Publication No. 171 The Massachusetts Acid Rain Monitoring Project: Ten Years of Monitoring Massachusetts Lakes and Streams with Volunteers.
by Paul J. Godfrey Mark D. Mattson Marie-Françoise Walk Peter A. Kerr O.Thomas Zajicek Armand Ruby III The Massachusetts Acid Rain Monitoring Project: Ten Years of Monitoring Massachusetts Lakes and Streams with Volunteers.
by Paul J. Godfrey Mark D. Mattson Marie-Françoise Walk Peter A. Kerr O.Thomas Zajicek Armand Ruby III
November 1996 Abstract
Between 1983 and 1985, the Acid Rain Monitoring (ARM) Project used as many as 1000 citizen volunteers to collect and help analyze more than 40,000 samples from 2444 lakes and 1670 streams, respectively 87% and 69% of the named lakes and streams in the state, and monitored a representative 453 randomly selected and 119 special interest lakes and streams for eight successive years (1985-1993) with approximately 300 volunteers. This report details the organizational effort that permitted maintenance of this ten year effort, the quality control efforts needed to insure that resulting data met scientific standards for quality, and the findings of ten years of extensive monitoring.
Results for the nearly comprehensive initial phases of the project show that 5.5% of lakes and streams in Massachusetts are acidified (pH < 5.0 and ANC < 0 µeq/l); 57.4% were sufficiently low in acid neutralizing capacity to be considered threatened by acid deposition (0
While results are more representative of Massachusetts lakes than the Eastern Lakes Survey because of planned exclusions in the latter’s site selections, the differences are relatively small. Comparisons between ARM results and those from other areas and countries suggest that Massachusetts surface waters are comparable to those from other areas of the world that are very sensitive to acid deposition.
Most of the differences in water chemistry can be related to the underlying geology. The western edge of the state is substantially higher in most cations and anions than average while southeastern, north-central and the Berkshire Mountain region are lower. In three of six regions, sulfate levels were twice as high as anticipated from deposition and evapotranspiration. Massachusetts lake and stream sulfate levels were comparable to eastern Canada and other high deposition sites. Streams were higher than lakes in calcium, magnesium, potassium, sulfate, aluminum and silicon dioxide and lower in sodium, chloride, organic acids, iron, and manganese. The combination of relatively high organic acids and base cations may account for the somewhat smaller percentage of acidified surface waters than in other high deposition areas.
The coastal region has a higher proportion of sodium and chloride resulting from sea salt influences. Statewide, sodium and chloride levels exceeded the amount expected solely from sea salt. A substudy of 162 randomly selected streams suggested that only 4% of the salt concentration of these streams could be attributed to sea salt. The remainder was highly correlated with the number of road lane miles in each stream’s watershed. Approximately 63% of the variance in stream sodium concentration were explained by the number of lane miles. Class 1 and 2 (interstate highways and major state roads) made the highest contribution with urban streets next. Rural roads contributed relatively little.
Analyses of long-term trends in pH, ANC and selected ions from 10 years of data on 330 streams and 181 lakes showed a significant but small increase in average pH and increase in ANC. For streams, the median slope after correction for hydrologic variation was +0.021 pH/yr and, for
i ANC, it was +2.4 µeq/l/yr. For lakes, the median slope (hydrologically corrected) was +0.016 pH/yr and +1.9 µeq/l/yr for ANC. Streams with higher ANCs show a faster rate of recovery than low ANC streams (4.15 µeq/L/yr vs 1.33 µeq/L/yr).
For pH, high ANC streams showed a more gradual increase than low ANC streams (0.018 ñH units/yr vs. 0.026 ñH units/yr. Sulfate declined by 1.8 and 1.4 µeq/L/yr for streams and lakes, respectively. There was no trend for base cations in streams but a significant trend for base cations in lakes (+1.5 µeq/L/yr). The latter was counteracted by an increase in chloride of 1.4 µeq/L/yr.
Most lakes and streams exhibited no significant trend for the 10 years of the study. However, 70 of 330 streams showed statistically significant increases in ANC, 11 showed decreases, 43 of 181 lakes increased in ANC while 7 decreased. Most of the streams and all of the lakes exhibiting statistically significant declines occurred in the southeastern portion of the state. In this group of lakes and streams only two lakes and two streams became acidic (dropped below 0.0 ANC and ñH 5.0) during the ten years of the study.
The thousand or more volunteers participating in the project were profiled through a questionnaire distributed in 1989. Of the 40% responding, roughly 50% were original participants, 69% were male, 79% were married, 70% had a college degree, 63% lived in small towns, 54% were native to Massachusetts, and 75% had household incomes between $20,000 and $75,000. In total, they listed 76 hobbies, sports and interests. They also listed membership in 69 civic organizations and 20 different religions or religious beliefs. Ninety-seven percent considered their efforts on the project to be worthwhile.
