to predict lake eutrophication. Four lessons seem apparent: Liebig's (1 840) "law of the minimum" is a useful paradigrn within which the "cause" and "control" of eutrophication can be sought; the evidence for general P-limitation appears convincing because it is denved Crom empincal observations, bioassays and whole-lake experiments (Hecky and Kilharn 1988); standing stocks of total nutnents are more successful in predicting phytoplankton biomass than inorganic nutrient concentrations (Schindler et al. 1971 ), reflecting the extremely rapid turnover of phosphate in lakes (Rigler 1964); the elegance and accuracy of regression and mass-balance models have demonstrated the snengths of both comparative approaches and simple rnodels in predicting eutrophication. The accuracy of regression models should be qualified. While the confidence limits of some regression models may span as much as an order of magnitude, this error should be considered relative to that of simulation models. In providing a synopsis of the 1974 workshop "Modelling the Eutrophication Process" that occurred at the sarne time that Dillon and Rigler (1974) published their regression model, McGaughey described the then state-of-the-art in modeling as far too primitive to produce either the univenal model or to evaluate correctly the inputs which should go into it ... in our ability to model a lirnnological situation, we are as infants newly discovering that modeling clay will 'squish' in our hands (McGaughey 19742).

More recently, Van Straten (1986), another simulation modeler, also indicated that if predictive accuracy is the goal, empincal models should be chosen over simulation models. The residual error of Ch1:TP regressions has been attributed to a number of physical, chernical and biological factors (Prairie et al. 1989). Moreover, predicting Ch1 fiom a TP value that was itself calculated fiom a mas-balance model in which P-loads were denved from land-use adds enor at every stage. As models are chained together like this, error cornpounds rather than increasing linearly (Reckhow and Simpson 1980). A major challenge for limnologists remains the reduction of this error. Coastal Eutrophication Coastal eutrophication has been slow to rnanifest itself on research and management agendas compared to lake eutrophication (Nixon 1995). Like lakes, excess phytoplankton biornass is associated with increased incidence of noxious and toxic algal blooms (Paerl 1988). decreased oxygen concentrations and anoxia (Turner and Rabalais 1994), and changes in benthos and fish biomass (Cederwall and Elmgren 1990, Desprez et al. 1992). Ryther (1 954) fint documented the effects of increased nutrient loading fiom duck fmsin Great South Bay, NY that resulted in nuisance blooms, high turbidity and declines in the oyster industry. While there were two estuarine papen included in the National Academy of Sciences eutrophication symposium (l969), it would be 25 years until the influential "Estuaries and Nutrients" symposium was held in 1979 (Neilson and Cronin 198 1). However, problems associated with coastal eutrophication are increasing and, correspondingly, interest in coastal and marine eutrophication exploded in the early 1990's (Rosenberg et al. 1990, Colombo et al. 1992, Vollenweider et al. 1992). The US Federal Govemment initiated the National Estuarine Eutrophication survey (Bricker and et al. 1995) which has resulted in a qualitative assessment of the major US estuaries. Coastal and lake eutrophication research share a number of common characteristics. As in lake eutrophication, the paradigrn of limiting nutrients has govemed much coastal eutrophication research. Based on Redfield's (1958) work, geochemical oceanographers generally consider the oceans to be P-limited. Redfield (1 958) demonstrated that the ratio of C:N:P in particulate matter in the ocean is 106: 16: 1 hypothesized that over geological tirne, in the absence of terrestrial material, this ratio reflects cellular needs. Given atmospheric pools of C and N, he hypothesized that P was most likely to be limiting. However, Ryther and Dunstan (1 97 1) used empirical observation of in situ nutrient depletion and bioassays to demonstrate nitrogen limitation in coastal waters off of NY.They argued that nitrate disappeared from the water column prior to phosphate and that only nitrate addition caused an increase in biomass in their assays. This work was seminal and nitrogen is still generally considered to be the nutrient most limiting coastal phytoplankton (Howarth 1988). Reflecting nitrogen's paradigmatic position, most coastal eutrophication research focuses on nitrogen. A review of the 1995-1 997 biological abstracts shows that of 596 articles on estuaries and nutrients, 52% consider only nitmgen, 32% refer to both nitrogen and phosphorus, and 16% consider only phosphorus. Demonstrations of nitrogen limitation include: observations of in situ nutrient concentrations and ratios (Ryther and Dunstan 197 l), bioassays (Ryther and Dunstan 1971) experiments (Oviatt et al. 1995) and, relatively rarely, comparative studies predicting Ch1 as a function of DM load (Boynton et al. 1996). Despite the preponderance of evidence for nitrogen limitation, other researchers have demonstrated limitation by phospliorus, carbon and silica. Phosphorus is considered to be limiting in some tropical estuaries (Lapointe and Clark 1992), the North Sea (Brockmann et al. 1990) and in the Mediterranean (Krom et al. 1991). Phosphorus is also considered to be seasonally limiting in some temperate estuaries: these systems are characterized by P limitation in the spring followed by N limitation in the summer (D'Elia 1986, Dodds and Priscu 1990, Granéli et al. 1990). Smith (1989) argued for carbon limitation in the net heterotrophic Tomales Bay, California. Silica limitation is usually important in prompting shifls fiom silica-requinng diatoms to dinoflagellates (Conley and Malone 1992). Like limnologists, coastal researchers have also explored the relationship between land-use and coastal eutrophication (Correll et al. 1992). This interest in part reflects the belief that nitrogen is the key limiting nutrient in coastal waters and that the majority of nitrogen is delivered to coastal waters fiom nonpoint sources such as fms, making a focus on land-use essential. As there are similarities between lake and coastal eutrophication research, there are also differences. First, coastal researchers have focused on inorganic nutrients rather than total nutnents. This focus is problematic because the conclusions of nutrient depletion are ambiguous. Are arnbient inorganic nutrient concentrations low because the system is naturally oligotrophic or are they low because phytoplankton are blooming and have dnven the concentrations to low levels ? Moreover, the application of Redfield ratios to ambient inorganic concentrations is problematic because the analytical measurement limit of phosphate is very close to the concentration indicating nutrient limitation (Fisher et al. 1995). Thus, we cannot be sure whether we are seeing nutrient limitation or simply the limits of our technology. A second difference lies in the scales at which nutrient limitation is tested in coastal systems. The majority olevidence for N-limitation cornes fiom in situ observations of inorganic concentrations and bioassays. The difficulties with inorganic nutrient concentrations have already been identified. Bioassays are problematic in that 1) it is unclear how results fiom small volume, short term (ml; hrs) experiments cmbe scaled up to whole ecosystems on longer time periods; 1) they measure increased growth rate rather than increased biomass yet biomass is ofien the response variable of concem and 3) there may be experimentai artifàcts associated with small bottles. Since the review by Hecky and Kilharn (1988)' a series of mesocosm experiments on N vs. P limitation have been conducted (e.g. Oviatt et al. 1995), demonstrating nitrogen limitation. To date, whole system experiments such as those of Schindler (1977) that provided such convincing evidence of P limitation in lakes are absent in coastal areas. A third di fference between lake and coastal researchers lies in the choice of modeling approaches. Coastal researchers have generally eschewed the comparative approach of limnologists. To my knowledge, there are no relationships that predict coastal Ch1 as a function of either RI or TP in the literature analogous to the work of Dillon and Rigler (1974). Examples of comparative models in the coastal literature include Nixon's (198 1) model predicting benthic remineralization as a function of primary production and organic input (i=0.94),Boynton et al's (1982) models predicting Ch1 as a function of nitrogen load (i-0.87) phosphorus load (8=0.87), Monbet's (1 992) model predicting Chl as a function of DM in micro and macrotidal estuaries, and Boynton et al's (1996) model predicting Chl as a function of nitrogen load (?=0.96). Such comparative work however remains rare. Coastal ecologists have however made much use of the mas-balance approach of Vollenweider in estuaries. Examples include Boyle et al. (1974), Jordan et al. (1991), Nixon et al. (1995) and Howûrth et al. (1996). However many of these mas-balances bcus on inorganic nutrients and thus differ in scale fiom those based on total nutrients. Mass-balances of inorganic nutrients reflect water column processes affecting primary productivity as opposed to whole-ecosystem level processes that affect biomass. These estuary mas-balance models have also not been used predictively. instead, they are generally used to identiQ important sources and sinks in estuaries and to estimate export of nutrients to ocean waters. Despite the use of some mas-balances, much of the modeling in coastal systems seems to focus on simulation models that are site-specific, data-intensive, complex and expensive (e.g. Stigebrandt and Wulff 1987, Visser and Kamp-Nielsen 1996). The net results of these differences are that: 1. the evidence for general N-limitation in coastal waters is less convincing than the evidence for general P-limitation in lakes (Hecky and Kilham 1988); and 2. coastal ecologists have not developed general models that predict coastal eutrophication across systems Given the ecological impacts of eutrophication and the economic and ecological value of coastal systems, there is a clear need for models that predict eutrophication accurately. These models must also be generalizable, given that eutrophication is widespread, simple, cheap. Useful models must also quantify the error associated with the predictions such that the risk associated with management decisions is known.

Tbesis Goal and Structure The goal of this thesis is thus to develop simple models that predict Ch1 as a function of land-use. The central hypothesis is that estuarine eutrophication cm be accurately predicted fiom land-use disturbance using statistical regression models. I have chosen Chl as the response variable in order to predict eutrophication on large spatial (whole systems) and temporal (growing season) scales. This is consistent with Fisher et al. (1995) who distinguish between nutrient limitation of growth rates which occurs on small temporal and spatial scales (days, litres) and nutrient limitation of biomass accumulation which occurs on larger scales. Thus biomass appears to be the better choice for large scale models. By choosing Chl, I am also able to compare the Ch1:total nutrient models developed for estuaries with existing models for lakes. Primary production is also frequently chosen as a response variable (OECD 1982, Nixon 1995). However, primary production rnay confound growth and biomass limitation as it is the product of growth rate and biomass. Moreover, primary production also varies less among estuaries than does Ch1 (LU) thus regression models may be more difficult to develop. However, it should be noted that by choosing Chl, my ability to discuss pmcesses that occur within the water column (e.g. grazing) is limited. As the independent variables, 1 chose land-use because: Ch1 increases directly with nutrients (TP and TN) which are themselves partly a function of land-use (eg agriculture, urban development). in other words, nitrogen and phosphonis are proximate causes of eutrophication but land-use is the ultimate cause of eutrophication predicting Ch1 directly from land-use rather than from chained models linking land-use to nutrients should be more accunte than models predicting Ch1 fiom total nutrients because error does not compound in the more direct models land-use planning is the central approach to manage non-point source nutrient loading the relative importance of M and TP in detennining Ch1 is unclear in estuaries thus variables such as land-use that integrate the effects of TN,TP, turbidity and other variables should more accurately predict Ch1 than any one of these variables it is unclear whether eutrophication is dnven by a single nutrient, TN or TP, thus integrative variables should predict Ch1 more accurately than single- nutnent models. That there may not be a single nutrient limiting phytoplankton biomass is a critical point. Liebig formulated his "Iaw of the minimum" with respect to theyield of a crop over a growing season and over a nurnber of growing seasons (Le. reflecting the long- term impoverishrnent of soils) (Liebig 1840). In other words, the law applies to biomass not production, to populations rather than communities, and at the large, ecosystem scale rather than at the experimental scale. Because eutrophication is a comrnunity response, it is not clear how Liebig's original formulation applies. Moreover, it is not clear how bioassays, which are essentially instantaneous and measure growth rate, can be used as evidence for nutrient limitation which, according to Liebig, reflected biomass over a lengthy period. Moreover, the terni limiting factor has becorne much extended since Liebig's formulation. Liebig very specifically included only nutrients that detemiine biomass rather than variables that effect growth rate such as temperature and light. It was Blackman (1905) who expanded the concept to include such growth-rate effecting variables. Odum (1 954) has expanded the terni to mean "any condition which approaches or exceeds the limits of tolerance" (Odum 1954 as cited in Rigler and Peters 1995). Essentially, the term limiting factor has come to mean that "organisms are affected by their environment" (Rigler and Peten 1995). Despite the looseness of the terni, it remains the 16th most popular ecological concept in a survey of 645 members of the British Ecological Society (Cherrett 1988 1989, cited in Peters 1991) An alternative to modeling the combined effects of individual limiting factors is to mode1 the effects of integrative variables such as land-use. This is particularly critical in estuaries where the Ch1:nutrient relations are relatively poorly understood. However, land-use should also predict Ch1 in lakes better than models based on single nutrients because such models do not include the compounded error associated with serially chained models and because there is some evidence for both P and N limitation in lakes. In 1, 1 thus test the hypothesis that land-use predicts lake phytoplankton biomass (measured as Chl) more accurately than does TP. This analysis is based on data fiom 60 lakes compiled by the Lake Biwa Institute. Ch1 is predicted for each of the lakes from either land-use models or via the traditional models that combine a Ch1:TP relation with a P mas-balance equations. LI extends this modeling to estuaries in which 1 test the hypothesis that land-use predicts estuarine phytoplankton biomass (measured as Chl) more accurately than does TP or TN. Fifleen estuaries in Prince Edward Island, Canada were sampled for phytoplankton biomass, TN and TP fiom May to Augusi 1996. Morphometry and land- use data were suppiied by the provincial govemment's Department of Agriculture. Because Ch1:total nutrient relations have not been developed for estuaries, we first developed models that predict Ch1 as a function of either TN or TP. Land-use models were then developed and compared to the TN and TP relations. A cornparison of lake and PEI estuarine Ch1:TP or Chl:TN relations shows that the yield of Ch1 per unit nutrient is approximately two orders of magnitude lower in estuaries than in lakes. This result was unexpected as other data compiled fiom the literature show similar Ch1:nutrient relations for the subestuaries of the Chesapeake and estuaries of North and South Carolina. Moreover, phytoplankton nutrient requirements do not appear to differ between lakes and estuaries (Hecky and Kilharn 1988) Top-down grazing is one possible explanation for low Chl yields. Oficrr (1982) speculated that shellfish may be a natural eutrophication control in estuaries. In Mes, Mazurnder (1994), Quiros (1990) have both demonstrated that filter feeding Daphnia substantially decrease the yield of Ch1 per unit TP. Effects of shellfish grazing are potentially large in estuaries compared to lakes as estuaries are generally shallow with greater physical energy that ensure oxygenation of bonom waters. Thus estuaries support relatively large benthic biomass compared to lakes, many of which are filter feeders (Nixon 1988). In 6 of the 15 estuaries sampled in PEI. blue mussels (Mvtilrcs edzdis) are fmed intensively. 1 therefore tested the hypothesis that grazing by M. edulis accounts for low Ch1 yields in these estuaries (III). 1 used a mass-balance approach to calculate whether the estimated losses due to herbivory are sufficient to account for the low Ch1 yields. Lakes and estuaries are considered to be distinct aquatic systerns. Lakes are typically relatively closed systems. freshwater and nontidal. Estuaries are relatively open systems. brackish to saline and tidal. Based on this dichotomy, one would not necessarily expect similar patterns in Ch1 response to total nutrients and land-use across lakes and estuaies. To test the generality of Ch1 :land-use and Chknutrient relations across aquatic systems and whether these responses Vary systematically with hydrodynamic openness, salinity or tidal regime, 1 compiled data for approximately 16 lakes and 22 US estuaries that include Chl, TP, TN,morphometry and land-use. Sirnilar data have been compiled for 19 Finnish estuaries which are intermediate between lakes and true estuaries: like lakes, the Fimish estuaries are non-tidal; like estuaries, they are brackish and hydrodynarnically open. 1 first tested the hypothesis that TP and land-use models can predict Ch1 in Fimish estuaries (IV). I then compared the Ch1:TP and Ch1:land-use models for lakes, Finnish and US estuaries to test for systematic differences across these systems. in summary, these 5 manuscripts test the hypotheses that 1) estuarine eutrophication cm be accurately predicted fkom land-use using linear regression models, 2) lakes and estuaries respond similarly to nutrients and land-use, and 3) a comparative approach rooted in empirical regression and mass-balance models is equally effective across aquatic systems. Literature Cited:

Anderson, N.J. 1995. Naturally eutrophic lakes: reaiity, myth or myopia ? TREE lO(4): 137-138. Anon. 1969. Report to the international Joint Commission on the pollution of Lake Erie, Lake Ontario und the International section of the St. Lawrence River. Vol 1, Summary. Ottawa and Washington. Bierhuizen. J.F.H.and Prepas. E.E. 1985. Relationship between nutrients. dominant ions, and phytoplankton standing crop in prairie saline lakes. Can. J. Fish. Aquat. Sci. 42: 1588-94. Blackrnan, F.F. 1905. Optima and limiting facton. Ann. Bot. 19: 281-295. Boyle, E.; Collier, R.; Derigler, A.T.; Edmond, J.M.; Ng, A.C.; Stallard, R.F. 1974. On the chemical mass balance in estuaries. Geochimica et Cosmochintica Acta 38: 1719-1728. Boynton, W.R.; Kemp, W.M.; Keefe, C.W.1982. A comparative analysis of nutnents adn other facton influencing estuarine productivity. In: Kennedy, V.S. (ed.). Estuarine comparisons. NY: Academic Press. pp 60-90. Boynton, W.R.; Murray, L.; Hagy, J.D.; Stokes, C.; Kemp, W.M. 1996. A comparative analysis of eutrophication patterns in a temperate costal lagoon. Estuaries. 19: 408- 421. Brandt, K. 1899. Uber den Stoffivechsel in Meere. Wissenschafiliched Meeresuntersuchunger. Abt. Kield. Bd. 4: 2 13-230. Bricker, S.B.; Clement, C.G.; Frew, S.; Hannon, M.R.; Pirhalla, D.E. 1995. NOAA's Estuaries Eutrophication Survey Volume 1 : Mid-Atlantic Region. ORice of Ocean Resources Conservation and Assessment, National Ocean Service, National Oceanic and Atrnospheric Administration. Silver Spring, Maryland. Brockrnann U.H.;Laane, R.W.P.M.;Postma H. 1990. Cycling of nutrient elements in the North Sea. Netlterlands J. Sea Research 26 : 23 9.264. Canfield, D.R.and Bachma.R. W. 198 1. Prediction of total phosphorus concentrations, chlorophyll-a, and Secchi depths in naturai and arrificial lakes. Can. J. Fish. Aquat. Sci. 38: 414-423. Carpenter, S.R.; Kitchell, J.F.; Hodgson, J.R. 1985. Cascading trophic interactions and lake productivity. BioScience 3 5 : 634-639. Cederwall, H.; Elmgren, R. 1980. Biomass increase of benthic macrofauna demonstrates eutrophication of the Baltic Sea. Ophelia. suppl. 1: 287-304. Cherrett, J.M. 1988. Ecological concepts - a survey of the views of the members of the British Ecological Society. Biologist 35:64-66. Cherrett, J.M. 1989. Key concepts, the results of a survey of our memben' opinions. British Ecologicul Society Svmposium. Oxford: Blackwell. Cole, G.A. 1979. Textbook of lininology 2"edition. St. Louis: The C.V. Mosby Co. Colombo, G.; Ferrari. 1.; Ceccherelli, V.; Russi, R. (eds.). 1992. Marine eutrophication und population dvnamics. Proceedings of the 25" European marine Biology Symposium. Fredensbord: Olsen and Olsen. Conley, D.J.and Malone, T.C. 1992. Annual cycle of dissolved silicate in Chesapeake Bay: implications for the production and fate of phytoplankton biomass. Mar. Ecol. Prog. Ser. 81:121-128. Cooper, S.R. 1995. Chesapeake Bay watershed historical land use impact on water quality and diatom communities. Ecological Applications 5(3): 703-23. Correll, D.L.;Jordan, T.E.; Weller, D.W.1992. Nutrient flux in a landscape: effects of coastal land-use and terrestrial cornrnunity mosaic on nutrient transport to coastal waters. Estiruries: 15(4): 43 1-442, D'Elia, C.F.,Sanders, I.G.; Boynton, W.R.1986. Nutrient e~chmentstudies in a coastal plain estuary: phytoplankton grown in large-scale, continuous cultures. Can. J. Fish. Aquat. Sci. 43 : 397-406. Desprez, M.; Rybarczyk, H.; Wilson, J.G.; Ducrotoy, J.P; Sueur, F.; Olivesi, R.; Elkaim, B. 1992. Biological impact of eutrophication in the Bay of Somme and the induction and impact of anoxia. In: Heip, C.H.R.and Nienhuis, P.H.(eds.). Effects of disturbances on estuarine and coastal marine environments. Fredensbord: OIsen and Olsen. pp. 149-159. Dillon, P.J. and Kirchner, W.B. 1975. The effects of geology and land use on the export of phosphorus nom watersheds. Water Research 9: 135-148 Dillon, P.J. and Rigler, F.H. 1974. The phosphorus-chlorophyll relationship in lakes. Limnoi. Oceanogr. 19: 767-773. Dillon, P.J.; Scheider, W.A.; Reid, R.A.; Jefies, D.S. 1994. Lakeshore capacity study: part 1 - test of effects of shoreline development on the trophic status of lakes. Lake and Reservoir Management 8(2): 12 1 - 129. Dodds, W.K. and Priscu, J.C. 1990. A cornparison of rnethods for assessrnent of nutrient deficiency of phytoplankton in a large oligotrophic lake. Can. J. Fish. Aquafic Sci. 47: 2328-2338. Duarte, C.M. and Kalff, 1. 1989. The influence of catchment geology and lake depth on phytoplankton biomass. Arch. Hvdrobioi. 1 15( 1):27-40. Fisher, TA.; Melack, J.M, Grobbelaar, J.U.;Howarth R.W. 1995. Nutrient limitation of phytoplankton and eutrophicaiion of inland, estuarine, and marine waters. in: Tiessen, H. (ed.). Phospltoms in the global environment. Chichester: John Wiley. pp. 301 -322. Granéli, E.; Wallstrom, K, Lanson, U; Granéli, W; Elmgren, R. 1990. Nutnent limitation of primary production in the Baltic Sea area. Ambio 19: 142-5 1. Hasler, A.D. 1 947. Eutrophication of lakes by domestic drainage. Ecology 28:383-95. Hecky, R.E. and Kilharn, P. 1988. Nutrient limitation of phytoplankton in fieshwater and marine environrnents: a review of recent evidence on the effects of enrichment. Limnoi. Oceanogr. 33(4/2): 796-822. Hendenon-Sellers, B. 1987. Deca-ving lakes: the origin and controi of nttrophication. Chichester, NY: Wi ley. Home, A.J.; Goldman, C.R. 1994. Limnology 2"dedition. N'Y: McGraw-Hill Inc. 576. Howarth. R.J. 1988. Nutrient limitation of net primary production in marine ecosystems. Ann. Rev. Ecol. Svs. l9:89-lIO. Howarth, R. W.; Billen, G.; Swaney, D.; Townsend, A.; Jaworski, B.; Lajtha, K.; Downing, J.A.; Elmgren, R.; Caraco, N.; Jordan, T.; Berendse, F.; Freney, J.; Kudeyarov, V.; Murdoch, P.; Zhu Zhao-Liang. 1996.Regional nitrogen budgets and riverine N and P fluxes for the drainages to the North Atlantic Ocean: natural and human influences. Biogeochemistry 35 :75- 1 39. Hutchinson, N.J.; Neary, B.P; Dillon, P.J. 199 1. Validation and use of Ontario's trophic status mode1 for establishing lake development guidelines. Lake and Resenloir Management 7(l):13-23. Jones, J.R. and Knowlton, M.F. 1993. Limnology of Missouri reservoirs: an analysis of regional patterns. Lake and Reservoir Managenient 8(1):1 7-30. Jordan, T.E.;Correll. D.L.;Miklas, J.; Weller, D.E. 1991. Nutrients and chlorophyll at the interface of a watershed and an estuary. Limnol. Oceanogr. 36(2): 15 1-267. Kerr, P.C.;Paris, D.F.;Brockway, D.L. 1970. The interrelation of carbon and phosphoms in reguiating heterotrophic and autotrophic populations in aquatic ecosystems. U.S. Dep. Inter. Fed. Water Pollut. Control Adm. Water Pollut. Control Res. Ser. 16050 FGS 07/70. Kirchner. 1975. An examination of the relationship between drainage basin morphology and the export of phosphoms. Limnol. Oceanogr. 20(2):267-270. Kirchner, W.B. and Dillon, P.J. 1975. h empirical method of estimating the retention of phosphoms in lakes. Water Resources Res. 1 l(1): 182-1 83. Klein, T. 1993. Impact on lake developrnent of changed agricultural watershed exploitation during the last three centuries. In: Hillbricht Ilkowska, A. and Pieczynska, E. (eds.). Phosphorus and nitrogen dynamics and relention in ecotones of fowfandtemperate lakes and rivers. vol. 25 1(1/3): 279-308. Krom, M.D.; Kress, N.; Brenner, S.; Gordon, L.I. 1991. Phosphorus limitation of primary productivity in the eastem Mediterranean Sea. Limnol. Oceanogr. 36(3):424-432. Kuentzel, L.E. 1969. Bacteria, carbon dioxide and algal bloom. J. Water Poliut. Contr. Fed. 41: 1737-1747, Lange, W. 1967. E ffect of carbohydrates on the symbiotic growth of planktotiic blue- green algae with bacteria. Nature (London). 21 5: 1277-1278. Lapointe, B.E. and Clark, M.W.1992. Nutrient inputs fiom the watenhed and coastal eutrophication in the Florida Keys. Estuaries 15(4): 465-76. Liebig, J. von. 1840. Organic chemistry in its application to vegetable physioiogy and agriculture. In: Komondy (ed.). Readings in ecology. Englewood Cliffs, NJ: Prentice-Hall. pp. 12-14. Mazumder, A. 1994. Patterns of algal biomass in dominant odd- vs. even-link lake ecosystems. Ecology 75(4): 1141 -9. McCauley, E.; Downing, I.A.; Watson, S. 1989. Sigrnoid relationships between nutrients and chlorophyll arnong lakes. Can. J. Fish. Aquatic Sci. 46:1 17 1- 11 75. McGaughey, P.H.1974. Synopsis of workshop on modeling of the eutrophication process. in: Middlebrooks, E.J.; Falkenborg D.H.;Maloney, T.E. (Eds.). Modeling the eutrophicution process. AM Arbor, Michigan: Ann Arbor Science. pp. 1-4. Mellina, E.; Rasmussen, J.B.;Mills, EL. 1995. Impact of zebra mussels (Dreissena po!vmorpha) on phospliorus cycling and chlorop hyll in lakes. Can. J. Fish. Aquur. Sci. 52: 2553-2579. Mills, EL. 1989. Biological oceanographv: an early history. 18 70-1 960. Ithaca, NY: Comell University Press. Molot, L.A.; Dillon, P.J. 1991. Nitrogen/phosphoms ratios and the prediction of chlorophyll in phosphorus limited lakes in central Ontario. Can. J. Fish. Aquat. Sci. 48(l):140-145 Monbet, Y. 1992. Controls of phytoplankton biomass in estuaries - a comparative analysis of microtidal and macrotidal estuaries. Estuaries. 15: 563-571. Murphy, T.P.;Hall, K.J.; Yesaki, 1. 1983. Coprecipitation of phosphate with calcite in a naturally eutrophic lake. Limnol. Oceanogr. 28: 5 8-69. National Academy of Sciences. 1969. Eutrophication: causes. consequences, correctives. Proceedings of a symposium. Washington: NAS. Naumann, E. 1919. Nana synpunker angaende planktons okologi. Med sarskilde hansyn till @oplankton. Svensk bot Tidstr. 13: 12% 158. Neilson, B.I. and Cronin, L.E. (eds.). 198 1. Estuaries and nutrients. Clifton, NJ: Humana. Nixon, S.W. 1988. Physical energy inputs and the comparative ecology of lake and marine ecosystems. Limnol. Oceanogr. 33(4/2):1005- 1025. Nixon, S.W. 1995. Coastal marine eutrophication: a definition, social causes, and future concems. Oplielia: 4 1: 199-2 19. Nixon, S.W.; Ranger, S.L.; Nowicki, B.L. 1995. An assessrnent of the annual mass balance of carbon, nitrogen, and phosphorus in Nanagansett Bay. Biogeochemistry 3 l:IS-61. Odum, E.P. 1954. Fundamentufs of ecology. 2"6ed. Toronto: W.B. Saunders Co. OECD 1982. Eufrophication of waters: monitoring assessrnent and control. Paris: Organization for Economic Cooperation and Development Officer, C.B.; Smayda T.J.;Mann, R. 1982. Benthic filter feeding: a natural eutrophication control? Mar. Ecol. Prog. Ser. 9: 203-210. Outridge, P.M.;Athington, A.H.; Miller, G.I. 1989. Limnology of naturally acidic, oligotrophic dune lakes in subtropical Australia, including chlorophyll-phosphorus relationships. Hydrobiologia 179:39-5 1. Oviatt, C.; Doering, P.; Nowicki, B.; Reed, L.; Cole, J.; Frithsen, J. 1995. An ecosystern level experiment on nutrient limitation intemperate coastal marine environments. Mar. Ecol. Prog. Ser. 116:171-179. Paerl, H.W. 1988. Nuisance phytoplankton blooms in coastal, estuarine, and inland waters. Limnol. Oceanogr. 33: 823-847. Pennock, J.R. 1985. Chlorophyll distribution in the Delaware estuary: regulation by light limitation. Est. Coastal SheffSci. 2 1: 7 1 1-725. Peters, R.H.1986. The role of prediction in lirnnology. Limnol. Oceanogr. 3 l(5): 1 143- 1159. Peters, R.H.1991. A critique for ecology. Cambridge: Cambridge University Press. Prairie, Y.T.;Duarte, C.M.;Kalff, J. 1989. UniQing nutrient-chlorophyll relationships in lakes. Con. J. Fish. Aquat. Sci. 16: 1 176-82. Prairie, Y.T. and Kalff, J. 1986. Effect of catchent size on phosphonis export. Wuter Res. Bulletin 22(3): Pridmore, R.D.; Vant, W.N.; Rutherford, S.C. 1985. Chlorophyll-nutrient relationships in North Island lakes (New Zealand). Hydrobiologia 12 1: 18 1-89. Quiros, R. 1990. Factors related to variance of residuals in chlorophyll - total phosphonis regression in lakes and reservoin of Argentina. Hydrobiologia 2001201 : 343-355. Reckhow, K.H. and Chapra, S.C. 1983. Engineering approachesfor lake management. Boston: Butterworth. Reckhow, K.H. and Simpson, J.T. 1980. A procedure using modeling and error analysis for the prediction of lake phosphocus concentration fiom land use information. Can. J. Fish. Aquat. Sci. 37: 1439-1448. Redfield, A.C. 1958. The biological control of chernical factors in the environment. Am. Sci. 46: 205-222. Rigler, F.H.1964. The phosphonis fractions and the turnover time of inorganic phosphorus in different types of lakes. Limnol. Oceanogr. 951 1-5 18. Rigler, F.H.and Peters, R.H. 1995. Science and limnology. OldendoNLuhe: Ecology Institute. Rosenberg, R.; Elmgren, R.; Fleischer, S.; Jonsson, P.; Persson. G.; Dahlin, H. 1990. Marine eutrophication case studies in Sweden. Ambio. 19: 102- 108. Rowan, D.J.and Ka1 ff, J. 199 1. The lirnnolopical implication of catchment sediment load. Verh. Internat. Verein. Limnol. 24: 2980-2984. Ryther, J.H. 1954. The ecology of phytoplankton blooms in Monches Bay and Great South Bay, Long Island, NY. Biol. Bull. 106: 198-209. Ryther , J.H. and Dunstan, W.M. 1971. Nitrogen, phosphonis, and eutrophication in the coastal marine environment. Science. 17 1 : 1008- 10 1 3. Schindler, D. W.; Armstrong, F.A.J.; Holngress, S.K.;Brunskele, G.I. 1971. Eutrophication of lake 227, experimental lakes Area, northwestem Ontario, by additions of phosphate and nitrate. J. Fish. Res. Bd. Con. 28: 1763-82. Schindler, D.W.1977. Evolution of phosphonis limitation in lakes. Science. 195: 260- 262. Sharpley, A.N.; Chapra, SC.; Wedepohl, R.; Sims, J.T.; Daniel, T.C.;Reddy, K.R. 1994. Managing agicultural p hosphonis for protection of surface waters : issues and options. J. Environ. &al. 23: 437-5 1. Smith, S.V. 1989. Tomales Bay, California: a case for carbon-controlled nitrogen cycling. Limnol. Oceanogr. 34(1):37-52. Stigebrandt, A. and Wulff, F. 1987. A mode1 for the dynarnics of nutrients and oxygen in the Baltic Proper. J. Mar. Res. 45:729-59. Strem, K.M. 1928. Production biology oftemperate lakes. Int. Rev. d. ges. Hydrobiol. u. Hydrogr. 19: 329-348. Sumrner, R.M.;Alonso, C.V.; Young, R.A. 1990. Modeling linked watershed and lake processes for water for water quality management decisions. J. Environ. @al. 19: 42 1-427. Thienemann, A. 1925. Die Bimengewbser Mitteleuropas: eine Lirnnologische Einfuhrung. Die Binnengewaser. 1: 1-255. Turner, R.E. and Rabalais, N.N. 1994. Coastal eutrophication near the Mississippi river delta. Nature. 368: 6 19-62 1. Vallentyne, I.R. 1974. The nlgal bowl: lakes and man. Ottawa: Department of the Environment, Fishenes and Marine Service. Miscellaneous Special Publication 22. van Straten, G. 1986. Hypothesis testing and parameter uncertainty analysis in simple phytoplankton-P models. in: Somlyody, L. and van Straten, G. (eds.). Modeli~igand managing shallow lake eutrophication: with application ro Lake Balaton. Berlin: Springer-Verlag. pp. 287-308. Visser, A.W. and Karnp-Nielsen, L. 1996. The use of models in eutrophication studies. In: K. Richardson and B.B. Jergensen (eds.). Eutrophicarion in coastd marine ecosystems. Coastal and Estuarine Studies, Volume 52. American Geophysical Union. pp. 221-242. Vollenweider, R.A. 1975. Inputsutput models with special reference to the phosphorus loading concept in limnology. Schweiz. Zeit. Hydrol. 37: 53-84. Vollenweider, R.A.; Marchetti, R.; Viviani R (eds.). 1992. Marine coastal eutrophication. Science ofthe Total Environment. Supp. Weber, C.A. 1907. Aufbau und vegetation der moore norddeutschlands. Beibl. Bot. Jahrb. 90:19-34. Welch, P.S. 1935. Limnology. McGraw-Hill. NY. Welch, P.S. 1952. Limnology. 2" Ed. McGraw-Hill. NY. Wetzel, R.G. 1975. Limnology. Philadelphia: W.B. Saunders Co. pp. 743 Woolf, H.B (ed.). 198 1. Webster f New Collegiate Dictionary. Springfield: G and C. Memam Company. CIRCUMVENTINCPHOSPHORUS IN LAKE MANAGEMENT:

A COMPARISONOF CHLOROPHYLL-APREDICTIONS FROM LAND-USEAND PHOSPHORUS-

LOADING MODELS.

Meeuwig, J.J. and R.H.Peten.

Cunadian Jozinial of Fisheries and Aquatic Sciences 53(8): 1795- 1 806

Abstract Prediction of lake chlorophyll-a (Chl) fiom phosphorus (P) loading models is based largely on measurements of land-use and lake morphometry. Empincai models based on simple characteristics of the lake and watershed should thus predict Ch1 as effectively as the more complicated P-loading models, but more directly and with bener definition of the error associated with predicted values. To test this hypothesis, we compared the predictive power of empirical land-use models generated by regression to that of an empincal P-loading mode1 and a set of literanire-derived, two-step P-loading models. These hvo-step P-loading models combine a mass-balance equation that calculates TP with a regression equation that estirnates Ch1 fiom calculated TP. Cross-validation and jackknife analyses demonstrated that, overall, empincal land-use models predict Ch1 more accurately. more precisely and with less bias than do the two- step P-loading models. Patterns in error among two-step P-loading models suggest that 1) choice of both the mas-balance and Ch1:TP equations affect predictive power and 2) interactions between mass-balance and Ch1:TP equations make it difficult to partition error withn two-step P-loading models.

Introduction Vollenweider (1975) and Dillon and Rigler (1974a) developed dynamic and steady state equations to predict total phosphorus (TP)concentration in lakes as a function of phosphorus load (P-load), phosphorus retention in the sediments, mean depth and hydrologic flushing of each lake. As a second step, such mass-balance equations are frequently linked to empincal TP-response equations, such as those that predict chlorophyll (Chl). These combined equations, here tenned "two-step P-loading models", are the basis of modem eutrophication control (OECD 1982; OMNR 1986). Although elegant, two-step P-loading models cm be difficult to apply with confidence. The mas-balance component requires information on land-use, lake morphometry, phosphorus export and phosphorus retention (Dillon and Rigler 1975). Phosphorus export and retention coefficients are expensive to mesure. Borrowed coefficients may not reflect the climatic, geological. and land-use characteristics of the watershed or lake of interest (Dillon and Kirchner 1974), so emr is introduced through the choice of phosphorus export and retention coefficients. This error may be as high as *50% (OECD 1982). Moreover, the serial calculations associated with mass-balance equations compound error which is further exacerbated when, in the second step, calculated TP is used to predict Ch1 via an additional empincal model. Two-step P- loading models may compensate for their statistical shortcomings because they reflect underlying mechanisms i.e. phosphorus limitation of algal biomass. The extent of such compensation has never been assessed. An alternative to the two-step P-loading model would predict Ch1 directly from land-use characteristics using empincal regression models. Conceptually, such empincal models recognize the influence of anthropogenic activities on natural systems and the need to manage such activities. Practically, they eliminate the error associated with choosing phosphorus export and retention coefficients. By directly predicting Chl, empirical land-use models also eliminate error associated with the serial linking of mas-balance and Ch1:TP equations. This study tests the hypothesis that empincal regression models that predict Ch1 directly fiom land-use charactenstics are more accurate, precise and unbiased in their predictions than the traditional two-step P-loading models.

Data Set This analysis is based on data for 63 laices cornpiled by the Lake Biwa Institute (1988 1989 1990). This data set was selected following an extensive search of the literahue because it included data for CM,TP, P-loads and land-use, al1 of which are required for this analysis. Although these three volumes provide data for 147 lakes, oniy the 63 lakes listed in Table 1 had the requisite information. Other variables (Table 2) were included in the analyses based on two cntena: 1) that previous work had identified the variable as relevant to eutrophication; and 2) that the variable was reported for at least two thirds of the lakes in the data set. Growing season averages for TP and Ch1 were generated by avenging monthly values over a defined growing season. First we divided the lakes into three geographical regions based on ScraSkraba (1993): north temperate lakes between 35% and 60%; south temperate lakes between 35"s and 65%; and a tropical region between 35% and 35"s that combines StraSkraba's tropical and dry regions. For north temperate lakes, the growing season was defined as April - September because most "seasonal averages" are based on sarnples taken between May and September (Marshall and Peters 1989). We extended the range to include April since Apnl values were reported for many of the lakes and we were thus less likely to miss the spring tumover and bloom. Lacking a similar study for south temperate lakes, the growing season was defined as October to March, the reciprocal of the north temperate season. Annual averages were used for the tropical Mes where annual temperature and irradiance Vary less than in temperate zones. Lakes in the data set range widely in terms of morphornetry and trophic status, reflecting the geographical breadth O € the data: the majority of lakes are fiom Europe ( 1 9). Japan (16) and North Arnerica (1 6), but fica(1 ), New Zealand (1), the Middle East (1), South Amenca (2) and Southeast Asia (7) are also represented. Lake turnover time and surface amspan three and four orden of magnitude respectively. Land-use varies fiom catchrnents that are heavily populated (2657 km") and urbanized (60%) to remote catchments with linle human aciivity. Chl and TP concentrations span two orders of magnitude, thus the data include oligotrophic, mesotrophic, and eutrophic lakes. Of the 57 lakes for which sewage treatment information is available, six receive no sewage and are considered pristine. Of the remaining lakes, 47 receive a degree of sewage treatment ranging from septic fields to tertiary treatment. Only four have no treatrnent (Table 1; Lake Biwa Institute 1988 1989 1990). For the 37 lakes for which TN:TP ratios can be caiculated, the average ratio is 29: 1 (Table 2), suggesting phosphorus- rather than nitrogen-limitation of algal biomass (Smith 1979). Models Three classes of models were compared in this study: empincal land-use models, empincal P-loading models, and two-step P-loading models. i) Empirical Land-Use Models Empirical land-use models (Table 3:i) that predict Ch1 directly from land-use and morphometry were developed using the Lake Biwa institute data set (Lake Biwa Institute 1988 1989 1990) and multiple regression techniques. Al1 variables, except those indicating land-use, were log-transfomed to stabilize variance (Zar 1984). The land-use variables, reported as the percentage of the catchment area covered by forest (FOR-p), by agriculture (AG-p) or by urban development (URB-p) were included in the analysis both as untransformed estirnates and their logarithms. Arcsine transformations, which are ofien recornrnended for percentage data (Kleinbaum et al. 1988), were not used as they did not substantially reduce the skew in the data. The land-use variables were also expressed as absolute areas (FOR-a, AG-a and üRB-a in km'). These areal land-use variables were also included in the analysis both as untransformed estimates and their logarithms. The SAS procedure "Proc REG, selection = nquare" (Sas Institute 1985) was used to regress Ch1 against al1 lake morphomenic and catchment variables listed in Table 2. The rsquare option generates models for al1 combinations of variab les, allowing direct cornparison of these combinations. When several combinations of variables have similar coefficients of determination, the user can choose those combinations most relevant to the purpose at hand. In this case, relevant variables are those that are readily available to land-use plannen and environmental managers. The nquare option decreases sample size as it uses only those lakes for which al1 entered variables are reported. Therefore, once several promising combinations of variables were identified using the rsquare option (Le. those regression models with the highest coefficients of determination utilizing variables of interest), regressions were re-run without the aquare option to increase the sample size. A correlation analysis was then conducted to ensure that collinearity in independent variables was minimized (Table 4). The independent variables are combined in a mode1 only if their correlation coefficient is less than 0.6. Although this cut-off level is only a rule of thumb, a power analysis (Cohen 1977) indicated that an improvement in the coefficient of determination of a model with three independent variables could only be identified when any two of the independent variables were correlated at the ~0.6level or less. If the correlations between independent variables are higher, the power of statistical tests is insuficient to detect improvements in the model's coefficient of determination. Because model sample size varies arnong models according to which independent variables are included, and because rnodel standard errors are less sensitive to variation in rnodel sarnple size than the coefficients of detemination. the model standard erron were used as the principle cntenon of fit. Those models with minimal collineanty and the lowest mode1 standard errors of the estimate were chosen as the "best" empirical land-use models. ii) Empincal P-loading Mode1 An empirical P-loading model (Table 3:ii) was also developed fiom the Lake Biwa Institute data set (Lake Biwa Institute 1988 1989 1990). This empirical P-loading model is a multiple regression formulation of the three key variables in the mass-balance equation: areal P-load (APL), lake mean depth (Z,J and lake turnover time (9).The empirical P-loading model was developed using multiple regression techniques: the variables (Chl, R,, Z, and APL) were log-ü-ansformed to stabilize variance and the SAS procedure "Proc REG" (Sas Institute 1985) was used to regress logChl against IogAPL, log2, and logR,. This model was developed to test whether P-load is a better predictor of Ch1 than land-use through the direct comparison of two regression models as opposed to the cornparison of a regression model and a P-loading model. Because log, and logR, have a correlation coefficient greater than 0.6, logR, did not enter the regression at the p=0.15 level. A second empirical P-loading rnodel was consequently developed in which IogChl was regressed on logAPL and Io&. This model perfoned poorly compared to the three variable model and was subsequently discarded. iii) Two-step P-loadinp: Models Two-step P-loading rnodels (Table 3:iii) predict Ch1 by (1) calculating TP via a mass-balance equation and (2) estimating Ch1 from the calculated TP via a Ch1:TP model. Seventeen mass-balance equations drawn fiom the literature (Table 5) were used to calculate a suite of 17 TP values for each lake. Al1 of the mass-balance equations derive fiom the mas-balance approach of Vollenweider (1975) and Dillon and Rigler (1974a). Most equations assume that phosphorus retention (R) is either a hinction of lake volume and thus of the hydraulic flushing rate, p (yfl), and the phosphorus sedimentation coeflicient, (yf'),or that R is a fùnction of lake area and thus of areal water loading, q (myf'), and phosphorus settling velocity, v (m.yfi)(Table 5). Other variations include the absence of a defined R (Table 5, equations 1, 2 and 16) and the definition of R in terms of P-load (Table 5, equations 6 and 7). This large number of mas-balance equations was included in the analysis as there is no obvious way to select an appropriate equation for a given purpose. The second step in the two-step P-loading models predicts Ch1 using calculated TP as the independent variable in a Ch1:TP regression equation. There are myriad Ch1:TP equations and no clear guidelines as to their use. As choosing the "nght" relationship is there fore a hi t-and-miss exercise, this study uses several alternatives (Table 6).The Dillon and Rigler equation (1974b) was chosen because this well-known mode1 is perhaps the most Frequently used (Reckhow and Chapra 1983) and has a very high correlation between Ch1 and TP (r=0.96). Vollenweider's OECD (1 982) mode1 wuchosen because, like our models, it was developed from a data set with a broad geographic range. A "Fitted" Ch1:TP relation was also generated by regressing observed Ch1 against observed TP in the Lake Biwa Institute data set. These three Chl:TP equations are not seasonally compatible: essentially, the Dillon and Rigler equation predicts summer Ch1 from spring TP; the Fitted equation predicts mean growing season Ch1 from mean growing season TP; and the Vollenweider-OECD equation predicts mean annual Ch1 from mean annual TP. in eutrophic lakes, values for mean annual Ch1 are indistinguishable fiom values for mean gmwing season Ch1 (Marshall and Peters 1989) thus growing season and annual data for such lakes can be used interchangeably. In oligotrophic lakes, mean growing season Ch1 may be higher than mean annual Ch1 (Marshall and Peten 1989) in which case the Vollenweider-OECD equation would systematically underestimate Ch1 in oligotrophic lakes. No such systematic bias was evident in our results suggesting that the differences between growing season and annual Ch1 were small enough to be disregarded. The 17 TP values calculated fkom the equations in Table 5 were used to predict Ch1 in each of the 3 Ch1:TP equations. Observed TP was also used to predict CM in each of the 3 Ch1:TP equations. To facilitate reference to the models, they are coded according to the Ch1:TP equation used (D=Dillon and Rigler, F= Fitted model, and V=Vollenweider-OECD) and by the nurnber of the mas-balance equation fiom Table 5. Thus. mode1 F6 uses equation 6 fiom Table 5 to calculate a TP value which is then inserted in the Fitted Ch1:TP equation. The three models using observed TP are coded according to the letter of the Ch1:TP equation (D, F or V) and the subscnpt "O"for

O bserved.