The ARM database is available on disk from the Water Resources Research Center, Blaisdell House, University of Massachusetts, Amherst, MA 01003-0820 or may be downloaded from the Center’s website at http://riga.fnr.umass.edu/~tei/wrrc/index.html.
ii Foreword
This report has three intended functions: first, to assemble the various reports that have been published on the scientific results in one common format; second, to provide full documentation of all aspects of the project, most of which have not been included in shorter journal publications, and third, to provide a "road map" for others who may want to try to create a long-term volunteer monitoring effort. Many of the parts of this report have been either published or presented elsewhere, but in the latter case, presentation may have been limited to conference attendees, ARM newsletter readers, or other restricted groups. Consequently, the ultimate goal of this report is to present the entire context of the ten year long Acid Rain Monitoring (ARM) Project in a way that makes clear the many benefits and the special requirements of this approach to environmental monitoring so that others may use the results and replicate the process.
We were novices when ARM started. Many of the errors that we caution against were lessons learned the hard way. Perhaps the most important lesson to be learned is that errors can be made without being fatal, but not very many and not very large. Another lesson which is not expressed in the text is the error of understaffing. For all volunteer efforts, there are significant requirements for coordination, lab and data analysis, reporting, etc. All too often the sense of volunteerism carries over to the coordination in the form of trying to do the many necessary things with too few people working too many hours. It can be done but the risks of mistakes and burnout are much greater. The ARM staff have shown extraordinary dedication to the Project; working long hours and, after hours, they volunteered additional time to meet project's needs. More importantly, citizen monitoring demands rapid information turn-around and science demands beating the competition to the publication. Always understaffed and usually underequipped, ARM did well at the former, but less well at the latter.
It is our hope that ARM provides an example of citizen monitoring, quality control, and scientific objectives that helps promote a new era of environmental science. It is also our hope that the story of ARM creates that last bit of necessary motivation for many, who might not otherwise participate for whatever reason, to become involved in the improvement of our environment.
iii Acknowledgments
Many people had a part in preparing this final report of the Acid Rain Monitoring Project, but these acknowledgements are a personal statement of the Project's Director, Dr. Paul Jos. Godfrey.
The Acid Rain Monitoring Project would not have been possible without the help of thousands of individuals and many organizations, agencies and institutions. It is impossible to personally acknowledge each individual's efforts, yet each has played a key part in accomplishing the Project's objectives.
Two individuals, both volunteers, stand out for their exemplary efforts to make the Project a success. Leon Ogrodnik was the principal architect of the Project's organization network. He worked with local organizations to identify key people within each county who would actively coordinate each district and county, and worked hand in hand with them to organize volunteers, find laboratories, establish the logistical arrangements, and initiate the publicity for our presentations and sampling days in the first year of the Project. In the early months of the Project, he travelled throughout the state, helping where help was needed, and he spent long hours on the telephone finalizing details. His dynamism, energy and genuine concern for the environment set an outstanding example of citizenship for us all.
In 1983, Dr. O.T."Tom" Zajicek responded to my plea for help in starting the ARM Project. Neither he nor I had any idea what we were getting into, but Tom has never flagged in his full commitment to the Project. Like Leon Ogrodnik, ARM never could have happened or lasted without Tom's involvement. He conducted the initial lab training sessions, travelling all over the state. He patched together the initial extended chemical analyses by either doing the work himself or enlisting volunteers from nearby chemistry students. He oversaw the creation of a fully equipped laboratory for Phase II, along with all the pains of startup. He trained and supervised the three lab supervisors. He worked elbow to elbow with hundreds of lab volunteers and lab staff in analyzing samples on ARM collection days (and he bought the pizza!). He has responded to volunteer lab analytical problems around the state, often driving to the site to solve the problem. He has sought ways to integrate ARM with the University educational purpose by encouraging involvement of student interns, student assistants, and various kinds of science demonstrations. Like Leon, he has always been an unselfish, trusted advisor on all aspects of the Acid Rain Monitoring Project.
Trout Unlimited, and particularly the Massachusetts Chapter representatives, made the key decision to provide initial funding for the Project. The concept was a bold one; its chances of success could not be predicted. In the face of the potential risks, the decision by the Massachusetts Trout Unlimited Chapter was courageous. They convinced the national Trout Unlimited organization to match their funding and set to work identifying members to work as coordinators in districts across the state. This early monetary support was absolutely crucial to the Project, but their membership support, which continued unabated throughout the Project, was even more crucial to the Project's long-term goals.
Numerous environmental, sportsmen's and other organizations have provided support for the Project and assisted in developing and maintaining the grass roots network of volunteers. These include: the Massachusetts Audubon Society, Massachusetts Sportmens' Council, Appalachian
iv Mountain Club, Berkshire Regional Planning Commission, Pioneer Valley Regional Planning Commission, Massachusetts Grange, Nashua River Watershed Association, Berkshire Garden Club, League of Women Voters of Massachusetts, Connecticut River Watershed Council, and Peter Menard of Berkshire County. IBM provided early support through a community development grant.
Initial state funding for the Project was the result of efforts by the Massachusetts Legislature's Environmental Caucus, the Joint Committee on Agriculture and Natural Resources and other members of the legislature, particularly former Senator Atkins, Senator Wetmore, former Representatives Guernsey and Suhoski and Representative Nagle. Over the duration of the Project, these members and others, especially former Senator and present U.S. Representative John Olver, Senator Stan Rosenberg, Representative Ellen Storey, the late U.S. Representative Silvio Conte, former U.S. Representative Chet Atkins, U.S. Representatives Edward Markey and Gerry Studds, and Senators Edward Kennedy and John Kerry, have provided valuable support for the Project.
Principal funding over the ten years of the Project was provided by the Massachusetts Division of Fisheries & Wildlife. Initially responding to a legislative directive to administer the Project, over the decade of their involvement, they strongly argued for continued legislative budgetary support. When fiscal times forced budget cutbacks, the Division assumed the cost of the project on funds provided by license sales. Notable for their vigorous efforts on behalf of the Project were the late Director Richard Cronin, late Chief Aquatic Scientist Peter Oatis, present Director Wayne MacCallum, and Assistant Director Jack Buckley. Fishery Biologists Richard Keller, Ken Simmons, David Halliwell and Robert McHaig provided a variety of valuable assistance during the Project.
Others in public service have played important roles in the success of the Project. These are former Governor Michael Dukakis, former Lieutenant Governor and present U.S. Senator John Kerry, former Environmental Affairs Secretary James Hoyte, former Commissioner of the Department of Environmental Quality Engineering (DEQE) Anthony Cortese, former Commissioner of the Department of Fisheries, Wildlife and Environmental Law Enforcement Walter Bickford and former staff members in the DEQE Division of Air Quality Alan Van Arsdale and Kenneth Hagg.
Significant funding was provided by the U.S. Department of Interior through the Massachusetts Water Resources Research Center's annual Water Resources Institute Program grant to initiate chemical analyses in addition to pH and ANC, demonstrating the need for creation of an equipped and staffed laboratory for continuing extended analyses. The same program funded the orderly closeout of the project in 1994.
No less critical to the success of the ARM Project has been the support of the University of Massachusetts, particularly Dr. Samuel Conti, former Vice Chancellor for Research, Dr. Joseph Larson, Director of the Environmental Institute, the Department of Chemistry that provided laboratory space and Dan Keedy who patched the equipment together beyond any reasonable expectation.
Finally, in ten years, there have been many, many people who have participated directly in the management and conduct of the ARM Project at the Water Resources Research Center. The principal author expresses his deeply emotional appreciation for the extraordinary efforts and creativity of these individuals and his coauthors in what has been an extraordinary life experience.
v Statewide coordinators have played a key role in ARM's success. There have only been two: Armand Ruby and Marie-Françoise Walk, and they have been exceptional. There have been four database managers/programmers: Louis Cyr, Nick Layzer, Donald Boy, and James Scace. They have made it possible for us to manage more than 40,000 data records in a reasonably efficient manner. As the project developed, it became clear that the statewide coordinator needed an assistant; Susan Hancock, Cass Mason and Rita Reinke made a potentially chaotic situation orderly and manageable by a small staff. Later in the project, the principal author recognized the need for help in data analysis and report preparation. Dr. Mark Mattson has filled that role admirably and pitched in to clarify and improve analytical data quality control. The ARM laboratory, from its creation in 1984 to present has operated at minimal expense and with minimal staff but has managed to meet very high standards of performance. The credit belongs to the lab supervisors: Irene Ellis, William Brooks, and Peter Kerr and the assistants and students that have worked with them.
There have been dozens of others who have helped as staff, student assistants and volunteers in the conduct of the ARM Project at the Water Resources Research Center. They are listed below. I hope that none have been missed, and further hope that all found something from their participation in ARM that has been important in their lives.