Analysis 1 : Cross-validation We could not make direct comparisons of observed vs. predicted values of Ch1 from the empincal land-use, empirical P-loading and two-step P-loading models since they were not denved fiom the same data set. The empirical land-use and empirical P- loading models were fit to our study data set using regression techniques that minimize residual error. and thus these empirical models would compare favourably to the two-step P-loading models derived from independent data sets if compared directly. Cross-validation is one technique that allows an independent test of rnodel performance. In cross-validation, the data are randomly divided into a "development" data subset and a "test" data subset. The development data are used to estimate regression model parameten for the mode1 of interest. The test data, when entered in the model, provide an independent test of model accuracy as they are not used in estimating model parameten (Kleinbaum et al. 1988). The disadvantage of cross-validation is that it decreases the number of observations available to estimate model parameters. In this analysis, the 63 lakes were randomly divided into a test data set of 10 lakes and a development data set of 53 lakes. The lakes in the test data set must include the variables required for both the empirical and two-step P-loading models (lake turnover time, mean depth and P-load; see Table 3). Of the 63 lakes, 36 lakes have complete information and the test data set was generated by randornly choosing 10 of these 36 lakes. The development data set is thus composed of the remaining 26 lakes with cornplete P-loading data and the 27 lakes with incomplete P-loading data. This procedure was repeated ten times, generating ten independent test data sets and ten model development data sets. This procedure does mean that the test data are not randomly drawn from the entire set of 63 lakes. To determine whether the randomization procedure introduces any serious bias, pararnetric and nonparametric analyses of variance were completed. Using ANOVA, we could evaluate whether the 36 lakes with complete P-loading data differed from the 27 lakes with incomplete P-loading data. The SAS ~rocedureGLM (SAS Institute 1985) was used to test for significant differences between the complete and incomplete P-loading data subsets for al1 the morphometric and lake chemistry variables which were log-normally distributed. The Wilcoxon test (Proc MIPAR1WAY; S.AS Institute 1985) was used to test for significant differences between the complete and incomplete P-loading data subsets for the land-use variables which were neither norrnally nor log-normally distributed. Lakes for which complete P-loading data is reported tend to be cooler (9.1°C vs. 13.7"C; p

Predicring Ciil froin the Empirical Land-use, Empincal P-loading and Two-step P-lwding Models Only empirical land-use models that perfomed well in al1 ten runs (Le. low mode1 standard errors and minimum collinearity) were evaluated as predictors of Ch1 from land-use. This critenon ensured that models with the sarne variables, albeit differing in parameters, were available for eac h of the ten runs. For each run, the ernpincal land-use models based on the development data were used to predict Ch1 for each lake in the corresponding test data set. Similarly, each run generated an empirical P-loading mode1 based on values of P-load, turnover time and mean depth in the development data set, and this mode1 was used to predict CM in the same test lakes. The empirical land-use models, empirical P-loading models, and the values for their regression statistics and parameter estimates averaged over the ten runs are listed in Table 7. In each of the ten runs, the 17 TP values calculated fiom the 17 equations in Table 5 were each substituted into the Dillon and Rigler, Fitted, and Vollenweider-OECD equations. This procedure generated 5 1 predicted Ch1 values fiom the two-step P-loading models for each m. in each run, observed TP was also substituted into the Dillon and Rigler, Fitted and Vollenweider-OECD equations. The entire procedure generated a total of 57 predictions of Ch1 for each lake: 5 1 Ch1 values from two-step P-loading models, 3 Ch1 values fkom observed TP, 2 Ch1 values frorn empirical land-use models and L Ch1 value fiom the empincal P-loading model. Repeating this process for the ten runs generates 570 predictions. Criteriafor Cornparhg Models A modei's utility is reflected in its ability to predict the dependent variable of interest accurately, precisely and without bias. One measure of the accuracy of a model is the mean of the squared residuals (MSR) defined as:

MSR = [Z(Chl, - ~hl,),']1 n [es- 11 where (Chl, - Chlp)iis the difference between observed Ch1 and predicted Ch1 for the ith observation and n is the number of observations. This critenon is based on the criterion of fit used in ordinary least squares regression - the squared residual. The smaller the MSR, the closer are the predicted and observed values over the range of predictions. Precision is the consistency with which the model makes accurate predictions and is reflected in the magnitude of the variance associated with the squared residuals. Precision cm thus be defined as the variance of the squared residuals (vSR): vSR = [~((Chl,- ~hl,),'- MSR)]' / (n-1) [es* 21 Models with low variance in the squared residuals are more precise than models with high vanance. Bias provides information on how the model behaves across the range of the dependent variable. Mode1 bias has two major forms. A model may be uniformly biased across the range of values by consistently over- or underestimating observed values. Altematively, it may be non-uniformly biased in that it responds differently in different parts of the variable range. For instance, a model may overestimate Ch1 in oligotrophic lakes while showing no particular bias in more eutrophic systems. Unifom bias cmbe defined as the mean error (ME) where: ME = [T(Chl. - Chui]/ n [es*31 Values near zero indicate no bias, positive values indicate underestirnation and negative values indicate overestimation. Non-uniform bias cm be evaluated by plotting the error against the predicted value and looking for patterns in the residuals. Results of the Cross- Valdation One-tailed, pairwise t-tests and F tests are the appropriate parametric tests to identiQ statistically significant differences in the accuracy and precision of the empirical land-use, empincal P-loading, and two-step P-loading models (Zar 1984). However, with a sample size of 10, airnost a two-fold difference between means and a three-fold di fference between variances are required for a statistically signi ficant di fference at the Pe0.05 level (b,os(1,9=1.83 and F,.05,,,,,,9=3.18; Zar 1984). Power analysis (Cohen 1977) demonstrated that with a sample size of 10, and = 0.05, the difference between bvo group means would have to be at lest 0.35, or 3 times the MSR to achieve a power of 0.80. Since the MSRs and vSRs generally did not differ to this extent, statistically significant differences could not be identified. Reducing the critical level to P4.IO level still requires 1.5 to 2.5 fold differences (t,.,,,,,, 9=1 .38 and Fo,,,,2,,9,u=2.44;Zar 1984), thus the small increases in sensitivity of the test are not worth the increased uncertainty. To increase the number of lakes in the test data set is equally unsatisfactory because that would decrease the number available for model development. Moreover, power analysis indicates a sarnple size of at least 50 lakes in the test data set would be required to identify a difference between the MSRs of 0.15 at a power level of 0.80.As the aim of this analysis is not to dismiss two-step P-loading models, but only to demonstrate that empirical land-use models provide an effective complement, formal statistical differences are Iess relevant than the relative petformances of the two model types. Thus. rather than look for "statistically significant" differences, the analysis focuses on identifjmg pattems in the ranking of the empirical land-use, empirical P-loading and two-step P-loading models across the ten runs. To identim patterns, the MSRs, vSRs. and MES for each model were averaged over the ten runs. The models were then ranked by these averaged MSRs to assess relative accuracy and by the average vSRs to assess relative precision. Table 8 reports the averages and standard deviations of the MSRs, vSRs, and the MES for the two empirical land-use models, LU1 and LU2, the empirical P-loading model (EP), and the three models using observed TP (Do,Fm V,). in addition, the two-step P-loading models with the lowest average MSRs for each Ch1:TP equation are reported (D6,F7, V7). The most important result is that the empirical land-use models provided more accurate estimates of Ch1 than did any of the two-step P-loading models (Table 8, Fig. 1). Models LU 1 and LU2 had average MSRs notably smaller than any of the two-step P-loading models. The empirical P-loading model, EP, had an average MSR similar to that of the best two-step P-loading models. Surprisingly, the equations using the observed values of TP were consistently the lest accurate; they had noticeably higher average MSRs than any of the empirical land-use or two-step P-loading models. Mode1 precision (vSR) followed a similar pattern in that the most accurate models were also the most precise, and the most inaccurate models were the least precise (Table 8). The standard deviations around the MSRs and vSRs were slightly smaller for F7 and V7 than those of the empirical land-use models, suggesting that across the runs, based on equation 7, there is Iess variation in the accuracy and precision of the two-step P-loading predictions. Bias was evaluated by creating categories of "degree of bias". The range of the absolute values of the mean error of each model was calculated and this range was then divided into keecategones of bias. Models were then assigned to a category by the absolute value of their mean error. Both empirical land-use models and the empirical P- loading model lay at the low end of this scale (Table 9). On the whole, the Dillon and Rigler, Fitted and Vollenweider-OECD equations also showed little bias: only 7 of the 54 models lay at the high end of this scale. In these cases, the models tended to underestimate observed Chl. Plots of error against predicted Ch1 for the empirical land- use, empirical P-loading and two-step P-loading models indicated no apparent non-uni form bias. Patterns within the Empirical Land-use Models Several patterns cm be identified within the empirical land-use models, LUI and LU2. Both empirical land-use models contain a morphometric variable: R, in LUI and Z,,, in LU2. Both models also include a variable reflecting human population in the catchent: total catchment population in LUI and population density in LU2. In both models, land-use is represented by the amount of forest in the catchment: the logarithmic transformation of forested area in LUI and the logarithmic transformation of percent-forested in LU2 (Table 7). The morphometnc variables in al1 three models are negatively correlated with Ch1 suggesting that Ch1 increases as lakes become shallower or more rapidly flushed. The relationship between human population and Chl is positive. Area forested is negatively correlated with Ch1 indicating that as the amount of cleared, non-forested area increases, so does Chl. Patterns within Two-step P-loading Models The two-step P-loading models (D 1-D 1 7, F 1-F 17 and V 1-V 17) can be compared in ternis of both the CMTP and mas-balance equations used. Patterns among the Ch1:TP and mass-balance equations cm be compared graphically by plotting the average MSRs for each two-step P-loading rnodel against the Ch1:TP equation used. Lines joining points represent the same mas-balance equation and reveal the separate effects of the ChI:TP and mass-balance equations on the MSR of the two-step P-loading model. Figure 2 demonstrates three alternative patterns that could hypothetically occur. If al1 Ch1:TP equations are equally accurate in predicting Chl, but there are differences in the accuracy of the mass-balance equations, the MSR for a given two-step P-loading model will be the sarne across the Ch1:TP equations and vary by a constant between mas-balance equations (Fig. 2a). If a11 mas-balance equations are equal but there are differences in the accuracy of the Ch1:TP equations, the MSRs for a11 two-step P-loading models will be the same for a given Ch1:TP equation and Vary only between the Ch1:TP equations (Fig. 2b). Finally, if there are differences between mass-balance equations and between Ch1:TP equations, the MSRs should Vary with both the mas-balance and Ch1:TP equations used (Fig. 2c). When the average MSRs are ploned for al1 5 1 combinations (1 7 mass-balance equations x 3 Ch1:TP equations), it is clear that both mass-balance and Ch1:TP equations affect the accuracy of predictions (Fig. 3). The Vollenweider-OECD and Fitted equations perfonned very similarly and both perform better than the Dillon and Rigler equation. This result was identified earlier in ternis of the best two-step P-loading model for each Ch1:TP equation but this analysis demonstrates that the pattern held for al1 17 mass-balance equations. Moreover, in comparing the variation within groups of mass- balance equations for a given Ch1:TP equation, Figure 3 demonstrates that the Fitted and Vollenweider-OECD equations tend to have much more precise estimates of Ch1 than the Dillon and Rigler equation, regardless of the mass-balance equation used. Of the mas-balance equations, equations 6 and 7 clearly perfonned better than the othen. An unexpected result of this analysis was the demonstrated interaction between Ch1:TP equations and rnass-balance equations. In Figure 3, the hes comecting the average MSRs for the mas-balance equations are not parallel, suggesting that for a given TP calculation, the choice of Ch1:TP equation could exacerbate or diminish the error accumulated in the calculation of TP. Thus, the partitioning of error between the mas-balance and Ch1:TP equations is not constant. Figure 3 also includes the MSRs for the empirical land-use models, the empirical P-loading model. and the Chl:TP models using observed TP. It is clear that the empirical land-use models are more accurate than the two-step P-loading models and that in some cases, the di fferences are extreme. Final 1y, Figure 3 demonstrates how poorly observed TP performs compared to the calculated TP values. In fact, the MSR for observed TP ranks 16th of 18 MSRs when the Dillon and Rigler equation is used, and 13th of 18 when either the Fitted or Vollenweider-OECD equations are used.

Anulysis 2: the jackknife A disadvantage of cross-validation is its reduction of sample size. Because the sarnple size of the test data set is small (n=10), statistical power is low and the results may be biased by the relatively large influence of individual observations. To test the robustness of the cross-validation results, a jackknife analysis was perfomed. The jackknife is a resampling technique that generates regression statistics that are less sample-dependent than ordinary least squares regression and thus should better reflect the population of interest (Efion and Tibshirani 1993). h the jackknife, single observations are systematically removed fiom the data set, and the regression model is fit on the remaining (n- 1) observations. The procedure is repeated for al1 observations to yield a set of (n-1 ) models and (n- 1) estimates of each observation's residual. The (n-1 ) residuds of each observation are then averaged to generate the jackknife residuals which are used to calculate the model standard error of the estimate and the coefficient of determination in the sarne manner as in ordinary least squares regression (SAS Institute 1985; SAS Proc Reg with the 'press' option). Thus, the jackknife estimate of the standard error of the estimate is calculated as: s,, s,, = (ZSS E, / ((n- 1)-k))O-" [es* 41 where ESSEj is the jacklmife error sum of squares calculated as the sum of the jackknife residuais, (n-1) is the number of observations in the jackknife data and k is the number of parameters estimated. The jackknife estimate of the coefficient of detexmination is: i = 1-( ZSSE, I ZSSJ [es*51 where zSS, is the total surn of squares. Jackknife residuals were generated for the empirical land-use models and the empincal P-loading model by regressing observed Ch1 on the respective sets of independent variables associated with each model (LUI, LU2 and EP; Table 7). Jackknife residuals were also calculated for the two-step P-loading models by regressing Ch1 on each of the 17 TP values calculated hmthe rnass-balance equations in Table 5. Finally, jackknife residuals were calculated for observed TP by regressing observed Ch1 on observed TP, analogous to models Do,F. and V,. JacWaiife estimates of the regression statistics were then calculated for al1 the models using equations 4 and 5 above. The jackknife is attractive because it uses al1 the data to estimate model parameters. The main weakness of the jackknife in tenns of this analysis is that, in practice, the jackknife estimates of the two-step P-loading models underestimate the total error associated with the approach. The analysis only evaluates the effect of the fiat step: the mass-balance equation. It does not account for the enor associated with choosing a Chl:TP equation as the TP values are fitted to observed Chl. The jackknife provides a contrast to cross-validation but also a conservative test of the relative strengths of the empirical land-use and two-step P-loading models. Results of the Jackknwe Al1 the variables required for the empirical land-use, empirical P-loading and two- step P-loading models were available for 32 of the 63 lakes in the total data set. When the jackknife estimates of the coefficient of determination are ranked, empirical land-use models LU1 and LU2 perfonn better than al1 the two-step P-loading models (Table 10). The empirical P-loading model also performed very strongly, ranking second if the model standard error is considered. As in the cross-validation analysis, rnass-balance equations 6 and 7 performed very well, appearing among the three best two-step P-loading models; observed TP perfomed relatively poorly, ranking 19th out of 22 models in the prediction of Chl. The coefficients of determination and model standard errors are, however, very similar for the top six models so their relative rankuig is rneaningless. Of greater relevance is the fact that the empirical land-use and empirical P-Loading models rank in the top half of the models analyzed rather than the bottom half. Despite the small differences in the top half of the rankings, the range of performance (coefficients of detemination ranging from 0.53 to 0.3 1) is notable as it demonstrates that some of the calculated TP values estimate Ch1 very poorly.

Discussion Our results suggest that empirkal land-use models based on catchent characteristics and lake morphometry predict Ch1 more accurately and precisely than two- step P-loading models. This study has not used statistics to distinguish arnong models due to the srna11 sarnple size and thus the empirical land-use models can not be described as more accurate than the empincal and two-step P-loading models. However, the conclusion seems robust: both the cross-validation and jackknife analyses support it. Moreover, the MSR of the best two-step P-loading models is over 30% greater than the MSR for the best empincal land-use model (0.1 17 vs. 0.152; Table 8) suggesting that the magnitude of the improvement is suscient to warrant use of the empincal land-use models. That the jackknife results are not as clear as the cross-validation results reflects the differences between the jackknife and cross-validation approaches: the jackknife only accounts for error associated with the mas-balance component of the two-step P-loading models whereas cross-validation incorporates enor in both the mass-balance and Ch1:TP components of the two-step P-loading models. Linking a Ch1:TP equation to a mas-balance equation is a major source of additional error. Of the three broad land-use categones (forest, agriculture, and urban), forest was the "best" land-use predictor of Chl. This result was unexpected as a~culturaland urban areas export more phosphorus than forested areas (Reckhow and Chapra 1983). Forest was a significant variable (p<0.05) in nine of ten runs in model LUI and in seven of ten runs in mode1 LU2. This inconsistency may in part reflect the poor resolution of the land-use data. Given that the catchments are composed pnmarily of agriculture, urbanized and forested areas in varying proportions, forest is the inverse of the sum of agriculture and wbanization. If the mode1 is unable to resolve differences between urban and agricultural areas, both sources of nutrients and sediments, it may be that forest, although a mediocre predictor of Chl, is better than either danor agricultural area alone. Separation of agriculture into Pasture and row crops (Prairie and Kalff 1986) and urban areas into residential, semi-industrial and industrial (Reckhow and Chapra 1983) may improve the performance of these variables in the models. That P-based eutrophication models do not predict Ch1 as well as the empirical land-use models does not question the role of phosphoms as a controlling factor in eutrophication (cf Schindler 1977; Edrnondson 1970). However, the relatively poor performance of the two-step P-loading models compared to the empirical land-use models in this data set suggests that calculated TP may not integrate processes affecting lake eutrophication as well as the land-use variables. For instance, sediment load affects Chl (Rowan and Kalff 199 l), and agriculture and urbanization may increase sediment load to a lake through increased erosion. At high phosphoms concentrations, nitrogen concentration also affects Ch1 levels in lakes (Smith 1979; McCauley et ai 1989), and human settlements can both increase TN and change the TN:TP ratio in lake water. The moderate correlation coefficients between TP and land-use variables (r = 0.27-0.69; Table 4) suggest that TP does indicate general disturbance in the catchment. However, it does not capture al1 the changes that affect Ch1 as well as the land-use variables do. The poor performance of obsented TP in comparison to calculated TP in both the cross-validation and jackknife analyses also suggests that lake eutrophication mûy be detennined by more than phosphorus. This result is unlikely to represent analytical or sarnpling variation because Ch1 and TP analyses are fairly robust (Griesbach and Peten 199 1; Hanna and Peten 199 1) and because Ch1 and TP values in this analysis represent mean values of between 6 and 12 sarnples, sufficient according to Marshall et al. (1988). A more intriguing possibility is that two-step P-loading models, like the empirical land-use models, account for some factor not capnired in the simple measurement of TP. For instance, anthropogenic P-load is likely to be higher in areas of human activity where more land has been cleared and where sediment and nitrogen loads to the lake are higher. Thus, "TP"calculated fkom P-load may represent more than TP, and although less effective than land-use variables, calculated TP may integrate variables driving eutrophication better than observed TP. Further support for the role of P-load as an indicator of catchment disturbance is provided by the empirical fomulation of the P-loading model. This empirical P-loading model perfonns comparably to the best of the two-step P-loading models and better than most of them suggesting that, although P-load is an important determinant, Ch1 is more effectively predicted directly from the deteminants of P-load than via an intermediate estimate of TP. That the empincal P- loading mode1 is less effective than the best of the empincal land-use models suggests that the determinants of P-load still capture only part of the effects of catchment disturbance and that land-use variables such as forested-area and population density may more comprehensively address the factors producing eutrophication. This study also provides insights into the use of two-step P-loading models. Clearly, some mass-balance equations perform better than others. In this study, equations 6 and 7 out-perform the other mass-balance equations (Table 8; Fig. 3). These two models are closely related. They are denved from similar, large data sets (n > 150; Canfield and Bachmann 198 1) and they both use volumetric derivations of phosphorus retention that incorporate P-load to the lake. It is not however clear whether the strong performance of equations 6 and 7 should be attributed to their volurnetric derivation or the large sample size. Moreover, the strong showing of the volurnetric denvations in this study disagrees with the suggestion of Chapra and Tarapchak (1976) that the areal denvation reflects the theoretical understanding of phosphorus retention. It is also notable that the three equations (Fig. 3; equations 1,6 and 7) that perform best also calculate phosphorus retention as a positive function of P-load. Prairie (1988) and Canfield and Bachmann (1 98 1) also concluded, on the basis ofempirical comparisons, that phosphorus retention is more closely related to P-loading than phosphorus in the lake. Choosing the nght Ch1:TP equation is also critical to effective prediction. Both the Vollenweider-OECD and Fitted equations are more accurate than the Dillon and Rigler equation (Fig. 3). The close agreement between the performance of the Vollenweider-OECD and Fitted equations suggests that there is no obvious benefit to denving a Ch1:TP equation specifically for this data set. Rather, it is more important to ensure that the range of variables in the lakes of interest match those used in the development of the chosen Ch1:TP equation. in ihis analysis, the lakes included in the Vollenweider-OECD Ch1:TP equation appear more shilar to those in the study data set: both groups include lakes fiom a broad geographic distribution and have relatively large ranges in TP and Chi. Altematively, the Dillon and Rigler equation is perhaps better-suited to more oligotrophic lakes in that it coven a lower range of TP values (Table 6). The relatively strong performance of the Vollenweider-OECD equation also suggests that models developed across bmad geographical ranges can be used effectively in prediction in addition to their more traditional heuristic role of identifying patterns. Empirical Land-use Models and Environmental Management Effective environrnental management and land-use planning rely on scientific input. Such input incliides models that predict the effects of alternative management decisions. The empincal land-use models developed in this study are one source of such information. Empincal land-use models are directly relevant to managen concemed with nonpoint sources of pollution and land management. Empincal land-use variables such as area forested and population density were chosen because these variables reflect human use and are amenable to management. The empincal land-use models are relevant to environrnental managers as they provide direct predictions of Ch1 which is a "true quality variable of concem" (Reckhow and Chapra 1983:24) rather than estimating Ch1 via an intermediate like TP. Empincal land-use models are also attractive to managen because they are cheaper and simpler than two-step P-loading models. The variables required for the empirical land-use models are few, simple to estimate and generally available fiom maps. Two-step P-loading models require more data (cf. Dillon and Rigler 1975), and some variables such as P-load are costly to measure accurately. Two-step P-loading models require greater expertise as managen must choose kom a plethora of phosphorus export coefficients (Reckhow and Chapra l983), mass-balance formulations and Ch1:TP relations. The literature provides little direction in making these critical choices. Error estimation is a centrai issue to managers concemed about the degree of risk associated with management decisions. Error estimation in the empirical land-use models is simple as it is approximated by the standard error of the model. The error term in two- step P-loading models is harder to estimate. Fint order error analysis can be used to estimate enor in the mas-balance equations (Reckhow and Simpson 1980) but it requires a number of assumptions as to how error is partitioned arnong variables in the mas-balance equation. Moreover, when an empincal relationship such as a ChI:TP equation is linked serially to the TP calculation, total error must be estimated. This calculation may not be a simple additive function as seen in Fig. 3 and will depend on which combination of mass-balance and Ch1 equations the manager chooses. That there are interactions between TP calculations and Ch1:TP equations resulting in non-additive errors suggests caution in chaining regression models to the mas-balance component. Despite the strong performance of the empincal land-use models in this analysis, they should be viewed as complementary to the traditional, two-step P-loading approach. The choice of models depends on the management target, goals, and resources. Two-step P-loading models facilitate identification of sources and sinks of phosphoms in the watershed and lake and are key to understanding lake responses to dynamic processes such as changes in nutrient loads. Altematively, empirical regression models such as the empincal land-use models developed in this study are best used in the prediction of long-tem averages for a lake district, and have proven less adept at making predictions for single Mes (Reckhow 1994). With caution, regression models can also be used in the management of single lakes where financial resources for more in-depth studies are unavailable. indeed, statistical tools such as random coefficient regression are being developed to facilitate the application of cross-sectional models to single lakes (Reckhow 1994). Empincal land-use models offer advantages in setting management guidelines, predicting under steady state conditions, and supporting management under scarce human and financial resources.

Ackaowledgments Support for this research was provided by NSERC and FCAR. This paper is a contribution fiom the Limnology Research Centre of McGill and the Quebec Groupe de Recherche Interunivenitaire en Limnologie. The authors would like to thank Tom Miller and Sue Watson for critical reviews of the manuscript. Literature Cited:

Canfield, DR.,and R. W. Bachmann. 198 1. Prediction of total phosphoms concentrations, chlorophyll-a, and secchi depths in natural and artificial lakes. Cm. J. Fish. Aquat. Sci. 38: 414-423. Chapra, S.C., and S.J. Tarapchak. 1976. A chlorophyll a model and its relationship to phosphoms loading plots for lakes. Water Resources Res. 12: 1260-1264. Cohen, J. 1977. Statistical power analysis for the behavioural sciences. Academic Press, New York, N.Y. 474 p. Dillon, P.J., and W.B. Kirchner. 1974. The effects of geology and land use on the export of phosphoms from watersheds. Water Research 9: 135-148. Dillon. P. J., and F.H. Rigler. 1974a. A test of a simple nutrient budget model predicting the phosphoms concentration in lake water. J. Fish. Res. Board Can. 3 1 : 177 1- 1778. Dillon. P.J., and F.H.Rigler. 1974b. The phosphorus-chlorophyll relationship in lakes. Limnol. Oceanogr. 19: 767-773. Dillon. P.J.. and F.H. Rigler. 1975. A simple method for predicting the capacity of a lake For deveiopment based on lake trophic status. J. Fish. Res. Board Can. 32: 15 19- 1530. Edmondson, W.T. 1970. Phosphorus. nitrogen. and algae in Lake Washington after diversion of sewage. Science 169: 690-691. Efron, B., and R.J. Tibshirani. 1993. An introduction to the bootstrap. Monographs on Statistics and Applied Probability 57. Chapman Hall, New York, N.Y. 436 p. Gnesbach, S.J., and R.H. Peters. 1991. The effects of analytical variations on estimates of phosphoms concentration in surface waters. Lake & Reserv. Man. 7: 97-106. Hanna, M., and R.H.Peters. 1991. Effect of sampling protocol on estimates of phosphonis and chlorophyll concentrations in lakes of low to moderate trophic status. Cm. J. Fish. Aquat. Sci. 48: 1979-1986. Jones, J.R., and R. W. Bachmann. 1976. Prediction of phosphonis and chlorophyll levels in lakes. J. Water Pollut. Contr. Fed. 48: 2 176-2182. Kirchner, W.B.,and P J. Dillon. 1975. An empirical method of estimating the retention of phosphonis in lakes. Water Resources Res. 1 1: 182-183. Kleinbaum, P.L., L.L. Tupper,, and K.E. Muller. 1988. Applied regression analysis and other multivariate methods. Second edition. PWS-Kent, Boston, M.A. 71 8 p. Lake Biwa Research Institute [Ed]. 1988. Data book of world lake environments. Otsu: International Lake Environment Committee. Lake Biwa Research Institute [Ed]. 1989. Data book of world lake environments. Otsu: International Lake Environment Committee. Lake Biwa Research Institute [Ed]. 1990. Data book of world lake environments. Otsu: International Lake Environment Committee. Lanen, D.P.,and H.T.Mercier. 1976. Phosphorus retention capacity of lakes. J. Fish. Res. Brd. Can. 33: 1742-1750. Marshall, C.T.,A. Morin,, and R.H.Peters. 1988. Estimates of mean chlorophyll-a concentration: precision, accuracy and sampling design. Water Resources Bull. 24: t 027- 1034. Marshall, C.T.,and R.H. Peters. 1989. General patterns in the seasonal development of chlorophyll a for temperate lakes. Limnol. Oceanogr. 34: 856-867. McCauley, E., J.A. Downing,, and S. Watson. 1989. Sigmoid relationships between nutrients and chlorophyll among lakes. Can. J. Fish. Aquat. Sci. 46: 1171-1 175. OECD. 1982. Eutrophication of waters: monitoring, assessment and control. OECD, Paris, France. 154 p. OMNR. 19 86. Lakeshore capacity study. Queen's Printer for Ontario, Toronto, Canada. 89 p. Ostrofsky, M.L. 1978. Modification of phosphorus retention models for use with lakes with low areal water loading. J. Fish Res. Board Can. 35: 1532-1536. Prairie, Y.T. 1988. A test of the sedimentation assumptions of phosphorus inut-output models. Archiv. Hydrobiol. 1 1 1 : 32 1-327. Prairie, Y.T., and J. Kalff. 1986. Effect of catchment size on phosphorus export. Water Res. Bulletin 22: 465-470. Reckhow, K.H. 1979. Uncertainty applied to Vollenweider's phosphorus criterion. J. Water Pollut. Cont. Fed. 5 1: 2123-2128. Reckhow, K.H. 1994. Water quality simulation modelling and uncertainty analysis for risk assessment and decision making. Ecological Modelling 72: 1-20. Reckhow, K.H., and S.C. Chapra. 1983. Engineering approaches to lake management. Vol. 1 : data analysis and empirical modeling. Butterworth, Boston, M.A. 340 p. Reckhow, K.H., and I.T. Simpson. 1980. A procedure using modeling and error analysis for the prediction of lake phosphorus concentration kom land use information. Can. I. Fish. Aquat. Sci. 37: 1439-1448. Rowan, D.J.,and J. Kalf'f. 199 1. The limnological implication of catchment sediment load. Verh. Internat. Verein. Limnol. 24: 2980-2984. SAS Institute. 1985. SasISTAT User's Guide, Release 6.03. SAS Institute Inc., Cary, N.C. 1028 p. Schindler, D.W. 1977. Evolution of phosphoms limitation in lakes. Science 195: 260-262. Smith, V.H. 1979. Nutrient dependence ofprimary productivity in lakes. Limnol. Oceanogr. 24: 1051-1064. StraSkraba, M. 1993. Some new data on latitudinal differences in the physical lirnnology of lakes and reservoin, p. 19-39. In A. Boltovskoy and H.L. Lopez [ed.] Conference on Limnology. Instituto de Lirnnologia "Dr. R.A. Ringuelet", La Plata. Vollenweider, R.A. 1975. Input-output models with special reference to the phosphoms loading concept in lirnnology. Schweiz. Zeit. Hydrol. 37: 53-84. Zar, J.H.1984. Biostatistical analysis. Prentice-Hall, Englewood Cliffs. 7 18 p. Table 1: List of lakes included in the study: TS indicates the trophic statu where E = eutrophic, M = mesotrophic and O = oligotrophic; RG refen to whether the lake is regulated (Y) or not or a reservoir (R); GS = growing season defined where NT is north temperate (April to September), ST is south temperate (October to March) and TS is tropical and subtropical (annual averages); SEW = sewage treatment where Y indicates treatment rangkg 60m septic fields io tertiary, N indicates no treatment, and P indicates "pnstine" with no treatment. A '-' indicates no information available. Lakes with complete P-loading Data Lakes with incomplete P-loading Data

LAKE COUNTRY TS RG GS SEW LAKE COUNTRY TS RG GSSEW San Roque Res. Argentins ER STY Represa do Lobo Brazil EY TS Y Neusiedlersee Austria MY NTY Great Central Canada OY NT P Koo tenay Canada OY NTY Muskoka Canada OY NTY Lac Saint Jean Canada OY NTY Southern Indian Canada OY NTP Paajiirvi Finland OY NTN Williston Canada OY NTY Paijiime Finland O Y hTY Ontano CanadwCS ON NT Y Pielinen Finland O N NTY Chao-hu China OY TS N Ammersee Germany MN NTY Miyun Res. Chma OR TS N Stamberger Germany ON NTY Inari Finland OY NT Y Balaton H~~wEY NTY Lago Trasimeno Ital y OY NTY Lough Derg ireland EY NTY Hac hirô-gata Japan EY NTY Lough Ree Ire land OY NTY Mashû-ko Japan O- NT P Kinneret Israel M- TSY Shumarinai-ko Japan OR NTY Chiizenji-ko Japan OY NT- Tow ada- ko lapan OY NTY [nawashiro-ko Japan OY NTY Tasek Bera Malaysia ON TS P Inba-numa Japan EY NTY Phewa Nepal MY TS - Kawaguchi-ko Japan MY NTY Rotorua N.Zea1and EY ST Y Kizaki-ko Japan O Y NTN Laguna de Bay Philippines OY TS Y Koj ima-ko Japan EY NTY Buhi f hilippines E- TS - Nagase-damu-ko Japan MY NT- Victoria Tanzan ia EY TS Y Ogawara-ko Japan MU NTY Songkhla Thailand ON TS Y Ogôchi-damu-ko Japan ER NTY Loch Ness LJK OY NTY Shinji-ko Japan E - NTY CaYWa USA OY NTY Suwa-ko Japan E- NT- Chicot USA EY NT- Tega-numa Japan E Y NTY Okeechobee USA EY TS Y Taupo NZealand OY STY Washington USA MY NT Y Sniardwy Poland OU NTY Valencia Venezuela EN TS Y Hjalmaren Sweden MY NTY Malaren Sweden O Y NTY Vanern Sweden OY NTY Vàttem Sweden O Y NTY Zurichsee Switz. OY NTY Lough Neagh UK E Y NTY Canandaigua USA O Y NTY Lower Twin USA O Y NTP Upper Twin USA O Y NTP Table 2: Variables used in the analysis where n = the nurnber of observations, min = the minimum value, max = the maximum value, sd = the standard deviation of the mean, and grnean = the geometnc mean.

Variable units n min max mean sd gmem Surface area (Ao) km' 63 Lake volume (W) lo7.m! 60

Lake turnover time (R) YS 53 Mean depth (Z,,,) m 59 Max depth (Z) rn 61 Chlorophy ll-a (Chl) rng.cn.' 63 To ta1 Phosphoms (TP) mgm" 56 TN:TP 37 TN Load (TNLD) kg. 1 Ob.yr.' 3 1 TP Load (TPLD) kg*10"yr' 41 areal P-load (APL) kg*lO"yP*m" 41 population (Pop) # 58 pop. density (Pden) 56 rain fa11 (Rain) mmy-' 5 3 watershed area(Ws) km: 62 %- forested (FOR-p) 63 area forested (FOR-a) km' 62 %-agriculture ( AG-p) 63 area agriculture (AG-a) km' 62 %-urban (URB-p) 63 area urban (URB-a) km2 62 latitude (Lat) O 63 longitude (Long) O 63 annual mean temp. "C 55 Table 3: Description of 3 classes of models to predict Chl: Enipitical Land-Use (i); Eiiipincal P-loading (ii); and Two-step P- loading (ii i)

Model hscription Coding ------

(i) Empirical Land-use Modcls logChl= Bo f BiXi 4- B2Xi+ B,Xi coded: LU 1 and LU2;

one-stcp prediction of Cht from where X, arc land-use and niorphoinetric sec 'l'able 7 for variables and coefficients land-use and marphometry variables (sce Table 2)

(ii) Empirical P-loading Models logch1 = Bo 4-BJogR, + BJogNL + BJogZ, coded: El'; (sec 'l'able 2) one-slep predic tion of Ch1 froni sec Table 7 for coefficients variablcs uscd in mass-balance equations

(iii) Two-step P-loading Models step 1 : 1'1' is calculated from rnass-balance codcd: equations 1- 1 7 (Table 5) two-step prediction of Chl D 1-Dl 7 for Clil fi-om Dillon & Rigler; step 2: Ch1 is esiiniated froni calculatcd ïPusing Dofor Chl from obscrved TP three Chl:*J'P regressian cquaiions (Table 6) F 1-F 17 for Chl frtm the Fitted equation; E:, for Chl from otiserved Ti' Chl is also cstiniated froni observed TP usiiig tJic liree Chl:TP equatioiis V 1 -V 1 7 for Clil kom Vollenweider-OECD; Vo for Ch1 from observed TP Table 4: Correlation coellicients Tor the variables used in the analysis. All variables log-transfonned. Numbers in bold indicaie a significant correlation at the p = 0.05 level.

Chl Pden Pop R, n' z., AG-p AG-a FORp FOR-a URR-p URB-a APL AO -0.20 -0.01 0.52 0.50

Chl 0.62 0.45 -0.57

Pden 0.73 -032

POP -0.08

R( TP Zm

AGp

AG-a

FOR-p

FOR-a

m-P

üRB-a Table 5: Mass balance equations used to calculate TP (mgm") where APL = areal P- loading (mg-m2.yf'), Z, = mean depth (m), p = hydraulic flushing rate (yfl),o = P sedimentation coefficient (yfl)q= areal water loading (m-yi'),v=settling velocity (myfl) and R = phosphoms retention. D designates the equation denvation where 'a' indicates P- retention as a function of lake area, 'v' indicates P-retention as a function of lake volume and '-' indicates that P-retention is not specified.

Mass Balance Equation D Ref. TP,=0.8APL 1 (zm(o.0942(APUz,")".422+P)) TP:= 0.49APL / (~(0.0926(APL/&,,)0~510+p)) TP,= APL( 1 -(v/(v+q)))1 Z,,,p where v = 1 1.6+1.2q TP,= APL( 1 -(v/(v+q)))1 &p where v = 12.4

TP,=APL(1-(a/(&p))) 1 Zg where = 0.94 TPr APL(1-(o/(&p))) 1 Z,p where o = (0.162(APUZ,)O"' TP,= APL(1 -(ai(&p))) 1 2.p where = (0.129(APUZ.)o'" TP,= APL( 1 -(ol(o+p))) 1 Zg where = 1O/& TP,= APL( 1 -(VI(v+q))) 1 Z+ w here v=2.99+ 1.7q TPl,= APL( 1-R) 1 Lpwhere R = l/(l+p") TPIl=APL(1 -R)1 Z,,,p where R = 1/(1+0.747p0-'03 TPp APL(1-R)1 &p where R=0.426e(""'~+0.574e(4~OOW9iv T'Pli=APL( 1 -(24/(3O+Q))) I Zg Ti',,= APL(L-R)1 Zmp where R=0.201e14"*+0.574e~4-~ Tels= APL(I-R)/ Zg where R= I/(l+0.614p0-"') TPIr 0.603APL 1 (2,,,(0.257+p))

Pl,=APL( 1 -(a/(o+p))) l Zg where 0 =0.65 refs: 1 - Canfield & Bachmann (1981); 2- Larsen & Mercier (1976); 3- Jones & Bachmann (1 976); 4- Vollenweider (1 975); 5- Kirchner & Dillon(1975); 6- Ostrofsky (1 978); 7- Reckhow (1 979); 8- Chapra & Tarapchak (1 976). Table 6: Chl:TP equations, in pg-l", used in combination with rnass balance equations. Mode1 and regression parameter estimates for the fitted mode1 are the mean values for the ten runs with standard deviations in parentheses. n is the sample size, SEE is the standard error of the estimate, R' is the coefficient of determination.

TF Ch1 n SEE R' Range Range

Dillon logChl= - 1.136 + 1.49(logTP) 77 0.21 4 0.96 3- 180 3.75-260 & Rigler

Vollenweider logChl= -0.553 + 0.96(logTP) 77 0.25 1 0.88 5.6- 1 120 0.2- 107 - OECD Table 7: Suniniary statistics for "best"crnpincal land-use iiiodels (LU 1 -LU2) and the cnipirical P-loadiiig inodel (EP).

Parameter estimates reported as the nieais for the ten nuis * standard deviations. il is the saiiiple sixe, SEE is the standard error of the estimate, R* is the coefficient of determination, Bo is the intercept and B,-B, are the partial regession coeîficienis. Nuinbers in pareiltheses are partial R' values averûged over ten mns * staiidard deviations and n.s. iiidicates a non-significant partial +. Variable narnes as in Table 3; al1 variables are log-tra~isfomied.Parameter estirnates for eacli run available on request.

MODEL n SEE 8 Bo BI oz B3 ------logChlLu,=BofB,log~+B,iagPop+B,logI:OR-a 38 0.338*0.028 0.72i0.05 0.4 18k0.163 -0.272k0.03 1 0.3 1gkO.012 -0.227+0.028 (0.37k0.03) (0.24k0.03) (O. 12k0.04) l~gChl~~=B~+U,lag~+B~l~gPden+B~logliOK-p4 1 0.4l9fO.02 1 0.55*0.06 O.%'f 0.W 1 -0.296k0.033 0.29 1*0.010 -0.309k0.045 (O. 1 O*O.OI) (0.41k0.04) (0.05*0.01) logChlw=&+B IlogR,+B210gJa+B,logZ, 26 0.336Tt-0.027 0.73k0.04 -0.130;tO.1 17 0.082i0.057 0.5 16kO.057 -0.468k0.057 n.s (0.57it0.07) (O. 14M.03) Table 8: Cross-Validation Cornparison - estimates of mode1 accuracy as the mean of the squared residuals (MSR), model precision as the variance of the squared residuals (vSR) and uniform mode1 bias as the mean error (ME) for the empincal land-use (LUI-LU2), empirical P-loading (EP), best two-step P-loading models (D6, F7, and V7) and observed TP models (D,, F. and VJ. Reported values are the means * standard deviations for the ten runs.

-- - MODEL MSR vSR ME Table 9: Results of bias analysis where ME is the absolute value of the mean error (indicating magnitude of the bias), LU, EP, DR, F and VO are the nurnber of models in each bias category for, respectively, the empirical land-use models, the empirical P- loading mode1 and the two-step P-loading models according to the Ch1:TP equation used (Dillon & Rigler, Fitted and Vollenweider-OECD). Table 10: Ranked jackknife estimates of regression statistics: log observed Ch1 as a function (0 of the empincal land-use models (LU1-LU2), empincal P-Ioading mode1 (EP), observed TP (TP.) and mass balance estimates of TP (TP1-TP17).R', is the jackknife estimate of the coefficient of determination, and SEE, is the jackknife estirnate of the standard error of the estimate, n = 32.

MODEL R, SEE, Figure 1: Mean squared residuals (MSRs) averaged over ten runs for the land-use models (LUI,LU?), the empincal P-loading model (EP), the "best"two-step P-loading models @6, F7, V7) and models using observed TP (Do, Fo, Vo). Error bars indicate standard errors.