Heidi Adler Marjorie Ferris Judy Lane Cheryl Ruby Zahiruddin Alim Jim Figlar Michael Linde Kathleen Sanquist Michael Azure Karen Freedman Cathy McDonald Kate Savage Ben Bailar Steve Ghim Ann McMenemy Jutta Schneider Douglas Baldwin P. Candee Gibbs Becky Menard Peter Scott Thomas Bales Mollie Godfrey Robin Mikleson Lynn Sedgwick Jim Barabe Paul Graves, Jr. K.C. Murphy Eric Shaw Douglas Barker Ira Grossman Lauren J. Murphy Anne Slepski Rafael Bradley Dan Hebner Cheryl Ottley J.P. Smith Huiyong Chen Mary Henry Susan Papierski Uan-Kang Tan John R. Cole Katherine Holbrow Carol Piacentini Diane Weir Jon Cronin Paul Iorio Sheila Pelczarski John Wentworth Jon Daniels Marcus Jetter Candace Purseglove Surasu Wati Geoff Dangerfield Edward Kaynor Joseph Randall Adrian White Annette Dion Dan Kelleher Sara Rausher Dora Wong Joseph Eno Jeremy Lafrancois Lori Rocchio Abdulkadir Yusuf
vi Candace Purseglove, Amherst, John McElwee, Dennis MA MA
Finally, the ARM Project relied on thousands of citizen volunteers who each cared about the quality of Massachusetts lakes and streams. We dedicate this final compilation of Project results to them and hope that the information they collected and their experiences as ARM volunteers will help protect our environment for future generations.
The Cohens, Medfield, Ma
vii Table of Contents
Abstract ...... i Foreword ...... iii Acknowledgment ...... iv Table of Contents ...... viii List of Figures ...... xi List of Tables ...... xv 1.0 Introduction ...... 1.1 1.1 Mission ...... 1.1 1.2 Budget History ...... 1.2 2.0 Methods ...... 2.1 2.1 Network Organization ...... 2.1 2.2 Sampling Site Selection ...... 2.3 2.2.1 Phase I ...... 2.3 2.2.1.1Testing for Bias in Phase I ...... 2.4 2.2.2 Phase II ...... 2.5 2.2.3 Phase III ...... 2.6 2.2.4 Site Selection Summary ...... 2.9 2.3 Field Collection ...... 2.10 2.3.1 Phase I ...... 2.10 2.3.2 Phase II ...... 2.14 2.3.3 Phase III ...... 2.14 2.4 Sampling Dates ...... 2.14 2.5 Local Laboratory Analysis ...... 2.16 2.6 Central Laboratory Analysis ...... 2.17 2.7 Quality Control ...... 2.19 2.7.1 Target Population of Lakes and Streams ...... 2.19 2.7.2 Sampling Sites ...... 2.20 2.7.3 Volunteer Collection ...... 2.22 2.7.4 Local Labs -- pH and ANC ...... 2.24 2.7.4.1Phase I ...... 2.24 2.7.4.2Phase II ...... 2.27 2.7.4.3Phase III ...... 2.31 2.7.5 Central Lab -- Inorganic Chemistry ...... 2.35 2.8 Conventions ...... 2.37 2.8.1 Analytical Units ...... 2.37 2.8.2 Number of Sites ...... 2.38 2.9 Data Management and Analysis ...... 2.39 2.10 Volunteer Motivation and Reward ...... 2.40
viii 2.10.1 Rock Concerts ...... 2.41 2.10.2 Media Reports ...... 2.41 2.10.3 Identification of Volunteers ...... 2.42 2.10.4 Buttons and Decals ...... 2.45 2.10.5 Newsletters ...... 2.45 2.10.6 Presentations to Volunteer Groups ...... 2.46 2.10.7 Publications, Presentations, Testimony Public Debate, and Awards ..... 2.46 2.10.8 Appreciation Parchment ...... 2.52 2.10.9 Calendar ...... 2.53 3.0 Comprehensive Survey -- Phases I &II ...... 3.1 3.1 pH and ANC ...... 3.1 3.1.1 General Results ...... 3.1 3.1.2 Seasonal Patterns of pH and Acid Neutralizing Capacity ...... 3.3 3.1.3 Sensitivity of Massachusetts Surface Waters to Acidification ...... 3.4 3.1.4 Comparison of Lakes vs. Streams ...... 3.6 3.1.5 Geographic Distribution of Water Bodies According to their Sensitivity . 3.10 3.1.6 Statistical Analysis of Geographic Variation in Lake ANC ...... 3.10 3.1.7 Comparison with other Large Scale Surveys ...... 3.12 3.2 Ion Chemistry ...... 3.25 3.2.1 Specific Methods ...... 3.25 3.2.2 Delineation of Regions ...... 3.25 3.2.3 Lakes and Ponds ...... 3.27 3.2.3.1Regional Chemistry, October 1984 ...... 3.27 3.2.3.2 Other Factors in Determining Regional Chemistry ...... 3.30 3.2.3.3Road-Salt Influence ...... 3.31 3.2.3.4Sulfate Sources ...... 3.31 3.2.4 Streams and Rivers ...... 3.33 3.2.5 Comparison With Other Surveys ...... 3.36 3.3 Identification of Road Salt Contamination Using Multiple Regression and GIS .. 3.41 3.3.1 Introduction ...... 3.41 3.3.2 Methods ...... 3.42 3.3.3 Results ...... 3.43 3.3.3.1Sea-spray Precipitation Inputs ...... 3.43 3.2.3.2Inputs from Roads ...... 3.44 3.3.3.3Combined Multiple Regression Model ...... 3.45 3.3.4 Discussion ...... 3.46 3.3.4.1.General ...... 3.46 3.3.4.2Model reliability ...... 3.48 3.3.5 Conclusions ...... 3.49 4.0 Trend Analysis - Phase III ...... 4.1 4.1 Introduction ...... 4.1 4.2 Methods ...... 4.1 4.2.1 Sample Selection ...... 4.1 4.2.2 Statistical Analysis ...... 4.2
ix 4.2.2.1Detrending for Runoff ...... 4.2 4.2.2.2Temporal Trend Analysis ...... 4.3 4.2.2.3Meta-Analysis ...... 4.6 4.3 Analysis and Interpretation of Trends ...... 4.7 4.3.1 Streams and Rivers ...... 4.7 4.3.1.1Raw Data ...... 4.7 4.3.1.2Trends in Raw ANC and pH Data ...... 4.7 4.3.1.3Trends in Residual ANC and pH Data ...... 4.7 4.3.1.4Meta-Analysis Results of Trends in ANC ...... 4.11 4.3.1.5Other Ion Trends in Streams ...... 4.14 4.3.2 Lakes and Ponds ...... 4.14 4.3.2.1 Raw Data ...... 4.14 4.3.2.2 Trends in Raw ANC and pH Data ...... 4.14 4.3.2.3 Trends in Residual ANC and pH Data ...... 4.19 4.3.2.4 Meta-Analysis Results of Trends in ANC in Lakes ...... 4.19 4.3.2.5 Other Ion Trends in Lakes ...... 4.23 4.4 Discussion and Synthesis ...... 4.25 4.4.1 Site Effects ...... 4.27 4.4.2 Relative Sensitivity of High vs. Low Alkalinity Waters ...... 4.28 4.4.3 Comparison with other Trend Studies ...... 4.32 4.4.4 Meta-analysis and SKT Power ...... 4.33 4.5 Summary of Trend Analyses ...... 4.33 5.0. A Profile of the Acid Rain Monitoring Project Citizen Volunteer ...... 5.1 6.0 Conclusions ...... 6.1 7.0. Literature Cited ...... 7.1 8.0. Appendices ...... 8.1 8.1 List of Laboratories Participating in the ARM Project between 1983 and 1993. .. 8.2 8.2 Alkalinity protocol for local laboratories ...... 8.8 8.3 Equipment care, use and maintenance procedures...... 8.9 8.4 ARM's Statewide Coordinator's How To Guide ...... 8.14 8.5 List of Equipment used on ARM Project — Environmental Analysis Laboratory ...... 8.16 8.6 Quality assurance procedures for volunteer collectors and laboratories excerpted from Project Quality Control Plan...... 8.17 8.7 U.S. EPA Water Supply and Water Pollution Audit Program Results ...... 8.19 8.7.1 pH...... 8.19 8.7.2 ANC...... 8.20 8.7.3 Sulfate...... 8.21 8.7.4 Calcium...... 8.22 8.7.5 Magnesium...... 8.23 8.7.6 Sodium...... 8.24 8.7.7 Cadmium...... 8.25 8.7.8 Chromium...... 8.26 8.7.9 Copper...... 8.27
x 8.7.10 Lead...... 8.28 8.7.11 Nickel...... 8.29 8.7.12 Zinc...... 8.30 8.7.13 Results of Other E.P.A. Water Pollution and Water Supply Audits...... 8.31 8.8 Database ...... 8.33 8.8.1 Database Structure ...... 8.33 8.8.2 Codes, Flags, and Limitations ...... 8.35 8.8.3 Massachusetts Town Codes -- Town, Alphabetic Code Number and County...... 8.36 8.8.4 dBase Quality Control Program ...... 8.40 8.9 Conversion Factors for Ions Measured in the Acid Rain Monitoring Project. ... 8.47 8.10 ARM Phase 3 sites...... 8.48 8.11 Responses to Volunteer Questionnaire ...... 8.71
xi List of Figures
2.7-1 Relative size frequency of ponds and lakes in the combined ARM phases I and II versus all named ponds and lakes...... 2.21 2.7-2 Comparison of pH results from volunteer collections and near-simultaneous (but unknown to the volunteers) professional staff collections. The regression is Vol = 0.45 + 0.93*Pro where N = 135, R2 = 0.949 and the root mean square error = 0.17...... 2.23 2.7-3 Comparison of Alkalinity (ANC) results from volunteer collections and near- simultaneous (but unknown to volunteers) professional staff collections. The regression is: Vol = 0.21 + 1.00*Pro where N = 135, R2 = 0.986 and the root mean square error = 1.94 ...... 2.23 2.7-4 Volunteer lab pH vs. expected pH...... 2.28 2.7-5 Volunteer lab ANC vs. expected ANC...... 2.29 2.7-6 Standard deviation of alkalinity (ANC) for the two regular samples over the course of the ARM Phase 3 program. The average standard deviation is 1.14 mg L-1, equivalent to a coefficient of variation of 9.7%...... 2.33 2.7-7 Standard deviation of pH for the two regular samples over the course of the ARM Phase 3 program. The average standard deviation is 0.14 pH units...... 2.33 2.7-8 Comparison of Gran plot ANC titration vs. double end point titration results on thirteen selected lakes from April 1991 and 37 randomly selected samples from April 1992. The regression is: Gran ANC µeq L-1 = 4.7 + 1.02 * (double end point ANC), n = 50, r2 >0.999...... 2.35 2.7-9 Comparison of spectrophotometrically determined color at 425 nm vs. dissolved organic carbon for 13 selected lakes from April 1991 and 37 randomly selected sites from April 1992. The regression is: DOC mg L-1 = 1.8 + 0.057 * PCU, n = 50, r2 = 0.94...... 2.38 2.10-1 Daily Hampshire Gazette announcement of ARM benefit concert, November 21, 1982 ...... 2.41 2.10-2 Ed Driscoll presenting Senator Edward Kennedy with an ARM t-shirt on Mt. Greylock with Ernie LeClair and Gene Chague (left to right) looking on. .... 