LUI LU? EP D6 FI V7 Do Fo Vo

model Figure 2a-c: Mean squared residuals (MSRs) for Nne hypothetical P-loading models ploned by Ch1:TP equations (1-3). Lines join points representing the same mas-balance equation. Panel 2a indicates that the mass-balance equation but not the Chl:TP equation afFects the MSR whereas panel 2b Uidicates the reverse situation where the Ch1:TP equation but not the mass-balance equation affects the MSR. Results such as those in Panel 2c indicate that both the Ch1:TP and mass-balance equations affect the MSR. Figure 3: Averages of the mean squared residuals (MSRs) for the 5 1 two-step P-loading models and observed TP (TPo)ploned by Chi:TP equation. Solid lines join points representing the 17 mass-balance equations; the numben refer to the mas-balance equation (Table 5). Averages of the MSRs for the land-use models (LUI, LU2; dotted lines) and empincal P-loading model (EP;dashed line) are included for cornparison.

-- -- D F V

Chl:TP model PREDICTINGCOASTAL EUTROPHICATION FROM LAND-USE:

AN EMPIRICAL APPROACH TO SklALL NON-STRATIFIED ESTUARIES.

lessica Jane Meeuwig

i\hrlne Ecology Progress Series: accepted 07.98

Abstract: Few models exist to directly and quantitatively predict the effrct of land-use on coastal water quality even though it is recognized that land-use is a major deteminant of coastal water quality. Such models are needed as 1) land-use is a major source of nutrients transported to estuaries; 2) land-use integrates multiple factors that may determine the mean algal biomass and 3) land-use is more easily managed than single nutrients when nutrient sources are non-point. Regression rnodeis accurately predict lake eutrophication as a function of land-use. Similar modeis do not exist for estuaries. A data set was compiled for 15 estuaries in Prince Edward Island (PEI), Canada, that includes phytoplankton biomass (as chlorophyll-a, Chl), total phosphonis (TP),total nitrogen (TN), estuary morphometry and land-use characteristics. A regression model predicting Chi as a function of estuary volume and area of agriculture was developed. This model accounts for 68% of the variance in Chl, a level similar to that of models based on TN (?-0.72) or total phosphonis (?=0.66). The estuary models based on land-use and total nutrients demonstrate low yields of Ch1 when compared to analogous lake models; this low yield is likely attributable to high levels of herbivory by suspension feeding mussels. Despite these low yields, the pattern between Ch1 and land-use is sufficiently accurate such that environmental managers can predict the effects of changing land-use on estuary water quality in PEI with a known level of error. Introduction Coastal eutrophication is an environmental concem along the US Atlantic Seaboard (Bncker et al. 1995), the Baltic (Rosenberg et al. 1990) and a nurnber of European coasts (c.f. Vollenweider et al. 1992). Much of the research investigating this phenornenon has focused on nutnent loading to coastal systems (c.f. Nixon 198 1, de Jonge et al. 1994, Boynton et al. 1996), an approach that assumes phytoplankton biomass, usually measured as chlorophyll-a (Chl), is conh.olled hy a single nutrient, frequently nitrogen (Hecky and Kilham 1988). However, this assumption must be questioned: Chl is also influenced by other nutrients such as phosphorus (Brockmann et al. 1990, Krom et al. 1991). and silica (Tumer and Rabalais 1994), the ratio of nutrients (Smith 1979, Prairie et al. 1989), the rate of nutrient turnover (Smith 1984) and turbidity (Fisher et al. 1988). Such multifactor control suggests that linear models based on single nutrients (e.g. Dillon and Rigler 1974), may not accurately predict Ch1 in coastal systems. One alternative is simulation models that allow the inclusion of multiple compatments (c. f. Stigebrandt and Wu1 ff 1987, Linker et al 1993). However, these models are site-specific, complex and expensive io develop, and thus may not be widely used. Another alternative is to develop empincal regtession models that utilize variables that integrate multiple factors. Such models retain the simplici ty and generality of the single nutnent empincal models while addressing the limitations inherent in assuming that there is a single factor limiting Chl. Land-use is one such integrating variable that reflects human disturbance in a catchment, thus combining the potential effects of, for example, nitrogen, phosphorus and sediment load on aquatic systems. With the increasing availability of geographic information systems (GIS),our ability to utilize spatial variables has improved and land-use is more easily and accurately incorporated into models. Unfortunately, land-use is not traditionally used to predict Ch1 directly. Instead, land-use is used to calculate a nutrient load (e.g. phosphorus) which is then used to predict phytoplankton biomass (Dillon and Rigler 1975; Field et al. 1996). This process reflects the assumption that Ch1 is more closely related to a single limiting nutrient than to land-use and thus is better predicted by nutrients than by land-use. To test this assumption. Meeuwig and Peters (1 996) compared the accuracy of empiricai models based on land-use and models based on total phosphorus (TP) to predict Ch1 in lakes. in the analysis, predictions derived from land-use were approximately 30% more accurate, suggesting that integrative variables such as land-use more effectively predict Ch1 in lakes. Chlmutrient relations in estuaries are not well defined. In their review of phytoplankton and nutrients, Hecky and Kilham (1988) conclude that the evidence for general nitrogen limitation of estuary Ch1 is weak compared to the evidence of general phosphoms limitation of lake Chi. Evidence from the Chesapeake Bay also suggests that estuaries may be limited first by phosphorus and then by nitrogen over the course of the growing season (Malone et al. 1996). Thus, land-use models that integrate the effects of both nitrogen and phosphorus as well as turbidity may predict estuary Chl more accurately than models based on either total nitrogen (TN) or TP. To evaluate the effectiveness of land-use as a predictor of estuary Chl, the following hypotheses were tested: 1) that Ch1:land-use models can be developed for estuaries; and 2) that models based on land-use predict Ch1 with greater accuracy and less bias than models based on TN or TP. Finally, as there are no a priori reasons to assume that estuary phytoplankton responds differently to land-use disturbance than freshwater phytoplankton, 1 tested the hypothesis that Chl:land-use relations for estuaries are similar to those for lakes.

Metbods Data Set A data set was compiled for fifieen estuaries in Prince Edward Island (PEI), Canada (Fig. 1) that includes water chemistry, coastal morphometry, and land-use variables (Table 1). The estuaries were chosen to represent a range of size and land-use charactenstics. They were sampled 6 times between May and August, 1996 at 3-5 locations in each estuary. Ch1 was determined spectrophotometrically following Bergmann & Peters (1 980). TP and M were also determined spectrophotometrically, adapting the methods of Menzel and Corwin (1965) and Solorzano & Sharp (1980), respectively. Details descnbing the sampling and analyses can be found in Meeuwig et al. (1998). Growing season averages were calculated for Chl, TP and TN in each estuary by averaging the values for the stations at a given samphg round and then averaging the values for the sarnpling round. Coastal morphometry (mean depth, surface area and volume) was determined fiom bathymetric charts published by the Canadian Hydrographie Service (1980). Land-use information derived from a GIS was provided by the PEI Department of Agriculture. For each watershed, land-use was divided into area under forest, area under crops and "other", which includes area under hay, grain, vegetables etc. GIS was also used to estimate human population size. Electoral districts were overlaid on the watersheds and the area of each electoral district in each watershed detennined. The Govemment of PEI provided the number of eligible voters in each electoral district which is approximately 7 1% of the total population (Thomson pers. comm.). The population in each watenhed was then estimated as

Population = (ZVden, AreaJ0.71 [es- 11 Where Vden, is the voter density in electoral district x and Area, is the area (m') of electoral district x in the watershed. Stutistical Analyses An estuary's response to land-use is a function of both the magnitude of the disturbance and the estuary's sensitivity to the disturbance. The estuary's sensitivity is determined by the morphometry of the estuary; for instance, small estuaries are more likely to demonstrate a response to a given disturbance than large estuaries, assuming other factors are constant. 1 therefore used multiple regession techniques to predict mean growing season Ch1 as a linear function of 1) land-use (e.g. area or percent forest, population density) as the disturbance variable and 2) coastal morphometry (e.g. mean depth, water residence tirne) to indicate sensitivity. Mean Ch1 over the growing season was used because temporal variability in the response variable should match temporal variability in the independent variables. As land-use does not Vary within the growing season, growing season Ch1 values must be used. Similarly, Ch1 values are averaged over the sarnpling stations because land-use is measured at the catchent scale and thus Ch1 values must be estimated at the scale of the whole estuary. Al1 variables were log- transformed to reduce non-normality and stabilize variance. The SAS procedure "Proc REG was then used with the "rsquare" option (SAS institute 1985) as this option calculates the coefficient of determination for a11 possible combinations of variables and thus, as an exploratory technique, identifies a suite ofpotential models. Of the possible models, 1 chose the "best" bivariate mode1 which is that model with minimal correlation between independent variables (r < 0.5). the highest coefficient of determination (i)and the lowest standard error of the estimate (SEE). The analysis was limited to bivariate models because such models allow the inclusion of one disturbance variable and one sensitivity variable. Fitting additional variables would likely result in over-parameterizhg the model given the sample size of 15. In addition to describing the pattems between estuary Chl and land-use, the rnodels should predict the effect of changing land-use pattems on Chl. Thus, confidence limits for predicted values (hence prediction limits) of Ch1 were also calculated (Zar 1984). The estuary Ch1:land-use model developed in this study was compared to estuary Ch1: nutnent models and fieshwater Chl :land-use models Frorn the li terature. Estuary Ch1:nutrient models (Table 2) are those reported in Meeuwig et al. (1998); freshwater Chl:land-use models (L 1, LZ) are those reported in Meeuwig and Peters (1996) based on a set of lakes fiom amund the world compiled by the Lake Biwa institute (1988, 1989, 1990). In addition. the lake data (Meeuwig and Peten 1996) were used to fit bivariate Chkland-use models using the same variables that appeared in the estuary model (Table 2; L3). Cornparisons were made between the accuracy, precision and bias of 1) the estuary and lake land-use models and 2) the lake land-use models in predicting estuary Chl. These cornparisons are problematic because traditional goodness of criteria fit are inappropriate. Departure fiom the 1 :1 line of an observed vs. predicted plot allows only a qualitative cornparison of accuracy (is the slope significantly different fiom 1 or not) and the coefficient of detemination is sensitive to model sample size. Following Meeuwig and Peten 1996,I therefor used the mean squared residual (MSR),variance of the squared residuals (vSR) and mean error (ME) as the indicaton of accuracy, precision and bias: MSR = Z(lCh1, - Ch$)' n*' [es- 21 where IChl, - ]Ch$ is the difference between the log values of observed and predicted Ch1 and n is the number of observations vSR = L[(IChl, - 1ChlJ2 - MSRI2 . (n - 1)-' [eq*31 ME = Z(lCh1, - lCh$) -n" [eq- 41 in addition to comparing the fit of estuary data by lake land-use models, 1 tested whether a single model could be fit to both lakes and estuaries. Regression analyses were re-nui on a combined lake/estuary data set with a dumrny variable (DV) set to O for lakes and 1 for estuaries.

Results Estuary Ch1 is most accurately predicted by estuary volume (Vol) and the area under agriculture (Ag-a) (Fig. 2). The model describing this relationship has the smallest SEE (SEE=0.135)of al1 the bivariate models estimated and the independent variables account for 68% of the variance in Ch1 (Table 2). Despite the small sample size (n = 15), the model is robust. The independent variables are orthogonal: they are not correlated (r = 0.12; p = 0.67) and individually, Vol and Ag-a account for only 45% and 16% of the variability in Chl respectively, which is less than the total variance accounted for by the bivariate model. The bivariate model thus does not overfit the data. Al1 of the other bivariate rnodels estimated had larger SEEs and/or incorporated independent variables that were correlated. Prediction intervals around Ch1 values for the estuary Chl:land-use model suggest that the absolute magnitude of the error increases as the Ch1 value increases; however, the relative error remains approxirnately the sarne. On an arithrnetic scale, the 95% prediction interval is constant across the values with the lower predicted limit approximately half of the predicted value and the upper limit 2 fold greater (Table 3). Thus, for example, the land-use mode1 predicts a Ch1 value for Mill River of 2.7 pg - lm'Ch1 and the lower and upper predictive limits are 1.3 pg . 1-' to 5.5 pg 1'' respectively. Land-use, TP and TN are comparable in the degree of estuary Ch1 variability for which they account (68,65 and 72% respectively; Table 2; Fig. 3). Even though the accuracy of the models is similar, the land-use model is more slighily more precise and less biased then the total nutrient models (Table 2). The prediction intervals for Ch1 as a fhction of TN and TP are also similar to the intervals for Chl as a function of land-use (Table 3) with the prediction intervals fiom TN marginally mialler. The SEE of the estuary Ch1:land-use model is smaller than those of the lake Ch1:land-use models (L k.338, L2= 0.4 19 and L34.470; Table 2). This result suggests that the accuracy with which Ch1 is predicted by land-use in estuaries is likely greater than that seen in lakes. The SEE of the estuary Ch1:land-use model is almost 4 times srnalier than the SEE for the analogous lake Ch1:land-use model (L3) suggesting that volume and agriculture predict estuary Ch1 more accurately than they predict lake Ch1 (Table 2). These analyses also show lake and estuary land-use rnodels cannot be used interchangeably. Lake Ch1:land-use models are inaccurate, imprecise and biased in their prediction of estuary Chl. In models L 1, L2 and L3, the MSR are 1-2 order of magnitude greater than the MSR of the estuary-specific model. Model L2 is the most accurate in predicting estuary Chl with a MSR of 0.105; this is however still more than double that of the estuary specific model (Table 4). Precision is also low with the vSRs of the lake land- use model predictions 2-3 orders of magnitude greater than the vSR of the predictions lrom the estuary model (Table 4). Model LZ is the most precise of the land-use models. The lake models are also highly biased in their prediction of estuary Ch1 (Fig. 4) with MES 3 orders of magnitude greater than the estuary mode1 value (Table 4). Model L2 is the least biased (Table 4). The dummy variable analysis also shows that a single model does not apply to lakes and PEI estuaries. In al1 the land-use models, the added dummy variables were significant at p < 0.05. Thus, although the lake and estuary data are fit by models with the sarne variables, the relationship between Ch1 and these variables differs in the estuaries. The dummy variable coefficients are negative (DV=O for lakes; DV=l for estuaries; Table 5), which indicates that Ch1 values are lower in estuaries than in Mes. In al1 cases, inclusion of a DV increased the coefficients of determination and decreased the SEE (Table 5) suggesting that it is inappropriate to fit a single model without accounting for differences between lakes and estuaries. However, once these differences are included via the dmy variable, the model accuracy is much improved. The combined land-use models have higher coefficients of determination and lower SEE than models without the DV. However, despite the inclusion of the dummy variables, the combined models still likely predict estuary Ch1 less accurately than the estuary-specific model given the higher SEE of the combined model (Tables 2 and 5) thus the system-specific model is probably generally preferred to a combined model with dummy variable. Discussion A robust model predicting estuary Ch1 as a hction of volume and area under agriculture was developed. This model explained 68% of' the variance in estuary Ch1 and demonstrates that even with their relatively short residence time (mean = 0.35 years vs 6 years for these lakes (Meeuwig and Peten 1996), estuaries show the influence of watershed-based activities. This result is consistent with the river study of Basu and Pick (1996) who demonstrated a strong relation between Ch1 and TP, despite residence times as short as 3 days. This mode1 has the same structure as the models used to manage lake eutrophication (OECD 1982): there is a variable that quantifies the anthropogenic disturbance (area under agriculture) and a variable that quantifies the sensitivity to that disturbance (estuary volume). Once the model accounts for an esniary's sensitivity, the elfect of agriculture is clear. For exarnple, in a small estuary like Savage Harbour (volume = 18.5 10h3),the Ch1 biomass at a given level of agriculture is higher than that in a larger estuary like Murray Harbour (volume = 35 1. 10' m') (Fig. 5). Increasing the areal extent of agriculture in the watershed also results in a steeper increase in Ch1 in Savage Harbour than in Murray Harbour (Fig. 5). Human population and population density did not enter into any of the land-use models. This result was unexpected because they both were significant in the lake Ch1:land-use models (Meeuwig and Peters 1996). Peierls et a1 (1 99 1 ) also demonstrated strong correlations between human population and nitrogen concentration. However, in PEI, populations are generally small and there is little variation in population density (Table 1); population density varies by a factor of 4 for these estuaries whereas population density varies over Four orden of magnitude in the lake study (Meeuwig and Peten 1996). Although land-use predicts estuary Chl, it does not predict Ch1 more accurately than either TP or TN. This result contradicts those of a similar analysis of lakes in which land- use predicted Ch1 30% more accurately than TP (Meeuwig and Petea 1996). In lakes, it is generally accepted thai TP is the nutrient limiting Ch1 (Schindler 1981) and thus one might expect that it is the best predictor of Ch1 and that variables such as land-use that integrate watershed disturbance (e.g. nitrogen loading, erosion etc.) would be less effective than TP. That land-use more accurately predicted Ch1 in lakes suggests that other factors such as nitrogen (Smith 1979) and sediment (Rowan and Kalff 1991) are important despite the strong pattern between Ch1 and TP. In estuaries, the relative importance of TP and TN in controlling Ch1 appears to be more variable than in lakes (Hecky and Kilham 1988; Jordan et al. 1997), thus 1 expected integrative variables such as land-use to be more accurate in predicting Ch1 than either TN or TP. Two possible explanations exist for the similar performance of land-use and total nutnents in predicting Ch1 in PEI estuaries. First, estuaries are thought to be more complex than lakes and if may be that estuary hydrodynarnics and seaward processes uncouple Ch1:land-use relationships relative to Ch1:total nutrient relationships. This explanation seems unlikely for a number of reasons. The estuaries in this study are small (mean volume 98 IO6 m3 and mean depth of 8.7 m), are vertically well-mixed and show no stratification. Mean-variance relationships cmalso be used to assess the arnount of variability in a system (Marshall et al. 1988) and an analysis of mean-variance relationships for Ch1 and TP in the PEI estuaries (Nayar and Meeuwig n.d.) shows similar temporal patterns in variability to those seen in lakes, suggesting that estuaries are no more variable temporally than lakes. More relevant, the estuary spatial mean-variance patterns indicate that spatial variability is less than temporal variability within the PEI estuaries. This evidence suggests that these estuaries are not sufficiently spatially variable to uncouple the regression models. Moreover, one would expect that complexity per se would uncouple both land-use and total nutrient relations. Seaward processes might however di fferentially uncouple land-use and total nutrient relationships. If this is the case, the weakness of Ch1:land-use relationships relative to Ch1:total nutrient relationships should increase as one sarnples Merseaward. The estuary Ch1:land-use and Ch1:total nutrient models were refit using only the data fiom either 1) the most land-ward station or 2) the most seaward station. The coefficients of detemination for the Ch1:land-use and Chktotal nutrient relationships for the landward and seaward stations showed no difference suggesting no change in the relative strength of the relationships. If seaward processes uncouple the relationship between Ch1 and land-use relative to Ch1 and total nutrients, one would also expect that the relationships between total nutrients and land-use would be uncoupled. There is a strong relationship between TN and land-use (?=0.72; Table 6) however the best predictor of TP is estuary volume and no land-use variables are correlated with TP. Thus there may be an influence of seaward processes on TP but it is less likely to be the case with TN. Strong collinearities between land-use, RI and TP may also account for the similar performance of land-use, RI and TP in predicting Chl. There are indeed strong correlations between TN and TP (r = 0.82; p = 0.0002), between TN and land-use but not between TP and land-use (Table 6). This result is consistent with other studies that show no relationships between land-use and TP: Meeuwig and Peters (1996) found no correlation between lake TP and land-use, and the strongest correlate of P-export in subestuaries of the Chesapeake is sediment load (Jordan et al. 1997). Nitrogen has more frequently been correlated with land-use (Jordan et al. 1997). This difference is thought to reflect the mechanisms by which nitrogen and phosphorus are transported to estuaries. Nitrogen is transported primxily via ground water whereas phosphorus is transported pnmarily via surface runoff (Peterjohn and Correll 1984). Thus the influence of agriculture on TP export may be sufficiently modi fied by catchment geology and soi1 characteristics to eliminate the correlation (Jordan et al. 1997). The mechanistic explanation for the lack of a correlation between TP and land-use does not, however, explain how Chi can be correlated with both TP and land-use in the absence of a relationship between TP and land-use. These correlations suggest that TP and land-use account for different components of the variability in Chl. This conclusion is supported by a multiple regression includes both land-use and TP:

lChl= 1.745 -0.12. lVoi+0.15 Mga+ 1.14- ITP [es- 51 where lChl is log Chlorophyll, lVol is log estuary volume, lAga is log area of agriculture and ITP is log T'P.The coefficient ofdetermination for this relationship is 0.81, an increase of 0.13 over the model without TP ($=O.68, Table 2). The model standard error has also decreased fiom 0.135 to 0.108. This model must be viewed cautiously as exploratory rather than predictive as fitting 3 independent variables to a data set of 15 pushes the limits of the data; indeed, estuary volume enters the model at p = 0.06. These analyses do however suggest a role for TP in predicting coastal eutrophication separate fkom TN and land-use and it may be this additional role that results in the similar strengths of the land-use and total nutrient models. This analysis does not however indicate whether Ch1 is pnmarily correlated with land-use or TN. The low yield of Ch1 per unit TP, TN or agricultural land-use in the PEI estuaries as compared to lakes raises additional questions as to which factors control phytoplankton biomass. A cornparison of the estuary Ch1:land-use and Chknutrient relations to comparable lake relations shows that Ch1 yields per unit TP or 'M in the PEI estuaries are consistently an order of magnitude lower than yields in lakes (Meeuwig et al. 1998). A similar pattern holds for the models that predict Ch1 as a fiction of volume and area under agriculture; the estuary model intercept is an order of magnitude lower than the lake model intercept (Table 2). When lake land-use models were applied to estuary data, observed estuary Ch1 values were consistently overestimated (Fig. 5) and the bias is negative (Table 4). The dummy variables for the combined lake/estuary data are always signi ficant and negative (Table 5) indicating a lower yield in the estuaries. The low yield cannot simply be an estuary phenornenon reflecting the tidal, saline nature, or relatively short residence time of these systems. Data for Ch1 and TP from a set of Carolinian estuaries show yields comparable to those seen in lakes (Table 7) and t-tests show no significant difference between the intercept of the Carolinian model and the lake models. The low yield in PEI estuaries suggests that although nutrient concentrations are high relative to phytoplankton biomass, phytoplankton are unable to take advantage of the nutrients which in tum suggests that nutnents are not limiting algal biomass. Herbivory by suspension-feeding bivalves and interactions between iron and phosphorus may be responsible for the low Ch1 yield in PEI estuaries (Meeuwig et al. 1998). Oficer (1982) first speculated that suspension feeden might prove to be natural eutrophication controls and that this was key in estuaries which have, compared to Mes, high biomass of suspension feeden. in lakes and nven, Manimder (1994), Quiros (1 990), Mellina et al (1 995) and Basu and Pick (1 996) have also invoked herbivory by zooplankton and zebra mussels to explain low Ch1 yields. In the PEI estuaries for which 1 have estimates, musse1 biomass is strongly correlated with Ch1 (r = -0.92). Thus top-down effects appear to occur yet not to the extent that the relationships between CM and bottom-up factors, such as nutrients and land-use, and the physical characteristics of the estuary, are uncoupled. This likely reflects some covariance between the presence of musse1 aquaculture and land-use as aquaculture operations tend to be established in estuaries surrounded by relatively low levels of agriculture. From the model cornparison and the dummy variable analysis, it is clear that PEI estuaries respond to land-use differently than lakes. However, this difference may well be a function of herbivory rather than an intrinsic difference in phytoplankton response to nutrients or watershed disturbance across these systems (Meeuwig et al. 1998). There may also be some additional seaward effect of TP such that land-use and total niitrient rnodels are comparable in their prediction of Chl. A cross-system regression model for lakes and estuaries can be developed as a heuristic tool in which the coefficient of the durnmy variable indicates the magnitude of the difference due to, for exarnple, herbivory. However, they should not be used for predictions as combining the data decreases the accuracy of the models.

Management Implications Estuary Ch1 in PEI is likely detennined by a combination of bottom-up and top- down factors. With respect to bottom-up factors, the management focus should be on land-use because nutrient loading in PEI is essentially nonpoint due to the importance of agriculture and the low population density. Not only is it difficult to establish single nutrient control prograrns for nonpoint source nutrient loading, but in the event of success, another nutrient frequently becomes limiting. Nutrients must thus be managed together (D'Elia et al. 1986, de Jonge et al. 1994, Turner and Rabalais 1994) and management efforts should focus on land-use which integrates multiple effects. Thus, the preoccupation with the relative importance of TN and land-use in the bottom-up control of estuary Ch1 is likely irrelevant to management. Ch1:land-use models are not uncoupled by herbivory thus these land-use models provide estimates of the degree to which agricultural development cm proceed in a given watenhed if Ch1 is to be maintained at or below a given limit, given the presence of herbivory. The most likely use of these models is to predict the effect of land-use change on an estuary of interest. Mode1 accuracy can first be determined by measuring Chl and land-use for the estuary of interest and then comparing the model predictions to observed Ch1 values. If the prediction is within a confidence interval considered acceptable by the managers, the mode1 can then be used to predict the expected Chl value with a change in land use. For instance, in Savage Harbour, observed Ch1 is 1.6 mg . m" and 18% of the catchment is agricultural. The mode1 predicts 1.8 k 0.13 mg mJ Ch1 under current conditions. If the percentage of the catchment under agriculture were to increase to the provincial average, Ch1 values would be expected to increase to 2.2 t 0.1 3 mg m" Chl, an increase of 14%. These models thus provide plmers with a range of likely outcornes to support their management decisions.

Acknowledgements: 1 thank A. deBmyn and an anonymous referee for their critical reviews. Field and laboratory assistance was provided by E. Sunderland and D. Chamberland. In PEI, B. Penak (Bedeque Bay Environmental Management Association), B. Raymond (PEI Dept. Fisheries and Environment) and B. Thomson (PEI Dept. of Agriculture and Forestry) provided essential support. This work was supported by a McGill Faculty of Graduate Studies Grant and the National Science and Engineering Research Council of Canada and is a contribution from the McGill University Limnology Research Group. Literature Cited:

Basu B, Pick F (1996) Factors regulating phytoplankton and zooplankton biomass in temperate riven. Limnol. & Oceanogr. 41(7): 1572- 1577 Bergmann M, Peters RH (1980) A simple reflectance method for the measurement of particulate pigment in lake water and its application to phosphorus-chlorophyll- seston relations. Can. I. Fish. Aquat. Sci. 37: 1 11-1 14 Boynton WR, Murray L, Hagy JD, Stokes C, Kemp WM (1996) A comparative anaiysis of eutrophication patterns in a temperate coastal lagoon. Estuaries 19:408-42 1. Bncker SB, Clement CG, Frew S. Hannon MR, Pirhalla DE (1995) NOAA's Estuarine Eutrophication Survey Volume 1 : Mid-Atlantic Region. Office of Ocean Resources Conservation and Assessment National Ocean Service National Oceanic and Atmospheric Administration Maryland Silver Spring. Brockrnann UH, Laam RWPM, Postma H (1990). Cycling of nutnent elements in the North Sea. Netherlands J. Sea Research 26: 239-264. Canadian Hydrographic Service ( 1980) Bathymetric charts (nos. 4420,442 1 4422 4425 4459 4467 4491 4492). Minister of Supply and Services, Ottawa De Jonge VN, Boynton W. D'Elia CF, Elmgren R, Welsh BL (1994) Responses to developments in eutrophication in four different north Atlantic estuarine systems. In: Dyer KR, Orth RI (eds.) Changes influxes in estuaries. Olsen and Olsen, Fredensborg D'Elia CF, Sanden JG, Boynton WR (1986) Nutrient e~chmentstudies in a coastal plain estuary: phytoplankton growin in large-scale, continuous cultures. Cm. J. Fish. Aquat. Sci. 43:397-406. Dillon PJ, Rigler FH (1 974) The phosphorus - chlorophyll relationship in lakes. Limnol. and Oceanogr 19: 767-773 Dillon PL Rigler FH (1975) A simple method for predicting the capacity of a lake for development based on lake trophic status. J. Fish. Res. Board Can. 32: 1519- 1530. Field CK, Siver PA, Lott AM (1996) Estimating the effects of changing land-use patterns on Conneticut lakes. J. Env. Qual. 25(2): 325-333 Fischer TR,Harding LW, Stanley DW, Ward LG (1988). Phytoplankton, nutrients, and turbidity in the Chesapeake, Delware, and Hudson estuaries. Estuarine, Coastal and Shelf Science 27:6 1-93 Hecky RE, Kilham P (1988) Nutrient limitation of phytoplankton in fieshwater and marine environments: a review of recent evidence on the effects of enrichment Limnol. Oceanogr. 33 :796-822 Jordan TE. Correll DL. Weller DE ( 1 997) E ffects of agriculture on discharges of nutrients from coastal plain watersheds of Chesapeake Bay. J. Environ. Qual. 26: 836-848. Krom MD. Kress N. Brenner S. Gordon LI (1991). Phosphorus limitation of pnmary productivity in the eastern Mediterranean Sea. Lirnnol. Oceanogr. 36: 424-432 Lake Biwa Research Institute [Ed]. (1988). Data book of world lake environments. Otsu: International Lake Environment Committee. Lake Biwa Research Institute [Ed]. (1989). Data book of world lake environments. Otsu: International Lake Environment Committee. Lake Biwa Research Institute [Ed]. (1990). Data book of world lake environrnents. Otsu: International Lake Environment Cornmittee. Linker LC, Stigall GE, Chang CH, Donigian AS (1993) The Chesapeake Bay watershed mode[. CSC.MD 1 : 3-9 Malone, TC.Conley, D.J.,Fisher, T.R., Glibert, P.M.,Harding, L.W., Sellner, K.G. ( 1996). Scals of nutrient-limited phytoplankton productivity in Chesapeake Bay. Estuaries 19(28):371-385. Marshall, C.T.,Morin, A., Peters, R.H. 1988. Estimates of mean chlorophyll-a concentration: precision, accuracy, and sarnpling design. Water Resources Bull 24(5) Manunder A ( 1994). Patterns of algal biomass in dominant odd- vs. even-link lake ecosystems. Ecology 75(4): 114 1 -49 Meeuwig JJ, Peten RH (1996). Circumventing phosphonis in lake management: a cornparison of chlorophyll a predictions fiom land-use and phosphorus-loading models. Can. J. Fish. Aquat. Sci. 53: 1795-1806 Meeuwig JJ, Rasmussen, J, Peters RH (1998). Twbid waters and clari@ng mussels: their moderation of Ch1:nutrient relations in estuaries. Marine Ecology Progress Series. in press. Mellina E, Rasmussen JB, Mills EL (1995) Impact of zebra mussels (Dreissena polymorpha) on phosphorus cycling and chlorophyll in lakes. Can. J. Fish. Aquat. Sci. 522553-2579 Metuel DW. Convin N (1965) The measurement of total phosphorus in seawater based on the liberation of organically bound fractions by persulfate oxidation. Lirnnol. Oceanogr. 10:280-82 Nayar T. Meeuwig JJ (n.d.) Charactenzing estuaries: a mean-variance approach to designing estuarine sarnpling programs. in prep. Nixon SW (1 98 1) Remineralization and nutrient cycling in coastal marine ecosystems. In: Neilson BJ. Cronin LE (eds.) Estuaries and nutrients. Humana Press, Clifion, NJ OECD (1982) Eutrophication of waters: monitoring assessrnent and control. Organisation for Economic Cooperation and Development. Paris Officer CB, Smayda TJ, Mann R (1982) Benthic filter feeding: a natural eutrophication control? Mar. Ecol. Prog. Ser. 9203-210. Peierls BL, Caraco NF, Pace ML. Cole JJ (1991) Human influence on river nitrogen. Nature 350: 386-87. Petejohn WT, Correll DL (1 984) Nutrient dynamics in an agricultural watenhed: observation on the role of a nparian forest. Ecology 65: 1466- 1475. Prairie YT. Duarte CM, .Kalff J (1989) Unifjmg nutrientshlorophyll relationships in lakes. Can. J.Fish. Aqua. Sci. 46: 1 176-82 Quiros R (1990) Factors related to variance of residuals in chlorophyll - total phosphorus regression in lakes and reservoirs of Argentina. Hydrobiologia 200120 1:343-355 Rosenberg R, Ehgren R, Fleischer S, Jonsson P, Persson G, Dahlin H (1990). Marine eutrophication case studies in Sweden. Ambio 19: 102-1 08 Rowan DJ, Kalff I (1 99 1) The limnological implications of catchment sediment load. Verh. internat. Verein. Limnol. 24: 2980-2984. Sakarnoto M (1966) Primary production by phytoplankton community in some Japanese lakes and its dependence on Iake depth. Arch. Hydrobiol. 62: 1-28. SAS Institute Inc (1985) SASSTAT user's guide release 6.03. SAS Institute inc. Cary N.C Schindler DW (198 1) Studies of eutrophication in lakes and their relevance to the estuarine environment. p. 71 -82 In: Neilson BJ, Cronin LE (eds.). Estuaries and nutnents. Clifton New Jersey: Humana Press Smith SV (1984) Phosphorus versus nitrogen limitation in the manne environment. Limnol. Oceanogr. 29: 1 149- 1 160 Smith VH (1979) Nutnent dependence ofprimary productivity in lakes. Limnol. Oceanogr. 24: 105 1- 1064 Solorzano L, Sharp IH (1980). Determination of total dissolved nitrogen in natural waters. Lirnnol. Oceanogr. 25:75 1-754 **Stigebrandt Wulff 1987 Thomson. B. Persona1 Communication. PEI Department of Agriculture and Forestry. Turner RE, Rabalais NN (1994) Coastal eutrophication near the Mississippi river delta. Nature 368:6 19-62 1 US EPA (1997). Unpublished data retrieved fiom STORET. Vollenweider RA, Marchetti R. Viviani R. (1992) Marine coastnl eutrophication. Elsevier NY 1310pp Zar .Fi (1984) Biostatistical analysis. Prentice-Hall. Englewood Cliffs NI Tabb 1: Data for 15 estuaries (abbreviatioiis in parentheses) froni Prince Edward Island used in lhis anûlysis. Musse1 wcight is in grams dry weight.

Site Chl Total Total Secchi Menn Vulunie Resideiice Saliiiity Watershcd Agriculture Pop. Mussels Ph~sphOnis Nilrogen Dcpih Ti me Area Density - - - - . ------.------. ------~nks.rn~mi*' rng.~' m m rn'-l~~ Days 960 km2 Boughton (BO) 1.398 57.9 209 3.5 18.0 140 332 2 1.2 39.5 10.3 9.3 7.29 Bnidenell (BK) Cardigan (C Darnley Basin (DB) Dunk (D) Foxtey (F) Grand (G) Mumy Harbor (M) Mill River (ML) North Lake (NL) Percival (P) Rustico (R) St. Peters (SP) Savage Harbor (SV)

SD Min Max 5.915 94.8 48 1 4.1 18.0 351 356 22.2 292.9 140.4 7.2 9.63

Table 3: Observed Ch1 and Ch1 predicted from the land-use mode1 (Chl-land-use). TP (Chl-TP) and TN (Chl-TN)with corresponding 95% lower and upper prediction limits. Numbers in bnckets indicate the x-fold decrease or increase of the lower or upper limit over the predicted value. Al1 units in pg -*l*'

Table 4: Application of lake models to estuary data where accuracy is estimated by the mean squared residual (MSR),precision is estimated as the variance of the squared residuals (vSR) and bias is estimated as the mean error (ME). ModeIs are as in Table 2.

Model acc uracy precision bias MSR vSR ME

Lake land-use (L 1) 0.903 0.126 -0.935

Lake land-use (L2) 0.105 0.0 18 -0.247

Lake land-use (L3) 0.860 0.084 -0.916 Table 5: Models for conibined lake and estuary data including a duniniy variable (DV) where n is the sample size of the combined data, 2 is the coellicient of determination, SEE is the mudel standard error and CV is the coemcient of variation. 'Ihe values in parentheses are the siatistics for the models fit to the combincd data without a dumniy variable.

I Mode1 Quation n r' SEE Lake land-use (LI) lChl = 1 -029-0.189(1Kt)+0.086(IPop) 4.201 (IFor-a) - 0.793 (DV) 52 0.57 (0.42) 0.40 (0.50)

Lake land-use (L2) lChl = 1 .#8-0.5 IO(1Zm) + 0.206(1Pden) -0.260(1For-p) - 0.475(DV) 50 0.74 (0.64) 0.32 (0.37)

Lake land-use (L3) lChl = 1 -704.339 (lV1) + 0.10 (!Ag-a) - 0.97(DV) 56 0.50 (O. 13) 0.44 (0.56) Table 6: Regression equations predicting total nutrients as a function of estuarine volume (VI) and land-use (Ag-a) for total nitrogen (TN) and total phosphorus (TP);the coefficients of determination

(9)and mode1 standard error (SEE) are also reported. All variables are log transfomed.

Mode1 Equation ? SEE TN 1TN = 2.49-0.11 7(1VI)+O.107(1Ag-a) 0.72 0.062

TP ITP = - 1.O?-0.102(IVI) 0.46 0.077

Table 7: Ch1 yields per unit TP and TN.Al1 yields except that for TN fkom the Carolinian esniaries were calculated tiom regression equations from the [ 1 ] OECD (1982), [2] Sakamoto (1966). [3]

Carolinian estuaries based on data from the US EPA (1997) and [4] PEI esniaries (this snidy) using mean values for TP (66.7~1") and TN (287 pg 1") found in PEI estuaries. The yteld for the

Carolinian TN data was simply calculated as the rnean of individual ratio of Ch1 and TN for each observation as the regression of Ch1 on TN was not statistically- significant. Mode1 yield (TP) yield (TN) Lake 0.237 ' 0.030 '

Carolinian estuaries 0.1 17 ' 0.038 '

PEI estuaries 0.035 " 0.008 " Figure 1 : Map of Prince Edward Island, Canada, showing the location of the 15 estuanes included in the analysis. The provincial capital, Charlottetown, is indicated by the astenx.

Figure 2: Observed vs. predicted chlorophyll (Chl; mg m") as a function of estuary volume and area of agiculture. Solid heis the 1: 1 line.

2 3 4 5 predicted Ch1 (mg m-3) Figure 3: Observed vs. predicted chlorophyll (Chl; mg - m") as a function of (3a) total phosphoms (TP;mg - m'l) and (3b) to ta1 nitrogen (TN; mg - m"). Solid lines are the 1: 1 lines. O 1 2 3 4 5 6

predicted Ch1 (mg ni3) Figure 4: Observed vs. predicted chlorophyll (Chi; mg m") as a function of land-use models (4a) L 1, (4b) L2 and (4c) L3. Filled circles are the lake data to which the models were fit, open circles are the estuaries to which the models were applied. Solid lines are 1 :1 lines.

predicted log (Chl (mg m'3))

-1 O 1 2 3 predicted log (Chl (mg m")) Figure 5 : Predicted chlorophyll (CH;mg - rnJ) vs. area of agriculture (kd) for a mal1 estuary (Savage Harbour, open circles) and a large estuary (Murray Harbour, filled circles) demonstrating differences in system sensitivity to a aven level of disturbance.

10 15 20 agriculture - km' TURBIDWATERS AND CLARIFYINC MUSSELS: THEIR MODERATION OF EMPIRICAL CHL:NUTRIENTRELATIONS IN ESTUARIES IN PRINCEEDWARD ISLAND, CANADA

Meeuwig, I.J., J.B. Rasmussen and R.H. Peten

Marine Ecology Progress Series: ac cep ted 06.98

Abstract: Coastal eutrophication has been identified as an important ecological problem in many regions. Yet simple, generalizable models, such as those available for the management of lake eutrophication. do not exist for estuaries. As a fint step in the development of estuarine eutrophication models, we generated chlorophyll-amutnent regession models for 15 estuaries in Prince Edward Island, Canada. Total phosphorus and total nitrogen account for 65% and 72% of the variance in chlorophyll (Chi), respectively. However, when these models are compared to similar models for lakes, the yield of Chl per unit nutnent is between 1 and 2 orders of magnitude lower in estuaries than in lakes. As herbivoiy and turbidity are likely contributon to this low yield, we used a mas-balance approach to model phytoplankton biomass as a function of pnmary production and losses due to flushing, sedimentation and herbivory. In the six estuaries with musse1 aquaculture, 45-88% of the Ch1 deficit could be accounted for by herbivory. In the remaining 9 estuaries, turbidity accounted for 35-74% of the Ch1 deficit. Considenng both herbivory and turbidity, the mass-balance accounted for 68% of the Chi deficit for the 15 estuaries on average. We also generated an empirical model predicting the deficit as a hction of herbivory and turbidity; this model accounted for 50% of the variation in the deficit. The analysis suggests that Chknutrient relations can be generalized across fresh and estuarine aquatic systems if turbidity and herbivory are considered. Introduction Coastal eutrophication is recognized as an important ecological problem in many regions (e.g. US Atlantic Coast (Bricker et al. 1995) and the Baltic (Rosenberg et al. 1990)). Nutrient loading to coastal waters resulting in increased phytoplankton biomass has been linked to increased incidence of toxic phytoplankton blooms (Paerl 1995) and increased anoxia both locally (Cooper 1995) and on the continental shelf (Turner and Rabalais 1994). Concem over coastal eutrophication is reflected in the US Estuarine Eutrophication Survey (Bricker et al. 1995), and two primary journals have dedicated issues to the topic (Rosenberg et al. 1990; Vollenweider et al. 1992). Because eutrophication is defined as an aquatic system's response to increased nutrient loading (Edmondson 199 1), identification of the key nutrient controlling coastal phytoplankton biomass is considered essential. Based on the stochiometric work of Redfield (1958). phosphorus (P) has been considered a key limiting nutnent in marine systems. Moreover, P control of phytoplankton biomass in many fieshwater systems and similarities in phytoplankton physiology and nutnent requirements in both coastal and Freshwater systems (Hecky and Kilham 1988) make P control of coastal systems intuitively appealing. Nevertheless, following Rhyther and Dunstan's (1971) influential work, nitrogen (N) is generally seen as the limiting nutrient in coastal systems and has received the bulk of research interest. A review of the 1995-1997 biological abstracts shows that of 596 articles on estuaries and nutrients, 52% consider only nitrogen, 32% refer to both nitrogen and phosphorus, and 16% consider only phosphorus. Despite the preponderance of research on N, the evidence for general N limitation of coastal systems is weak compared to the evidence for general P limitation of Freshwater systems. In their comparative review of nutrient limitation in aquatic systems, Hecky and Kilham (1988) argued that although N may well be the key limiting nutrient in coastal waters, the evidence is inconclusive as it is mostly derived fkom bioassays and observations of inorganic nutrient concentrations. Since this review, a series of mesocosm experiments on N vs. P limitation have been conducted (e.g. Oviatt et al. 1995, Tamminen 1995). However, whole system experiments such as those of Schindler ( 1977) that provided such convincing evidence of P limitation in lakes are absent in coastal areas. Coastal and marine ecologists have also generally eschewed the comparative approach of limnologists (e.g. Dillon and Rigler 1974), which has been instrumental in establishing the generality of P-limitation in lakes. Although limited exarnples exist of comparative work in the coastal literature (c.f. Nixon 1981, Monbet 1992, Boynton et al. 1W6), the relative strength of patterns between phytoplankton biomass and total N (TN) and total P (TP) in estuaries has not been tested. To evaluate patterns between phytoplankton biomass and nutrients, the following hypotheses were tested: 1 ) estuarine phytoplankton biomass (measured as chlomphyIl-a (Chl)) is primarily a function of bottom-up nutrient control; 2) Ch1:nuhient relations can be established using a comparative, empincal approach; 3) the relation between Ch1 and TP is stronger than that between Ch1 and TN;and 4) Ch1 responds to total nutrient concentrations similarly in estuaries and lakes. We also examine the role of herbivory and turbidity in modenting the relation between nutients and coastal phytoplankton biomass.

Methods Study Location This analysis is based on data fkom fifieen estuaries in Prince Edward Island (PEI), Canada. PEI is a small island (575,000 ha; Environrnent Canada 1990) located in the Gulf of the St. Lawrence River, approximately 15 km fiom the New Brunswick Coast (Fig. 1). Maximum salinity surrounding PEI is approximately 29%0,reflecting the strong influence of the Gulf of St. Lawrence. The island is heavily bisected by rivers flowing into approximately 25 coastal embayments along a 1600 km coastline (Environrnent Canada 1990). These embayments include coastal plain estuanes (Fairbridge 1980) as well as lagoons that form behind barrier sandban on the Island's north shore. Tides are semi-diurnal with a mean high tide of 0.9 m and mean low tide of 0.2 m on the north shore and a mean high tide of 2.4 m and mean low tide of OSm on the south shore (Dept. of Fisheries and Oceans 1996). Agriculture is an important economic activity on the island: approximately 35% of the land base is agricultural (MacDougall et al. 1988). The soils are generally acid, well drained podzols (MacDougall et al. 1988). These podzols erode easily and soi1 loss can be as hi& as 40-45 tonnes ha-' yeaf' (Hirnmelman and Stewart 1979). Because podzols are low in organic matter (< 3%; MacDougall et al. 1988) and because of the intensity of agricultural activity, synthetic fertilizers are heavily used (1100- 1600 kg ha-' of 13:20:20 N P K fertilizer; Thompson, pers. comm.). Nitrogen leaches into ground water through podzols whence it is transported to estuaries; phosphorus is transported to surface waters adsorbed to particles of the easily eroded podzols. Data Set and Analyses A data set was compiled for fifieen estuaries in PEI (Fig. 1) that includes water chemistry, shellfish biomass, coastal morphometry, and land-use variables (Table 1). The estuaries were chosen to include a range of size, land-use charactenstics and both the presence and absence of musse1 aquaculture. Water Chemise The fifieen estuaries were sampled from May to August, 1996. Al1 estuaries larger than 4 km' were sampled at five stations along a land-sea salinity gradient; the remaining four estuaries (Darnley Basin, North Lake, Savage Harbour and Wilmot; Table 1) were sampled at three stations. At each station, the location, time of day, muimurn depth (Zrnx) and Secchi depth (SD) were recorded. Salinity and temperature were measured 0.5 m below the surface and 0.5 m above the bottom. At stations where the maximum depth was less than 1 m, single salinity and temperature readings were taken at 0.5 m depth. Integrated water sarnples were taken at each station fiom the surface to 5 rn; if the depth was less than 5 m, integrated samples were taken from the surface to 0.5 m above the bottorn. Each estuary was sarnpled six times during the sampling season at approximately two week intervals generating 18 (3 stations per estuary) or 30 (5 stations pet estuary) observations per estuary. Tnplicate samples were analyzed for Chl, TP and TN. For Chl, 0.5 1 water was filtered ont0 Gelman AIE which were then fiozen. Within six months of sampling, the filten were extracted in 90% ethanol and Chl was detemined spectrophotomeûically following Bergmann and Peters (1980). TP was also determined spectrophotometrically following the persulfate digestion method of Menzel and Convin (1965). The method was modified in that, after digestion, a 2 ml subsample was taken from each replicate. These subsamples were then frozen to be read on an Alpkem autoanalyzer (Alpkem 1992) within six months of sampling. TN, like TP, was deterxnined on the autoanalyzer using two ml subsamples from sarnples that were digested following Solorzano and Sharp (1980). Shellfish biomass The estimated production (total weight in kg yf') for the six estuaries in which blue mussels (Mvtilus edulis) are hedwas provided by the PEI Department of Agiculture, Fisheries and Forestry (Table 1). Musse1 spat are placed in grow-out bags when approxirnately three months old and 20-30 mm long; they are harvested in the 50- 60 mm size range. A total of 13 km' are under production in PEI (Table 1; Department of Fisheries and Oceans 1997). Musse1 filtration rates are frequently reported as clearance rates (la hi1)as a function of dry weight (Table 2). We converted a11 clearance rates to specific filtration rates (f,;m3* day"@ g~") using conversion factors calculated from length. total weight, and dry weight measurements for 50 PEI mussels (Table 2).

Coastai rnorphometrv and Land-Use Coastal morphornetry was detennined from bathymeaic charts published by the Canadian Hydrographie Service (1980). Land-use information was provided by the PEI Department of Agriculture. For each watershed, land-use is divided into area under forest, area under potato crops and "other" which includes area under hay, grain, and vegetables. Statistical Analyses Chl: nutrient relations were developed using least squares regression techniques (SAS Institute 1985, Zar 1984). Mean values for Chl, TP and TN were calculated by averaging the values for the stations for each of the six sampling rounds. Means and standard deviations were then calculated fiom these six values to yield growing season averages. The growing season averages were log-transformed to stabilize variance.

Results Chl: nutrient relations The Ch1:TP and Ch1:TN relations are both highly significant, accounting for 65% and 72% of the variation in Ch1 respectively (Fig. 2). The estuarine Chl: nutrient relations were then compared to lake Chl: nutrient relations. For the TP cornparison, we chose the Dillon and Rigler (1974) equation for the strength of its correlation and the OECD (1982) equation as it was developed for lakes fiom a wide geographical range (Table 3). For the TN comparison, we used the only two lake Ch1:TN rnodels we could find: Sakamoto (1966) and Prairie et al. (1989) (Table 3). When the PEI Ch1:nutrient relations are compared to the lake relations, we see that although the dopes are similar, the intercepts of the estuarine models are approxirnately one to two orders of magnitude lower than those of the lake models (Table 3). Estuarine phytoplankton yield (Ch1:TP and Ch1:TN) is much lower in PEI estuaries than in lakes. Assuming that estuarine phytoplankton can respond to nutrients similarly to lake phytoplankton, we calculated a potential Ch1 estimate for each estuary using the observed TP values in the Dillon and Rigler equation (Fig. 3). The difference between these potential Ch1 values and the observed Ch1 values cmbe considered as a "phytoplankton deficit". As herbivory and turbidity are two possible contributors to this low yield, we then used a mas-balance approach to estimate how much of the phytoplankton deticit could be accounted for by these two factors. Mass-balan cing Phytoplankton Bioniass The equation used to mode1 phytoplankton growth is: dB/dt = PP - B kL [eq* 11 where B is phytoplankton biomass (mgCe m"), PP is primas, production (mgC rn-' day") and kL is the specific loss rate (day"). At steady-state, dB/dt is zero and equation 1 can be rewritten as: PP=B kL [es- 21 To detemine whether we could assume that phytoplankton biomass is at steady- state, we plotted biornass as a function of time over the six sampling rounds for each estuary and then looked for trends in the data. In almost all the estuaries, Ch1 fluctuated without trend around the mean (Fig. 4). We thus assumed that biomass is at steady state and that we can calculate a steady state mass-balance for phytoplankton. in our phytoplankton mass-balance, two main losses in al1 15 estuaries are losses via flushing and sedirnentation. Thus equation 2 cm be rewritten as: B = PP (kt+ kJ' [es- 31 where kr,and k, are the loss coefficients (day-') for flushing and sedimentation. The loss coefficient due to tlushing, kfis calculated using the salt water fraction method (Bowden 1980): kf= Q V-I [(SI - Sm) s~"]" [es*41 where Q is the fieshwater load (m3m day-'), V is the eshiary volume (m'), Sm is the mean salinity in the estuary (%O)and S, is the salinity (YM)of open water around the island. For each estuary, the fieshwater load was calculated as the product of the average daily rainfall during the period sampled (Raymond pers. comm.) and the area of the watenhed. The loss coefficient due to sedimentation, k, was calculated fiom the specific settling rate (O. lm*day"; O'Connor, 1981) divided by the mean depth of the estuary. To estimate the PP in the estuaries, we collected data for five estuanes along the US east coast for PP (Boynton et al. 1982) and Ch1 and TP (US EPA 1997). We calculated the means and coefficients of variation for these three variables: while Ch1 varied 6 fold, PP varied by a factor of 2 (Table 4). This is consistent with results fkom Oviatt et al. (1986) demonstrating that experimental nutrient additions resulting in a 32 fold increase in nutrients produced only a 3.5 fold increase in PP. Given the small range of Chl in the PEI estuaries, it is unlikely that PP varies greatly across the systems. We chose a value of 300 gC m'2 year" as this approximates the average value of the estuaries represented (Table 4). It also corresponds to mesotrophic status (Nixon 1995) which is consistent with Our TP values. To test whether the mass-balance calculation provides a reasonable estimate of the biomass, B. observed in the estuary, we calculated B for al1 the estuaries including only sedimentation and flushing as losses. These mass-balance estimates of B (BMB)should be comparable to the B estimates from the Dillon and Rigler (1974) equation (BDR;we converted Ch1 to C using a C:Chl ratio of 50; Nixon et al. 1986). We ploned the Dillon and Rigler equation and its 95% confidence bands for a predicted value (Zar 1984). The values ofBMeal1 fa11 within the prediction bands suggesting that the values of BMeare no less precise than the BDRestimates (Fig. 5). We thus felt that the phytoplankton mass- balance and the chosen parameten adequately estimate B for these estuaries. Herbivory and the Mass-balance To estimate how much of the phytoplankton deficit was attributable io musse1 herbivory, the impact of musse1 aquaculture on phytoplankton biomass was estimated for the six estuaries with extensive mussel fms(Table 1). The loss coefficient due to herbivory, kh was added to the loss term in equation 4. It is calculated as the fraction of the estuary filtered each day by the mussels: kh = VF V-' [eq- 51 where VF is the volume filtered by the mussels per day and is equal to the product of the mussel biomass (gD) and the specific filtration rate, f, (Table 2). Filtration rates reported in the literature Vary over an order of magnitude (0.03-0.4 m3 day-' depending on the size of the animals used, temperature and the expenmental design. We chose a value in the middle of the range of 0.108 m3 day-' gD'' (Vahl 1973). To determine the phytoplankton deficit attributable to musse1 herbivory, we recalculated B including the herbivory loss coefficient, kh,in the mass-balance. The difference between the phytoplankton biomass under herbivory, BH, and that calculated in the absence of herbivory (BMe)is the phytoplankton deficit attributable to herbivory. herbivory accounts for between 45 and 88% of the phytoplankton deficit (Table 5). Tu rbidiîy The estimates of PP used to calculate BMBand BHabove were volumetric and assume that PP occun throughout the water column. Such an assumption overestimates PP and turbidity will further decrease the depth to which PP occurs. Secchi depth can be used to estimate the depth of the euphotic zone (ZE,Cole 1994). and convention frequently uses a &:Secchi ratio of 2 (Dillon and Rigler 1974). However, turbidity and colour can reduce PAR by 66-99% within a meter of water (James and Birge 1938 in Wetzel 1982). Thus for turbid estuaries where Secchi depth was less than 2 m (the Dunk, North Lake, Percival and the Wilmot), we used a ratio of &:Secchi of 1; for the remaining estuaries which are less turbid we used the conventional ratio of 2. The mass-balance equation thus becomes: B = PPE (kr+ k, + kt,)-' [es*71 where PPE is the amount of PP occumng in the euphotic zone: PPE= PP (ZE zh(-') Les*8l and ZM is the mean depth of the estuary. Recalculating B to include losses due to herbivory and the effect of twbidity, BT,we hdthat turbidity accounts for 8-35% of the phytoplankton deficit in the estuaries with mussels and 3574% of the deficit in the other estuaries (Table 6). The corn bined effects of herbivory and turbidity in erplaining the Phytoplankton deficit The combined effects of herbivory and turbidity account for between 35 and 96% of the phytoplankton deficit (Fig. 6). nie average remaining phytoplankton deficit is 32%. The mas-balance is most effective in accounting for the phytoplankton deficit in estuaries with musse1 aquaculture (78-96% of the deficit). In the remaining estuaries, the mass-balance accounts for 35-74% of the phytoplankton deficit and on average 46% of the deficit remains. The highest remaining deficit (65%) is in North Lake, a very small, shallow estuary that is remarkable clear for its depth, thus the mass-balance had linle effect. An empirical alternative to the mass-balan ce We have argued that the Ch1 deficit is a function of the loss rates (flushing, sedimentation and herbivory) and decreased primary productivity due to turbidity. Thus, it should be possible to predici the deficit empirically as a function of these loss rates and turbidity. We calculated the deficit both arithmetically: DefA= ChlE - Chlo [es*91 and geometrically: 1Ol~gChlE - logch10 DefG = [eq* 101 The arithmetic deficit, Def*, indicates the absolute difference between expected Ch1 (CME)and observed Ch1 (Chio) while the geornetric deficit, DefG, indicates the relative difference in terms of factors. We combined the effects of the loss rates and turbidity as follows: Total Losses = k~o~* (ZE ZM-') [esw1 11 where kTo~is the sum of kf, k, and kh and ZE &-' is the ratio of the euphotic depth to the mean depth as in equation 8. In other words, if the euphotic depth is half of the mean depth of the estuary, the losses are effectively doubled. We estimated a linear model predicting these deficits as a function of the total loss and were able to fit the model: DefA= 55.7 + 16.0 log (Total Losses) [eq* 121 This mode1 accounted for 50% of the variation in DefA(Fig. 7; n = 15, pc0.003). We also fit other linear and nonlinear models, using Secchi depth and other combinations of kror and turbidity. None of these models had stable residuals.

Discussion The strong relations between Ch1 and total nutrients (Fig. 2) support our initial hypotheses that estuarine phytoplankton biornass is tightly correlated to bottom-up nutrient factors and that the relations cmbe captured using a comparative empirical approach. The applicability of the comparative approach to estuarine systems is encouraging as this approach has provided simple ecological models that have contnbuted to effective lake management (e.g. OECD 1982). In contrast, there have been few ecological models available to support coastal management efforts and those that exist are frequently estuary-specific, information-intensive and expensive (cf. Linker et al. 1993, Bricker and Stevenson 1996). These results also demonstrate that this empirical approach can successfully identify simple pattems across estuaries. Such cross-system relations require that each system, whether a lake or an estuary, cm be represented by "characteristic" values. This requirement has meant that researchers implicitly assume that the systems are at steady- state and are relatively homogenous to minimize sampling effort. Estuaries have complex hydraulic regimes, are considered spatially heterogeneous and generally have shorter water residence times than lakes due to their relatively open exposure to the sea (Bowden 1980). These characteristics could constrain the use of empirical approaches. However, with a slightly more intense sampling regime than that typically used in lakes, representative values for estuaries can be estimated (Nayar and Meeuwig in prep). The strength of the relations also appean insensitive to residence time, which ranged between 3 and 356 dgs. Provided the residence tirne exceeds the specific growth rate of phytoplankton, patterns between Ch1 and nutrients are possible. This result is consistent with those of Basu and Pick (1996) who demonstrated a strong relation between Ch1 and TP (34.76) in nvers with residence times ranging between 3 and 19 days. That the relation between Ch1 and TN is marginally stronger than that between Ch1 and TP (Fig. 2) suggests that TN rather than TP limits estuarine Ch1 and thus our third hypothesis should be rejected. This interpretation reflects the assurnption that the relative strength of patterns indicates the relative importance of their respective independent variables to the dependent variable (Smith 1979). The average TNTP ratio of 4.5 also supports the argument for RI as the key limiting nutrient in these estuaries. Despite the general applicability of a limnological approach to estuaries, the relations between Ch1 and nutrients differ in lakes and estuaries. While the slopes of the Ch1:TP and Ch1:TN relations are similar in lakes and estuaries. the intercepts of the estuarine nutnent models are approximately one to two orders of magnitude lower than those of lake nutnent models. Thus, the yield of Ch1 per unit TP or TN is much lower in these estuaries. This result was unexpected as a Chl:TP relation developed for estuaries in North and South Carolina (unpubl. data, US EPA 1997) is indistinguishable fiom lake relations (Fig. 8), suggesting that estuarine phytoplankton can respond to nutrients similarly to lake phytoplankton. Other researchen have also noted a lower phytoplankton yield in costal waters (cl. Contreras and Kerekes 1993; Boynton et al. 1996) on the order of a 2 - 8 fold discrepancy. A discrepancy on the level of hvo orders of magnitude was unexpected as phytoplankton element requirements are fairly similar in lakes and estuaries (Hecky and Kilham 1988). The result cannot be amibuted to estuarine flushing as this would result in a general dilution w ith low concentrations of both phytoplankton biomass and nutrients rather than a low yield. Herbivory as a top-down control of phytoplankton biomass or light limitation as a fùnction of turbidity are the two most likely explanations. Since Carpenter et al. (1985) defined iheir trophic cascade model, interest in top- down control of phytoplankton biomass has increased. Mazumder (1 994) demonstrated that the Ch1:TP relation is weaker in lakes where large filtenng Daphnia are present than in Mes lacking Da~hnia,and that the yield of Ch1 per unit TP is lower in lakes with Da~hniaby approximately a factor of 3. Quiros (1990) also demonstrated that herbivory by macrozooplankton in Argentinean lakes and resewoirs strongly decreased the intercept of his Ch1:TP relation by almost an order of magnitude (-1 -9 to -2.6). Mellina et al (1995) showed similar results for zebra mussels, Dreissena ~oivmomha,in Lake Erie, Lake St. Clair and experimental aquaria. Our mas-balance calculation of the impact of musse1 farms on the standing stock of algae suggests that phytoplankton biomass in the six estuaries with musse1 fmshas been reduced by 45-88% (Table 5). This level is sirnilar to that found in an enclosure study in which M. edulis reduced phytoplankton biomass by 54 to 90% of controls (Riemann et al. 1988). It is likely that suspension feeden are also exerting pressure on the phytoplankton biomass in the other nine estuaries as al1 support natural clam and oyster (S.vireinica) populations. For instance. the Dunk and Wiimot estuaries in particular, provide 60-70% of the 1.6 10' kg annual oyster harvest. Estimating total biomass fiorn Sephton and Bryan (1989), and using a filtration rate of

0.002 m'*day-'* ~WETWEI~HT *' (Bacher et al. 1995), an additional 29% of the phytoplankton deficit in the Dunk can be explained. The turbidity analysis suggests that a small euphotic zone can sufficiently decrease PP to account for 8 to 74% of the phytoplankton deficit (Table 6). Light limitation as a function of turbidity appears feasible, particularly in the deeper estuaries such as the Grand and Mill Rivers. These estuaries show no evidence of stratifying with respect to salinity and temperature thus it is likely that the phytoplankton spend time below the euphotic zone. Generally, herbivory and turbidity account for the deficit between observed phytoplankton biomass and that expected fiom lake models, with an average deficit of 32% remaining. This is as accurate as we cm expect because PP itself has a coefficient of variation of approximately 30% and we cannot estimate B with greater accuracy than PP. In cornparison, the empirical model lefi 50.4% of the deficit unexplained. These two approaches to explaining the deficit complement each other: the mas-balance model allows a more refined breakdown of the loss rates but requires an estimate of PP; the empirical model is a black-box in tems of loss rates but does not require an estirnate of PP. Because both approaches demonstrate the importance of herbivory and turbidity, together they make a strong case for the deficit as a fùnction of these two factors. The effectiveness of the mass-balance model suggests that once herbivory and turbidity are included in the analysis, lakes and estuaries show similar patterns between phytoplankton biomass and nutrients. Although the approach is particularly effective in the estuaries where herbivory is a dominant factor, it is less effective in the estuaries in which no musse1 aquaculture occurs, decreasing the deficit by only 54% on average. It is particularly ineffective in very small estuaries (North Lake) and in the very shallow, turbid estuaries (Dunk, Percival, Wilmot). In these shallow, turbid estuaries, a decrease in PP due to light limitation seems less feasible than in the deeper estuaries because these shallow estuaries support a large biomass of the benthic algae (Ulva lactuca) despite the very high turbidity. If light were limiting, it is unlikely that U. lactuca would be thriving to this extent. Thus although we have argued that turbidity indicates light limitation, it may be a surrogate for another variable controlling phytoplankton biomass. We would speculate that this variable is iron. Iron is most usually thought of as a limiting nutrient in certain areas of the ocean (Martin and Fitzwater 1988). However. in areas of excess iron, iron scavenging of phosphoms may make phosphoms unavailable to phytoplankton and thus limit phytoplankton biornass (Froelich 1988). Natural clays with iron hydroxides adsorb phosphoms under acid conditions such as those found in the iron-rich soils of PEI (MacDougall et al. 1988). Soi1 particles with P adsorbed to them would then be transported into the estuaries where this P may remain relatively unavailable to the phytoplankton. Thus, TP values which do not discriminate between available and unavailable P may result in a low yield in systems where an unusually hi& proportion of P is unavailable to the algae. As the correlation between particulate TP and particulate iron is strong (r = 0.95; Eyre 1994), the proportion of unavailable P should be constant and a function of iron and thus of turbidity. Control of P availability by iron was considered by Schindler (198 l), who suggested that P limitation in lakes might in fact be a function of high iron concentrations recycled from the sediments as a result of the reducing conditions of the hypolimnion. in coastal systems, Krom et al (1991) also invoked adsorption of P by iron resulting in P limitation of the eastem Mediterranean which receives high levels of iron-laden dust hm the Sahara. iron control of P may thus partly account for the low Ch1 yields in turbid estuaries where the mass-balance analysis accounted for only a small proportion of the phytoplankton deficit. Lon control of P may also account for the tight correlation between Ch1 and nutrients. In the absence of nutrient limitation, the pattern between Ch1 and nutrients should only occur if the variable that limits phytoplankton biomass covaries with nutrients. It is likely that iron covaries with nutrients in the PEI estuaries: iron is correlated with turbidity as it enters the estuaries on soi1 particles and turbidity is correlated with nutrients (Secchi: TN r = - 0.68; Secchi: TP r = -0.57) . If iron controls the arnount of P available to phytoplankton, an excess of iron would result in a reduced phytoplankton yield and lead to effective P limitation irrespective of the TN:TP ratio. Indirect P limitation via excess iron would contradict the earlier conclusion that these estuaries are N limited. However, the evidence for N limitation is not that compelling. First, the coefficient of determination for the Ch1:TN relation (? = 0.72) is only rnarginally greater than that for the ChI:TP (? = 0.65) and it is unclear whether such a small difference is sufficient to indicate the relative importance of N and P in limiting Chl. Moreover, because TN and TP covary (t=0.82), Chl will be strongly correlated with both nutrients. Second, the low TN:TP ratios (3.1-8.7) that suggest N limitation may be a function of the loading ratio of N and P to the estuary rather than a fùnction of phytop lankton uptake. Given the relatively short residence times of the estuaries, i t is unlikel y that ambient nutrient ratios are determined by phytoplankton uptake (Smith 1984). Such indirect control of phytoplankton biomass has been suggested by other researchers as well; Smith and Hollibaugh (1989) argue that carbon control of net heterotrophic systems is masked by an apparent N-limitation. Ln conclusion, strong pattems can be identified between Ch1 and total nutrients in estuaries. However, the relative strength of these pattems and the TN:TP ratio cannot be used to infer which nutrient is limiting phytoplankton biomass. Thus, although such models are useful toois for costal managers to predict phytoplankton biomass fiom nutrient concentrations, they cannot be used to support decisions with respect to single nutrient reduction strategies. The low yield of Ch1 per unit nuûient points to the importance of other factors such as herbivory and turbidity, and potentially to indirect control by iron, in determining phytoplankton biomass. The comparison of these estuaries to lakes, and the use of a mass-balance and empiricai mode1 to account for the deficit suggests that once the effects of herbivory and turbidity are accounted for, phytoplankton response to nutrients is similar in lakes and estuaries. Acknowledgemeots: We thank M. Trudel for his generous advice on mass-balance models and an anonymous reviewer for suggesting the empincal mode1 . Field and laboratory assistance was provided by E. Sunderland, D. Chamberland and T. Nayar. In PEI, B. Penak (Bedeque Bay Environmental Management Association), R. Angus (Dept. of Fishenes and Oceans), and B. Raymond (PEI Dept. Fishenes and Environment) provided essential logistic support. This work was supported by the National Science and Engineering Research Council of Canada and is a contribution frorn the McGill University Lirnnology Research Group. Literature Cited Alpkem (1 992) User's manual. Alpkem, Wilsonville, Oregon. Bacher C, Bioteau H, Chapelle A (1995) Modelling the impact of cultivated oyster populations on the nitrogen dynamics: the Thau Lagoon Case (France). Ophelia 42:29-54 Basu B, Pick F (1996) Factors regulating phytoplankton and zooplankton biomass in temperate rivers. Limnol. & Oceanogr. 4 l(7): 1572-1577 Bergrnann M, Peters RH (1980) A simple reflectance method for the measurernent of particulate pigment in lake water and its application to phosphorus-chlorophyll- seston relations. Can. J. Fish. Aquat. Sci. 37: 1 1 1- 1 14 Bowden K.F (1980) Physical factors: salinity, temperature, circulation, and mixing processes. In: Olausson E. and Cato 1. (eds.) Chemistry and biogeochemistry of estuaries. John Wiley and Sons, Chichester, p.37-70 Boynton WR, Kemp WM, Keefe CW (1 982) A comparative analysis of nutrients and other factors influencing estuarine productivity. In: Kennedy VS (ed). Estuarine Comparisons. Academic Press, NY, p. 60-90. Boynton WR, Murray L, Hagy JD, Stokes C, Kemp WM (1996) A comparative anaiysis of eutrophication patterns in a temperate coastal lagoon. Estuaries 19:408-42 1. Bricker SB, Clement CG, Frew S, Hannon MR, Pirhalla DE (1995) NOAA's Estuarine Eutrophication Survey Volume 1 : Mid-Atlantic Region. Office of Ocean Resources Conservation and Assessrnent National Ocean Service National Oceanic and Atmospheric Administration Maryland Silver Spring. Bricker SB, Stevenson, JC (1996) Nutnents in coastal waters: a chronology and synopsis of research. Estuaries 1g(2B): 33 7-34 1. Canadian Hydrographie Service (1980) Bathymetric charts (nos. 4420,442 1, 4422, 4425, 4459, 4467, 449 1, 4492). Minister of Supply and Services, Ottawa Carpenter SR, Kitchell JF, Hodgson JR (1 985) Cascading trophic interactions and lake productivity. BioScience 35 :634-639 Cole GA (1994) Textbook of Limnology 4" edition. Waveland Press, Prospect Heights Illinois Contreras F, Kerekes J, (1993) Total phosphorus - chlorophyll relationships in tropical coastal lagoons in Mexico. Verh. Internat. Verein. Limnol. 15:448-45 1 Cooper SR (1995) Chesapeake Bay watershed historical land use impact on water quality and diatorn communities. Ecological Applications 5(3):703-23 Department of Fisheries and Oceans (1996) Canadian Tide and Current Tables 1996: Volume 2 - Gulf of St. Lawrence. Fisheries and Oceans Communications Directorate, Ottawa Department of Fisheries and Oceans (1 997) Unpublished data. Halifax: Canada Dillon PJ, Rigler FH (1974) The phosphorus - chlorophyll relationship in lakes. Limnol. and Oceanogr 19: 767-773 Edmondson WT (1991) The uses of ecology: Lake Washington and beyond. U. Washington Press, Seattle Environment Canada (1990) Activities affecting surface water resources: a general overview. The Canada-PEI Water Management Agreement. Minister of Supply and Services, Ottawa Eyre B (1994) Nutrient biogeochemistry in the tropical Moresby River estuary system North Queensland Australia. Est.Coast. and Shelf Sci. 39: 15-3 1 Fairbridge RW (1980) The estuary: its definition and geodynarnic cycle. In: Olausson E. and Cato 1. (eds.) (eds.) Chemistry and biogeochemisry of estuaries. John Wiley and Sons, Chichester, p. 1-35 Famme P, Riisghd HU, Jergensen CB (1 986) On direct mesure of pumping rates in the musse1 M. edulis. Mar. Bio. 92323-7 Froelich PN (1 988) Kinetic control of dissolved phosphate in natural nvers and estuaries: a primer on the phosphate buffer mechanism. Limnol. Oceanogr. 33:649-668 Hecky RE, Kiham P (1 988) Nutrient limitation of phytoplankton in freshwater and marine environrnents: a review of recent evidence on the effects of enrichment Limnol. Oceanogr. 33 :796-822 Himmelmnn D, Stewart NE (1 979) Soil erosion studies in PEI. Soil erosion workshop. Truro, Nova Scotia: Nova Scotia Agricultural College. 9 pp. James HR, Birge EA (1938). A laboratory study of the adsorption of light by lake waters. Trans. Wis. Acad. Sci. Arts Lett. 3 1: 1-154. In Wetzel RG (1982) Limnology. 2"d edition. Saunders College Publishing. Philadelphia Jergensen CB (1 966). Biology of suspension feeders. Pergamon, Oxford Krom MD, Kress N, Brenner S, Gordon LI (1991). Phosphorus limitation of primary productivity in the eastem Mediterranean Sea. Limnol. Oceanogr. 36: 424-432 Linker LC, Stigall GE, Chang CH, Donigian AS (1993) The Chesapeake Bay watershed rnodel. CSC.MD 1: 3-9 MacDougall JI, Veer C, Wilson F (1988) Soils of Price Edward Island: Prince Edward Island Soi1 SweyNO. 83-54.0ttawa: Ministry of Supply and Services Martin, IH, Fitzwater SE (1 988) Iron deficiency bitsphytoplankton growth in the north- east Pacific subarctic. Nature 33 1:Ml -343 Mazumder A (1994). Patterns of algal biomass in dominant odd- vs. even-link lake ecosysterns. Ecology 734): 1 141-49 Mellina E, Rasmussen SB, Mills EL (1995) Impact of zebra mussels (Dreissena polyrnorpha) on phosphoms cycling and chlorophyll in Mes. Can. J. Fish. Aquat. Sci. 52:2553-2579 Menzel DW, Corwin N (1 965) The measurement of total phosphorus in seawater based on the liberation of organically bound fractions by persulfate oxidation. Limnol. Oceanogr. 1O:280-82 Mshlenberg F, Riisghd HU (1979) Filtration rates using a new indirect technique for 13 species of suspension feeding bivalves. Mar. Biol. 54: 143-147 Monbet Y (1992) Controls of phytoplankton biomass in estuaries - a comparative analysis of microtidal and macrotidal esniaries. Estuaries l5:563-57 1. Nayar T, Meeuwig JI. Characterising estuaries: a mean-variance approach to designing estuarine sampling programs. In prep. Nixon SW (1981) Remineralization and nutrient cycling in coastal marine ecosystems. In: Neilson BJ, Cronin LE (eds.) Estuaries and nutrients. Humana Press, Clifton, NJ p 11 1-138 Nixon SW (cd.) (1990) Marine eutrophication: a growing international problem. Ambio 19. Nixon SW (1995) Coastal marine eutrophication: a definition social causes and future concems. Ophelia 4 1: 199-2 19 Nixon SW, Oviatt CA, Frithsen J, Sullivan B (1986) Nutrients and the productivity of estuarine and coastal marine ecosystems. I. Lirnnol. Soc. South. Agr. l2(1/2):43-7 1 O'Connor DI (1 98 1) Modelling of eutrophication in estuaries. in Neilson B.J. and Cronin L.E. (eds). Estuaries and nuüients. Humana Clifion NJ. 183-224 OECD (1982) Eutrophication of waters: monitoring assessrnent and control. Organisation for Economic Cooperation and Development. Paris Oviatt C, Doenng P, Nowicki B, Reed L, Cole J, Fnthsen J (1995) An ecosystem level experiment on nutrient limitation in temperate coastal marine environrnents. Mar. Ecol. Prog. Ser. 116: 17 1 - 1 79. Oviatt C, Keller AA. Sampar PA, Beatty LL (1986) Patterns of productivity during eutrophication: a mesocosm expenment. Mar. Ecol. Prog. Ser. 28:69-80. Paerl H (1995)Coastal eutrophication in relation to atmospheric nitrogen deposition: current perspectives. Ophelia 41 :237-59 Prairie YT, Duarte CM, .Kalff J (1989) Unifjmg nutrient-chlorophyll relationships in lakes. Cm. J.Fish. Aqua. Sci. 46: 1 176-82 Prince Edward Island Department of Agriculture Fishenes and Forestry (1 997) unpublished data. Quiros R (1990) Factors related to variance of residuals in chlorophyll - total phosphorus regession in lakes and reservoirs of Argentina. Hydrobiologia 200/201:343-355 Raymond B. Pnce Edward Island Department of Fisheries and Environment. Personal Communication. Redfield AC (1958) The biological control of chemical factors in the environment. Am. Sci. 46:205-222 Rhyther JH, Dunstan WM (1971) Nitrogen phosphorus and eutrophication in the coastal marine environment. Science 17 1: 1008- 13 Riemann B, Nielsen TG, Horsted SJ, Bjmsen PK, Pock-Steen 1(1988) Regulation of phytoplankton biomass in estuarine enclosures. Mar. Ecol. Prog. Ser. 48:205-215 Rosenberg R., Elmgren R, Fleischer S, Jonsson P, Persson G, Dahlin H (1990). Marine eutrophication case studies in Sweden. Ambio 19: 102- 108 Sakamoto M (1966). Primary production by phytoplankton community in some Iapanese lakes and its dependence on lake depth. Arch. Hydrobiol. 62: 1-28 SAS Institute Inc (1985) SASISTAT user's guide release 6.03. SAS Institute Inc. Cary N.C Schindler DW (1977) Evolution of phosphorus limitation in lakes. Science 195:260-262. Schindler DW (198 1) Studies of eutrophication in lakes and their relevance to the estuarine environment. p. 71-82 In: Neilson BS. Cronin LE (eds.). Estuaries and nutrients. Clifton New Jersey: Hurnana Press Sephton TW. Bryan CF (1989) Changes in the abundance and distribution of the Amencan oyster population of the Dunk River Public Fishing Area of Bedeque Bay Prince Edward Island. Can. Tech. Rep. Fish. Aquat. Sci. 1677:v1+21 pp Smith SV (1984) Phosphotus venus nitrogen limitation in the marine environment. Limnol. Oceanogr. 29: 1 149- L 160 Smith SV Hollibaugh JT (1989) Carbon-controlled nitrogen cycling in a marine 'macrocosm': an ecosystem-scale mode1 for managing cultural eutrophication. Mar. Ecol. Prog. Ser. 52: 103- 109 Smith VH (1979) Nutrient dependence ofpnmary productivity in Mes. Lirnnol. Oceanogr. 24: 1051 - 1064 Solorzano L, Sharp JH (1 980). Determination of total dissolved nitrogen in natural waters. Limnol. Oceanogr. 25:75 1-754 Tamminen T (1995) Nitrate and ammonium depletion rates and preferences during a Baltic spring bloom. Ma.. Ecol. Prog. Ser. 120(1-3): 123- 133. Thompson B. Prince Edward Island Department of Agriculture and Forestry. Persona1 Communication. Turner RE, Rabalais NN (1994) Coastal eutrophication near the Mississippi river delta. Nature 368 :6 19-62 1 United States Environmental Protection Agency (1997) Unpublished data retrieved from STORET (EPA's Central Data Respository). Vahl O (1973) Pumphg and oxygen consumption rates of M. edulis L. of different sizes. Ophelia 1245-52 Vollenweider RA, Marchetti R, Viviani R. (eds). (1992) Marine coastal eutrophication. Sci. of the Tot. Env. Suppl. Walne, PR (1972) The influence of current speed, body size and water temperature on the filtration rate of 5 species of bivalves. J. mu.biol. Ass. UK 52:345-374 Widdows J, Fieth P, Worrall CM (1979) Relationships between seston available food and feeding activity in the cornmon mussel M. Edulis. Mar. Biol. 50: 195-207 Winter SE (1973) The filtration rate of M. Edulis and its dependence on algal concentration measured by a continuous automatic recording approach. Mar. Biol. 22:3 17-28 Zar JH (1984) Biostatistical analysis. Prentice-Hall. Englewood Cliffs NJ Table I :Data for 15 estuaries froni Prince Edward lsland used in tliis analysis. Musse1 weiglit is iii grains dry weighl.

Site Ch1 Total Total Secchi Mean Volume Residence Salinity Watershed Agriculture Mussels phosphorus nitrogen Dept h Ti rnc Area

Units mgni3 n~gm-~ mgm" ni n~ m'-106 days %O k ni2 km2 -gr- -10' Boughton 1,398 57.9 209 3.5 18.0 140 332 2 1.2 39.5 10.3 7.29 Brudenell Cardigan Darnley Basin Dunk Foxley Grand Murray River Mill River North Lake Percival Rustico St. Pcters Savagc Wilmot 3.475 94.8 416 1.1 2.2 2.3 3 19.2 206.6 65.1 O Mean 2.354 66.7 287 2.4 8.7 98 129 19.6 105.7 38.7 5.80 SD 1.437 15.7 79 1 .O 5 .O 1 06 118 1.5 8 1.3 38.1 2.90 Min 1 .O9 7 45.8 209 1.1 2.2 2.5 3 16.4 25.1 2.4 2 .O3 Max 5.915 94.8 48 1 4.1 18.0 351 356 22.2 292.9 1 40.4 9.63 Table 2: Calculations of Specific Mwsel Filtration Rates (f,; m3 g," day-') where CR = clearance rate in 1 hr-', and g~ = gnms dry weight .

fi Equation Range of a Reference

0.029-0.043 0.84- 125 Vhr for 55 mm animal$ 0.7 Jargensen 1966 O .O63 CR = 2.4 10 go "" 0.003- 1.186 Winter, 19?3 0.108 CR = 3.90 g~ O*' 0.008-1.0 Vahl 1973 O. 121 CR = 3.846 go "" 0.5-4.0 Walne 1972 0 202 CR = 7.15 gD 0.0 1 1- 1.36 1 Mshlenberg & RiisgArd 1979 0.427 Average of data 0.07-0.39 Farnme et al. 1986

:Conversions as bllows: A 5 cm animal Iength weighs approxirnately 0.7 (Widdcws et al., 1979; this study)

Table 3: A cornparison of lake Chknutnent relations from other studies to those for the PEI estuaries where Lake and Eshianne Ch1:nuûient relations. SEE is the mode1 standard error of the estimate: Nr: not recorded.

5Mode1 r SEE ref LogCh1:logTP 1.449 -1.136 77 0.96 0.2 14 Dillon and Rigler 1974 LogCh1:iogTP 0.96 -0,553 77 0.88 0.25 1 OECD 1982 Log Chl: log TP 1.76 -2.890 15 0.65 0.136 Thisstudy

LogCh1:logTN 1.445 -3.131 133 0.69 nr Prairie et al. 1989 Log Chl: log TN 1.40 -2.5 21 IU IU Sakamoto 1966 Log Chl: log TN 1-78 4.06 15 0.72 0.121 This study Table 4: A cornparison of primas, production (PP; Boynton et. al 1982), Ch1 and TP (US EPA unpubl. data) data for estuaries along US eastern Coast.

Site PP PP Ch1 TP

Chincoteague 0.553 202 7.17 68.3 Pamlico 1.256 458 10.50 61.2 Mid-Chesapeake 0.95 1 347 13 .O0 73 .O Neuse 0.938 342 20.16 162.0

standard deviation Coefficient of variation

Table 5: Phytoplankton biornass deficit attributable to rnussels: obsewed B (Bo; mg C ma),

biomass expected from the mass-balance (BhlB)calculated tiom equation 3 where the loss tems are

flushing (kf) and sedimentation (k),biomass expected in the presence of mussels (BH) where

herbivory (h)is included in the loss tm of equation 4, and percent reduction in the phytoplankton

deficit ( 10Oe(BM~ - B,$(Bbi,re- Bo)).

Site Bo BMB BH % reduction Boughton 70 5328 704 88 Brudenel 1 98 1920 1105 45 Cardigan 77 4865 111 78 Murray 65 6417 1936 71 Rustico 99 1424 1424 69 St. Peters 86 1057 1057 67 Table 6: Phytoplankton biomass deficit attributable to turbidity: observed B (B.; mg C m"), biomass expected from the mass-balance (Bue: mg C - m") biomass expected given reduction in primary production as a hinction of turbidity (BT;mg C . m"), and percent reduction in the phytoplankton deiïcit (for estuanes with mussels = (lOO*(BH- BT)/(BLIB- Bo)); for estuaries without

~USS~S= (lOO*(BMB - BT)/(BbIB- Bo)).

ESTüARY Bo BMB BT % reduction Boughton Bmdenell Cardigan Darnley Basin Dunk Foxley Grand Mill Murray River North Lake Percival Rustico St. Peters Savage Harbour Wilmot Figure 1 : Map showing location of Prince Edward Island, Canada and the 15 study estuanes; asterix indicates the provincial capital, Charlottetown. Figure 2: Chl:TP and Chl:TN relations for PEI estuaries

A) log(Ch1) = -2.89 + 1.76 log (TP) r' = 0.65 see = 0.136

1.60 1.65 1.70 1.75 1.80 1.85 1.90 1.95 2.00 log(TP (mg m-'1)

B) log (Chl) = - 1.0 1 + 1.77 log (TN) f = 0.72 see = 0.12 l?

Figure 4: Variation in Ch1 over thefor Boughton Eshiaiy

1 1 k 2 3 4 5 sampling round Figure 5: A cornparison of Ch1 estimates ftom Dillon and Rigler and the mas-balance estimate (filled circles) demonstmting that the mas-balance estimates lail within the 95% confidence bands for predicted values (dashed lines; Zar 1984). Filled circles are the original data used to fit the model.

log (TP (mg m-3)) Figure 6: Proportion of phytoplankton deficit accounted for by herbivory and turbidity

site Figure 7: Empincal mode1 predicting the arithmetic Chi-deficit @et, mg m')) as a function of log total losses (day-');dashed lines are 95% prediction intervals.

Def, = 55.7 + 16.0 (log (Total Losses)) ?=49.6 see=7.44

-2 -1 log (total losses (day-')) Figure 8: Ch.i:TP relations hmDillon and Rigler (1974), OECD (1982) and Carolinian estuaries (US EPA 1997).

OECD

2 Mode1 n SEE r bo D8R 46 0.214 0.96 -1.136 OECD 77 0.251 0.88 4.553 NC&SC 68 0.245 0.50 4.86 PREDICTING COASTALEUTROPHICATION IN THE BALTIC:

A LIhl NOLOGICAL APPROACH

kssica I. Meeuwig, Pirkko Kauppila and Heikki Pitkhen

Canadian Journal of Fishenes and Aquatic Sciences: submitted 08.98

Abstract Coastal eutrophication is s key environmental concem in Finland. The highly indented coastline with myriad small estuaries, and a well settled coastline mean that eutrophication occurs at numerous localities. There is thus a need for general models that predict eutrophication for a number of estuaries. Lake eutrophication has been successfully predicted using a combination of a ChI:TP regression model and phosphorus rnass balance model (OECD 1982). We appiied this limnological approach to 19 Fi~ish estuaries. The Ch1:TP regression was highly significant, accounting for 67% of the variation in Chl. When combined with a TP mas-balance equation, log observed and predicted Ch1 were within 28% of each other on average. Accuracy was irnproved by dividing the estuaries into those dominated by nonpoint source (NPS) loading (n=l 1) and those dominated by point source (PS) loading (n=7). A land-use regression model based on percentage of the catchment forested and estuary mean depth then best predicted Ch1 in the NPS dominated estuaries (*=0.74; MSR = 0.01 1). The mas-balance approach remained the rnost accurate model for the PS estuaries (MSR = 0.107) The land-use model and mass-balance approach are complementary tools in that their use maximizes accuracy for both NPS and PS dominated estuaries. This hi& level of accuracy demonstrates the relevance of typically lirnnological approaches to estuaries. Moreover, the use of a lake-based mass-balance model suggests that lakes and these estuaries respond similarly to phosphorus loading. Introduction Coastal eutrophication has been identified as a global problem from temperate estuaries (Boynton et al. 1996, Rosenberg 1990) to tropical waters (Lapointe and Clark 1992). It manifests as an increase in phytoplankton and macroalgal biomass, increased incidences of toxic and noxious blooms, hypoxia, anoxia, and fish and benthos kills (Fisher et al. 1995, Vollenweider et al. 1992). The Baltic Sea is perhaps one of the fint coastal systems in which eutrophication was identified (Rosenberg 1990) and it continues to be a key environmental concem (Wallstrom 1991, Nehnng 1992, HELCOM 1997). Finland's coastal waters are particularly sensitive to coastal eutrophication: they are generally shallow and water exchange with the open Baltic is restricted due to the complex coastal morphometry (Bonsdofl et al. 1997). Nutrient loads are denved from agriculture (Rekolainen et al. 1999, municipal and industnal wastes (Pitkiinen 1994) and fish farms (Bonsdorffet al. 1997). Finland has a long and highly indented coastline (39.000 km; Grano and Roto 1986) that includes approximately 50 river-fed estuaries as well as numerous embayments. Human settlement, while concentrated in the south and south-west, occurs along the entire coastline. Such a coastline and seulement pattern mean that eutrophication is not a localized problem; rather it occurs in a number of estuaries from the eastem Gulf of Finland to the NE Bothnian Bay. The Baltic Sea have received much attention in terms of modelling eutrophication (Savchuk 1986, Wulff et al. 1990, Koponen et al. 1992). However, much of Finland's coastline is comprised of myriad smdl estuaries for which eutrophication models have not been developed. Given constraints on human and financial resources, it is also unlikely that site-specific models will be developed for the rnajority of small estuaries. Thus there is a need to develop general models that predict eutrophication for a variety of systems. To be relevant to environmental managers, these models must 1) be accurate, 2) link the eutrophication response with variables that can be managed, and 3) quanti& the error associated with predictions so that decision-risk cmbe evaluated. Such models have been developed to address lake eutrophication. OECD (1 982) used a mass-balance approach to estimate total phosphorus concentrations in lakes as a fùnction of total phosphorus loading (estimated from land-use activities), water residence time and phosphorus sedirnentation. The TP estimate fiom a mas-balance equation is then used in a regression mode1 (e.g. Dillon and Rigler 1974, OECD 1982) that predicts response variables of concem, typically phytoplankton biomass (as Chl). This approach has formed the basis of successful lake eutrophication management both in Europe and in North Arnenca (Edmondson 1991, OECD 1982). Meeuwig and Peten (1996) demonstrated that regression models based on land-use also accurately predict Ch1 and are an alternative to the phosphorus-based mas-balance approach. In Finland. riverborne nutrients have been identified as the major source of excess nutrients both within estuaries and in the open coastal waters (e.g. Pitkiinen 1994). These nutnents are denved primarily from agicultural activities and to a lesser extent from point source waste waters (Rekolainen 1989). To date, strong relationships have been developed to predict total nutrient loads (N and P) as a function of land-use (?=0.80 and 0.73 respectively; Pitkiinen 1994). However, no empincal models analogous have yet been developed to predict phytoplankton biomass as either a function of ambient nutnent concentrations, nutrient Ioads or land-use in Fimish estuaries. Nor has the mass-balance approach, so successful in lakes, been applied to these systems. We have compiled a data set for 19 estuanes and their watersheds to test the hypotheses that the rnass-balance approach accurately predicts Ch1 in Finnish estuaries. To test this hypothesis, we must demonstrate that 1) total nutrients (RIand TP) accurately predict estuary Chl; 2) mass- balance models accurately estirnate TP; and 3) TP estimated from mass-balances accurately predicts estuary Chl. We also tested the hypothesis that land-use, which integrates a number of factors affecting phytoplankton biomass, better predicts Ch1 than TP and the mass-balance approach.