2.42 2.10-3 1987 T-shirt design, color on white background. Towns with median sensitivities in acidified or critical category - red: in endangered to sensitive category - orange; in not sensitive category - green (see Fig. 3.1-7 and Table 3.1-2), Design by ARM staff ...... 2.43 2.10-4 1988 T-shirt design, red on yellow background. Winner of statewide school contest. Design by Mariah Peelle, Shutesbury Elementary School...... 2.43 2.10-5 1990 T-shirt design, purple on light green. Original art by Scott Landry, University of Massachusetts student...... 2.44 2.10-6 Pin-on Button (blue on white)...... 2.45 2.10-7 Stick-on decal (blue on white)...... 2.45 2.10-8 President George Bush and Dr. Paul Godfrey at the Rose Garden reception for National Environmental Achievement Award Winners ...... 2.49
xii 2.10-9 Reception at the Norwegian Embassy, Washington, D.C. for National Achievement Award Winners...... 2.50 2.10-10 Rose Garden reception for National Environmental Achievement Award Winners . 2.50 2.10-11 Example of the Certificate of Appreciation presented to volunteer sample collectors...... 2.51 2.10-12 ARM Calendar ...... 2.52 3.1-1 Lake pH vs. ANC for Lakes with ANC Less Than 800 ueq L-1. The upper curved line represents the relationship expected for lakes with carbonate ANC in -3.5 equilibrium with atmospheric carbon dioxide (pCO2 = 10 atm.). The lower line -2.55 represents the curve of best fit to the pH and ANC data (pCO2 = 10 ) ...... 3.2 3.1-2 Stream pH vs. ANC for streams with ANC Less than 800 ueq L-1. The upper curved line represents the relationship expected for streams with carbonate ANC -3.5 in equilibrium with atmospheric carbon dioxide (pCO2=10 atm.). The lower -2.4 line represents the curve of the best fit to the pH and ANC data (pCO2=10 atm.)...... 3.2 3.1-3 Median monthly pH (March 1983 - April 1984, October 1984 & April 1985) compared to seasonal means and medians...... 3.3 3.1-4 Median monthly ANC (March 1983 - April 1984, October 1984 & April 1985) compared to seasonal means and medians ...... 3.4 3.1-5 Cumulative frequency of pH for ARM Phase II, April and October Sites...... 3.5 3.1-6 Cumulative Frequency of ANC for ARM Phase II, April and October sites...... 3.5 3.1-7 Comparison of Lotic vs. Lentic Sites for Phase I and II Combined...... 3.7 3.1-8 Mean Alkalinity by Town of Massachusetts Surface Waters ...... 3.11 3.1-9 Location of Lakes in the State Within Five Alkalinity Classes (data from October 1983 and 1984)...... 3.12 3.1-10 Assessment of EPA statistical site selection process on the development of the target population for potential sampling (based on Johnson et al., 1989) versus lakes in the PALIS listing (Ackerman et al., 1984) categorized by area. ... 3.13 3.1-11 Comparison of the ARM sampled and estimated lake population with that of the EPA ELS...... 3.14 3.1-12 Comparison of ARM double endpoint ANC vs. ELS Gran titration ANC for lakes sampled by both in the fall of 1984. Actual sampling dates may differ by as much as three weeks...... 3.15 3.1-13 Comparison between ARM pH and ELS closed head space pH for lakes sampled by both in the fall of 1984. Actual sampling dates may differ by as much as three weeks...... 3.16 3.1-14 Comparison of ARM pH vs. ELS air equilibrated pH for lakes sampled by both in the fall of 1984. Samples were collected as much as three weeks apart. ... 3.16 3.1-15 Number of lakes and ponds in low pH categories from ARM (October, 1984) and EPA/ELS ...... 3.18 3.1-16 Numbers of Lakes and Ponds in Low ANC Categories from ARM (October) and EPA/ELS...... 3.18
xiii 3.2-1 Regions of similar Bedrock Type in Massachusetts. The regions and major rock types are listed from west to east: (1) Stockbridge Region — composed of calcitic and dolomitic marble, with small amounts of quartz phyllite; (2) Berkshire Region -- is mainly quartz gneiss with large areas of schist and mica quartzite. We have included some small isolated areas located to the west of the Stockbridge Region; (3) Connecticut Valley Region -- consists of conglomerate sedimentary rocks of arkose, siltstone, and sandstone; (4) Central Highland Region -- comprised mainly of gneiss, quartz, granite, and granofels with a wide band of sulfitic schists oriented north-south; (5) Boston Region -- contains a mix of schist, gneiss, granite, tonalite, and diorite with sedimentary rocks including conglomerate sandstone, siltstone, greywacke, and shale in the south; (6) Cape Cod Region -- this area is mainly granite, gneiss, and schist which is overlain by cretaceous sediments of clay, silt, sand, and gravel deposits and includes portions of the mainland and the offshore islands...... 