Methods Description of Data Set Data were compiled for 19 estuaries located on the Baltic Coast of Finland fiom the north east Bothnian Bay to the eastem Gulf of Finland, near Russia (Fig. 1). Each estuary is hence refened to by the name of the river ("joki" in Finnish) with which it is associated (Table 1). The data set includes information on: phytoplankton biomass (as chlorophyll a (CM)),water chemistry, coastal morphometry, land-use and total nutrient loads (Table 1). Ch1 and water chemistry data are fiom monitoring surveys between 1989 and 1993 conducted by the Regional Environmental Centea of Finland and coordinated by the Fimish Environment Institute (FEI). Total phosphorus (TP), total nitrogen (TN), and conductivity were measured along vertical profiles while Ch1 was measured from integrated water samples from the surface to 5 m. TP and TN were analyzed from unfiltered samples using Fimish standard methods (Koroleff 1976, 1979). Ch1 was measured afier filtering (Whatman GFK) according to Lorenzen (1967). Conductivity (C,

ms + m" ) was also rneasured and used to estimate salinity (S in ppt) (National Board of Waters 198 1): S = (C - 60.91) 163-' [es- 11 There are altogether approximately 500 sampling stations in the coastal waters monitoring program. Of these 500 stations, we included 176 stations that were located "within" the estuaries. Estuary boundaries are typically difficult to define. We used a geographical approach, drawing the outer limit of the estuary across the nmwest part of the outermost headlands (Peson and Hakanson 1996). Growing season averages were calculated for Chl, TP, TN and salinity using al1 measurements taken ai depths less than 5 m between June and August. nie growing season was defined as June to August because the high latitude of these sites (60" to 65") means that the spnng bloom generally occurs in May (Niemi 1973) and the inclusion of May Ch1 values increased the vuiability in the growing season estimates. Most estuaries included in the study were well sampled (Table 2). Of the 19 estuaries, 12 included data for five years. Spatially, there was an average of 4.2 stations per estuary with five estuaries represented by a single station. These five estuaries were generally the smailer systems (mean surface area = 23 km' vs. 48 km2for al1 estuaries). Although represented by a single station, three of these five estuaries were sampled 3-6 times dunng the growing season. Temporally, the sampling effort in a given area varies depending on the source and amounts of loading and local hydrographical conditions. The sampling frequency ranges in general nom 2 to 20 times per year, being most typically from 4 to 6 times per year. in our data set, the mean nurnber of sarnples per growing season was 3.4 and the mode was 2. Phytoplankton biomass and nutrients are strongly effected by variation in hydrodynamics (Bennett et al. 1996, Mallin et al. 1993, Schaub and Gieskes 1991) that Vary annually. We thus chose an averaging approach that ernphasizes interannual variability. Growing season averages were calculated, using Ch1 as an example, as follows: 1. Chl,, is the mean Ch1 for year-x and station-y, using the values of Ch1 fkom June to August of year x 2. Chl,, is the mean Ch1 for year-x, using the values of Chl,, for al1 the sampling stations located in estuq-z 3. Chl, is the mean Ch1 for estuary-z, using the values of Chl,, for al1 the sampling years for estuary-z This averaging approach gives equal weight to each year even if the year is represented by a single station. By emphasizing yean, we were also able to better match the water chemistry data to the hydrologie flow and nutrient loading data which were calculated on a yearly basis. Estuary surface area and mean depth were calculated from 1:50,000 bathyrnetric charts (Fimish Institute of Navigation, 1996, 1997a, 1997b. 1997c, l998a l998b). Surface area was measured by planimetry. Mean depth was estimated using a grid technique whereby the depth under each square of the grid covenng the estuary was recorded and the average of these values taken. The mean depths estimated using this approach may overestimate the tme mean depths as the available bathymetric charts did not indicate depths for the shaliower fnnging areas at the mouths of the rivers. These areas were however generally small relative to the total surface area. Water residence time, R, (yean), was calculated as: Rt = V (QR+ Q~)-' [eq*21 where V is the estuary volume (calculated as surface area multiplied by mean depth; m3),

QR is the mean annual river inflow (m3 - sec") and Qzis the mean annual inflow of the deep layer (m3 - sec"). QRhas been estimated hmdaily water level recordings using calibrated flow-rating curves. For those nven which are not monitored, indicated by 't' in Table 1, QR was extrapolated either on the basis of the catchent areas and the water flows of the small monitored nvers (or, in a single case (Temmesjoki) estimated on the bais of seasonal mean water flows). QIwas calculated using Knudsen's flow equation as described in Pitkhen (1994): QI= Si . QR (SZ. s~).' [es- 31 where Si and St are the surface and bottom salinity respectively. The use of Knudsen's flow equation requires a difference between surface and bottom salinity thus it was not possible to calculate water residence time for al1 of the Finnish estuaries. In particular, estuaries with low inflow and estuaries in the North where offshore salinity is very low were not arnenable to this calculation. Thus we also calculated residence tirne as if the estuaries are lakes: R[ = v . Q~-I [eq- 41 QRand Qzwere calculated for each year from 1989 to 1993. Mean values for each estuary were then calculated by averaging values for those years for which Ch1 and nutrient data were available. Catchent size and land-use data from the early 1990s were obtained from the data bases of the FEI. If the catchrnent associated with an estuary had a lake area greater than 5%, only the lower, lake-poor subcatchment was used in the calculations (Pitkiinen 1994, indicated by 't' in Table 1). Human population density was estimated using Arc Info. Population data are collected by municipality (Finnish Environment Institute 1998) rather than by watenhed. Thus, watershed population (POPw) was estimated by overlaying municipalities with watersheds and surnming the product of municipal (M) population density, (PdenM)and the area (AWM)of the municipality within the watershed (W) for al1 the municipalities within the watenhed: POPw = UPdenM AwM) [es*51 Watenhed population density was then calculated as POPw divided by total watershed area. TP and TN loads for the catchrnent were calculated f?om the annual river loads (LR)by multiplying the monthly concentration (Cm)by the monthly water flow (Qm)and sumrning the monthly loads:

LR= UQm Cm) [eq- 61 The fkequency of flow proportional sampling was usually 12 times per year, ranging from 4 to 20 times per year. With low sampling fiequencies the estirnates of river fluxes are expected to underestimate the true loads (Walling and Webb 1985). However, this effect is mitigated by using multiple years of data The loads for non-rnonitored river catchments (Table 1, $) were extrapolated on the basis of small coastal rivers within the four main catchent areas of the Fimish coastal waters (Pitkhen 1994). Six of the estuaries also receive point source loading fiom municipalities and industry. These loads, obtained fiom the FEI data bases, were calculated as annual means. As with the river inflows, river and point source nutrient loads were calculated for each year fiom 1989 to 1993. Mean values for each estuary were then calculated by averaging the values for years for which Ch1 and nutrient data exist. The Mass-Balance Approach The mass-balance approach to predicting eutrophication involves two steps. First, TP is estimated fiom a mass-balance equation. At steady state, TP is essentially a function of TP load, and losses via sedimentation and flushing (Vollenweider 1975). The TP load is either measured or estimated fiom land-use via phosphonis export coefficients (Dillon and Rigler 1975). P sedimentation is usually approximated via an empincally denved equation as it is difficult to meaure; losses via flushing are estimated by the water residence time. The second step uses the estimated TP value to predict Ch1 via an empirical regression model such as that of Dillon and Rigler (1 974) or the OECD (1982). There are a number of mass-balance equations in the literature with few guidelines as to their use. They are similar in structure, usually taking the form oE TP = TPLA . (1 - (a(a + p)")) (Z, )" [es*71 where TPLA is the areal TP load (mg-me2y-'), a is the P sedimentation coefficient (yf'), 2, is the mean depth (m) and is the hydraulic flushing rate (yf:). They differ pnncipally in their estirnates of P-sedimentation. in a review of 17 mass-balance equations, Meeuwig and Peten (1996) demonstrated that Canfield and Bachman's (198 1) mass-balance equation estimated TP values that most accuraiely predicted Chl. We thus chose this equation, which takes the form of equation 7 and estimates the P sedimentation

We chose to use an existing mas-balance equation as we lacked information on general patterns in P-sedimentation rates in the Finnish estuaries. The TP values estimated bom this mass-balance were then used in a Ch1:TP regression model developed in this study specific to the Finnish estuaries. Statistical Analyses and Goodness of Fir Criteria Standard least squares regression techniques were used to develop the regression models predicting Ch1 as a fùnction of total nuirients and land-use (SAS Institute 1985, Zar 1996). All variables were log-transformed to stabilize variance (Zar 1996). Land-use variables that were calculated as percent of the watershed were transformed as loglo(X + 1) due to the presence of zeros in some of the land-use categories. Whereas the Ch1:total nutrient relations are univariate, the land-use models include two independent variables. Aquatic systems respond differently to disturbance as a function of their sensitivity , usually determined by morphornetry. This can be conceptualized in terms of a load-sensitivity-effect relationship (Hakanson 199 1). Thus the land-use models include one variable indicating the load or disturbance (eg the amount of agriculture, forested land or human population density), and one variable indicating their sensitivity (e.g. mean depth or water residence time). Together, these two variables predict the effect, or response variable, Chl. A set of preliminary models was identified using exploratory regression techniques (Proc Reg; selection rsquare; SAS Institute 1985); the 'bbest" model was then chosen based on model standard errors and the significance levels of partial regression coefficients. To quantitatively compare the accuncy, precision and bias of Ch1 values predicted from the mass-balance approach and the land-use model, we used critena based on the least-squares goodness of fit critenon: the mean squared residual. Following Meeuwig and Peters (1996). Accuracy was estimated as the mean squared residual (MSR): MSR = X(ICh1, - i~hl,)~- n" [eq. 91 where (lChl, - lChlp) is the difference between the log values of observed and predicted Chl, and n is the number of observations. The variance of the squared residuals (vSR) and the mean error (ME) were used as critena of precision and bias (Meeuwig and Peters 1996):

vSR = x[lChl, - l~hl,)~. MSR] -(n - 1)-' [eq- 101 ME = ~(IChl,- lChl,) n-' [es* 111

The MSR, vSR and ME were calculated for each estuary for Ch1 predicted via the mass- balance equation and Ch1 predicted fiom the land-use model. These individuai estuary values were then averaged to evaluate the overall accuracy, precision and bias of the predictions of the two types of rnodels.

Results The 19 estuaries encompass a wide range of conditions (Table 1). Size varies fiom mean depths of 3 to 18 m and surface areas of2 to 145 km'. Growing season Ch1 ranges ffom 3.9 to 45.9 mgm" and TP and TN range corn 10 to 97.2 mgm4 and 320 to 2133 mgmm3.hua1 phosphorus and nitrogen loads vary across two orders of magnitude frorn 5 to 447 ty" and 130 to 10433 ty8. Salinity ranges from O in Kyronjoki in the oligohaline Bothnian Bay to 6.1 in Paimionjoki in the western most Fi~isharchipelago. Lunar tides are nonexistent but occasional wind-driven water level changes can be as high as 50 cm. The hydrological water balance in the Baltic also affects water levels with fluctuations of ca. 7 cm in the Bothnian Bay and 16 cm in the Gulf of Finland (Ehlin 198 1 ). Coastal morphometry is also highly variable and includes relatively enclosed systems such as Virojoki, the winding, island-rich systems of the Finnish archipelago such as Paimionjoki, and relatively simple pocket estuaries such as Temmesjoki. Land- use is also highly variable: the percentage of the watenhed under agriculture ranges from 9.5 to 42.9% with a mean value of 23.9% and percentage of forest ranges fiom 54.3 to 87.2% (Table 1 ). Predicting Ch1from Total Nutrients Regression models predicting log Ch1 as a fùnction of log TP and log TN were highly significant (Fig. 2a and 2b). The relationship between log Ch1 and log TP was stronger than that based on log TN, explaining 67% of the variance in log Ch1 as compared to 53%. The TP relationship is similar to those generated in lakes: the coefficients are intemediate between those of the Dillon and Rigler (1974) and OECD (1982) equations (Table 4). Moreover, the yields are more similar to those generally seen in lakes (ChüTP = 0.25 in Finnish estuaries vs. 0.45 in lakes Meeuwig. n.d.) than the yields seen in PEI esniaries (ChVTP = 0.034, Meeuwig et al. 1998). The estuarine Ch1:TN relationship is less similar to those in lakes: both the intercept and slope are shallower (Table 3). It is unclear however whether this reflects a true difference between lakes and Finnish estuaries, a difference in range, or the relatively weaker fit of the model. Predirring Ch1/rom the Mass-Balance We first used the mass-balance (equation 7) to estimate TP for sites for which the inflow of bottom sea water could be calculated, using equation 1 to estimate water residence time. The estimated TP values were on average an order of magnitude lower than observed values for these sites (Table 4). This discrepancy suggests that either the TP loads are underestimated, that there are large intemal loads or that the water residence times are underestimated. Because we are confident in the estimates of TP loading and because we have little information on intemal loading, we first re-calculated the mas- balance, estimating water residence time via Bowden's (1980) salt water fraction method and as freshwater replacement time (via equation 3). TP was best estimated by treating the estuaries as lakes (TP-FW; Table 4): the average percentage difference between log vaiues was 9.7%. Calculated as the MSR, the difference between lob observed and predicted TP is 0.038. The only problematic estuary was Perhonjoki in which TP was overestimated by 3 fold. Perhonjoki is a relatively srnall, open system and TP was best predicted by Bowden's salt water fraction method. Given the relatively close agreement between observed and estimated values of TP in general, we predicted Ch1 as a function of TP estimated fkom the mas-balance, using the Finnish Ch1:TP equation (Table 3). The MSR for Ch1 predicted fiorn the mas- balance approach was 0.095 (Table 4) suggesting that the agreement between observed and predicted Ch1 is less than that between observed and predicted TP. Thus, although the mass-balance accurately predicts TP, this accuracy declines when the Ch1:TP mode1 is chained to the mass-balance to predict Chl. This is typical of chained models (Reckhow and Simpson 1980). Predicthg Chl from Land-Use No significant relationships existed between Ch1 and land-use variables for 17 estuaries (2 of the 19 estuaries were missing morphometric data and thus could not be included in the land-use models). However, land-use models pnmarily capture nonpoint source human influences and of the 17 estuaries, 7 had large municipal or industrial point sources irnrnediately on the shore of the estuary that were not included in the riverine nutrient load estimates. We thus temporarily removed these estuaries, hence referred to as point source (PS) estuaries hmthe data set. The remaining 10 estuaries, hence referred to as nonpoint source (NPS) estuaries , then formed the core data set for the land-use regression modeling. A number of significant regression models were then generated. Of these models, the model predicting log Ch1 as a fùnction of log mean depth (1Zm) and log percentage of the catchment forested (For-P) had the highest coefficient of determination and lowest model standard error. Both partial regression coefficients were also significant (p < 0.05). We then added the PS estuaries to the model to see which fit the sarne pattern. Of the 7 PS estuaries. only Kokemaenjoki did not substantially decrease the model coefficient of determination or increase the model standard error when included. Given the small size of the data set available for the regression modeling, we included Kokemaenjoki in the regression model. The best model is thus: lChl = 5.44 - 0.96 Km - 2.09 . [For-P [es- 121 with n = 1 1, p = 0.005, i = 0.74 and MSE = 0.108. One must be cautious in estimating 3 parameten fiom a sarnple size of 1 1. However, both the p values for the partial regression coefficients were significant (p = 0.0014 for Km and p = 0.01 for IFor-P) and the coeffxient of determination sums to more than the individual coefficients of determination (? = 0.44 for 1Zm and = 0.01 for 1For-P). Thus, the model coefficients are likely robust. Comporing the Accuracy of Ch1 Predictions from the Mass-Balance and Lund-Use Models We compared the accuracy, precision and bis of the Ch1 predictions generated via the mass-balance model and the land-use rnodel for al1 the estuaries, the NPS estuaries and the PS estuaries (with Kokemaenjoki included in both groups) (Table 4). When considenng al1 the estuaries, the predictions of Ch1 via the mass-balance model were more accurate, precise and less biased than the predictions fiom the land-use model. Accuracy (MSR)was 0.095 when the mus-balance model was used as opposed to 0.143 with the land-use model. Precision (vSR) was respectively with 0.01 1 as opposed to 0.073 and bis (ME) was 0.023 vs. 0.203. The positive bias value for the land-use model suggests that it tends to underestimate observed Ch1 . The most accurate, precise and unbiased estimates are to be achieved by separating the estuaries into NPS and PS groups (Table 4). The land-use model most accurately and precisely predicts Ch1 for the NPS estuaries with a MSR and vSR of 0.01 1 and 0. Bias is only 0.022. If the mas-balance is used to predict Ch1 in the NPS estuaries, the MSR and vSR increase to 0.079 and 0.0 1 1 and bias increases to -0.1 1. This can only partially be attributed to the poor estimate of TP for Perhonjoki (for which it was difficult to estimate TP) as its rernoval only reduced the MSR to 0.68. The most-accurate, precise and unbiased predictions of Ch1 for the PS estuaries were generated by the mas-balance model. The MSR is 0.107 which is substantially smaller than the MSR of 0.334 for the predictions based on the land-use model. Despite this improvement, the mas-balance does not predict Ch1 in the PS estuaries with a degree of accuracy comparable to the accuracy with which the land-use model predicts NPS estuary Chl. The mass-balance model performed particularly poorly in Virojoki, Porvoonjoki and Vantaa. For reasons explored in the discussion, we recalculated the MSR, vSR and ME for the PS estuaries without these three. For the four remaining estuaries in the PS group, the mass-balance model accurately predicts Ch1 with a MSR of 0.024, vSR of 0.001 and a bias of 0.132 (Table 4). All of these are smaller than the values generated for the sarne estuaries using the land-use model.

Discussion Predicting Coastal Eutrophication in Finnish Estuaries The mass-balance approach can be used to accurately predict coastal eutrophication in Finnish estuaries. The approach requires a Ch1:TP relationship and strong patterns were identified between Ch1 and both TP and TN (Figs. 2a and 2b; Table 3). TP was accurately estimated: predicted and observed values were within 9.7%. When combined, the mass-balance and Ch1:TP regression predict Ch1 with a MSR of 0.0% (Table 4), or a rnean absolute percentage error of 29%. Although Meeuwig and Peters (1 996) demonstrated that regression models predicting Ch1 directly fiom land-use are more accurate than the two-step mass-balance models, we were unable to generate a land-use model for the entire data set of 19 estuaries. However, the land-use rnodel developed for the NPS estuaries predicted Ch1 more accurately than the mas-balance model (Table 4) withh this group. This result is similar to that demonstrated by Meeuwig and Peters (1996) and likely reflects the increase in error as equations are chained together (Reckhow and Simpson 1980). The mass-balance approach remains the best approach to predicting Ch1 in the PS estuaries. This likely reflects the different sources and thus composition of TP in PS and NPS estuaries. In the PS estuaries, there are generally two sources of nutrients: point sources derived fiom municipal and industrial activities and, in one case, fish farms, and nonpoint source P denved primarily from agricultural activities in the catchment. In the NPS estuaries, the majority of TP is denved from agricultural activity (Rekolainen et al. 1995). TP from municipal wastes contains a geater proportion of bioavailable P than TP derived fiom agricultural sources (Ekholm 1994, Pietilainen & Rekolainen 199 1). In addition, the point sources are proximate to the estuaries compared to NPS TP which is derived from the entire catchment. The combination of a greater proportion of bioavailable P that is rapidly deiivered to the estuary may explain why the mas-balance model performs better than the land-use model in the PS estuaries. The land-use model gives equal weight to al1 disturbance, regardless of proximity to the estuary and thus ernphasizes diffuse disturbance. The eflect is also likely exaggerated in these estuaries which have a relatively large catchmeni:surface area ratio (70). This analysis indicates the importance of considering both the nature of sources and their spatial distribution. The PS category remains problematic as the improvement in accuracy generated by dividing the estuaries into PS and NPS categories was not as substantial as that seen with the NPS estuaries. The remaining inaccuracies in the PS group could primarily be attributed to Virojoki, Porvoonjoki and Vantaa. The mean Chl:TP ratios of these three estuaries is almost twice that of the other PS estuaries (0.47 vs. 0.26) and more than twice that of the NPS estuaries (0.11). This suggests that a larger proportion of the TP is bioavailable in these estuaries than in the other PS estuaries. Such a difference may reflect the source and distribution of nutrient loading in these estuaries. Virojoki is unique arnong the estuaries in receiving a large proportion (30%) of its total phosphorus load from fish fanning. The phosphorus denved fiom this fish fami is likely highly available as it is introduced directly into the euphotic zone throughout the entire growing season. Vantaa was for decades a recipient of Helsinki's sewage. The sewage has been diverted since 1987 but intemal loading is thought to be very high. This is consistent with the mass-balance calculation for Vantaa which underestimated TP. There are no available estimates of P sediment fluxes since the diversion however, we used the mass-balance to back-calculate the intemal load, assuming 1) complete agreement between observed and estimated TP and 2) an underestimate of 10%. The intemal load thus calculated is 39- 53% of the total load and amounts to 5 to 9 mg P mm'day-'. This is much lower than Bostrum et al.3 (1988) maximum estimates for shallow enriched waters of 50 mg P day-' but may reflect the high concentrations of P in the water (TP = 91 mg*m4)or a decrease in intemal loading over tirne. Phosphoms that is intemally loaded is inorganic and likely highly available to phytoplankton. Porvoonjoki has no "unusual" sources of nutrient loading however, like Virojoki and Vantaa, a relatively large proportion of the total load is point source (20% vs. a mean of 4.8% in the other PS estuaries). If these charactenstics of Virojoki, Vantaa and Porvoonjoki are accepted as sufficient to remove them from the PS group, the mass-balance predicts PS Ch1 within 2.4%' a level comparable to that of the land-use model applied to the NPS estuaries. It is admittedly dangerous to subdivide one's data until "accurate" predictions are attained. The division between NPS and PS dominated systems among these estuaries does however highlight some limitations of the modeling approaches. The land-use approach is less effective in the presence of point source loads directly on the Coast whereas the mass-balance approach is less effective in integrating disturbance on a watershed scale that may include factors other than TP such as sediment (Rowan and Kalff 1991 ), and TN (McCauley et al. 1989). The division also highlights di fferences arnong the estuaries such as sources of P and their relative contribution to the TP load. The analysis suggests the following guidelines for model choice in predicting Chl: 1. for estuaries dominated by nonpoint source loading (PS Load:Total Load< 0.01), Ch1 is best predicted by the land-use regression model. 2. for estuaries receiving point-source loading between 1 and 15% of the Total Load, Ch1 is best predicted by the mass-balance model. 3. for estuaries receiving point-source loading greater than 15% of the Total Load or with Ch1:TP ratios greater than 0.4, Ch1 is best predicted by the mass-balance model. However, the mass-balance should be used with caution, recognizing that Ch1 will likely be underestimated. These guidelines can also be used to test the relevance of these divisions by predicting Ch1 in estuaries not included in the analysis, using both the land-use regression and the mass-balance. The relative accuracy of the predictions can then be compared along with their correspondence to the above guidelines. To this end, we have made the guidelines quantitative to reduce ambiguity as to which mode1 should perform best. It may be that the above guidelines refiect the quirks of the present set of data rather than any general patterns in the applicability of the models. in this case, the most conservative position would predict Ch1 via the mass-balance. Such predictions should still be within 10% of observed values and thus represent more prediciive power than yet exists for Finnish estuaries.

Coastal Limnology: Finnish Estuaries as Salty Lakes

There are essential differences between these two [fresh and marine] types of systems that prevent us from simply applying knowledge gained through limnological studies to the marine environment .. . the more variable hydrodynarnic properties .. . and the fact that rates of and processes leading to nitrogen and phosphorus sedimentation are generally not well documented for marine systems makes the immediate application of Vollenweider type models diffcult (Richardson and Jergensen 1996:s).

The above quotation aptly captures the accepted dogma that freshwater and coastal systems are fundarnentally different. In addition to differences in inherent variabiliiy referred to above, freshwater and coastal systems di ffer in: water residence tirne, water chemistry, turbidity, grazing, morphometry, physical energy, and limiting nutrients. Whether these differences translate into differential eutrophication response to nutnents and anthropogenic disturbance is an assumption. Moreover, our ignorance of specific rates should noi prevent us examining whether known rates apply. In fact, this is the strength of the mass-balances: they clearly indicate the inappropriateness of the various components when the equation refuses to balance. Our results suggest that a mass-balance approach cm in fact be applied to coastal systems such as the Fimish Baltic estuaries. In considering this mas-balance application to coastal systems, we will evaluate the Ch1:TP mode1 and the mass-balance equation separately as the di Eerences listed above do not always affect both components. Et is perhaps less surpnsing that Chi:TP equations are similar in lakes and estuaries. Phytoplankton have similar elemental composition and requirements in both systems (Hecky and Kilham 1988) thus the Ch1:TP yield should be similar. Water residence time and morphometry should have little affect on Ch1:TP yields as one would not expect differential dilution of these components. For instance, Basu and Pick (1996) developed regression models predicting Ch1 as a function of TP in nvers with residence times as short as 3 days. Their relationship is strong (+=0.76,n=3 1) and shows yields similar to those found in lakes despite the short residence times. Water chemistry could affect the Ch1:TP yield if the bioavailability ofTP changes. For example. in their review of phosphorus in aquatic systems, Howarth et al (1995) describe the process whereby flocculation and prediction of riverborne humic complex iron occurs at low salinity (2-3 ppt). Such flocculation results in the coprecipitation of P. However, such precipitation removes P fiom the water colurnn and thus should have no effect on Ch1:TP yields. Similarly, the release of P when river water meets sea water due to cornpetition for sorption sites increases P but still allows for a corresponding increase in Chl. Of the cited di fferences between fieshwater and coastal systems, turbidity, herbivory and differences in nutrient limitation are most likely to change the Ch1:TP yield. While light limitation reduces the yield of Ch1 as phytoplankton are unable to take advantage of available P. light limitation has been identified as important in a number of estuaries (Fisher et al. 1988). However, light limitation occurs pnmarily in the spring and in permanent turbidity maxima. Our data exclude spnng values and to Our knowledge, turbidity rnavima have not been documented in Finnish estuaries. Herbivory has also been shown to decrease Ch1 yields (Mellina et al. 1995, Meeuwig et al. in press). The zoobenthos of Finnish estuaries includes phytoplanktivores such as Macomo bafthica and Mytilus edulis (Lax et al. 1993). However, although abundance of bivalves has increased in general in the Baltic (Cederwall and Elmgren 1980), inshore areas show declines due to pollution and hypoxia (Mattila 1993, Andersin and Sandler 1991). Prairie et al. (1989) demonstrated systematic changes in the regression coefficients with changing N:P ratios, thought to indicate nutrient limitation. However, although the Baltic is thought to be primarily nitrogen limited (Wulff et al. 1990), the estuaries are considered to be primarily P limited (Pitkiinen and Tamminen 1994, Yurkovskis et al. 1993). Given neither turbidity, herbivory or N limitation seem to be major factors in these estuaries, it is likely that the Finnish estuaries would show similar Chl:TP yields to lakes. It appean easier to argue that the application of the mas-balance is inappropriate. The key tems in the mas-balance are TP Load, water residence time and P sedimentation. Differences in physical energy and morphometry that result in shorter residence times in coastal systems should however be incorporated in the mass-balance mode1 via the water residence time tem. The issue is then whether the residence times we calculated are appropriate. When residence time was modified to include the inflow of seawater from offshore via Knudsen's equation, estimates of TP were an order of magnitude lower than those observed (Table 4). When the estuaries were treated as lakes and residence time calculated only fiom river FW load, estimated log TP values were within 9.7% of observed log TP values on average. This is perhaps not surprising as some of the estuaries have very restricted exchange with the open Baltic due to the complex coastal morphometry and the presence of islands. However, even for the relatively open pocket estuaries such as Temmesjoki, estimated TP was within 3% of observed TP (on a log scale). This suggests that for the purpose of estimating growing season TP, water exchange can be treated similarly to lakes and that the mass-balance is able to accommodate differences in water residence time. This is consistent with Nixon et al. (1996) who were also able to effectively use Freshwater replacement times as a substitute for water residence times in their mass-balance equations for major estuaries of the North Atlantic. The other terni in the mass-balance likely affected by estuary characteristics is P- sedimentation. Morphometry rnay affect P-sedimentation as estuaries tend to be shallower than lakes and shallower systems are more vulnerable to resuspension of P sediments. However, the mass-balances include a depth term so this difference, like that of residence time, should be addressed. Differences in water chemistry are also most likely to affect the sedimentation term: flocculation, as described by Howarth et al. (1995) should increase sedimentation while release of P due to cornpetition for sorption sites should decrease sedimentation. The former should be important in Finnish estuaries as runoff is relatively high in humic substances and iron (Rekolainen pers. comm.). Cornpetition for sorption sites should also be important as even the low salinity waters of these estuaries represent a large increase in ionic concentration. It is however unclear what the outcome of these two opposing processes is in tems of the sedirnentation rate. The slight bias in the mass-balance calculations is positive (0.64%) suggesting that the mass-balance equation slightly underestimates TP. We tested the robustness of our mass-balance choice and P sedimentation term by estimating TP using the other 16 mas-balance equations cornpiled by Meeuwig and Peten (1 996). The ranking of the top 3 mass-balance equations in predicting Ch1 was identical in both the lakes and Finnish estuaries. This result supports the generality of the rnass-balance equation as well as the similûnties between lakes and Fimish estuaries. This generality is also supported by the mas-balance models calculated by Nixon et al. (1996). In the absence of estimates for P md N sedimentation in estuaries, they also borrowed lentic estimates and demonstrated that lakes and estuaries show similar patterns between net transport of TN and TP and water residence times, consistent with the mas- balance calculations.

Summary A combination of land-use modeis and mass-balance models accurately predict Chl in Fimish estuaries. An average degree of accuracy for the land-use models (e.g. MSR=0.014 ) would estimate Ch1 in Perhonjoki as 9.3 mgm" where the observed is 7.5 rngqm". An average degree of accuracy (e.g. MSR= 0.04) for the mass-balance mode1 would predict Ch1 in Karjaanjoki as Ch1 level of 4 mgam" where the observed value is 6 rng~n'~.This degree of accuracy suggests that the mass-balance approach and other limnological models such as the land-use regression effectively predict costal eutrophication in Finnish estuaries. Moreover, this analysis demonstrates that Finnish estuaries and lakes respond similarly to total nutrients and nutrient loads. Finnish estuaries are however not typical estuaries as they are essentially non-tidal and have lower salinity than most estuaries . It thus remains to be demonstrated whether The mass-balance approach cm be effectively applied to the estuaries of North America and Atlantic Europe. Thus the generality of The mass-balance approach, while here extended, should be fiuther tested. Acknowledgements: This study would have been impossible without the hospitality of the Finnish Environment hstitute and the extensive data set that they meticulously collect and maintain. We are grateful for the help of Anu Hakala, Yki Laine, Kati Manni and Mika Ristimaki in compiling the data and thank Petri Eckholrn and Oiva Rekolainen for their insights into phosphorus export. This research was supported by a CM0research fellowship to IM and is a contribution fiom the McGill Lirnnology Research Center. Literature Cited

Andersin, A.B. & Sandler, H. 1991. Macrobenthic fauna and oxygen deficiency in the Gulf of Finland. Memoranda Soc. Farina Ffora Fennica 67: 3-1 0. Basu, B. and Pick, F. 1996 Factors regulating phytoplankton and zooplankton biomass in temperate nvers. Limnol. Oceanogr. 41(7): 1572- 1577 Bennett, J.P; Woodward, J.W.; Shultz, D.J. 1996. Effect of discharge on the Chlorophyll a distribution in the tidally-influenced Potomac River. Estuaries 9(4A):250-260. Bonsdorff, E.; Blomqvist, E.M.; Mattila, J.; Norkko, A. 1997. Coastal Eutrophication: Causes, consequences and perspectives in the Archipelago Areas of the Northem Baltic Sea. Estuarine. Coustal and ShelfScience 44 (Suppl. A): 63-72. Bostrom, B.; Andersen, J.M.; Fleisher, S.; Jansson, M. 1988. Exchange of phosphorus across the sediment-water interface. Hydrobiologia 1 70: 229-244. Bowden K.F. 1980 Physical factors: salinity temperature circulation and mixing processes. In: Olausson E. and Cato 1. (eds.). Chemistty and biogeochemistv of estuaries. Chichester: John Wiley and Sons 37-70 Boynton, W.R.; Murray, L.; Hagy, J.D.; Stokes. C.; Kemp, W.M. 1996. A comparative analysis of eutrophication patterns in a temperate coastal lagoon. Estuaries. 19: 408-42 1. Canfield, D.R., and Bachmann, R.W. 198 1. Prediction of total phosphorus concentrations, chlorophyll-a, and Secchi depths in natural and artificial lakes. Can. J. Fish. Aquat. Sci. 38: 4 14-423. Cederwall, H. and Elmgren, R. 1980. Biomass increase of benthic macrofauna demonstrates eutrophication of the Baltic Sea. Ophelia Supp 1:287-304. Dillon, P.J., and Rigler, F.H.1974b. The phosphorusshlorophyll relationship in lakes. Limnol. Oceanogr. 19: 767-773. Dillon, P.J., and Rigler, F.H.1975. A simple method for predicting the capacity of a lake for development based on lake trophic status. J. Fish. Res. Board Can. 32: 1519- 1530. Edmondson, W.T. 199 1. The uses of ecology: Lake Washington and beyond. Seattle: University of Washington Press. Ehlin, U. 198 1. Hydrology of the Baltic Sea. Ln: Voipio, A. (ed.). The Baltic Sea. Elsevier Oceanography Senes no. 30. L 23- 134. Ekholm, P. 1994. Bioavailability of phosphorus in agriculturally loaded rivers in southem Finland. Hvdrobiologia 287: 179- 194. Finnish Environment Institute. 1998. Database of the monitoring system of spatial structure in major Finnish urban regions. Helsinki: Finnish Environment Institute. Finnish Institute of Navigation. 1996. Merenkulkulaitos Kartta (Sea charts). Volorne A. Helsinki. Fimish lnstitute of Navigation. 1997a. Merenkulkulaitos Kartta (Sea charts). Volume B. Helsinki. Fimish Institute of Navigation. 199%. Merenkulkulaitos Kartta (Sea charts). Volume E. Helsinki. Fimish Institute oflriavigation. 1997c. Merenkulkulaitos Kartta (Sea charts). Volume G. Helsinki. Fimish Institute of Navigation. 1998a. Merenkulkulaitos Kartta (Sea charts). Volume D. Helsinki. Finnish Institute of Navigation. l998b. Merenkulkulaitos Kartta (Sea charts). Volume F. Helsinki. Fisher, T.R.;Harding, Jr., L. W.; Stanley, D.W.; Ward, L.G. 1988. Phytoplankton, nutnents, and turbidity in the Chesapeake, Delaware, and Hudson estuaries. Esruarzne, Coastal and Sheif Science 27: 6 1-93. Fisher, T.R.;Melack, LM,Grobbelaar, J.U.; Howarth R.W.1995. Nutrient limitation of phytoplankton and eutrophication of inland, estuarine, and marine waters. in: Tiessen, H. (ed.). Phosphorus in the global environment. Chichester: John Wiley. 30 1-322. Grano, O. & Roto, M. 1986. Rantaviiva. (Atlas of Finland). Helsinki: National Board of Surveying. Hecky, R.E. and Kilharn, P. 1988. Nutrient limitation of phytoplankton in Freshwater and marine environments: a review of recent evidence on the effects of enrichment. Limnol. Oceanogt. 33 (4/2): 796-822. HELCOM - Baltic Marine Environment Protection Commission - Helsinki Commission 1997. Third periodic assessrnent of the state of the marine environment ofthe Baltic Sea. 1989-1 993. Background document. Baltic Sea Environrnent Proc. no. 64 B. Helsinki: HELCOM. Howarth, R.W. 1988. Nutrient limitation of net primary production in marine ecosystems. Anri. Rev. Ecol. 19:89-110. Howarth, R.W.; Jensen, H.S.; Marino, R.; Postma, H. 1995. Transport to and processing of P in near-shore and oceanic waters. in: Tiessen. H. (ed.). Phosphorus in rhe global envirocimenr. Chichester: John Wiley. 323-343. Hakanson, L. 1991. Ecometric and dynarnic modelling: exemplified &y cesium in lakes afler Chernobyl. Berlin: Springer. Koponen, J., Alasaarela, E., Lehtinen, K., Sarkkula, J., Sirnbierowicz, P., Vepsa, Hl, & Virtanen, M. 1992. Moilelling the dynamics of a large sea area. Helsinki: Publications of the Water and Environment Research Institute no. 7. Koroleff, F. 1976. Determination of nutrients. In: Grasshoff, K. (ed). Methods of seawclter ana!vsis. Weinheim, NY: Verlag Chemic 1 17- 133. Koroleff, F. 1979. The general chernical ana!vsis nrerhods of sea water. Helsinki: Institute of Marine Research. Lapointe, B.E. and Clark, M.W. 1992. Nutrient inputs from the watershed and coastal eutrophication in the Florida Keys. Estuaries 15(4): 465-76. Lau, H.G., Kangas, P., StorgArd-Envall, C. 1993. Spatio-temporal variations of sedimentation and sofi bottom macrofauna in the coastal waters of the Gulf of Bothnia. Aqua Fennica 23(2): 177- 186. Lorenzen,C.J., 1967. Determination of chlorophyll and phaeopigrnents: spectrophotometric equations. Limnol. Oceanogr. 12: 343-346. Mallin, M.A.; Paerl, H.W.;Rudek, I. Bates, P.W. 1993. Regulation of estuarine primary production by watershed rainfall and river flow. Mar. Ecol. Prog. Ser. 93:199-203 Mattila, J. 1993. Long-tenn changes in the bottom fauna along the Finnish coasts of the southem Bothnian Sea. Aqua Fennica. 23(2):143-1 52. McCauley, E., J.A. Downing,, and S. Watson. 1989. Sigmoid relationships between nutrients and chlorophyll among lakes. Cm. J. Fish. Aquat. Sci. 46: 11 7 1- 11 75. Meeuwig, J.J. n.d. Predicting eutrophication in lakes and estuaries: a quantitative comparison of system response across a range of tidal energy, openness, and salinity. submined: Lirnnol. Oceanogr. Meeuwig J.J. and Peters R.H. 1996. Circumventing phosphorus in lake management: a comparison of chlorophy11 a predictions hmland-use and phosphorus-loading models. Can. J. Fish. Aquat. Sci. 53: 1795- 1806 Meeuwig J.J.: Rasmussen. J.B.: Peten R.H. 1998. Turbid waten and clarifjmg mussels: their moderation of Ch1:nuûient relations in estuaries. Mar. Ecol. Prog. Ser. in Press. Mellina E.; Rasmussen J.B.;Mills E.L. 1995. impact of zebra mussels (Dreissena polymorpha) on phosphorus cycling and chlorophyll in lakes. Can. J. Fish. Aquat. Sci. 522553-2579. National Board of Waters. 198 1. Vesihallimon analyysirnenetelmat (Analytical methods used by the Finnish water authonty). Report 213. Helsinki: National Board of Waters. Nehring, D. 1992. Eutrophication in the Baltic Sea. Science of rhe Total Environment Suppl.: 673-682. Niemi, A. 1973. Ecology of phytoplankton in the Tvihminne area, SW Coast of Finland. 1. Dynamics of hydrography, nutrients, chlorophyll a and phytoplankton. Acfa Botanica Fennica. Nixon, S. W. 1988. Physical energy inputs and the comparative ecology of lake and maine ecosystems. Limnal. Oceanogr. 33(4/2): 1005- 1025. Nixon, S.W.; Ammerman, S. W.; Atkinson, L.P.; Berounsky, V.M.; Billen, G.; Boicourt, W.C.;Boynton, W.R.; Church, TM.;Ditoro, D.M.;Elmgren, R.; Garber, J.H.; Giblin, A.E.; Jahnke, R.A.; Owens, N.S.P.; Pilson, M.E.Q.; Seitzinger, S.P. 1996. The fate of nitrogen and phosphorus at the land-sea margin of the North Atlantic Ocean. Biogeochemistry 35 :14 1- 180. OECD. 1982. Eutrophication of waters: moniroring, assessrnent und control. Paris: OECD. Persson, J. and Hakamon, L. 1996. A simple empirical mode1 to predict deep water turnover time in coastal waten. Cm. J. Fish. Aquat. Sci. 53(6): 1236-1245. PietilZIinen, O.P. and Rekolainen, S. 1991. Dissolved reactive and total phosphoms from agricultural and forested bains to surface waters in Finland. Aqua Fennica 21: 127- 136. Pitkhen, H. 1994. Eutrophication of the Finnish coastal waters: Origin.fate und effects oJriverine nutrientflwes. Publications of the Water and Environment Research Institute no. 18. Pitkhen, H. & Tamrninen, T. 1994. Nitrogen and phosphorus as production limiting factors in the estuarine waters of the eastem Gulf of Finland. Mar. Ecol. Prog. Ser. 129: 253-294. Prairie, Y.T.;Duarte, C.M.; Kalff, J. 1989. UniQing nutrient-chlorophyll relationships in lakes. Cm. J. Fish. Aqitar. Sci. 46: 1 176-82. Reckhow, K.H., and J.T. Simpson. 1980. A procedure using modeling and error analysis for the prediction of lake phosphorus concentration fiom land use information. Can. J. Fish. Aquuf. Sci. 37: 1439-1448. Rekolainen, S. Finnish Environment Institute, Helsinki, Finland. Rekolainen, S.. Pitkiinen, H., Bleeker, A. & Sietske, F. 1995. Nitrogen and phosphorus fluxes frorn Fimish agricultural areas to the Baltic Sea. Nordic Hvdroiogy 26: 55- 72. Rekolainen, S. 1989. Phosphorus and nitrogen load fiom forest and agricultural areas in Finland. Aqua Fennicu 68: 40-54. Richardson, K. & Jergensen, B.B. 1996. Eutrophication: definition, history and effects. in: Richardson, K. & Jergensen, B.B.(eds.). Eutrophication in coastal marine ecosystems. Coastal and Estuarine Studies. Volume 52. Amencan Geophysical Union. Rosenberg, R.; Elmgren, R.; Fleischer, S.; Jonsson, P.; Penson, G.; Dahlin, H. 1990. Marine eutrophication case studies in Sweden. Ambio. 19: 102-108. Rowan DJ, Kalff J (1 991) The limnological implications of catchment sediment load. Verh. Internat. Verein. Limnol. 24: 2980-2984. SAS hstitute. 1985. SAUSTAT User's Guide, Release 6.03. Cary, NC: SAS Institute Inc. Savchuk, O. P. 1986. The study of the Baltic Sea eutrophication problems with the aid of simulation models. Baltic Sea Environment Proc, 19: 52-6 1. Schaub, B.E.M. and Gieskes, W.W.C.1991. Trends in eutrophication of Dutch coastal waters: the relation between Rhine river discharge and chlorophyll-a concentrations. In: Elliott, M and Ducrotoy, J.P (eds.). Estuaries and coasts: spatial and temporal intercomparisons. Fredensborg: Olsen and Olsen. 85-90. Vollenweider, R.A. 1975. Input-output models with special reference to the phosphorus loading concept in limnology. Schweiz. Zeit. Hydrol. 37: 53-84. Vollenweider RA, Marchetti R, Viviani R. (eds). 1992. Marine coastal eutrophication. Science ofthe Tord Environment. Supp. Walling, B.E. & Webb, B. W. 1985. Estimating the discharge of contaminants to coastal waters by rivers: some cautionary cornrnents. Murine Pollutioti Bulletin 16: 488- 493. Wallstrom. K. 199 1. Ecological studies on nitrogen-fixing blue-green algae and on nutrient limitation in the Baltic Sea. Ph.D. Thesis, Uppsala University. Wulff, F., Stigebrandt, A. & Rahm, L. 1990. Nutrient dynamics of the Baltic Sea. Ambio 19: 126-133. Yurkovskis, A.; Wulff, F.; Rahrn, L.; Andwaitis, A.; Rodnguex-Medina, M. 1993. A nutnent budget of the Gulf of Riga - Baltic Sea. Estuarine, Coastul und Sher Science 37: 113-127. Zar, J.H. 1984. Biostaristical anafysis.Englewood Cliffs: Prentice-Hall. Table I : Descriptive data for the estuaries, including growing season averages of chlorophyll (Chl), total phosphorus (TP), total nitrogen ('IN), Secchi depth (Sec) and salinity. Estuary morphometry is descnbed by mean depth (Zmn),surface area (Ao), volume (Vol), river water loading (QR) and water residence time (Res). Land-use is described by percentage of the watershed that is urban, agicultural, or forested (Urb-P, ApP, For-P), human population density (Pden) and watershed area (Wshed). Nutrient loads include the river derived loads (nonpoint source) for TP and TN (TPL-R TNL-RI. and the direct (point source) loads (TPL-D, TNL-D). Missing values for the direct loads indicate estuaries where the direct load is less than 0.1. Rat is the ratio of direct load to total load for TP. - -- Esluiuy Name Ch1 TP TN Sec Sal Ziiin Ao Vol QH Res (Finnish Code) mgm4 mgm" mg ni" ni PPt ni hn2 10~n1' ni3s-' YrS Virojoki (1 1) 17.3 Velikajoki (1 2) t 5.6 Summanjoki (1 3) t 3.9 Kymijoki (14) $ 9.8 llolanjoki ( 1 7) 5.8 Porvoonjoki ( 18) 30.5 Mustijoki (1 9) Y .4 Sipoonjoki (20) t 12.4 Vantaanjoki (2 1) 45.9 Karj aanjoki (23) 6.3 Halikonjoki (25) 15.7 Paimionjoki (27) 4.5 Hirvijoki (29) t 5.5 Laajoki (30) t 7.9 Kokemaeiijoki (35) $. 13.4 NlirpiBnjoki (39) t 4.2 Kyronjoki (42) 23.2 Perhonjoki (49) 7.5 Temmesjoki (58) t 9.9 37.8 b- - Mean 12.6 44.5 613 1.53 7.1 46.9 422 40.9 1.61 Stdev Min Max Table 1 cont.

Estuary Name Urb-P Ag-P For-P Wshed Pden Tpload-R TPL-D TNL-R TNL-D Rat (Fimish Code) km2 km-' t yr-' t yr*' 1 yr'l t yr-' 1 Virojoki (1 1) Vehkajoki (12) t Summanjoki (1 3) t Kymijoki (14) $ ilolanjoki (1 7) Porvooiijoki ( 18) Mustijoki (1 9) Sipoonjoki (20) t Vantaanj oki (2 1 ) Karjaanjoki (23) Halikonjoki (25) Paimionjoki (27) Hirvijoki (29) t Laajoki (30) t Kokemiienjoki (35) $ NWiGnjoki (39) t Kyriinjoki (42) Perhonjoki (49) Temmesjoki (5 8) t L- i Mean 1.7 24.5 72.6 1365 43.6 73.7 6.6 1623 109.2 12 Stdev 1.54 8.36 8.59 1719 58.3 1 13.0 7.9 2730 114.6 11.1 Min 0.35 9.50 54.34 116 9.0 5.3 0.9 131 23.6 2.0 Max 6.73 42.97 87.16 6817 265.1 446.6 20.0 10433 337.4 30.0 Table 2: Distribution of sampling effort for chlorophyll where years is the number of yean sampled, stations is the number ofstations sampled in each year and season indicates the mean, minimum (min) and maximum (mu)number of samples taken at a given station during the growing season, averaged for dl stations in the estuary.

1 / years 1 stations l season estuary ; 1989 1990 1991 1992 1993 mean min max I Table 3: A cornparison of observed TP to TP calculated using water residence times calculated as fresh water replacement time (TP-FW,Rt-FW), via Knudsen's equation (TP-K, Rt-K) and via Bowden's (1980) saltwater fraction method (TP-B,Rt-B). Estuaries where TP is better estimated via either Knudsen's equation or Bowden's equation are indicated with an *.

estuary obs TP TP-FW TP-K TP-B Rt-FW Rt-K Rt-B code rng-m'-' mg-m" mgm" mgm" months months months Table 4: A cornparison of the Ch1:TP and Ch1:TN regression equations in Finnish estuaries and lakes. Al1 variables transfonned to loglo,r? is the coefficient of determination and n is the sarnple size and nr indicates not reported.

equation r' n

d Finnish estuaries lChl= - 1 .O3 + 1.26 ITP 0.67 19 Lakes (OECD 1982) lChl= -0.55 + 0.96 ITP 0.88 77 Lakes (Di1lon and Rigler 1974) lChl = - 1.14 + 1.45 ITP 0.96 77 Finnish estuaries LChl=-2.10 + 1.13 ITN 0.53 19 Lakes (Sakamoto 1966) lChl= -2.5 + 1.4 ITN nr 2 1 Lakes (Prairie et al. 1989) lChl=-3.13 + 1.45 1TN O .69 133

Table 5: Standardized goodness of fit criteria for accuracy (MSR), precision (vSR) and bias (ME) for the rnass-balance and land-use regression models applied to al1 estuaries (ALL),nonpoint source dominated estuaries (NPS), estuaries with point source loads (PS) and the estuaies with point source loads lcss than 15% of the total load (PS45). Kokemaenjoki is included in both NPS and PS categories. Al1 values expressed as percentages.

Criteria Mode1 Type Al1 NF5 PS PS45 (n=i7) (n=ll) (n=7) (n=4)

- ---- Accurac y Mass-balance 0.095 O ,079- O. 107 0.024 (MW Land-Use 0.143 0.01 1 0.334 O. 102 Precision Mass-balance 0.0 11 0.01 1 0.0 12 0.00 1 @SR) Land-Use 0.073 0.000 O. 123 0.009 Bias Mass-balance 0.023 -0.110 0.275 O. 123 (ME) Land-Use 0.203 0.022 0.486 0.267 Fig. 1: Map of Finland indicating locations (by code; Table 1) of estuaries included in this study . Fig. 2: Regression equations for chlorophyll (Chl) as a function of total phosphorus (TP) and Chl as a function of total nitrogen (TN) for the 19 estuaries where i is the coefficient of determination and SEE is the mode1 standard error of the estimate.

r' = 0.67 SEE = 0.180 a

= 0.33 SEE = 0.213 O/

2.4 2.6 2.8 3 .O 3.2 3.4

log CM (mg rn'3)) IV.

Fig. 3: Observed vs. predicted values for Ch1 as a fiction of log mean depth (Kmn) and log percentage forest (lFor-P) for the Finnish estuaries dominated by non-point source loading where i is the coefficient of determination and SEE is the mode1 standard enor of the estimate.

7 rœ = 0.74 see = 0.108

0.8 1.O

log (CM (mg mo3)) - predicted Fig. 4: Observed vs. predicted values for Ch1 estimated by the mas-balance model for the Finnish estuaries dominated by point source loading where 8 is the coefficient of detemination and SEE is the model standard error of the estirnate. The numbers hdicate the estuaries Virojoki (1 l), Porvoonjoki (1 8) and Vantaa (21) which are discussed in the t ext.

1.O 1.2 1.4

log (CM (mg mo3)) - predicted PRED~CT~NCEUTROPHlCATION IN LAKES AND ESTUARIES:

A QUANTlTATlVE COMPARISON OF SYSTEM RESPONSE ACROSS A RANGE OF TlDAL

ENERCY, OPENNESS AND SALINITY.

Jessica J. Meeuwig Limnology und Oceanography: submitted 08.98 Abstract Eutrophication is a phenornenon common to both kesh and coastal waters yet its science and management have developed largely independently. Cross-system cornparisons have tended to focus on system differences relating to tidal energy (Nixon 1988b), openness (Visser and Kamp-Nielsen 1996) and water chemistry (Froelich 1988, Howarth et al. 1995). However, the responses of lake and estuary phytoplankton to nutnents and human disturbance has not been quantitatively compared. A data set was compiled for lakes, Finnish estuaries and US estuaries that represents a continuum of tidal range (O to Zm), openness (water residence times 0.01 to 8 yem) and salinity (O to 24 ppt), w ith the Finnish estuaries intermediate between lakes and US estuaries. Regression models were developed for phytoplankton biomass (as chlorophyll-a (Chl)) as a function of total phosphorus (TP), and as a function of land-use (percent of the catchment forested) and system morphometry (mean depth). TP accounted for a significant proportion of the variance in Ch1 in the lakes (?=0.68),Finnish estuaries (i=0.68)and US estuaries (i=0.49).ANCOVA indicated there were no significant differences arnong the three system-specitic Ch1:TP models and the Chl:TP model generated fiom the combined data accounted for 62% of the variation in Chl. Models based on percentage of the catchment forested and mean depth predicted Ch1 more accurately than the TP based models. The system-specific land-use models accounted for 56 to 84% of the variation in Chl. ANCOVA again indicated no significant differences among system-specific models and the model based on the combined data accounted for 80% of the variation in Chl. These analyses suggest that Ch1 responds similarly to human distuhance in lakes and estuaries, despite differences in tidal energy and water chernistry. These analyses also demonstrate that Ch1 is best predicted by models that integrate numerous factors (e.g. turbidity, other nutrients) that may influence phytoplankton as opposed to single nutrient models. Introductioa Since the Rim of the century, aquatic scientists have explored the relationship between phytoplankton biomass and nutrient concentrations in both lakes (Naumann 1919 cited in Home and Goldman 1994) and the sea (Brandt 1899 cited in Mills 1989). Despite these common historical roots, the science and management associated with excess aquatic primary production have developed Iargely independently in Ereshwater and coastal systems. Lake eutrophication has been widely recognized as an important environmental concem since the rarly 1960's (Vallentyne 1974). Limnological approaches to eutrophication are typified by whole-system experimenis (Schindler 1977), mass-balance models (Vollenweider 1975) and empirical models (Dillon and Rigler 1974); the latter two form the basis of lake eutrophication management (c.E OECD 1982). Recognition of coastal eutrophication has been slower, coming only in the 1980's (Nixon 1995). Coastal research has emphasized bioassays, mesocosm experiments (Hecky and Kilham 1988) and simulation modeling (Aksnes et al. 1995, Visser and Kamp-Nielsen 1996, Stigebrandt and Wulff 1987). Mas-balance models have also been used extensively in coastal research (Boyle et al. 1974, Boynton et al. 1995, Nixon et al. 1996) but generally to identify nutrient sources and sinks rather than to estimate arnbient concentrations of nutrients as in limnology. There are a number of similarities arnong freshwater and coastal approaches to eutrophication. The concept of a "lirniting nutx-ient" is central to both (Schindler 1977; Nixon et al. 1986) as is the recognition of the link between land-based human activities and water quality (Correll et al. 1992, Valiela 1992, Correll et al. 1992). Several researchers have also considered the application of the OECD lake management approach to coastal and marine systems (Giovanardi and Tromellini 1992, Lee and Jones 198 1) and empirical relations are increasingly published in the coastal literature (Monbet 1992, Pitkiinen 1994, Boynton et al. 1996). There have also been a number of comparisons of lakes and coastal systems (Schindler 1981, Nixon 1988a, Hassett et al. 1997). However, while some similarities are recognized, such as the nutritional requirements of phytoplankton (Hecky and Kilharn 1988), and levels of primary production (Kilham and Hecky 1988), most cornparisons seem to emphasize the differences between freshwater and coastal systems. These inc lude di fferences in: tidal energy (Nixon 1988b); hydrography and the effects of shorter water residence times in estuaries (Visser and Kamp-Nielsen 1996); salinity and its effects on water chemistry (Howarth et al. 1995); and nitrogen vs. phosphorus limitation (Howarth 1988, Fisher et al. 1995). To date, there have however been few quantitative comparisons of lakes and coastal waters. Nixon (1982) demonstrated that the dope of the relationship between fishenes yield and prirnary production is sirnilar in lakes and lagoons but that the yields (elevations) differ. Nixon et al. (1996) also demonstrated that the relationships between TN and TP export and water residence time are sirnilar in lakes and estuaries. The yield of Ch1 per unit TP in small estuaries in Prince Edward Island, Canada, has also been show to be an order of magnitude Iower than that typically seen in lakes (Meeuwig et al. in press). However, this difference was attributed primatily to intense grazing (as a result of suspended musse1 farms), a phenomenon that also occurs in lakes (Mellina et al. 1995), rather than any estuary-specitic charactenstics. This study in Prince Edward Island, however, is to our knowledge the only quantitative comparison of Chknutrient relationships in the literature. This lack may in part reflect the different "currencies" used in freshwater and marine research. Limnologists predict phytoplankton biomw ffom total nutrient stocks; oceanographen have emphasized pnmary production and inorganic "bioavailable" nutrients (c. f Boynton et al. 1982, Nixon 1995). There are also relatively few empirical patterns between phytoplankton biomass and total nutrients in the marine literature with which to make comparisons. This lack of quantitative cornparisons also likely reflects a conviction that "there are essential differences between these two [fresh and marine] types of systems that prevent us Crorn simply applying knowledge gained through limnological studies to the marine environment" (Richardson and Imgensen 1996: 5). To compare quantitatively the response of lake and estuary phytoplankton to nutrients and watershed disturbance in freshwater and coastal systems, data were compiled for a set of lakes, Baltic estuaries and estuaries fiom the Atlantic Coast of the United States. The lakes and US estuaries represent opposite ends of a continuum of tidal energy, salinity, and openness, characteristics considered key in distinguishing kesh water and coastal systems. The Baltic estuaries are intemediate between lakes and US estuaries as they are non-tidal, open systems with low salinity (O-Sppt). They thus share characteristics of both lakes and US estuaries. This study examines 1) whether there are systematic differences in the relationships between phytoplankton biomass, measured as Chl, and total phosphorus (TP) across the three categories (lakes, Finnish estuaries and US estuaries); and 2) whether there are systematic differences in the relationship between phytoplankton biomass and land-use across these systems.

Description of Data Sets Data were compiled for 1) 22 north temperate lakes; 2) 19 Baltic estuaries along the Coast of Finland; and 3) 22 mid- and south Atlantic estuaries in the US (henceforth US estuaries). The datasets include information on: growing season mean phytoplankton biomass (as Chl) and TP, morphornetry, and land-use (Table 1). The lakes were chosen from the data set used by Meeuwig and Peters (1996) to develop Ch1:land-use models. Lakes were chosen fiom a larger data set of 63 (Meeuwig and Peters 1996) so that the values for the independent variables fell in the sarne range as those of the estuaries. Thus, lakes were included with: 1) TP concentrations less than the maximum seen in the estuaries (TP-167 mg*m"); 2) a mean depth (Zmn) falling in the estuary range (1 - 18 m); and 3) percentage of the catchment forested (For-P) falling in the estuary range (33- 100%). Mean depth and For-P were chosen as delimiting variables iteratively: they were the variables ultimately included in the land-use model. The fmal data set includes lakes fiom Europe (n=6), Japan (n=6), North Arnenca (n=5) and the southem hemisphere (n=5), onginally denved fiom the Lake Biwa database (1988, 1989, 1990; for details see Meeuwig & Peten 1996). Nineteen Finnish estuaries fiom the nofi eastem Bothnian Bay to the eastem Gulf of Finland were included in the Ch1:total nutrient analyses. The eleven estuaries dominated by nonpoint source disturbance were included in the Chl:land-use analyses (for details, see Meeuwig et al. subm.). Growing season means were calculated for Chl, TP and TN using al1 measurements between June and August for the yem 1989-1993. May was excluded from the growing season average because the high latitude of these systems meaw that the spring bloom occurs in May. The mean values were calculateci, using Ch1 as an example, as: 1. Chl,, is the mean Ch1 for year-x and station-y, using the values of Ch1 fiom June to August of year x 2. Chl,, is the mean Ch1 for year-x, using the values of Chl,, for al1 the sampling stations located in estuary-z 3. Chi, is the mean Ch1 for estuary-z, using the values of Chl,, for al1 the sampling years for estuary-z This averaging approach gives equal weight to each year even if the year is represented by a single station. Such an approach incorporates interannual variability in Ch1 due to variation in climatic factors (Bennett et al. 1996, Mallin et al. 1993, Schaub and Gieskes 199 1 ). Catchment size, human population and land-use data were taken fiom the databases of the Finnish Environment Institute. Twenty-two US estuaries locaied along the Atlantic Coast fiom Long Island Sound to Charleston Harbour were included (Table 2). For 1 1 of the "NOAA estuaries, growing season averages for Ch1 and nutrients were calculated from data provided by the US Environmental Protection Agency's central data repository, STORET (US EPA n.d.), for stations sarnpled between 1989-1994. For Long Island Sound and Hudson River, growing season averages were calculated using STORET data for stations sarnpled between 1980 and 1985. Because these yean do not match the yean for which land-use data exists, these two estuaries were used only in the Chl:TP modeling. Estuary mean values were calculated for these 13 estuaries as for the Finnish estuary data. As the spnng blooms generally occur in Apnl in these estuaries, the growing season was defined as May to August. To expand the data set, data for Delaware Inland Bays were also included for both the Ch1:TP modeling (n=4; ref) and the Chl:land-use models (n=5; Boynton et al. 1995). Estuary morphometry for the "STORE'ï" estuaries was compiled from NOAA (1996, 1997). Water residence time was estimated using a modified salt water fiaction method (Bowden 1980): R, = v [(s, - s,) &] Q'~ [es* 11 where V is the estuary volume (m.'),Sm is the mean salinity in the estuary (%O), S, is the salinity (%O) of open water (approxirnately 35%~)and Q is the fieshwater load (m3e day"). We used the freshwater load estimated by NOAA (19xx). The mean estuary salinity was calculated as:

Sm = [(a, O%O) + (aM 12%0) + (asw a 28%0)] (a+aM+asw)-' [es- 21 where a, is the freshwater tidal area of the estuary, a~ is the mixing zone area of the estuxy and asw is the seawater zone area of the estuary (NOM 1996, 1997). The values of 12%~and 28960 represent the mean salinity values of the mixing and seawater zones as defmed by NOAA (1996, 1997). Such a calculaiion is only an approximation of the residence time. as it assumes that the mean salinity can be estimated using the mean values of the zones. However, it was not possible to find literature values for water residence times in al1 the estuaries and this technique provides a consistent method. This method generated values comparable to the literature values for Delaware Bay, Chesapeake Bay and the Neuse River and underestimated estimates for the Potomac and Albemarle Parnlico Sound (Table 3). The values are always lower than those of Nixon et al. (1996) as would be expected since they calculated residence time based on freshwater replacement times. Land-use for 1990 and human population data ( 1990) were taken from (EPA population study and EPA land-use document; refs still to corne**).

Statistical Analyses Standard Ieast squares regression techniques and ANCOVA were used to analyze the data (Zar 1984, SAS Institute 1985). Al1 variables were log-transformed to stabilize variance (Zar 1984). Land-use variables that were calculated as percent of the watenhed were transformed as logio(X + 1) due to the presence of zeros in some land-use categories. System-specific Ch1:TP and Chl-land-use models were generated for each of the individual data sets (lakes, Fimish estuaries, US estuaries). The land-use models took the form of Ch1 as a function of one morphometric variable, indicating system sensitivity, and one land-use variable, indicating human disturbance. Mer system-specific models were fit, the data for the lakes, Finnish and US estuaries were combined and models were generated to predict Ch1 for the combined dataset as a function of TP and land-use. ANCOVA and Tukey tests were used to determine whether there was an effect of system type (lake, Finnish estuary or US estuary) on the relationship behveen Ch1 and TP and Ch1 and land-use. The ANCOVA tested for both differences in slope (the interaction term) and intercepts (system type). To assess the interchangeability of models developed for specific systems, each system's mode1 was used to predict Ch1 in the other two systems. Following Meeuwig and Peten (1996), the models were compared in terms of accuracy, precision and bias. Accuracy was estimated as the mean squared residual:

MSR = ~ICM,- IC~I,)? n*' [es. 31 where IChl, - lChlp is the difference between observed and predicted log Ch1 and n is the number of observations. The variance of the squared residuals (vSR) and the mean error (ME) were used as criteria of precision and bias: vSR = (X((lCh1, - l~hl,)'- MSR))' (n - 1)-' [es- 41 ME = Z(lCh1, - lChlp) n'l [es- 51

Results The three systems reflect the range of physical characteristics which were sought. Mean tidal range varies from O in the lakes to occasional wind driven tides of 0.5 m in the Fimish estuaries to 0.7 (max. 2 m) in the US estuaries. Water residence times are longest in the lakes (mean 1.6 years) and the Fimish estuaries (mean 1.6 years) and shortest in the US estuaries (mean 0.2 yean). The Finnish estuary residence times are likely overestimated as they were calculated as fieshwater replacement times and the mean is highly influenced by a single long residence time (Meeuwig et al. 1998). Mean salinity ranges from O ppt in the lakes to 3.8 ppt in the Finnish estuaries to 15.8 ppt in the US estuaries. Thus, the systems can be treated categoncally and the rnodels compared via ANCOVA. Chi: TP relations Significant Ch1:TP relationships were generated for ail three types of systems with TP accounting for 49 to 68% of the variation in Ch1 (Table 4a). ANCOVA indicated no significant difference among systems in slope or elevation. However, the p values were close to significance (p=0.08 for the slope and p=0.09 for the intercept). The Finnish estuaries appear most distinct with an intercept an order of magnitude lower than those of the lake and US estuary relationships and a steeper slope. The relation between Ch1 and TP was also highly significant for the combined data (p=0.0001, % = 0.62; Table 4% Fig. 1). Chkland-use relations Significant relationships were developed between Ch1 and land-use variables for al1 three systems. In the lakes, Finnish and US estuaries, mean depth (Zm) and percentage of the catchent forested (For-P) were the best predictors of Ch1 (r?=0.84, r2=0.75 and ?=OS6 respectively: Table 4b). Al1 the systern-speci fie models were robust with the two independent variables each contnbuting significantly to the model. The ANCOVA indicated no difference between either the partial regression coefficients (dopes) for Zmn (p=0.23) or For-P (p=0.85) or the intercept @=0.77). The land-use model was tlius refit to the combined data; this model accounted for 80% of the variation in Ch1 (Table 4b; Fig 2). The land-use models account for a greater proportion of the variance in Ch1 than the TP models for the combined data set and in system specific data sets (Table 4). There is an improvement of 6-7% for the system-specific models and 11% for the combined data set. This difference is statistically significant as indicated by an F test comparing the MSR of the Ch1:TP and Ch1:land-use models (Fo.(jui p0.05). A coniparison of predictions: are system-speciflc models interchangeable ? In terms of the Ch1:TP relations, lake and Finnish estuary Ch1 is best predicted by the system-specific rnodels. The MSR, vSR and ME are consistently smallest for the system-specific model (Table sa). However, in the case of the US esniaries, the combined model predicts Ch1 most accurately (MSR = 0.027 vs. 0.073), precisely (vSR = 0.001 vs. 0.003) and with the least bias (-0.052 vs. 0.22) (Table 5a). In terms of the land-use models, lake, Finnish and US estuary Ch1 is most accurately and precisely predicted by the appmpnate system-specific model (Table Sb). However, the accuracy and precision of the combined model is very close to that of the system-specific models. This pattern is consistent with the results of the ANCOVA which discemed no differences among the systems in terms of Ch1:land-use models. This analysis is consistent with the earlier cornparison of the Chl:TP and Ch1:land- use models in terms of their coefficients of determination and the F test. The Ch1:iand-use models predict Ch1 more accurately and with less bias than the Chl:TP models. Discussion The results of these analyses fail to support the hypotheses of systematic differences in Ch1:nutrient and Chkland-use relationships across these categories of tidal energy, openness and salinity. There are no significant differences among the system types in either the Ch1:TP or Ch1:land-use model. In tems of the land-use models, the systems relate so similarly to land-use that the individual equations are to some degree interchangeable. In both the TP and land-use models, the lakes and US estuary models are interchangeable to a greaier degree than the Finnish models despite the "intermediate" nature of the Finnish estuaries. A Iake-estuary dichotomy based on tides ? Monbet (1992) demonstrated an effect of tidal range on the yield of Ch1 per unit dissolved inorganic nitrogen (DM). The Ch1:DiN yield was lower in rnacrotidal(>2m tidal range) estuaries than in microtidal estuaries. This pattern may not have ernerged in this analysis due to the relatively small range in tide compared to that of Munbet's study. nie maximum tidal range is 2 m (Table 1) thus this study only includes "micro" tidal estuaries. This result suggests that the tidal effect on Chknutrient yields may only occur after a certain threshold of tidal energy. Nixon (1988b)argued that the presence of tidal energy should increase biomass of zoobenthos which in Nni decreases the net storage of nutrients in estuarine sediments. Nixon's conclusions focused on the greater transfer efficiency of pnmary production in estuarine environments. Decreased nutrient storage should also result in higher phytoplankton biomass in tidal systems due to greater arnbient nutrient concentrations. However, decreased sediment storage cm increase arnbient concentrations of nutrients without changing the Ch1:nutrient yield and is thus consistent with the results of this study. Decreased sediment storage of nutrients should nonetheless affect the relationship between Ch1 and land-use: a greater proportion of the load derived from land-use should remain available to phytoplankton in systems with tidal energy. There is no evidence of such an effect. 1 would speculate that the inclusion of mean depth in the model incorporate this "tidaldecreased-nutrient-storage" effect dong with the other effects associated with shallowness such as increased resuspension and nutrient recycling. It is also not clear to what degree the relationship between sedimentation (and thus storage) differs in lakes and estuaries. For instance, Nixon et al. (1996) found that the large lakes and estuaries demonstrate the same pattern between N and P export and water residence time. A lake-estuary dichotomy based on openness ? nie openness of a system affects water residence tirne. The minimum water residence time in these systems is about 5 days which is greater than the generation tirne of phytoplankton. It is thus unlikely that short residence times would decrease the yields. For instance, Basu and Pick (1996) demonstrated that the Ch1:TP relationship in nvers, some with residence times as short as 3 days, was similar to the yield in lakes. However, water residence time more likely affects the relationship between Ch1 and land-use in that rapidly flushing systems are Iess sensitive to a given level of disturbance than slowly flwhing systems. This is a generally observed pattern and is, for instance, a key component of phosphorus mass-balance equations (Vollenweider 1975). Mean depth is frequently correlated with water residence time and thus it may be that mean depth so consistently enters the land-use models as a surrogate for water residence time. A lake-estuary dichotomy based on saliniîy ? Salinity aflects phosphate availability in several ways (for a review, see Howarth et al. 1995). Phosphate adsorbed ont0 particulate matter is released as river water meets estuarine water due to increased ionic cornpetition for adsorption sites and phosphonis concentrations consequently increase (Froelich 1988). Ambient concentrations of phosphonis should also be higher in estuaries as estuarine sediments bind phosphorus less tightly than fieshwater sediments (Howarth et al. 1995). Altematively, phosphonis concentrations may decrease in estuaries due to flocculation etc. These processes are likely to either increase or reduce the amount of phosphonis in the water but should not necessarily change the yield of Ch1 per unit phosphoms. It should however affect the relationship between Chl and land-use by altering the availability of phosphorus denved hmland-use. There is no evidence however that this is the case as the land-use models for the lakes, Finnish and US estuaries are indistinguishable. Meeuwig et al. (1998) also demonstrated that lake-derived coefficients for sedimentation could be used in mas- balances to accurately estimate TP fiom phosphorus loads in Finnish estuaries, despite differences in salinity. As P recycling is related to depth (Fisher et al. 1982), any such differences may again be integrated in the land-use mode1 via the mean depth variable. A Iake-estuary synthesis based on graring and trophic status: evidence /rom diferenfial yields ? Although ANCOVA did not indicate a statistically significant difference among the Ch1:TP relationships. ANOVA and Tukey tests indicate that the yield of Chl pet unit TP differs significantly between the lakes (mean Chl:TP = 0.45) and the estuaries (Fimish Ch1:TP = 0.25 and US Ch1:TP = 0.18). Grazing is one possible explanation for the lower yields in the estuaries. Officer et al. (1982) suggested that bivalves may be a natural eutrophication control and they have been implicated in low yields of phytoplankton both in estuaries (Cohen et al. 1984, Meeuwig et al. 1998) and lakes (Mellina et al. 1995). Although the yields in the Finnish and US estuaries are lower than the lake yields, they are still rnuch larger than those seen in the PEI estuanes, where musse1 grazing was implicated (rnean yield = 0.03, n = 15; Meeuwig et al. 1998). This is perhaps not surprising. US estuaries such as the Delaware (Maurer et al. 198 L), Neuse River (Lenihan and Peterson 1995), and Bogue Sound (Sumrnenon and Petenon 1990) have seen major declines in bivalve populations due to hypoxia and anoxia, over- exploitation and disease. Although available information for the Baltic suggests that zoobenthos have generally increased with eutrophication due to increased food supply, bivalve biomass appears to have decreased in the more eutrophic coastal waters of Finland due to hypoxia, anoxia and pollutants (Andersin and Sandler 199 1, Mattila 1993). Thus, grazing may not be an important contributor to the lower yields in the Finnish and US estuaries. The differences in yields may also reflect differences in the range of TP. Prairie et al. (1989) demonstrated that Ch1:TP yields decrease as trophic status increases. The same pattern exists in these data (Fig. 3): Ch1 yields decrease as TP increases with the lowest yields in the US estuaries where TP concentrations exceed 100 mg*mS3.The lakes span a lower range of TP values than the two types of estuaries (Fig. 1; Table 1) and have a higher yield. Womuiately, 1 could fïnd no examples of oligotrophic estuaries to test whether these systems have higher yields, depressing support for Nixon's (1995?) description of estuaries as the most heavily fertilized ecosystems on the planet. The importance of trophic status in determining yields has also been identified by Howarth et al. (1995) and Downing (1997). Both argue that P vs. N limitation is a function of trophic status rather than system-based charactenstics such as salinity. They conclude that oligotrophic systems, Ereshwater, coastal or oceanic, tend to P-limitation while eutrophic systems tend to N-limitation. Iawonki and Howarth (1995) and Billen et al. (1 99 1) also demonstrated relationships between N:P ratios and human population density and land-use respectively. suggestine that that aquatic -stems move From P towards N limitation as human disturbance increases. This conclusion is consistent with these data: the US estuaries have the highest mean population density (122 km*' vs. 82 km**and 44 km" in the lakes and Finnish estuaries) and the lowest amount of forested area (58% vs. 73% in the lakes and Finnish estuaries).

Summary The differences among the Ch1:TP models are however generally small. Al1 of the slopes and intercepts fa11 within the range of coefficients seen for lake models currently in the literature. The most likely reason for the small differences in Ch1:TP relations is variation in trophic status among the systems. Trophic status, although used categorically, is determined From continuous variables. It is thus likely that the shift fiom P to N limitation, (assurning such pure endpoints exist), occurs across a gradient of CO-limitation by P and N. Such a conclusion indicates the importance of considering other nutrients and helps explain why the land-use models predict Ch1 more accuraiely than the TP models. Recognition of CO-limitationis increasing in the literature (Elser et al. 1990, Morris and Lewis 1988, Dodds and Priscu 1990, Fisher et al. 1992) and it appears that land-use is an effective way to integrate both nutrients (CO-Limitation)and other factors such as turbidity (Harding et al. 1986, Fisher et al. 1988, Rowan and Kaiff 1991) that affect phytoplankton biomass, Among the land-use models, there are no differences among the three types of systems. Thus differences due to tidal energy, openness or salinity seem to be either irrelevant or integrated within the model. Once differences in system sensitivity are included (via mean depth in this case), human distuhance in the form of deforestation is the best predictor of Chl. On the surface, this is surprising. In addition to the physical and chemical differences of the receiving waters discussed above, there are important regional differences in nutrient loading both between the Finnish and US estuaries and among the US estuaries. For instance, nutrient sources Vary strongly: in the Baltic region, 76% of nitrogen is denved from agricultural activities as compared to 17% and 39% in the NE and SE US respectively (Howarth et al. 1996). Molar N:P ratios also vary highly among these systems with N:P ratios of 23, 17 and 47 in the Baltic, NE and SE US respectively (Howarth et al. 1996). However, if, as suggested by the Ch1:TP models and others (Howarth, et al. 1996, Downing 1997), nutrient limitation is a hction of trophic level and if land-use models integrate nutrient factors, this result is less surpnsing.

Acknowledgements This research could not have been completed without the generosity and help of the US EPA staff at STORET in facilitating access to their data. I particularly thank Louis Hoelman This research was supponed by a CIMO scholanhip and is a contribution fiom the McGill Limnology Research Centre. Literahire Cited

Aksnes, D.L.;Ulvestad, K.B.; Balino, K.B.; Bemtsen, J.; Egge, J.K.; Svendsen, E. 1995. Ecological modeling in coastal waters: towards predictive physical-chemical- biological simulation models. Ophelia 41 5-36, Andersin, A.-B.& Sandler, H. 199 1. Macrobenthic fauna and oxygen deficiency in the Gulf of Finland. Menloranda Soc. Fatrna Flora Fennica 67: 3-10. Basu, B. and Pick, F. 1996 Factors regulating phytoplankton and zooplankton biomass in temperate riven. Limnol. Oceanogr. 4 l(7): 1572- 1577 Bennett, J.P; Woodward, J.W.; Shultz, D.J.1996. Effect of discharge on the Chlorophyll a distribution in the tidally-influenced Potomac River. Estuaries 9(4A):250-260. Billen, G.; Lancelot, C.; Meybeck, M. 1991. N, P, and Si retention along the aquatic continuum from land to ocean. In: Mantoura, R.F.C, Martin, J.M.; Wollast, R. (eds). Ocean nlargin processes in global change. Chichester: W iley pp 19-44. Bonsdorff, E.; Blomqvist, E.M.; Mattila, J.; Norkko, A. 1997. Coastal Eutrophication: Causes, consequences and perspectives in the Archipelago Areas of the Northem Baltic Sea. Est. Coast. and SheifScience 44 (Suppl. A): 63-72. Bowden K.F. 1980 Physical factors: salinity temperature circulation and mixing processes. In: Olausson E. and Cato 1. (eds.). Chemistry and biogeochemistry of estuaries. Chichester: John Wiley and Sons 37-70 Boyle, E.; Collier, R.; Dengler, A.T.; Edmond, LM.; Ng, AC.; Stallard, R.F. 1974. On the chemical mass balance in estuaries. Geochimica et Cosmochimica Acta 38: 1719-1728. Boynton, W.R.; Garber, J.H.; Summers, R.; Kemp, W.M. 1995. Inputs, transformations and transport of nitrogen and phosphorus in Chesapeake bay and selected tributaries. Estuaries 18(1B):285-3 14. Boynton, W.R.; Kemp, W.M.; Keefe, C.W. 1982. A comparative analysis of nutrients and other factors influencing estuarine productivity. In: Kennedy, V.S. (ed.). Estuarine comparisons. NY: Academic Press. pp 60-90. Boynton, W.R.;Murray, L.; Hagy, J.D.;Stokes, C.; Kemp, W.M. 1996. A comparative analysis of eutrophication patterns in a temperate coastal lagoon. Estuaries 19(4B):408-421. Brandt, K. 1899. Uber den Stofiechsel in Meere. Wissenschaftliched Meeresuntersuchunger, Abt. Kield. Bd. 4:z 1 3-230 cited in Nixon et al. 1986. Christian, R.R.; Boyer, J.N.; Stanley, D.W.1991. Multi-year distribution patterns of nutrients within the Neuse River Estuary, North Carolina. Mar. Ecol. Prog. Ser. 7 1:259-274. Cohen, R.R.H.;Dresler, P.V; Philips, E.J.P; Cory, R.L. 1984. The effect of the Asiatic clam. Corbictrfafluminea. on phytoplankton of the Potomac River, Maryland. Limnol. Oceunogr. 29(1):170-1 80. Correll, D.L.;Jordan, T.E.; Weller, D.W.1992. Nutrient flux in a landscape: effects of coastal land-use and terrestrial community mosaic on nuirient transport to coastal waters. Estrraries 15(4):431-442. Dillon, P.J., and Rigler, F.H.1 974. The phosphorus-chlorophy Il relationship in lakes. Limnol. Oceanogr. 19: 767-773. Downing, I.A. 1997. Marine nitrogen: phosphorus stoichiometry and the global N:P cycle. Biogeochemisty 37237252. Elser, I.J.; Marzol f, E.R.;Goldman, C.R. 1990. Phosphorus and nitrogen limitation of phytoplankton growth in freshwaten of North Amenca: a review and critique of expenmental enrichments. Can. J. Fish. Aquatic Sci. 47: 1468- 1477. Field, C.K.;Siver, P.A; Lott, A-M. 1996. Estimating the effects of changing land use patterns on Connecticut Lakes. J. Environmental Qitality B(2):325-333. Fisher, T.R.; Carlson, P.R.;Barber, R.T. 1982. Sediment nutrient regeneration in three North Carolina estuaries. Est. Coastal and Shelf Sci. 14: 10 1- 1 16. Fisher, T.R.; Harding, L.W. Jr.; Stanley, D.W.;Ward, L.G. 1988. Phytoplankton, nutients, and hirbidity in the Chesapeake, Delaware, and Hudson Estuaries. Est. Coast. and ShelfScience 2?:61-93. Fisher, T.R.;Melack, LM, Grobbelaar, LU.; Howarth R.W.1995. Nutrient limitation of phytoplanlaon and eutrophication of inland, estuarine, and marine waters. In: Tiessen, H. (ed.). Phosphonrs in the global environnent. Chichester: John Wiley. 301-322. Fmelich, P.N. 1988. Kinetic control of dissolved phosphate in natural nven and estuaries: a primer on the phosphate buffer mechanism. Limnol. Oceanogr. 33(4/2): 649-668. Giovanardi, F. and Tromellini, E. 1992. Statistical assessrnent of trophic conditions. Application of the OECD methodology to the marine environment. P. 2 1 1-234 In Vollenweider, R.A. R. Marcheni and R. Viviani (eds.). Marine Coastal Eutrophication. Special issue of Science of the Total Environment .. .x.. Harding, L.W.;Meeson, B.W.; Fisher, T.R. 1986. Phytoplankton production in two East Coast estuaries. Photosynthesis-light fùnctions and patterns of carbon assimilation in Chesapeake and Delaware bays. Est. Coastal and SheffSci. 23:773-822. Hassett. R.P.; Cardinale, B.; Stabler, L.B.;Elser, J.J. 1997. Ecological stoichiometry of N and P in pelagic ecosysterns: comparison of lakes and oceans with an emphasis on the zooplankton phytoplankton interactions. Limnol. Oceanogr. 42(4):648-662. Hecky, R.E. and Kilham, P. 1988. Nutrient limitation of phytoplankton in freshwater and marine environrnents: a review of recent evidence on the effects of enrichment. Limtzol. Oceanogr. 33(4/2): 796-822. Horne, A.J. and Goldman, C.R. l994. Lirnnologv. 2" edition. NY:McGraw-Hill Inc. Howarth, R. W. 1988. Nutrient limitation of net primary production in marine ecosystems. Ann. Rev. Ecol. l9:89- 1 10. Howarth, R.W.;Jensen, J.S.; Marino, R.; Postma, H. 1995. Transport to and processing of phosphorus in near-shore and oceanic waters. In: Tiessen H. (ed.). Phosphom in the global environmeni. Chichester: John Wiley & Sons. Pp 323-345. Howarth, R. W.; Billen, G.; Swaney, D.; Townsend, A.; Jawonki, N.; Lajtha, K.; Downing, J.A.; Ehgren, R.; Caraco, N.; Jordan, T.; Berendse, F.; Freney, J.; Kudeyarov, V.; Murdoch, P.; 2. Zhao-Liang. 1996. Regional nitrogen budgets and nverine N and P fluxes for the drainages to the North Atlantic Ocean: natural and human influences. Biogeochemistry 35:75- 139. ECilham, P. and Hecky, R.E.. 1988. Comparative ecology of marine and keshwater phytoplankton. Limnol. Oceanogr. 33(4/2): 776-795. Lake Biwa Research Institute. 1988. Data book of world lake environments. Otsu: International Lake Environment Cornmittee. Lake Biwa Research institute. 1989. Data book of world fake environments. Otsu: International Lake Environment Cornmittee, Lake Biwa Research Institute. 1990. Data book of wodd lake environments. Otsu: International Lake Environment Cornmittee. Lee, G.F.and Jones, R.A. 198 1. Application of the OECD eutrophication modeling approach to estuaries. p. 549-568 In Neilson, B.J. and L.E. Cronin (eds.). Estuaries and nzrtrients. Clifton NJ: Humana. Lenihan, H. and C.S. Peterson. 1995. Mass mortality of oysten, Crassostrea virginica, in a wind-driven estuq: prolonged hypoxia and the restoration of a declining fishery. Abstract only. In Grassle, J.P and others (eds.). Twenty-third Benthic Ecology meeting. Rutgen, NJ. Mallin, M.A.; Paerl, H.W.; Rudek, J.; Bates, P.W.1993. Regulation of estuarine primary production by watershed rain fa11 and river flow . Mar. Ecol. Prog. Ser. 93:199-203. Malone, T.C.; Crocker, L.H.; Pike, S.E.; Windler, B.W. 1988. Influence of river flow in the dynamics of phytoplankton production in a partially stratified estuary. Mar. Ecol. Prog. Ser. 48:235-249. Manila, J. 1993. Long-tenn changes in the bottom fauna along the Finnish coasts of the southern Bothnian Sea. Aqtra Fennica. 23(2): 143-152. Maurer, D. Howe, S.; Leathem, W.. 198 1. Secondary production of benthos in an industrialized estuary. Estuaries 4(3): 302 (abstract only). Meeuwig J.J. and Peters R.H. 1996. Circurnventing phosphonis in lake management: a cornparison of chlorophyll u predictions from land-use and phosphorus-loading models. Can. J. Fish. Aquat. Sci. 53: 1795-1806 Meeuwig J.J.; Rasmussen, J.B.; Peten R.H. 1998. Turbid waters and clariwg mussels: their moderation of Ch1:nutrient relations in estuaries. Mar. Ecol. Prog. Ser. Accepted. Meeuwig, J.J.; Kauppila, P.; Pitkiinen, H. Predicting Coastal Eutrophication in the Baltic: Vollenweider Applied to Small Finnish Estuaries. Submitted Can. J. Fish. Aquat. Sci. Mellina, E.; Rasmussen, J.B.; Mills, E.L. 1995 Impact of zebra mussels (Dreissena po!vmorpha) on phosphorus cycling and chlorophyll in lakes. Can. J. Fish. Aquat. Sci. 52:2553-2579 Mills, E.L. 1989. Biological oceanography: an early history, IWO-1960.Ithaca, NY: Comell University Press. 378 pp. Monbet Y. 1992. Controls of phytoplankton biomass in estuaries - a comparative analysis of microtidal and macrotidal estuaries. Estuaries 15563-5 7 1. National Oceanic and Atmospheric Administration (NOAA). 1996. NOAA's estuarine eutrophication survey. Vol. 1: South Atlantic Region. Silver Spnng, MD. Office of Ocean Resources Conservation Assessment. National Oceanic and Atmospheric Administration (NOAA). 1997. NOAA's estuarine eutrophication survey. Vol. 2: Mid-Atlantic Region. Silver Spring, MD. Office of Ocean Resources Conservation Assessment. National Oceanic and Atmospheric Administration (NOAA). 1998. Coastal Assessment Framework. Silver Spnng, MD. Office of Ocean Resources Conservation Assessment. Data taken from: http://www-orca.nos.noaa.gov~projects/caVcaf.h~l Naumann, E. 19 19. Nana synpwiker angaende planktons okologi. Med sankilde hansyn till S.toplankton. Svensk bot Tidrtr 13:129-158. Nixon, S. W. 1982. Nutrient dpamics, prirnary production and fisheries yields of lagoons. Oceanologica Acta Proceedings International Symposium on Coastal Lagoons SCOR/IABO/UNESCO, Bordeaux 198 1: 357-37 1. Nixon, S. W. 1988a. Comparative ecology of freshwater and marine ecosystems: dedicated issue. Linznol. Oceanogr. 33(4 p. 2)- 1025 pp. Nixon, S.W.l988b. Physical energy inputs and the comparative ecology of lake and marine ecosystems. Limnol. Oceanogr. 33(4/2): 1005- 1025. Nixon, S.W.1995. Coastal marine eutrophication: a definition, social causes, and hiture concems. Gpheiia: 4 1: 199-2 19. Nixon, S.W.; Ammerman, LW.;Atkinson, L.P.; Berounsky, V.M.; Billen, G.;Boicourt, W.C.;Boynton, W.R.;Church, T.M.;Ditoro, DM.; Elmgren, R.; Garber, J.H.; Giblin, A.E.; Jahnke, R.A.; Owens, N.J.P.; Pilson, M.E.Q.; Seitzinger, S.P. 1996. The fate of nitrogen and phosphorus at the land-sea margin of the North Atlantic Ocean. Biogeochernistry 35: 141 - 180. Nixon, S.W.; Oviatt, C.A.; Frithsen, J.; Sullivan, B. 1986. Nutrients and the productivity of estuarine and coastal marine ecosystems. J. Limnol. Soc. M.Afi. 12(1/2): 43-7 1. OECD. 1982. Eutrophication of waters: monitoring, assessrnent and control. OECD, Paris, France. 154 p. Officer, C.B.; Smayda, T.J.; Mann, R. 1982. Benthic filter feeding: a natural eutrophication control ? Mar. Ecol. Prog. Ser. 9:203-2 10. Pitkiinen, H. 1994. Eutrophication of the Fimish coastal waters: Ongin, fate and effects of rivenne nutrient fluxes. National Board of Waters and the Environment, Finland. Publications of the water and environment research institute 18,45 p. Prairie, Y.T.;Duarte, C.M.; Kalff, J. 1989. Uni@ng nutrient-chlorophyll relationships in lakes. Can. J. Fish. Aquat. Sci. 46: 1 176-82. Reckhow, K.H., and Simpson, J.T. 1980. A procedure using modeling and error analysis for the prediction of lake phosphorus concentration from land use information. Can. J. Fish. Aquat. Sci. 37: 1439-1448. Richardson, K and Jergensen, B. B. 1996. Eutrophication: de finition, history and effects, p 1- 19. In: Richardson, K. and Jergensen, B.B. (eds.), Eutrophication in coastaf marine ecosysrems. Coastal and Estuarine Studies, Volume 52. Amencan Geophysical Union. Rowan, D.J. and Kalff, J. 1991. The lirnnological implication of catchment sediment load. Verh. Internat. Verein. Limnof. 24: 2980-2984. SAS Institute. 1985. Sus/STAT User's Guide, Release 6.03. SAS Institute Inc., Cary, N.C. 1028 p. Schaub, B.E.M.and Gieskes, W.W.C.1991. Trends in eutrophication of Dutch coastal waters: the relation between Rhine river discharge and chlorophyll-a concentrations. In: Elliott, M and Ducrotoy, J.P (eds.). Estuaries and cousts: spatial and temporal intercomparisons. Fredensborg: Olsen and Olsen. 85-90. Schindler, D.W.1977. The evolution of phosphotus limitation in lakes. Science 195:260- 262. Schindler, D.W. 1981. Studies of eutrophication in lakes and their relevance to the estuarine environment, p. 7 1-82. In: Neilson, B.J. and Cronin, L.E. (eds.). Estuaries and nutrients. Clifion NJ: Humana. Stigebrandt, A. and Wu1 ff, F. 1987. A mode1 for the dynamics of nutrient and oxygen in the Baltic proper. J. Mur. Res. 45729-759. Summenon, H.C. and Peterson, CH.. 1990. Recruitment failure of the bay scallop, Argopecten irradians concentricus, during the first red tide, Ptychodisnis brevis, outbreak recorded in North Carolina. Estuaries 13(3):322-331. United States Environmental Protection Agency. 1997. Unpublished data retneved from STORET (EPA's Central Data Repository). United States Environmental Protection Agency. 1998. "Surf your watershed" A website compiling landuse information for hydrologie cataloging units.Data found at: http://www .epa.gov/surf Valiela 1. 1992. Coupling of watenheds and coastal waters: an introduction to the dedicated issue. Estiraries 15(4): 429-430. Vallentyne, J.R. 1974. The algal bowf:lakes and man. Ottawa. Department of the Environment, Fisheries and Marine Service. Miscellaneous Special Publication 22. Visser, A.W. and Kamp-Nielsen, L. 1996. The use of models in eutrophication studies. P. 22 1-242. In K. Richardson and Jsrgensen, B.B. (eds.), Eutrophication in coastal marine ecosystem. Coastal and Estuarine Studies, Volume 52. Arnencan Geophysical Union. Vollenweider, R.A. 1975. input-output models with special reference to the phosphorus loading concept in lirnnology. Schweiz. Zeit. Hydrol. 37: 53-84. Zar, J.H. 1984. Biostatistical ana!vsis. Prentice-Hall, Englewood Cliffs. Table 1: Sumrnary descriptive statistics (sarnple size (n), mean, standard deviation (SD), minimum (min) and maximum (max)) for the combined data, and the system-specific data sets (Lakes, Finnish estuaries and US estuaries) for chlorophyll (Chl), total phosphorus (TP), total nitrogen (TN), the TN:TP ratio, salinity (SAL), surface area (Ao), volume (Vol), mean depth (Zmn), water residence time (Rt), tidal range (Tide), watershed area (Wshed), percentage of the watershed that is forested (For-P), agriculturai (Ag-P), urban (Urb-P), and human population density (Pden). . I I Combined Data l Lakes l uni@ i n mean stdev min. niax. n mean stdev min. max.