3.26 3.2-2 Selected Chemistry by Region for October Data. Vertical bars represent the interquartiles (25th and 75th percentiles), the horizontal line on the bar is the median and the line connecting the bars represents the means. Regions are as in Figure 1. Note change of scale between panels...... 3.29 3.2-3 Ionic balance for lakes within six regions (October 1984 mean data). The left and right side of each pair of bars represents cations and anions, - - respectively. Titration ANC is represented here as HCO3 ; R is calculated by difference in the charge balance and is assumed to be organic acid anions...... 3.30 3.2-4 Ionic balance for streams within the six regions (October mean data). The left and right side of each pair of bars represents cations and anions, respectively. Titration - - ANC is represented here as HCO3 ; R is calculated by difference in the charge balance and is assumed to be organic acid anions...... 3.32 3.2-5 Comparison of cations in lakes versus streams for the six regions (October 1984 median data)...... 3.33 3.2-6 Comparison of Iron, Manganese and Aluminum for lakes versus streams in the six regions (October 1984 median data)...... 3.35
3.2-7 Comparison of Color (solid) and SiO2 (hatched) for lakes versus streams in the six regions (October median data)...... 3.35 3.2-8 Comparison of Massachusetts lakes and streams sulfate data versus other sites as reported in Baker et al. (1990)...... 3.39 3.2-9 Comparison of Massachusetts lakes and streams base cation data versus other sites as reported by Baker et al. (1990)...... 3.39 3.2-10 Comparison of Massachusetts lakes and streams organic acid data versus other sites as reported by Baker et al. (1990)...... 3.40 3.3-1 Road salt study sites. The watersheds of the streams are shown as shaded polygons. Only the major highways and major state roads are shown...... 3.43 3.3-2 Map of stream sodium concentrations in Massachusetts. The radius of the circle is proportional to the sodium concentration...... 3.44
xiv 3.3-3 Sodium observed vs sodium predicted from road salt loading calculations. Predicted sodium concentration is based on the reported rate of 11,3000 kg NaCl/lane km (see text)...... 3.45 4.2-1 Example of steps in the analysis of trends in ANC shown for Cronin Brook, MA. (A) Quarterly stream water ANC concentrations vs. time. (B) ANC vs. daily runoff from nearby gaging station on the Quinsigamond River in North Grafton, MA. The hydrology data based on Water Resources Data Massachusetts and Rhode Island Water Year 1982-1992 publications and direct gage readings. The line is from nonlinear regression fit to the data as described in the text. (C) The detrended residuals vs. time. The SKT test shows significant increases in ANC at this site (Z=2.59, á<0.01)...... 4.4 4.2-2 Example of steps in the analysis of trends in ANC for Ramshorn Pond, MA. (A) Quarterly stream water ANC concentrations vs. time. (B) ANC vs. daily runoff from a nearby gaging station on the Quinsigamond River in North Grafton, MA. The hydrology data based on Water Resources Data Massachusetts and Rhode Island Water Year 1982-1992 publications and direct gage readings. The line is from nonlinear regression fit to the data as described in the text. (C) The detrended residuals vs. time. The SKT test shows significant increases in ANC at this site (Z=2.am water ANC vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes= +2.4 µeq/L/year and +1.5 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only...... 4.5 4.3-1 Median stream water ANC vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes= +2.4 µeq/L/year and +1.5 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only...... 4.8 4.3-2 Median stream water pH vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = +0.021 pH units/year and +0.014 pH units/year), representing detrended and raw data trends respectively, are shown for visual reference only...... 4.8 4.3-3 Median stream sulfate vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = -1.8 µeq/L/year and -1.6 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only...... 4.9 4.3-4 Median sum of stream base cations vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = +1.0 µeq/L/year and -1.4 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only...... 4.9 4.3-5 Frequency distribution of Z scores for streams from the seasonal Kendall test on raw ANC...... 4.10 4.3-6 Frequency distribution of z scores for streams from the seasonal Kendall test on residual ANC after detrending for hydrology...... 4.10