1-1' TN 'I'N :TP Sal A0 Vol Zmn Rt Tide Wshed For-P Ag-P Urb-P Pden km2 ! 50 06 173 O 1088! Table 1 mat.

i Finnish Esluaries i US Estuaries 8 1 uniis 1 n mean stdev min. max. 1 n inean stdev niin. max. Ch1 TP TN TN:TP Sa1 Ao Vol Zmn Rc Tide Wshed For-P Ag-P Urb-P Pden Table 2: Data for US estuaries included in the analysis. Abbreviations as in Table 1. QR is the fieshwater loading rate. Data sources indicated by superscnpts where 1) US EPA (n.d.), 2) NOAA (1996, 1997), 3)calculated, see text, 4) HUC**, 5) Population Document** The symbol, *, indicates estuaries where STORET data was averaged over the period 1980-85; these sites were not included in the land-use modeling. The estuaries indicated by the symbol t are those for which data for al1 variables were derived from **. Delaware inland Bays; those estuaries indicated by $ are those for which data for al1 variables were taken fiom Boynton et al. 1995. Estuary NOAA ~til' TP' TN' sa12 2inn2 AO* vol2 3 code mg in nig in3 mg m3 ppi 111 km' 10' rn3 Long Island Sound Hudson River Delaware Bay Chesapeake Bay Potomac Chester Albermarle / Pandico Sound Pamlico-Pungo Neuse Rivcr Bogue Sound New River Cape Fear Charleston Harbour Assa t eague Chincoteague Indian River Rehobolh Assateague (2) Chincotcague (2) Isle of Wight NewPorî Sinetuxent 16.7 Table 2 cont.

Estuary NOAA Q: ~t' tide2 l%r-p4 A~-P~LJrb-p4 wshed2 ~deti' code ni3 sec-' YYS 111 % % Yo kni2 kni2 Long lsland Sound Hudson River Delaware Bay Chesapeake Bay Potomac Chester Albennarle / Patnlico Sound Pamlico-Pungo Neuse River Bogue Sound New River Cape Fear Charleston Harbour Assateapuc Chi ncoteague lndian River Relio bo th Assateague (2) Chincoteague (2) Isle of Wight NewPot-î

Sinetuxent - - Table 3: A comparison between water residence times calculated in this study and those published in the literature. Note, the values âorn Nixon et al. 1996 are âeshwater replacement times and thus higher than the other estimates. They are included as a reference point.

-- -- Estuary This ~tu& Litmturr value Source yrs yrs 3 Delaware (M090) 0.25 0.22 Fisher et al. 1988 Delaware (M090) 0.32 Nixon et al. 1996 Chesapeake (M 110) 0.50 0.55 Fisher et al. 1998 Chesapeake (M 1 20) 0.59 Nixon et al. 1996 Potomac (M120b) 0.13 0.23 Bemettt et al. 1986. Potomac (M120b) 0.43 Nixon et al. 1996 Albennarle / Pamlico (SOI 0) 0.49 0.92 Christian et al. 1991 Neuse (SO 1 Ob) O. 19 O. 14 Christian et al. 1991 Table 4a: Ch1:TP regression equations for system-specific data sets and the combined data where n is the sample size, bo is the intercept, b1 (TP) is the regression coefficient of logloof TP, $ is the coefficient of determination, SEE is the model standard error of the estimate and p is the probability associated with the model. Data set n bo bl (TP) f SEE p

Lakes --9-1 -0.01 0.72 0.68 0.259 0.0001 Finnish estuaries 19 -1.03 1.26 0.68 0.173 0.0001 US estuaries 17 0.02 0.58 0.49 0.165 0.0018 Combined Data 58 0.036 0.62 0.62 0.224 0.0001

Table 4b: Ch1:land-use regression equations for system-speci fic data sets and the combined data where bo is the intercept, b 1(Zmn) is the partial regression coefficient of logloof mean depth (Zmn), b2(For-P) is the partial regression coefficient of the logloof percentage of the catchment forested +l. Other abbreviations are as in Table 4a.

Data set n bo bl(Zmn b2(For- i SEE P

Lakes ,,77 3.45 -0.83 -1.55 0.84 0.189 0.0001 Fimish estuaries 11 5.54 -0.97 -2.14 0.75 0.106 0.004 US estuaries 16 4.04 -0.39 -1.54 0.56 0.156 0.005 Combined Data 49 4.43 -0.55 -1.7 0.80 0.170 0.0001 Table Sa: Standardized goodness of fit criteria for accuracy (SMSR), precision (SVSR) and bias (sME) for the Ch1:TP models for the combined and system-specific models, as applied to the cornbined and individual data-sets. Mean log Ch1 values are included as îhey varied among data sets; n is the sample size.

Data Set: Combined Lake Finland US mcm lChl 0.960 0.770 0.990 1 .130 Criteria: ModeI: (n=5 8) (n=22) (n= 1 9) (n= 1 7) Accuracy Combined 0.049 0.070 0.043 0.027 (MW Lake 0.064 0.06 1 0.062 0.07 1 Finland O. 143 0.381 0.027 0.093 US O. 123 O. 179 O. 102 0.073 Precision Combined 0.004 0.006 0.003 0.002 (vSR) Lake 0.006 0.004 0.005 0.008 Finland 0.073 O. 156 0.00 1 0.01 1 US 0.020 0.032 0.017 O .O03 Bias Combined 0.007 0.079 -0.024 -0,052 (ME) Lake -0.115 -0.004 -0.155 -0.214 Finland 0.090 0.388 0.004 -0.199 US 0.273 0.337 0.243 0.220 Table Sb: Standardized goodness of fit criteria for accuracy (SMSR), precision (SVSR) and bias (sME) for the Ch1:land-use models for the combined and system-specific models, as applied to the combined and individual data-sets. Mean log Ch1 values are included as they varied among data sets; n is the sarnple size.

Data Set: Combined Lake Finland US mean lChl 0.994 0.825 0.849 1.170 -Criteria: Model: (n=49) (n=22)- (n= 1 1) (n= 16) Accurac y Combined 0.027 0.036 0.0 13 0.024 (MW Lake O.045 0.03 1 0.02 1 0.08 1 Finland 0.058 0.058 0.008 0.090 US 0.032 0,045 0.0 19 0.02 1 Precision Combined 0.00 1 0.00 1 0.000 0.001 (vSR) Lake 0.005 0.00 1 0.001 0.0 12 Finland 0.0 1 O 0 .O08 0.000 0.01 7 US 0.002 O .O03 0.000 0.00 1 B ias Combined -0.001 O .O25 -0.008 -0.033 (ME) Lake -0.091 -0.001 -0.109 -0.202 Finland -0.008 0.120 0.005 -0.194 US -0.018 -0.030 -0.032 0.008 Figure 1: Chlorophyll (Chl) vs. total phosphorus (TP) for the combineci data set of Mes (filled circles), Finnish estuaries (open circles) and US estuaries (triangles). The solid line is the regression he,r2 is the coefficient of determination, see is the mode1 standard error of the estimate and n is the sample size.

log Ch1 = 0.36 + 0.62 log TP r' = 0.62 see = 0.224 n = 58

0.0 0.5 1.O 1.5 2.0 2.5 log (TP (mg mm3)) Figure 2: Observed chlorophyll (CM)vs. predicted Ch1 for the land-use mode1 generated nom the combined data where Zmn is the mean depth For-P is the percentage of the catchment forested. The doned line is the 1 :I line. Other abbreviations and symbols as in Fig. 1.

log Chl = 4.43 - 0.55 log Zmn - 1.70 log For-P ? = 0.80 see = 0.170 n = 49

log (Ch1 (mg mD3))- predicted Figure 3: Ch1 yield per unit TP as a function of log TP for the combined data. Symbols as in Fig. 1.

0.0 0.5 1.O 1.5 2.0 2.J log (TP (mg m-3)) GENERAL CONCLUSION

Three hypotheses were the focus of this thesis: 1) that estuarine eutrophication can be accurately predicted from land-use using linear regression models, 2) that lakes and estuaries respond similarly to nutrients and land-use, and 3) that an approach rooted in empirkal regression models and mass-balance rnodels is equally effective across aquatic systems.

Predictiog estuarine eutrophication from land-use: support for co-limitation and tree plaoting. In al1 the studied estuaries, land-use accurately predicts Chl. In PEI, the land-use mode1 account for 68% of the variation in Ch1 (II). In Finland (IV) and the US (V), land- use models account for 75% and 56% of the variation in Ch1 respectively. While in PEI, the land-use, TP and TN models account for a comparable arnount of variation in Ch1 (68%. 65% and 72% respectively). in Finnish and US estuaries, the land-use models were more accurate than the Ch1:TP models (75% vs. 68% and 56% vs. 49%). This result was expected as algal biomass is likely determined by a number of factors. For instance, the relative importance of nitrogen and phosphorus is unclear: estuaries may appear to be limited by primdy nitrogen (Howarth 1988) or phosphorus (Krom et al 1991, Lebo and Sharp 1992) but seasonal switches between N and P limitation are increasingly demonstrated (D'Elia et al. 1986, Pemock and Sharp 1994, Elmgren and Larsson 1997). Turbidity is also fiequently cited as controlling Ch1 (Mallin et al. 1993, Fisher et al. 1988, Cloem 1998,). The greater accuracy of the land-use models (in general), is support for the hypothesis that Chl is multiply detemined at the ecosystem and growing season scale. The cornparison of TP and land-use models in lakes (1) also demonstrated that land-use predicts Ch1 more accurately than TP. The irnprovement in accuracy (30%) in part reflects a reduction in error associated with chaining models: the land-use models allow a 1-step prediction of Ch1 whereas the TP mass-balance approach requires 2 steps. However, the improvement also likely reflects the inclusion of other factors that are important in controlling lake phytoplankton such as TN (Prairie et al. 1989) and turbidity (Rowan and Kalff 1991). This suggests that despite its paradigrnatic position, phosphorus may not be the only "cause" of fieshwater eutrophication. That phosphorus is seen as the primary limiting nutrient in lakes may reflect a confusion of cause and control discussed by Vallentyne (1974) and in the introduction. P-control through sewage treatment and refomied agriculture may drive systems to P-limitation and thus result in decreased Ch1 levels but this does not demonstrate that P is the only "limiting factor" This conclusion is not new. Other researchen have argued that not al1 hshwater syçtems are phosphorus limited. Reservoir phytoplankton may be light limited (Canfield and Bachman 198 1). Nitrogen limitation in eutrophic lakes has been invoked to explain variation in Ch1:TP relationships and residual error (V. Smith 1982, McCauley et al. 1989, Prairie et al. 1989). However, such analyses suggest that within a data set of many lakes, different factors are important in different lakes. They do not allow for CO- limitation of phytoplankton biomass within a system as do the land-use models. Co-limitation occurs when two or more nutrients act synergistically on biomass such that the biomass is greater in the presence of both nutrients than in the presence of one or the other. Most evidence for CO-limitationinvolves a macronutrient and micronutrient e.g. nitrogen and nickel (Pnce and More1 199 1) or a nutnent and light (Pennock and Sharp 1994). However, evidence for CO-limitationin lakes and estuaries is abundant. Elser et ale's (1990) review of bioassays fiom 68 lakes and 80 years of whole- lake manipulations demonstrated that significant responses occur primarily when N and P are added as opposed to N or P alone. For instance, of the 17 lake years where P alone was added, only 2 (12%) showed a significant response. Of the 49 lake yean where P and N were added, 39 (80%) showed a significant response. In this light, Schindler's (1977) whole-lake experiment demonstrating limitation by P has to be reconsidered: the only response occurred when C, N and P were added while the addition of P alone resulted in no significant increase (Fee 1979). Such a result suggests CO-limitationby N, P and C on the scale of whole systems and gmwing seasons as opposed to P limitation alone. In estuaries, limitation by N and P or by P and silica is increasingly reported (Dodds and Priscu 1990, Fisher et al. 1992) as well as seasonal switching (Del Arno et al. 1997, D'Elia et al. 1986, Gallegos and Jordan 1997, Granéli et al. 1990, Mallin and Paerl 1994). While such seasonal switching does not indicate CO-limitationof individual algal cells, it does suggest system-wide CO-limitationover the course of a growing season. The tem "CO-limitation" suffen however fiom the same problem as the "limiting factor" concept: they are both poorly defined. As with "limiting factors", it is important to identiQ the scale at which CO-limitationis being invoked. For instance, instantaneous CO- limitation of an individual or population at a given moment in time is difficult to imagine. Liebig's law of the Minimum, as was intended, appean to adequately address limitation at this scale. However, once one shifts scale from either a population to a community or from a given moment in time to a growing season, CO-limitationappean reasonable. For instance, within a community of phytoplankton, specific nutrient requirements vas, (Reynolds 1984) suggesting at a given moment in time, species are limited by different elements. On longer time scales, nutrient availability varies thus the nutrient status of a given species will Vary. Co-limitation remains however an unpopular concept with only 4 references to in the biological abstracts (01.1995- 12.1997). This unpopularîty may reflect statistical difficulties in its testing: for instance, in the case of P and N, covariance is high thus multiple regression models including both variables are not robust (Prairie et al. 1989). Goldman (1972) also expressed concems with respect to the quantitatively modelling of situations in which two or more nutrients are simultaneously limiting. While Tilman's (1977) resource cornpetition mode1 addresses situations in which different populations of the cornrnunity are differentially nuirient limited, his analyses are restricted to two populations. The consideration of 5- 10 populations and greater than 2 limiting nutrients soon becomes a modelling nightmare. In these difficulties lie the advantage of the land- use models: by integnting factors determining Ch1 levels, they avoid the statistical problems of covariance and other modelling problems. The most obvious "weakness" ofthe land-use models is that they do not indicate which nutrient(s) to reduce and by how much in order to achieve a given trophic status. However, that this is perceived as a weakness retlects our focus on engineering solutions that address single causes e.g. sewage treatment facilities that reduce point source loading. While such solutions are important, and in the case of the point source dominated Finnish estuaries (IV) the appmpriate management response, addressing nonpoint source loading remains an outstanding problem. In the lake, Finnish and US estuary models, percentage forest is consistently the most important predictor of Chl. In other words, reducing eutrophication requires increasing the amount of green space in our watersheds. We know this. Buffer strips are a key environmental management tool in both forestry and agriculture (Petejohn and Correll 1984. Lowrance et al. 1984, Kmg 1993. Barling and Moore 1994). Admittedly, increasing the area of forest in coastal watersheds is a daunting task. This challenge should not however lead to the exclusion of such variables fiom models.

Al1 water is wet: are estuaries simply salty Iakes ? In general, the relationship between Ch1 and TP, and Ch1 and land-use are similar across aquatic systems despite differences in tidal energy, openness and salinity (V). That these systems generally respond similarly is further supported by the fact that the same variables (percentage of the catchment forested and mean depth) were the best predictors of Ch1 for models generated for the independent data sets (V). Ln addition, TP in Finnish estuaries was accurately estimated from a lake-based phosphorus mas-balance mode1 (IV). Significant differences among systems. such as between lakes and PEI estuaries, could be explained by pan-system characteristics such as herbivory and twbidity (III). Although no significant differences existed among the lakes, Fimish and US estuaries, there was a suggestion ofa difference among lakes and estuaries in terms of their Ch1:TP relations. This potential difference is likely a function of trophic status: oligotrophic systems tend to P limitation with high Ch1:TP yields while eutrophic systems tend to N limitation with lower Ch1:TP yields regardless ofsysiern type (Howarth et al. 1995, Downing 1997). Such results suggest that lakes and estuaries per se respond similarly to TP and land-use; existing differences are a function of pan-system properties (e.g. herbivory) rather than systern-specific properties such as tidal energy and salinity.

The application of empirical approaches to estuaries: whither complexity ? Simple and elegant models such as empirical regressions and mass-balances effectively predict Ch1 in estuaries (II, III, V) despite complex hydrodynamics and water chemistry. While the ermr associated with these modeis, in arithrnetic terms, is still larger than one would desire, these models provide the most accurate Ch1 estimates available for most of these estuaries. These analyses suggest that cornplexity, if it exists, is not important on the scale of whole estuaries and growing seasons with respect to the prediction of Chl. These analyses also beg the question as to what one means by complexity. Peters (1991: 1 18) argues that the complexity of a system lies in the eyes of the beholder: if a system is deemed complex it is complex; if deemed simple, it is simple. This is because al1 systems are ultimately complex. Thus the oceanographer's argument that regressions and mass-balances work in lakes but not in estuaries due to the (implied) simplicity of the former is invalid (Richardson and Jergensen 1996). Lakes too can be complex (cf. the models of Jargensen 1995). Lakes differ from estuaries only in that simple rnodels have been tried. The analyses in this thesis, along with recent studies by Nixon et a1 (1996) and Boynton et al. (1996) demonstrate that empirical approaches work well in estuaries if one will but consider the systems arnenable to simplification. Empiricisrn is nevertheless criticized. Rigler and Peten (1995) identiS, two main "weaknesses" of empincism: 1 ) predictions are restricted to the correlated variables and thus do not yield unexpected predictions into other aspects of the entities considered; 2) empincism, saictly speaking, is not explanatory. The first weakness is real. It is then a question of whether one's preference is For restricted predictions or for insight. The second weakness is also real. One can not infer cause-and-effect fkom empirical patterns. However, cause is a quagrnire for both empincists and experimentalists seeking explanation (Peters 1991). Hume (as cited in Russell 1935) argued that, while the notion of cause is important, cause cannot be demonstrated. This hypothetical nature of knowledge means that the degree to which explanatory theories approach the tmth remains as unclear as the degree to which empirical theones approach truth. More concretel y, experimentalists are constrained to a limited nurnber of systems (replicates) and must simpli& systems in order to establish a control. The generality of their results across a mge of complex systems is thus unclear. Empiricists sacrifice control for realism. Moreover, data sets used by empiricists usually are biased in that representatives of al1 combinations of variables are not included. This makes it difficult to statistically tease apart the effects of CO-variatesand detemine which is ultirnately "causal". Thus explanatory and empincal theories ultimately suffer from the same constraint: we can never really know. Given this, it is perhaps best to evaluate theones under other cntena, such as predictive power (Peters 199 1). In remfor the explicit acknowledgement that there are no explanations, empirical approaches offer a great deai. As this thesis demonstrates, empincal models allow for accurate predictions of variables of concem for a wide range of systems. They serve where predictions are otherwise absent. Although error remains larger than is desirable, it is quantified. Thus, decision-makea know the degree of risk associated with decisions. Moreover, researchen can quantitatively assess the improvement offered by alternative models. These models are also relatively cheap to develop and simple to use thus are accessible to environmental managers. This is critical given the extent of eutrophication problems. What next ?

Theses seem to generally conclude with a list of future research topics. Such lists indicate humility on behalf of the doctoral candidate (i.e. one has not provided the final answer), demonstrate that the area of research is vibrant and fertile, and capitalize on the intense involvement of the doctoral candidate as she finishes by generating a number of future research projects ready to hand for incoming students. The danger of such lists (and the pressure to generate hem) is that one never moves on to new research problems as there are always more details to resolve (Peten 199 1). While academia appears to reward the step-wise resolution of problems on these continuously evolving lists, the magnitude and extent of environmental problems require a willingness to accept that an answer is suficient, and willingness to move on to new problems. My list is tempered by this latter concem. Unresolved problemî 1. Tidal energv range. Although Ch1:TP and Ch1:land-use models were compared among aquatic systems spanning tidal ranges fiom (0-2m),the analysis did not include macrotidal estuanes as defined by Monbet (> 2m, 1992). Thus, it is unknown whether, as Monbet (1992) argued, there is an effect of tidal energy of Ch1:TP or Chi: land-use relations. 2. I.iiriation in Chl:TP-~ieldr.1 have argued that the low yields in the PEI estuaries are a function primarily of herbivory and nirbidity. 1 also speculated that there may potentially be an effect of iron due to its ability to reduce the availability of phosphorus. The cornparison of lakes and estuaries suggests that there is a difference in yields in the iakes and the two types of estuaries and suggests that this is likely due to trophy. Altematively, it may reflect variation in phosphorus bioavailability. 3. Conversion factors for total and inorganic nutrient loah This thesis demonstrates the applicability of limnological tools to estuaries. 1 argued that one reason these tools have not been adopted by coastal researchen is the use of different currencies (total vs. inorganic nutrients and loads). Although ambient nutrient concentrations indicate linle about trophic status (e.g. low nutrient concentrations can indicate low Ch1 or high Ch1 that has taken up most of the nuîrient), this is not the case with inorganic nutrient loads. One might assume that the relationship between total and inorganic loads is constant such that lake models based on total loads and estuary models based on inorganic loads can be interconverted. However, individual systems may respond differently to total and inorganic loads, if for instance, bioavailability changes.

New problems listed in order of increasing distancefrom th is thesis 1. Predicting algal blooms. Although eutrophication refen to increased algal biomass, managers are frequently concemed by blooms, particularly if noxious or toxic. Regression models appear ill-equipped to predict blooms as they appear to perform best when based on longer tenn averages. Such models also require sampling programs that are suficiently temporally intensive in order to catch blooms of short duration. There has however been some success predicting maximum Ch1 levels (Harris 1986) suggesting that, with sufficient data, such peaks can be predicted. 7 Predicting iypoxia und anoxia. Like algal blooms, hypoxia and anoxia are key concerns of environmental managers. While models predicting hypolimnetic oxygen demand exist for lakes (Comett and Rigler 1979), no similar models exist for estuaries. 3. Predicting population characteristics of endangered species: a conservation application. This thesis focuses on a community level property (Chl). However, it should also be possible to use regession models to predict charactenstics of populations within a given species in response to factors such as habitat quality or exploitation. Such models could provide support for conservation decisions. These models, as in the case of the eutrophication models, would be accurate, cheap, simple, and quantify error, key characteristics given the rate of species decline. Literature cited:

Barling, R.D. and Moore, 1.D. 1994. Role of buffer strips in management of waterway pollution: a review. Environ. Manag. 18(4): 543-558. Boynton, W.R.;Murray, L.; Hagy, J.D.; Stokes, C.; Kemp, W.M. 1996. A comparative analysis of eutrophication patterns in a temperate coastal lagoon. Estuaries. 19: 408-42 1. Canfield. D.R.& Bachmann. R. W. 198 1. Prediction of total phosphorus concentrations. chlorophyll-a, and secchi depths in natural and artificial Mes. Cm. J. Fish. Aquai. Sci. 38: 414-423. Cloern, LE. 1998. The relative irnportancew of light and nutrient limitaiton of phytoplankton growth: a siple index of coastal ecosystem sensitiity to nutrient enrichment. J. Aquat. (fiorn Conley) Cornett, R.H.and Rigler, F.H.1979. Hypolirnnetic oxygen deficits: their prediction and interpretation. Their prediction and interpretation. Science 205: 580-82. D'Elia, C.F., Sanders, J.G.; Boynton, W.R. 1986. Nutrient enrichment studies in a coastal plain estuary: phytoplankton growin in large-scrle, continuous cultures. Can. Fish. Aquat. Sei. 43:397-406. Del Arno. Y.; Le Paper, O.; Tréguer, P.; Quéguiner, B.; Ménesguen, A.; Aminot, A. 1997. Impacts of hi&-nitrate freshwater inputs on macrotidal ecosystems. 1 Seasonal evolution of nutrient limitation for the diatom-dominated phytoplankton of the Bay of Brest (France). Mar. Ecol. Prog. Ser. 16 1 :213-224. Dodds, W.K. and Priscu, J.C.1990. A cornparison of methods for assessrnent of nutrient defiency of phytoplankton in a large oligotrphic Iake. Can. J. Fish. Aquatic Sci. 47: 2328-2338. Downing, I.A. 1997. Marine nitrogen: phosphorus stoichiometry and the global N:P cycle. Biogeochemistry 37:237-252. Elser, J.J.; Marzolf, E.R.;Goldman, C.R. 1990. Phosphorus and nitrogen limitation of phytoplankton growth in Ereshwaters of North Amerka: a review and critique of experimental enrichments. Can. J. Fish. Aquatic Sci. 47: 1468- 1477. Fee, E. 1979. A relation behveen lake morphometry and primary productivity and its use in interpreting whole-lake eutrophication experiments. Limnol. Oceanogr. 24(3):4O 1-4 16. Fisher, T.R.; Harding, Ir., L. W.; Stanley, D. W.; Ward, L.G. 1988. Phytoplankton, nutrients, and turbidity in the Chesapeake, Delaware, and Hudson estuaries. Estuarine. Coastal and Shelf Science 27: 6 1-93. Gallegos, C.L.and Jordan, T.E. 1997. Seasonal progressionof factors limiting phytoplankton pigment biomass in the Rhode River estuary, maryyland (USA). II Modeling N venus P limitation. Mar. Ecol. Prog. Ser. 16 1 :199-2 12. Goldman, C.R. 1972. The role of minor nutrients in limiting the productivity of aquatic ecosystems. In: Likens, G.E.(ed.). Nîttrients and eutrophicarion: the limiting- nutrient controversy. Special symposia Volume 1 . 328 pp. Granéli, E.; Wallstrom, K. Lanson, U; Granéli, W; Elmgren, R. 1990. Nutrient limitation of primary production in the Baltic Sea area. Ambio 19: 142-51. Harris, G.P.1986. Phytoplankm ecology: stmaure,function andfluctuation.London: Chapman and Hall. Howarth, R.W.1988. Nutrient limitation of net primary production in marine ecosystems. Ann. Rev. Ecol. l9:89- 1 10. Howarth, R. W.; Jensen, J.S.; Marino, R.; Postma, H. 1995. Transport to and processing of phosphorus in near-shore and oceanic waters. In: Tiessen H. (ed.). Phosphonu in the global environment. Chichester: John Wiley & Sons. Pp 323-345. Jergensen, S.E. 1995. State-of-the-art management models for lakes and reservoirs. Lakes and Reservoirs: Research and Management. 1 :79-87. Krom, M.D.;Kress, N.; Brenner, S.; Gordon, L.I. 1991. Phosphorus limitation of primary productivity in the eastem Mediterranean Sea. Limnol. Oceanogr. 36: 424-432. Krug, A. 1993. Drainage history and land use pattern of a Swedish river system: their importance for understanding nitrogen and phosphorus load. Hydrobiologia 25 1: 285-296, Lebo, M.E.and Sharp, J.H.1992. Modeling phosphorus cycling in a well-mixed coastal plain estuary. Est. Cocrstal SheifSci. 35: 235-252. Liebig, J. von. 1840. Organic chemistry in its application to vegetable physiology and agriculture. in: Kormondy (ed.). Readings in ecology. Englewood Cliffs, NJ: Prentice-Hall. Pp. 12- 14. Lowrance, R.; Todd, R.; Fail, J. Jr.; Hendrickson, O., Jr.; Leonard, RI;Asmussen, L. 1984. Riparian forests as nutrient filten in agricultural watersheds. BioScience 34(6): 374-377. Mallin, M.A. and Paerl, H.W. 1994. Commentary on primary productivity and nutrient limitation in the Neuse River Estuary, North Carolina, USA. Mar. Ecol. Prog. Ser. 11 l:3ll-312. Mallin, M.A.; Paerl, H. W.; Rudek, J.; Bates, P. W. 1993. Regulation of estuarine pnmary production by watershed rainfall and river flow. Mar. Ecol. Prog. Ser. 93: 199-203. McCauley, E.; Downing, J.A.; Watson, S. 1989. Sigmoid relationships between nutrients and chlorophyll among lakes. Can. J. Fish. Aquat. Sci. 46: 1 171 - 1 175. Monbet, Y. 1992. Controls of phytoplankton biomass in esniaries - a comparative analysis of microtidal and macrotidal estuaries. Estuaries. 15: 563-57 1. Nixon, S.W.;Ammerman, LW.; Atkinson, L.P.; Berounsky, V.M.; Billen, G.; Boicourt, W.C.; Boynton, W.R.; Church, T.M.; Ditoro, D.M.;Elmgren, R.; Garber, J.H.; Giblin, A.E.; Jahnke, R.A.; Owens, N.I.P.; Pilson, M.E.Q.;Seitzinger, S.P. 1996. The fate of nitrogen and phosphorus at the land-sea margin of the North Atlantic Ocean. Biogeochemistry 35 :14 1- 180. Pemock, J.R. and Sharp, J.H. 1994. Temporal alternation between light- and nuhient- limitation of phytoplankton in a coastal plain esniary. Mar. Ecol. Prog. Ser. 1 1 1:275-288. Petejohn, W.T.and Correll, D.L. 1984. Nutrient dynamics in an agricultural watenhed: observations on the role of a nparian forest. Ecology 14664475. Peters, R.H. 1991. A critique for ecology. Cambridge university press. Cambridge. Prairie, Y.T.;Duarte, C.M.;Kalff, J. 1989. Unimg nutrient-chlorophyll relationships in lakes. Canadian J. Fish. Aqurrt. Sci. 46: 1 176-82. Price, N.M. and Morel, F.M.M. 1991. Colimitation of phytoplankton growth by nickel and nitrogen. Limnol. Oceanogr. 36(6): 1071-1 077. Reynolds, C. S. 1984. The ecology of fresh water phytoplankton. Cambridge University Press. Cambridge. Richardson, K. & Jergensen, B.B. 1996. Eutrophication: defmition, history and effects. In: Richardson, K. & Jsrgensen, B.B. (eds.). Eutrophication in coastal marine ecosystems. Coastal and Estuarine Studies. Volume 52. Amencan Geophysical Union. Rigler, F.H.& Peters, R.H. 1995. Science and Limnology. Ecology Institute. Oldendorfhhe, Germany. Rowan, D.J. & Kalff, J. 199 1. The limnological implication of catchent sediment load. Verh. internat. Vereirr. Lintnol. 24: 2980-2984. Russell, B. 1935. Sceptical essa-vs. London: George Allen and Unwin. Schindler, D.W.1977. Evolution of phosphorus limitation in lakes. Science. 195: 260- 262. Smith, V.H. 1982. The nitrogen and phosphorus dependence of aigal biornass in lakes: an empirical and theoretical analysis. Limnol. Oceanogr. 27: 1 101- 1 1 1 1. Tilman, D. 1977. Resource cornpetition between planktonic aigae: An experimental and theoreticai approach. Ecology. 58: 338-348. Vallentyne, J.R. 1974. The nlgd bowl: lakes und man. Ottawa: Department of the Environment, Fisheries and Marine Service. Miscellaneous Special Publication 22. Appendix 1: Data for lakes used in Chapters 1 and 5 where site is the lake narne, code indicates my numbering system and Ch. indicates whether the observation was used in Chapter 5 (x) or not. Variables include surface area (Ao), lake volume (Vol), water residence time (Rt), mean depth (Zrnn), maximum depth (Zmx), chlorophyll (Chl), total phosphorus (TP), total nitrogen (RI), total phosphorus load (TPLD), total nitrogen load ('TMD),percentage of the catchment forested (For-P). agicultural (Ag-P), urban (Urb- P), natural (forest and al1 other natural veas (e.g. rocky areas, manhes; Nat-P) and other areas (Other-P), watershed area (Wshed), and human population density (Pden). Data Sources: Lake Biwa Institute ( 1988, 1989, 1990). Site Code Ch O Vol Rt Zmn Zmu Ch1 TP TN km' 1obm3 yrs m m mg m.' mg m" mg m.' Lough Neagh Mus ko ka Kootenay W illiston S. [ndian Lake Great Cennril Buhi Shurnarinai-ko Hac hiro-gata Koj ima-ko Ogawara-ko S hinj i- ko Ro torua Neusiedlersee ,Malaren Hjaharen Vattem Vanem Inari Pielinen Paijanne Paajarvi Lough Ree Lough Derg Ammersec Starnberger sec Trasirneno Sniardwy Lac St. Jean Canandaigua C~YW Chicot O kcec hobec Upper Twin Valencia Lowcr Twin San Roquuc Songkhla Phcwa Chuzcnj i- ko Nagase Chao Miyun Kinncrct Laguna de Bay Inawashiro Tasek Bcra Mashu Ogochi Towado Kawaguc hi 297 x 6 55500 0.31 9.3 16.1 7.03 14 JO8 Tega 298 7 5.6 0.07 0.56 3.8 365 500 4867 Inba 299 12 277 0.08 1.7 2.5 140.33 100 1932 Suwa-ko 300 13 63 0.11 4.7 7.2 50.17 100 1360 Kiza ki 301 x 1 2 5 0.5 17.9 29.5 2.33 13 345 Taupo 302 616 60000 10.6 91 164 3.02 5 Balaton 305 593 1900 2 3.25 12.2 18.42 100 IO75 Zurichsee 307 65 3300 1.1 5 1 136 4.15 16 Ness 308 56 7450 2.81 132 230 1.2 1O Victoria 309 68800 3E+06 23 40 84 15.8 Ontario 311 19009 ZE+06 7.9 86 223 5.25 Washington 312 88 2890 3.1 32.9 65.2 7 73 269 Represa de lob0 3 13 x 7 220 0.14 3 12 16.83 14 Site Code TPLD TNLD For-P Ag-P Crb-P Nar-P Other-P Wshed Pden lob kg lobkg yr-' km' fi' Lough Neagh >luskoka 94.5 1 Koo tenay 99 0.3 W illiston O O S. Indian Lake 100 O Great Ccnml 100 O Buhi 39.4 38.9 Shumahai- ko 72.3 1.53 Hachuo-gata 49.6 32.6 Kojuna- ko 38 31.6 Ogawara-ko 47.9 20.6 Shinj i- ko 72.2 13.8 Rotorua 33.4 O Neusiedlersee 16.2 24.5 Ma laren 64.5 25.4 Hjalrnaren 51.9 25.4 Vattern 62.9 26.7 Vanern 67.9 18.5 Inari 89.1 0.02 Pielinen 56.6 6.1 Paijanne 78 'IO Paajarvi 56.5 17.6 Lough Ree 5 5 Lough Derg 5 5 Ammersee 31.6 17.5 Starnberger sec 36.4 15.3 Trasimcno 25.1 65.7 Sniardwy 36.7 35.9 Lac St. Jean 75.2 0.9 Cananhigua 49.7 47.6 Wuga 37.1 58 Chicot O 90 Okcechobee 7.5 33.4 Upper Twin 99 Valencia 22 20 Lowet Twin 99 San Roquuc 48.8 3.4 Songkhla 22.5 73 Phewa 27 18 Chuzenji-ko O 0.3 Nagase 98.6 1.4 Chao 6.8 62.1 Mi yun 27.7 7.5 Kinnerct 3.7 38.5 Laguna de Bay 23.8 52 Inawashiro 87.2 9.8 Tasek Bera 90 O khu 100 O Ogochi 94.6 1.7 Towado 99 O Kawaguchi 297 0.016 0.125 79.9 5.6 7.1 79.9 7.4 120 171.5 Tega 298 0.112 0.865 O 33.4 39.3 26.4 O 150 2657 Inba 299 0.144 1.472 O 42.5 23.8 27 O 487 1109 Suwa-ko 300 0.111 0.815 O 12.8 5 71.9 10.2 15 339.4 Kizaki 301 0.0008 0.014 O 5.1 2.9 84.4 7.6 22.4 63 Taupo 302 0.116 0.657 68 O O 68 31 3327 7.7 Balaton 305 0.314 3.148 25.9 46.9 8.4 27.9 16.7 5181 64 Zuric hsee 307 0.128 23 53 5 37 6 1740 217 Ness 308 92.6 1.9 O 92.6 5.1 1775 1.8 Victoria 309 15 23 6 71 O 184000 170 Ontario 311 55.8 10.6 3.5 85.1 0.8 75272 94.8 Washington 3 12 30 10 60 30 O 1274 1180 Represa de lob0 3 13 20 20 10 65 5 280 96.4 Appeadix 2: Raw data for sampling stations in Prince Edward Island, May - .4ugust, 1996. SR = sampling round ( 1-6 begiming May 5 and continuing approximately bi- weekly), SS = sampling station (1-5 where 1 is most seaward station and 5 is most landward; stations located at equal distances), Chi = chlorophyll-a, TP = total phosphorus, TN = total nitrogen, Zrnx = maximum depth at sampling location, Secchi = Secchi depth, Tl = temperature 0.5 m below surface. T2 = temperature 0.5 m above bottom, S 1 = salinity 0.5 m below surface, S1= salinity 0.5 m above bottom. Site SR SS Ch1 TP TN Zmx Secchi Tl T2 S1 S2

rngm*' mgm-' mgm" m m OC "C ppt ppt Boughton 223.9 1.25 1.25 3.5 4.5 26 26 Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton Boughton 6 3.828 56.3 288.0

Brudencil Brudendl Brudencll Brudencll Brudenel1 Brudeneii Brudeneu BrudencH Brudencll Brudencii BrudencU BrudeneIl Brudendi Brudencll Brudcncll Brudcncii Bntdcncii BrudencU Brudendl Site SR SS Ch1 TP TN Zmu Secchi TI T2 SI S2 mg me' mg m.' mg rn-> m m OC OC ppt ppt Brudenel 3 4 2.109 68.3 128.7 5 4.2 14 9.5 24.5 22.5 Bnidenell Brudenel1 Brudeneil Brudeneil Brudeneil Brudeneil Brudenell Brudene 11 Brudeneil 6 5 2.855 6 384.2 4 2 18.5 17.5 14.5 18.5 Cardigan 1 1 1.067 81.2 Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan Cardigan -C;irdigan Dunk Dunk Dunk Dunk Dunk Dunlr Dunlt Dunk Dunk Dunk Site SR SS Ch1 TP TY Zmx Secchi Tl T2 S1 S2 mg m" mg m" mg rn" m m 'C OC ppt ppt

Dunk 5 ' 5.093 5 245.0 i 1 20 20 18 18 Dunk Dunk Dunk Dunk Dunk Dunk Dunk Dunk Dunk Dunk Dunk Dunlc DudC Dunk Dunk Dunk Dunk Dunk

Dunk 6 - - - - - 0.5 20 . 16 . Darnley Basin 1 0.413 57.5 251.8 Darnley Basin Dadey Basin Damiey Basin Darniey Basin Darniey Basin Dadey Basin Dadey Basin Darnley Basin Darnley Basin DarnIcy Basin Darnley Basin Damley Basin Danilcy Basin Darnley Basin Darnley Basin Darnley Basin Darnley Basin Darnley Basin

Foxicy Foxley Foxky Foxlcy Foxlcy Foxley Foxky Foxiey Foxky Foxicy Site SR SS Ch1 TP TN Zmx Secchi Tl T2 SI S2 mg m.' mg m" mg rn-' rn rn 'C OC ppt ppt Foxley 5 2 3.130 65.6 191.1 9.5 1.5 20 19 17.5 18 Foxle y Foxley Foxley Foxley Foxley Foxley Foxle y Foxle y Foxle y Foxle y Foxley Foxiey Foxley Foxley Foxley Foxley Foxlcy 6 5 1.154 56.4 310.5 1.75 1.75 21 20 14 15 Grand River I 1 0.483 17.0 147.3 6.5 4.33 12 10 22.5 22.5 Grand River Grand River Grand River Grand River Grand River Grand River Grand Rivcr Grand River Grand Rivcr Grand River Grand River Grand Rivcr Grand River Grand Rivcr Grand River Grand River Grand River Grand River Grand Rivcr Grand River Grand Rivcr Grand River Grand River Grand River Grand River Grand River Grand Rivtr Grand River Gmd River 6 5 0.488 37.8 253.4 Munay Harbour 1 1 0.608 52.1 392.1 6.25 4.7 5 4 25 24 Site SR SS Ch1 TP T'Fi Zmu Secchi Tl T2 S1 S2

mg KI^' mg m-' mg m*' m rn OC 'C ppt ppt Murray Harbour Mumy Harbour Munay Harbour Murray Harbour Mmy Harbour Murray Harbour Murray Harbour Murray Harbour Murray Harbour Murray Harbour Munay Harbour Mwray Harbour Murray Harbour Murray Harbour Murray Hrirbour Murray Harbour MmyHarbour Mumy Harbour Murray Harbour Murray Harbour Murray Harbour Murray Harbour MmyRwbour Murray Harbour Mwray Harbour iMurray Harbour ~MurrayHarbour .Murray Harbour Mill River 1 1 1.260 Mill River Mill River Mill River MiH River Mill River Mill River Mill River Mill River Mill River Mili River Mill River Mill River Mill River Mill River Mill River Mill Rivcr MiH River Mill River Mill River Miu Rivcr Mill River Site SR SS Ch1 TP TN 2m.x Secchi Tl T2 Sl S2 mgm-' mgm" mg m" m m "C 'C ppt ppt ,Mill River 5 4 2.822 34.4 312.4 5.5 1.5 22 20 14.5 17 Mill River 6 4 5.839 96.9 378.8 2.75 1.25 20 18.5 15 15.5 .Mill River 1 5 4.766 70.8 548.0 3.5 2.5 9.5 9.5 13.5 19 Mill River 2 5 0.968 99.8 505.9 3 2.5 16 16 17.5 17 .Mill River 3 5 4.486 f 14.9 432.3 2.5 L 19.5 19 15.5 17.5 Mill River 4 5 2.621 t35.9 420.3 2 1.9 13 13.5 21.5 21.5 Mill River 5 5 4.031 35.3 263.3 3.5 1.25 22 21 13 15.5 Mill River 6 5 12.702 75.6 340.5 7.5 1 18.5 17.5 16 17 North Lake 1 1 1.703 Sorth Lake North Lake North Lake North Lake North Lake North Lake North Ldce North Lake Llorth Lakc North Lake North Lake North Lake North Lake North Lake North Lake North Lake North Lake North Lake North Lakc North Lakc North Lakc

North Lakc 243.6 - 2 10 9 20 20.5 Pcrcival 11 Pcrcival Pcrcivai Percival Pcrcival Pcrcival Percival Percival Pcrcival Pcrcival Pcrcival Pcrcival Percival Pcrcival Percival Percival Percival Percival Percival Site SR SS Ch1 TP TN Zmx Secchi Tl T2 Sl S2 mg m3 mg m" mg m.' m m 'C OC ppt ppt Percival 2 4 0.692 318.5 0.5 0.5 19 19 19.5 20 Percival Percival Percival Percival Percival Percival Percival Percival Percival Percival 6 5 2.855 79.3 372.9 1 1 19 . 11.5 . Rustico 1 1 0.970 f 14.0 356.0 3.5 2.5 10 10 23 23 Rustico Rustico Rustico Rustico Rustico Rustico Rustico Rustico Rustico Rustico Rustico Rustico Rustico Rustico ~ushco Rustico Rustico Rustico Rustico Rus tico Rustico Rustico Rwtico Rustico Rustico Rustico Rustico Rustico Rustico 6 5 4.878 24.3 355.8 1.5 1.4 19 . 16.5 . St. Pcters 1 1 1.042 45.9 407.4 4.5 2.5 10 9 22 23 St. Pctcn St. Peters St. Pctcn St. Pcters St. Pctcn SL Peten St. Pctcn St. Peters Site SR SS Ch1 TP T.rl Znx Secchi Tl T2 Si S2

St. Peten 4 2 1.673 173.4 324.1 5.5 3.5 St. Peters St. Peten St. Peten St. Perers St. Peten Sr. Peters St. Peters St. Peters St. Peters St. Peters St. Peters St. Peters St. Peters St. Peters St. Pcters St. Peters St. Peters St. Peters St. Petcrs St. Peten 6 5 1.723 92.1 263.0 3.75 2.2 19.5 18 16 17.5 Savage Harbour 1 1 0.682 73.9 356.1 2 2 7.5 6.5 23 23.5 Savage Harbour Savage Harbour Savagc Harbour Savagc Harbour Savage Harbour Savage Harbour Savagc Harbour Savagc Harbour Savagc Harbour Savagc Harbour Savage Harbour Savagc Harbour Savage Harbour Savagc Harbour Savagc Harbour Savage Harbour Savagc Harbour Savage Harbour Savage Harbour Savage Hiubour Savagc Harbour Savage Harbour 3 5 2.079 73.1 118.0 3.5 1.5 20.5 19 20 19.5 Wilrnot 1 1 1.111 606.9 3 1.5 10 IO 21 23.5 W ilmot 2 1 6.927 94.1 253.7 2.25 1 17 17 21.5 21.5 W ilmot 3 1 2.074 105.3 334.8 1 0.9 18.5 18 17.5 19 W ilmot 4 1 2.459 90.4 195.9 Wilmot 5 1 4.407 127.2 205.3 2 0.5 20 . 16 . Wilrnot 6 1 4.588 61.0 220.6 3 2 20 19.5 18.5 19 Site SR SS Ch1 TP TN Zmu Secchi Tl T2 S1 S2

Wiimot 1 2 !.165 45.3 310.2 1.5 1.33 10.5 11 20.5 22.5 Wilmot Wilmot Wilmot Wihot Wihot W ilrnot W ilmot Wiimot Wilmot Wihot Wihot 6 3 9.133 64.8 346.4 1.8 1.5 19.5 20 17.5 17.5 Appendix 3: Descriptive data for PEI estuaries inciuding growing season averages for chlorophyll (Ch!), total phosphorus (TP), total nitrogen (TN). TN:TP. Secchi depth, and surface and bonom temperatures (Tl,T?) and salinity (S 1, S2)). estuary morphometry as mean depth (Zmn), surface area (Ao), volume (Vol) and water residence time (Rt (days), and watershed /land-use as wateahed area (Wshed), areas of: potatoes, hay, grain. total agriculture (= sum of area of potatoes, hay and grain), other and forest, and population density (pden).

SITE Ch1 TP TN TN:TP Secchi Tt T2 mg IT? mg m.' mg m.' rn "C "C Boughton 1.398 Brudenel1 Cardigan Darnley Basin Dunk Foxley Grand River Murray Harbour Mill River North Lake Ptrcival Rustico St. Peters Savage Harbour W ilmo t min 1.097 45.8 209.0 3.1 1.1 13.7 11.2 rnax 5.915 94.8 480.9 8.7 4.1 18.3 17.6 mean 2.354 66.7 287.1 4.4 2.4 16.0 14.8 sd 1.437 15.7 78.6 1.4 1.O 1.3 2.1 SITE Sl S2 Zrnn Ao Vol Ret Mussels PPt PPt m rn2 106 m' days 10' go Baughton Brudenell Cardigan Darnley Basin Dunk Foxley Grand River Murray Harbour Mill River 'lorth Lake Percival Rus tico St. Peters Savage Harbow

min 15.5 17.3 2.2 1022272 2.3 3.00 2.03 max 21.4 22.9 18.0 35219478 350.8 356.00 9.63 rnean 19.1 20.1 8.7 9461811 98.1 129.00 5.80 sd 1.6 1.5 5.0 9247204 106.4 118.00 2.90 SITE Wshed Potatoes Hay Grain Total Other For Pden *p. h2 km' km? km' km' km? km'? Boughton Brudenell Cardigan Dmley Basin Dunk Foxley Grand lùver Murray Harbour &MillRiver North Lake Percival Rustiço St. Petcrs Savage Harbour W ilmot min 25.10 0.09 1.54 0.20 2.45 0.64 2.90 8.3 max 292.87 49.75 58.39 52.95 160.90 20.72 204.24 33.4 mcan 105.70 9.27 17.92 17.87 45.06 5.75 54.89 14.4 - sd 81.30 15.22 19.41 17.33 49.25 5.44 50.1 1 7.2 Appendix 4: List of Fimish estuary sampling stations used in Chapters 4 and 5 where site is the esmary, ID is the station id, Coord-X and Coord-Y are the coordinates used to locate the sampling station according to the coordinate system used by the Finnish Govemment. Site Station Name ID Coord-X Coord-Y 11 SUOMENL VIROLAHT 286 153 671082 354113 SUOMENL VIROLAHT 287 SUOMENL ViROLAHT 288 SUOMENL VIROLAHT 289 SUOMENL VIROLAHT 290 SUOMENL VIROLAHT 291 SUOMENL VIROLAHT 292 SUOMENL HAMINANI. 23 1 SUOMENL HAMiNANL 236 SCOMESL SL'MMASLX 192 AHVENKOS STROMMI SUOMENL AHVENKOSKYVY-9 SUOMENL PüROLANL 46 BJORKI~OL114 EMASALO ITA 5 EMASALON 10 EMASALON 24 HAiKONSE 12 HAKONSE 22 KUCiGSUND 25 ORRENKYL 8 PORVOONJ - 144 SKOLDVIKED. 30 SKOLDVMED-3 1 SKOLDVIKED. 36 SKOLDVIKED. 38 STENSEOL 16 SVMO LUODE 20 HEVY-1 1 PORVOO 32 HEVY-t 2 PORVOO 40 HEVY- !3 PORVOO 48 lLLVARDEKAAKK0 97 ILLVARDEKOILLINE 27 SILLW 1 16 SILLVlK 7 SKOLDVIKED. 33 SIPOONLA 59 SIPOONLA 61 SiPOONLA 64 VANHANKA4 HEW-16 POHJANLA 92 POHJANLA 12 POHJANLA 13 POHJ~2 POHJANLAETELAOSA 1 1 POHJANLAKESK. 5 POHJANLASTOR~1 HALA 100- HALA 105WOHENSAET HALA 1 1OFUtKKfLA HALA 1 15KAISAAIUPOW Site Station Name ID Coord-X Coord-Y 25 HALA 12OVILNIEMI 27 668522 244251 HALA 127TEIJONS LOUN HALA 130STROMM POHJ HALA 136sTRoM~ET HALA ~~~IURJHOLMLA HALA 130LEMUNS KOILL HALA 143KRUOPINNOKKA HALA t4JSTORA FURUHOLM HALA 2321SONOKKALOWN HALX ~SSL~GHOLICIET HALA 240KARHUS LUOT HALA 24SKOKKILANLOSSI HALA ZSOVARTSALASAHA PALA 1 1 STRYHOLM IT PALA 1 ZOPAMIONLWS PALA 124TAMMOSTAIT PIK IOSPRTTIKARI TLJRM 304PAPINLUOTO MYLA 3 1 SSIKALUOTIT MYLA 3 17SAARNINELOUN POME SOPUSSAANLUOTO POME 5 1 S~~SALARET POME 52WORIKEMIEHD POME 56KOLPPA POME 57REPOS SILTA POME ~~ETE~SEL~ POME 64LANNASK KOILL POME 67TAHKOL LUOT POME ~OKRISTISKU POME 7 1ARVENK POHJ POME 72ISO-VAKKLA POME 83ISOT Plokit La POME 86YYTERIN ED DICKSHOLITAP. SELKAMERFORELLIi VAV- 1 5 VU-3 HYLPETSZJND NABBVIKEN PRPiSTHotMSVIKEN TOlTESUM) VASSOR M 1 KOKKOLANEDUSTA X PE-1 KOKKOtANEDUSTA LIMINGAN 11 LlMINGAN 12 LIMTNGAN 13 LIMINGANPlO Appendix Sa: Chlorophyll (Chl) values for Finnish estuaries included in IV and V where site is the estuary (see Table 1, IV), station is the station code (Appendix 4), year is the sampled year, mean is the mean Ch1 value calculated by averaging al1 growing season values for the given station in the given year, n is the number of observations at the given station in the given year, and sd, min, mau and CV are the standard deviation, minimum, maximum and coefficient of variation of these observations respectively. Appendix Sa: Chlorophyll values Site station vear mean n SD min rnax CV Amendix Sa: Chloro~hvllvalues Site station year rnean n SD min max CV Appendix Sa: Chloro~hyildues Site station vear mean n SD min max CV 5.02 1.6177 2.7 7.2 32 Appendiv Sa: Chlorophyll values Site station year mem n SD min max CV Appendix 5a: Chlorophyll values Site station year mean n SD min max CV Appendi. 5a: Chlorophyll values Site station year mean n SD min max CV Appendk Sa: Chloro~hvlivalues Site station year mean n SD min rnax CV 42 Appendix Sb: Total phosphorus (TP) values for Finnish estuaries included in IV adn V; abbreviations as in Appendix Sa. 6 I Site station ~ear mean n SD min ma.. CV 153 1989 70.8 Append~~Sb: Total phosphorus Site station year mcan n SD min mx CV Appendiu 5b: Total uhos~horus Site station year mean n SD min max CV -- Appendk Sb: Total phosphoms Site station year mean n SD min max CV Appendix 5b: Total phosp homs Site station vear mean n SD min max CV Appendix 5b: Total phosphocus - Site station vear mean n SD min ma^^ CV 27 9 A~~end1.ujb: Total ~hosbhonis Site station year mean n SD min max CV -7 Appendix Sc: Total nitrogen (TN) values for Finnish estuaries inciuded in IV adn V; abbreviations as in appendix sa. A~ucndix* 6 Sc: Total NiuoeenY Site station year mean n SD niin max CV

1 ZOO 510 560 610 5 80 730 1 O00 Il00 640 620 700 660 1300 7 10 700 800 660 540 520 550 5 80 5 70 770 680 650 650 590 580 530 780 570 590 580 530 770 630 520 490 440 400 670 450 620 540 800 1100 600 490 1600 Appendix 5c: Total Nitrogen Site station year mean n SD min max CV Appendix 5c: Total Nitrogen Site station vear mean n SD min max CV 210 Appendix 5c: Total Nitrogen Site station vear mean n SD min max CV Appendix 5c: Total Nitrogen Site station vear mean n SD min ma.. CV 25 33 1989 440 -i 57 400 480 13 25 34 1989 428 8 90 270 5 10 21 25 36 1989 533 3 67 490 610 12 25 37 1989 612 5 275 450 1100 4 5 25 38 1989 538 4 15 520 5 50 3 25 25 1990 870 4 79 800 980 9 25 26 1990 568 5 19 550 600 3 25 28 1990 494 5 111 390 620 2 3 2 5 29 1990 540 3 130 460 690 24 Z 5 30 1990 490 3 57 430 590 1S 2 5 32 1990 497 3 194 370 720 39 25 36 1990 587 3 8 1 500 660 14 25 37 1990 648 5 203 510 1000 3 1 25 38 1990 560 4 70 500 660 12 25 25 1991 1405 4 516 940 1900 3 7 25 26 1991 618 5 64 530 710 10 25 28 1991 566 5 119 480 770 21 25 29 199 1 563 3 171 450 760 30 2 5 30 199 1 540 3 122 460 680 23 25 3 1 199 1 370 6 64 300 440 17 25 32 199 1 5 10 3 154 380 680 3 0 25 33 1991 455 2 7 450 160 2 25 34 1991 380 8 74 230 450 19 2 5 36 199 1 503 3 76 450 590 15 25 37 199 1 642 5 259 180 1100 40 25 38 1991 530 4 33 490 570 6 2 5 25 1992 988 4 103 850 1100 1O 25 26 1992 714 5 198 540 1O00 28 25 28 1992 782 5 248 420 1000 32 25 29 1992 433 3 12 420 440 3 2 5 30 1992 533 3 23 1 400 800 43 25 3 1 1992 330 3 61 290 400 18 25 32 1992 473 3 81 400 560 17 25 34 1992 355 4 87 270 440 25 25 36 1992 537 3 90 480 640 17 25 37 1992 698 5 249 380 1100 36 25 38 1992 705 4 1 60 470 830 23 25 25 1993 1670 4 869 680 2400 52 25 26 1993 1038 5 563 580 1700 54 25 28 1993 780 5 295 5 10 1100 38 25 29 1993 623 3 206 490 860 33 25 30 1993 50 3 44 450 530 9 25 3 1 1993 453 8 108 320 640 24 25 32 1993 570 3 243 410 850 43 25 33 1993 655 2 346 4 10 900 5 3 25 34 1993 426 10 75 340 550 18 25 35 1993 414 5 33 380 450 8 25 36 1993 513 3 42 480 560 8 25 37 1993 766 5 223 490 1100 29 25 38 1993 LOOS 4 486 560 l6QO 48 Apvend~xSc: Total Nitroeen - Site station year mean n SD min max CV 117 Appendix Sc: Total Nitrogen Site station year mean n SD min mx CV 58 7 1 1989 705 3 544 32 1 1327 77 Appendix 6a: Chlorophyll (Chl) values for Finnish esniaries where the mean value is the mean of al1 the station values (Appendix 5a) in a given year as described in Chapter 4 (Chl,,), n is the number of stations which were sampled in a given year, and sd, min, max and CV are the standard deviation, minimum, maximum and coefficient of variation of these station values. r - - r --a -- ED A YEAR FREO AVG2 STD2 MM2 MAX2 CV2 32.0 Appendt~6a: Chlorophy 11 EDA YEAR FREQ AVG2 STD2 MN2 MAX2 CVZ 29 93 1 1.83 4.83 Appendix 6b: Total phosphorus (TP) values for Fimish estuaries; abbreviations as in Appendix 6a. hppendix 6b: Total phosphorus EDA YEAR N MEAN SD MW MAX CV Appcndix 6b: Total phosphorus EDA YEAR N MEAN SD MIN WY CV

33.73 27.73 32.82 3 1.46 34.50 JO.O0 3 5 .4O 40.33 35.67 43 .O0 45.80 58.67 44.33 52.25 19.00 2 1.00 80.00 95.50 8 1 SO 93.50 110.50 22.33 29.50 20.78 25 J3 15.00 43.00 23.67 48.67 58.60 60.00 Appendix 6c: Total nitrogen (TN) values For Fimish estuaries; abbreviations as in Appendix 6a. ADD~~~x6c: Total nitroeen Y ED A YEAR FREQ AVG2 STD2 MIN2 MAX2 CV2 11 89 597.22 152.28 440.00 855.00 25.5 53.08 4 1 S8 19.45

10.25 3.54 38.01 73.74 107.83 NO.42 208.60 332.34 28.28 t 34.35

24 1 1.23 2227.39

335.30 334.57 367.19 363.78 498.17 5 2.44 43.11 64.65 35.54

239.05 170.72 3 15.72 205.48 340. 1 1

19.91 98.29 82.17 106.0 1 10 1.43 179.07 118.97 269.64 Appendiv 6c: Total nitrogen ED A YEAR FREQ AVGZ STD2 MM2 ,W CV2 Appendir 7: Final chlorophyll, TP, and TN values for Fimish estuaries included in IV and V where site is the code as in Table 1, IV, the mean value is the mean of al1 the year values for each estuary as descnbed in IV (Chi,), n is the number of years the estuary was sampled, and sd, min, max and CV are the standard deviation, minimum, maximum and coefficient of variation of the yearly values.

Site mean n s.d. min miu CV 11 17.3 3 3.8 13.2 20.8 --7 7 12 5.6 1 5.6 5.6 13 3.9 1 3.9 3.9 14 9.8 5 3.2 5.5 14.5 33 17 5.8 -7 1.Z 5.0 6.6 20 18 30.5 5 3.5 26.4 34.5 1I 19 9.4 5 4.9 1.9 17.5 5 2 20 12.4 4 4.2 8.3 17.5 3 4 2 1 45.9 5 10.8 33.3 56.1 23 23 6.3 5 2.0 4.6 9.7 3 1 25 15.7 5 0.9 14.3 16.9 6 27 4.5 5 0.3 4.2 5.0 6 29 5.5 5 0.8 4.7 6.5 14 3O 7.9 5 2.0 4.6 9.9 26 35 13.4 5 2 -5 9.2 15.8 19 3 9 4.2 2 1.4 3.2 5.2 34 42 23.2 5 8.6 11.7 32.5 37 49 7.5 5 1.8 5.4 10.1 25 5 8 9.9 5 3 .5 6.3 15.3 35 TP Site mean n s.d. min ma.. CV

na Site mcan n s.d. rnin max CV Appendix 8: Raw data fiom Finnish municipalities used to calculate human population density in the watenheds associated with the estuaries (IV). Site is the estuary code, municipality is the name O lthe rnunicipality that is located in the watenhed, ID is the municipal code as designated by the Fimish government, Area-W is the area of the municipality occumng in the Katershed, Pden-M is the population density of the -Municipality and Pop-MW is the number of people fkom the rnunicipality located in the given watenhed. Site Municipaliry ID Area- W Pden-M Pop-MW (km') (km") (#)

Anjalaniroski 754 11 Luumiiiu Miehikk616 Vehkalahti Virolahti Anjalankoski Hamina Vchkalahti Virolahti Anjalankoski Kotka Luum6 ici Vakeala Vehkalahti Aa6nckoski Alajarvi AnjaIankoski Asikkala Elimdki Haapajorvi Hankasalmi Hartola Kaukivuori Heinola Hirvcnsatmi Hollola H6meenkosici lisalmi Iitti laala l6msb l0ms~nkosk.i Ii3ppil6 f outsa lyv6skyl6 Jyvi5skyl6n dk Kangasnicmi Kannonkoski Karstula Karmila Kcitcle Kcuniu Kinnuia Kiunivcsi KivijOrvi Konnevcsi Korpilahti Kotka Kouvola Kuhmalahti Site Municipality ID Area-W Pden-M Pop-MW ikm2) (km-? (#I

Kuopio Kuorevesi Kuusankoski Kyyjorvi Lahti Lapinjorvi Lappeenranta Laukaa Leivonmdki Lemi Leppdvirta Lestijorvi Ldngelmdki Luhanka LuumOki .Maaninka Mikkeli Mikkclin rnlk iMi3nty hqu Multia Muuramc Nastoia Padasjoki Perho Pertunmaa Peti5j6vcsi Pieks6m6en mlk, PicksOrndki Piclavesi Pihtipudas Pyh6jdrvi Pyhtdi5 PyUcanmi3ki Riautalampi ReisjOrvi Ristiina RuotsinpyhtM SaarijOrvi Savitaipale Soini Sumiaintn Suolahti Suomcnniemi Suonenjoki SysmO Taipaisaari Tm0 Toivakka Site Municipality U> Area- W Pden-M Pop-MW (km') (km-') (#l

Vakeala Vesanto Vii tasaari Vistasah Xnjalankoski E 1i.dki 1itti Kotka Kouvola Kuusankoski Lapinjdrvi Pyhtdi3 Ruotsinpyhtdd Valkeala Askola Myrskyli3 Pernaja Porvoo Pukkila As kola Hollola Kdrkal6 Lahti Mi5ntsi5li5 Myrsky 16 Nastola Orimartila Pomainen Porvoo Pukùila Hausji3rvi HyvuikM Korkalti Monts616 Orimattila Pominen Polvoo Riklrila Sm Kerava M6ntsi5ld Pomainen Sipoo Vanta? Espoo Hausjorvi Helsinki Hyvinki%s Site Municipality ID Area-W Pden-M Pop-MW (km:, ci

Karkkila Karstula Karvia Keuruu Kihnia Kidcoinen Kiukaincn Kokemoki Korkalo Koski Tl Kuhmalahti Kuhmoinen Kullaa Kuorcvcsi Kuru Kylmdkoski Lammi Lavia Lehtimdiu LempMl6 Lo imaa Loimaan kunta Lhgelmoki Loppl Luopioinen Luvia Mamifa Mellili!! M6ntsdli5 Mann6 Mouhijihvi Multia Naklula Nokia Noocmarwru Nummi-Pusula oripaa Chivesi Paytya Padasjoki Parkano Ptt6javesi Pirkkala POlkiSnc Pori Rinkalaidun PylkaMloki Rcnko Site Municipaliry ID Area- W Pden-M Pop-MW (km? ( km') (#) 35 Riihirndki 694 23.3 2 12 4940 Ruovesi Sahalahti Soini sakyia Somero Suodennicmi Taysi3 Tammela Tampere Toijala Tuulos Ulvila Ujala Valkeakoski Vammala Vampula Vesiiahti Viiala Viljakkala Vilppula Virrat YlajOrvi YpZij6 Aets6 Harjavalta HiSrneenkyra Huittinen kaalinen Jamijarvi Kay fia Kiikoincn Kiukainen Kokcm6k.i KuUaa Lavia Luvia Mouhij6rvi Nakkila Nokia Noocmatkku Pori Pdaidun Suodcnnicmi Ulvila Urjah Vammala vcsilabti Alavw Site Municipality ID Area-W Pdcn-M Pop-bfW (km') (km-') (#)

isokyra laIasj6rvi lurva Karvia Kauhajoki Kihnia Kurikka Laihia Lapua blustasaari Nurmo Parkano Peroseindjok Seindjoki Teuva Va yn Vaasa Virrat Vohokyca YlihOnn6 Yiistaro Halsua Kaustinen Kivijdrvi KokkoIa Kolvi6 K~~uPYY Kyyjdrvi Lestijdrvi Perho To holampi Ullava Vctcli Vimpeli Kcmpelc Kcstilo Lirninka LumijolQ Muhos Rantsila RuW Tcmmes Tym6vO Appendix 9a: Mean flow rates (m3 - sec-') of rnonitored rivers associated with estuaries where site is the estuary code. RR is the ninoff region. and flow (year) indicates the mean flow in a given year. The mean value is the rnean flow rate used for the given estuary and is calculated as the means of' the years for which Chi. TP and TN data exist.

(1989) (1990) (1991) (1992) (1993) m'sec-' m'sec" m3sec-' m'sec-' m'sec" Appendix 9b: Mean flow rates (m3. sec") of unmonitored rivers associated with estuaries where site is the estuary code. RR is the runo ff region, Wshed is the watenhed area, Mt is the total drainage area for the runoff region (RR), Qn is the residual mean flow associated with a given RR for the years for which there exist Chi, TP and TN data, and mean QR is the mean flow rate.

Site RR Wshed Adt Qrs mean QR (kd) (km') m'sec" m'sec*' 12 1 380 5504 52.1 3.6 13 1 569 5504 52.1 5.4 t 7 1 309 5504 51.3 2.9 20 1 220 5504 53.5 2.1 29 2 284 5374 52.9 2.8 30 2 288 5374 52.9 2.8 3 9 3 992 5877 59.8 10.1 58 4 1181 10761 110.1 12.1 Appendix 10: Yearly nonpoint source loads of total phosphorus and total nitrogen (t-yf') delivered to the esniaries by nvers. Loads calculated as the mean annual concentration of TP or TN mulitplied by the mean flow (Appendices 9a and 9b). Blanks indicate years which were not included in the calculation of the mean as no Chl, TP or TN data exist for these years.

Total Phosphorus Loads Site 1989 1990 1991 1992 1993 mcan tyil tyi' t yfl t yr-' t yr-l t yY-l Appendix 10 cont.

Total Niûogen Loads Site 1989 1990 1991 1992 1993 mean t yr-' t yi' t yi1 t yr-l t yr-' tyil Appendix 1la: Chlorophyll values for "STORET" US estuaries included in V where site code is the code used by NOM (1996,1997) (see V, table 1), station is the EPA station code. mean is the mean Ch1 value calculated by averaging al1 growing season values for the given station in the given year, year is the sarnpled year, n is the number of observations at the given station in the given year, and sd, min, max and CV are the standard deviation, minimum, maximum and coefficient of variation of these observations respec tively. A~~endix1 la: Chloro~hvlivaiues for US estuaries Site Station year mean n sd min max CV MO40 ER-02 LI-25 ER- 1 1 ER-17 ER- 1 9 LI-3 1 LI-24 LI-26 LI-34 LB-O4 RI-02 w-22 ER-02 ER-09 ER- 1 5 JB-07 LI-24 LI-25 LI-26 LI-34 LI-35 RI-O2 ER-04 HA-02 UH-13 LB-O8 RB-16 LB-08 RB-16 91002 9 1005 91008 9101 1 91014 91017 9 1020 9 1023 332046 332049 332052 332055 332061 40101 1 40 1021 40103 1 401041 40 105 1 401061 Appendix 1la: C hloropbyll values for US estuaries Site Station year mean n sd min max CV MO90 401081 89 81.40 Appendir 11 a: Cblorophyll values for US estuaries Site Station vear mean n sd min max CV Appendix Ila: Chlorophyll values for US estuaries Site Station year mean n sd min max CV

LM120 MCB4.4 .M 120 IMCB5.1 Ml20 MC85.2 Ml20 MCB5.3 Ml20 MCBl.1 iM120 MCB2.2 Ml20 MCB3.1 Ml20 MCB3.2 .Ml20 WB3.3C Ml20 MCB3.3E Ml20 iICB3.3V Ml20 tiCB4.1C Ml20 kîCE4.11 Ml20 K84.1V Ml20 WB4.2C Ml20 liIC84.21 Ml20 KB4,2V Ml20 tiCB4.3C iM120 VlCB4.3E Ml20 dCB4,3V MI20 MCB4.4 .Ml20 MCBS.1 Ml20 MCB5.2 Ml20 MCB5.3 M120B MLE2.3 M120B MLE2.3 M12OB MLE2.3 M120B MLE2.3 M12OB MLE2.3 M12OB MLE2.3 M120F MCB2.1 M12OF MCB2.1 M120F MCB2.1 M120F m668C M120F MCB2.1 M120F MCB2.1 Ml2OF MCB2.1 SOlO )895000( Solo 1949000i Solo 19995004 SOlO J995000(: SOlO do500001 Solo d390000 SOlO 11161ûûOû SOlO d7053Oû SOlO )905900( SOlO 1985000( SOlO )895000( Appendix lla: Chlorophyll values for US estuaries Site Station year mean n sd min mar CV SOlO )949000( 90 3.00 17.00 69.4 SO 1O 1999500( SO 1O 1995000( SO 1O ~~050000 SOlO A390000 SOlO il610000 SOlO 4705300, SO l O 1905900t Sol0 3985000( SO 1O 18950001 SO 1O >9490004 SO 1O 19995004 SO 1O 1995000( Sol0 4050000, Sol0 4390000 Sol0 46100W SOlO 47053001 SOlO )905900( Sol0 i985000( SO 1O 18950004 Sol0 )949000( sot0 19995004 SO l O 1995000C Solo il050000 Sol0 4390000 SOlO 461000Q SOI0 4705300 SOlO 1905900( SOlO )985000( SOlO )895000( SOlO 1949000( Solo 39995004 SO 1O J995000C SOI0 40500001 SOlO 43900001 SOlO 46100001 SOlO 4705300 SOI0 497m Solo 19059004 SOlO 198S0001 SOlO 1895000( Solo )949000( SOLO 19995m SO 10 J995000( SOlO 4050000, SOI0 4390000 SOlO tf61OOOO Sol0 4705300 SOlO q970000( Appeadu 11a: Chlorophyil values for US estuaries Site Station year mean n sd min max CV 2.87 3.00 12.50 36.9 SOfO 1985000( SO 1OA 17650001 SO 1OA 17870001 SO IOA 18498001 SO IOA 1865000i SO IOA 1765000i SO 1OA 17870001 SO 1OA 18498004 soi ox )865ooor SO 1OA 17650001 SO 1OA 17870001 SO IOA 18498001 SO 1OA 18650001 SO 1OA 1765000( SO 1OA 1787000i SO 1OA 1849800( SO 1OA 18650001 soI OA 17650001 SO 1OA 1787000i SO 1OA 1849800( SO 1OA 18650001 SO 1OA 17650001 SO 1OA 1787000i SO 1 OA 18498004 SO 1 OA 186500Q SO 1OB 1857000C SO 1OB 1877000C SO 1OB 1890080C SO 1OB 1890250C SOlOB J891000C SO 1OB J953000( SO 1OB J993000ç SOlOB 19825W SOlOB J857000( SO 1OB J8770W SOlOB J890080C SO 1OB J890250C SOlOB J891000( SO 1OB J953OûûC SO 1OB J993ûûûC SO lOB )982500( SOlOB J857000( SO 10B J877000( SO 1OB J89008M SO 1OB J89025M SOlOB 1891000C SO 1OB J953000C SO t OB J993ûûM Appendix 11a: Cblorophyll values for US estuaries Site Station year mean n sd min mas CV SOlOB 19825004 91 SO 1OB J857000C SO 1OB J877000C SO IOB J890080( SO 1OB J89025O( SO 1OB J89 1000( SOlOB 1953000ï SO 1OB J993000C SO 1OB 19825004 SOlOB J857000C SOlOB J877000C SO 1OB J890080C SO 1OB 1890250C SOlOB 189 1000C SO 1OB 1953000C SO 1OB 1993000Ç SO 1OB 1982500( SO 1OB 1857000( SO 1OB J877000C SO lOB 18900806 SO lOB 1890250C SO 1OB 189 1000( SOlOB J953000C SO 1OB J969000C SO lOB J99300M SO 1OB )982500( S020 198 1OOOC S020 1981000C S020 J981000C 5020 J981OW S020 1981000C SO2O ?960000( S020 ?972000( S020 ?973000( S020 ?974000( S020 1981000( S020 J99380ûC S020 ?870000( S020 ?896550( S020 ?958000( S020 ?960000( S020 ?972000( S020 373m S020 ?974000( S030 ? 120000( S030 ?370000( S030 ?475000( S030 ? 120000( S030 ?370000( Amendis 1la: Chloroohvll values for US estuaries Site Station year mern n sd min max CV S030 ?475000( 90 1.04 5.00 7.00 16.9 Appendir I lb: Total phosphorus (TP)values for "STORET" US esniaries; abbreviations as in Appendix 1 la. Appeadix Ilb: TP values ... Site SCation year mean n sd min max CV MO40 ER-02 1980 120.0 120 LI-25 ER- 1 1 ER- 17 ER- 19 LI-3 1 LI-24 LI-26 LI-34 LB-04 RI-02 UH-22 ER-02 ER-09 ER- 15 BO7 LI-24 LI-25 LI-26 LI-34 LI-35 ER44 HA-O2 UH- 13 LB-08 RB- 1 6 LB-08 9 1002 91005 91008 9101 1 91014 91017 9 1020 9 1023 332046 332049 332052 332055 33206 1 40101 1 40102 1 40103 1 401041 401051 401061 401071 40108 1 40 109 1

MCB4.2E MCB4.2W MCB43C .MCB4.3 E MCB4.3W .MCB4.4 MCBS. 1 MCB5.2 MC85.3 MCB1.1 .MC82.2 MCB3.1 MCB3.2 MCB3.3C MCB3.3E .MCB3.3 W MCB4. IC MCB4.1E MCB4.1W MCB4.2C MCB4.2E MCB4.2W MCB4.3C IMCB4 $3E MCB4.3W MCB4.4 MCBS. 1 MCB5.2 MCB5.3 MCB1 .1 MCB2.2 MCB3.1 MCB3.2 MCB3.3C MCB3.3E iMCB3.3W MCB4.1C MCB4.1 E MCB4.1W MCB4.2C MCB4.2E MCB4.2W MCB4.3C MCB4.3E MîB4.3W MCB4.4 MCBS. 1 MCBS .2 MCBS .3 MCB1.l MîB2.2 MCB3.1 M 120 MCB3.2 Ml20 .MCB3.3C .Ml20 MCB3-3 E Ml20 .MCB3.3 W iMl2O MC84.1C LM120 MCB4.1 E LM120 MCB4.1 W Ml20 .MCB4.2C Ml20 .MCB4.2E Ml20 MCB4.2W Ml20 MCB4.3C Mt20 MCB4.7E Ml20 MCB4.3W M 120 MCB4.4 Ml20 MCBS. 1 Ml20 MCB5.2 Ml20 MCB5.3 M120B MLE2.3 M l2OB MLE2.3 M l2OB MLE2.3 M 120B MLE2.3 M l20B iMLE2.3 LM120B MLE2.3 M l2OF MCB2. 1 M 120F MCBZ. 1 M 1îOF MCBZ. 1 M120F XJH6680 M120F MCB2.1 M 12OF MCB2.1 Ml2OF MCBZ. 1 Solo D8950000 SOlO D9490000 Solo D9995000 so 1O f 995OOOO SO 1O MO500000 Solo M3900000 SOlO M6 100000 Solo M7053000 Solo 09059000 so 10 09850000 SOlO D8950000 SOlO Dg490000 SOlO Dg995000 Solo J9950000 Solo MO500000 Solo M390Q000 SOlO M6 100000 SOlO Ni7053000 Solo 09059000 Solo 09850000 sot0 08950000 Solo Dg490000 Solo SO 10 Solo Solo Solo Solo SOlO SOlO Solo SOI0 sot0 SOI0 Solo SOlO Solo SOI0 Solo SOlO SOlO so IO SOlO SOI0 Solo so 10 so 10 so 10 Solo sot0 Solo SO 1O Solo SOlO SO 1O so 1O sot0 so 1O Solo Solo Solo so 1O SOlOA SO 1OA SOIOA SO 1OA SOlOA SO 1OA SOlOA SOlOA SOlOA SOlOA SOlOA SOlOA SOIOA SOlOA SOlOA SOlOA SOlOA SOlOA SOlOA SOlOA SOlOA SO 1OA SO 1OA SO 10.4 SO lOB SOlOB SO 1OB SO lOB SO lOB SO 1OB SO 1OB SO 1OB SO 1OB SO 1OB SO 1OB SO 1OB SO 1OB SO 1OB SO 1OB SOlOB SOlOB SO 1OB SO 1OB SO 1OB SO 1OB SO 1OB SO 1OB S010B SO 1OB SOlOB SO lOB SO 1OB SO 1OB SO lOB SO 1OB SO 1OB SO 1OB SO lOB SOlOB SO 1OB SO 1OB SO 1OB SO lOB SO lOB SOlOB SOlOB SOIOB SO IOB SO IOB SOlOB SO 1OB SOlOB SOlOB S020 S020 S020 S020 S020 S020 sot0 S020 S020 S020 S020 S020 S020 5020 S020 S020 S020 S020 S030 S030 S030 S030 S030 S030 S030 S030 S030 S030 S030 S030 S030 S030 S030 S030 S030 S030 SM0 S070 S070 S070 S070 S070 S070

Appendix 1lc: Total nitrogen (TN)values for "STORET" US estuanes; abbreviations as in Appendix 1la. A~~endix1t c: Total Nitronen Site Station vear mean a sd min mu CV Appendix 1lc: Total Nitrogen Site Station vear mean n sd min max CV Appendu 1lc: Total Nitrogen Site Station vear mean n sd min max CV Appendix 12a: Chlorophyll values for "STORET" US estuaries where the mean value is the mean of al1 the growing season station values (Appendix L la) in a given year as descnbed in V (Chl,,), n is the number of stations which were sampled in a given year, and sd, min, max and CV are the standard deviation, minimum, maximum and coefficient of variation of these station values. Si te year mean n sd min max CV .MO40 80 3.50 2 0.7 1 3 .O0 4.00 20.2 MO60 1M040 MO60 MO40 MO40 MO60 .?dl040 MO60 MO90 MI20 Ml20 Ml20 Ml20 Ml20 Ml20 M120B M120B Ml2OB M120B M120B M 120B 34 l2OF M120F M 120F M120F M120F M120F SOI0 SOI0 sot0 SO IO SO IO SOI0 SO 1OA SO 1OA SO 1OA SO 1OA SOlOA SOlOA SOlOB SOlOB SO 1OB SO 1OB SOlOB SOlOB Appeadix 121: Ch1 year mean n sd min rnax CV 5020 92 6.63 1 6.63 6.63 Appendir 12b: Total phosphorus values for "STORET" US estuaries; abbreviations as in Appendix 1Za. . . Site vear mean n sd min mar CV 91.9 120 250 49.7 MO60 MO40 hi060 MO40 MO40 MO60 MOU MO60 .MO90 Ml20 Ml20 Ml20 Ml20 &M120 Ml20 Ml2OB M120B rM 120B M120B M l2OB M 1208 M 120F M l2OF M 120F M 12OF M120F M 120F so 1O so 1O Solo so 1O SO 1O sot0 SO 1OA SO 1OA SOlOA SO lOA SO 1OA SO t OA SO 1OB SO 10B SOlOB SOlOB SO 1OB SOlOB S020 S020 S020 S020

Appendix 12c: Total nitrogen values for "STORET" US estuaries; abbreviations as in Appsndix 1Za.

Site year mean n sd min max CV Ml20 89 11 14 20 273 712 1834 24.5 Ml20 90 902 21 247 622 lSl l 27.4 Ml 20 9 1 792 20 193 615 1457 21.4 Ml20 92 830 20 188 62 1 1318 23.6 Ml20 93 927 20 208 746 1637 22.5 Ml20 94 994 20 184 833 1539 18.5 M 120B 89 794 1 794 794 M120B 90 649 1 649 649 M120B 91 688 1 688 688 M 120B 93 710 1 710 710 M120B 93 780 1 780 780 M120B 94 920 I 920 920 M 120F 89 1790 1 1790 1790 M 120F 90 1419 1 1419 1419 M 120F 91 1142 -i 68 1094 1190 6.0 M120F 92 1213 1 1213 1213 M-t20F 93 1386 1 1386 1386 M120F 94 1259 1 1259 1259 Appeodix 13s: Final chlorophyll values for "STORET" US estuanes included in V where estuary is the estuary name, site is the code used by NOM (1 996, 1977), the mean value is the mean of al1 the year values for each estuary as descnbed in Chapter 5 (Chi,), n is the number of yean the estuary was sampled, and sd, min, max and CV are the standard deviation, minimum, maximum and coefficient of variation of the yearly values.

Estuary Site mean n sd min max CV Long Island Sound MO40 26.83 5 17.52 3.50 52.00 65.28 Hudson River MO60 23.63 4 18.06 9.50 39.00 76.45 Delaware Bay MO90 27.38 1 . 27.38 27.38 . Chesapeake Bay Ml20 11.12 6 2.25 8.13 15.06 20.2 Potomac M120B 10.40 6 4.26 6.42 16.86 31.0 Chester MI20F 15.63 6 14.93 5.82 44.91 95.5 Albermarle Pamlico 5010 8.26 6 2.98 4.53 11-27 36.1 Pamlico-Pungo SOlOA 16.59 6 7.78 8.28 27.03 46.9 Neuse SO10B 17.09 6 6.19 8.05 24.67 36.2 Bogue Sound S020 8.00 6 4.23 2.76 13.25 52.8 New River S030 22.98 6 13.87 6.08 33.44 60.4 Cape Fear S040 . 7.50 1 . 7.50 7.50 Charleston Harbour SO7 O 6.09 I . 6.09 6.09 Appendix 13b: Final total phosphorus values for "STORET" US estuaries included in V; abbreviations as in Appendix 13a.

Estuary Site mean n sd min max CV Long Island Sound MO40 153.1 5 44.0 93 203 28.8 Hudson River MO60 151.3 4 11.8 135 160 7.8 Delaware Bay iMO90 159.1 1 159 159 Chesapeake Bay M 120 39.8 6 7.5 35 54 18.8 Potomac M120B 31.8 6 5.1 27 3 1 16.0 Chester M12OF 46.3 6 13.1 30 66 28.3 Aibennarie Pamiico SO 1O 65.4 6 14.2 49 88 21.6 Padico-Pungo SOlOA 166.7 6 71.0 76 258 42.6 Neusc SO 1OB 135.3 6 25. 1 114 175 18.5 Bogue Sound S020 78.3 6 27.1 45 108 34.7 New River S030 148.1 6 18.3 130 179 12.3 Cape Fear 5040 35.0 1 35 35 Charleston Harbow S070 82.3 1 82 82

Appendix 13b: Final total phosphorus values for "STORET"US estuaries included in V; abbreviations as in Appendix 13a. Estuary Site rnean n sd min max CV Chesapeake Bay Ml20 926 6 116 792 1114 12.6 Potomac M120B 757 6 97 649 920 12.8 Chester M120F 1368 6 232 1142 1790 16.9