xv 4.3-7 Trends for streams in major ions and ANC by season with the overall yearly trend for comparison. The left and right side of each pair of bars represent the cations and anions, respectively. Titration ANC is shown as a line. Increasing trends are shown as positive values, decreasing trends as negative values, and nonsignificant values are not shown...... 4.15 4.3-8 Median lake water ANC vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes +2.4 µeq/L/year and +1.5 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only...... 4.17 4.3-9 Median lake water pH vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = +0.013 pH units/year and +0.010 pH units/year), representing detrended and raw data trends respectively, are shown for visual reference only...... 4.17 4.3-10 Median lake sulfate vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = -1.10 µeq/L/year and - 1.39 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only...... 4.18 4.3-11 Median sum of lake base cations vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = +1.51 µeq/L/year and +0.98 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only...... 4.18 4.3-12 Frequency distribution of Z scores for lakes from the seasonal Kendall test on raw ANC...... 4.20 4.3-13 Frequency distribution of z scores for lakes from the seasonal Kendall test on residual ANC after detrending for hydrology...... 4.20 4.3-14 Trends for lakes in major ions and ANC by season with the overall yearly trend for comparison. The left and right side of each pair of bars represent the cations and anions, respectively. Titration ANC is shown as a line. Increasing trends are shown as positive values, decreasing trends as negative values, and nonsignificant values are not shown...... 4.25 4.4-1 Distribution of significant (á<0.05) trends in adjusted ANC for 330 streams in Massachusetts. The solid triangles indicate upward trends, the open triangles represent downward trends and the dots represent no trend...... 4.27 4.4-2 Distribution of significant (á<0.05) trends in adjusted ANC for 181 lakes in Massachusetts. The solid triangles indicate upward trends, the open triangles represent downward trends and the dots represent no trend...... 4.28 4.4-3 The median trend slope for each of 327 streams is shown vs. the mean ANC of the stream. The best fit regression is shown. Three high ANC streams are omitted for clarity...... 4.30 4.4-4 The median trend slope for each of 181 lakes is shown vs. the mean ANC of the lake. The best fit regression line is shown ...... 4.30 5.1-1 Years of involvement of ARM volunteers...... 5.1 5.1-2. Phases of involvement of ARM volunteers for those who complete one or more phases ...... 5.2
xvi 5.1-3. Geographic distribution of volunteer's state of origin ...... 5.3 5.1-4. Volunteer occupations ...... 5.3 5.1-5. Estimated household income of ARM Volunteers ...... 5.4 5.1-6. Age distribution of ARM volunteers ...... 5.4 5.1-7. Volunteer satisfaction with effort by groups to solve acid rain problem...... 5.5 5.1-8. Volunteer promotion of ARM Project and solutions to acid rain problem...... 5.5
xvii List of Tables
2.1-1 Responsibilities of Participants in the ARM Project...... 2.2 2.2-1 Comparison of the number of priority waters sampled with the total priority waters listed and the total number of waters sampled (April, 1983)...... 2.5 2.2-2 Chi-square analysis of the frequency of sampling of priority water bodies. No significant difference indicates random sampling...... 2.5 2.2-3 Results of regression of various physical factors against pH, alkalinity and ARM alkalinity category. April, 1985 data were used for the regressions involving the pH and alkalinity parameters. Regressions involving the ARM alkalinity category parameter were done using data from both April, 1985 and the winter-spring months of ARM I (1983 and 1984)...... 2.8 2.2-4 ARM III Random Site Distribution by Category...... 2.9 2.3-1 Sampling Protocol ...... 2.11 2.3-2 Sample Site Location Form ...... 2.12 2.3-3 Data collection form...... 2.13 2.3-4 Sample collection and preservation protocols...... 2.15 2.4-1 Schedule of Collections...... 2.16 2.6-1 Analytical procedures...... 2.18 2.7-1 Test of difference between lake shore sites used by volunteers and lake center. All samples collected by Project staff on October, 1990. N=21, pH mean difference = 0.0067, pH standard deviation of differences = 0.1059, ANC mean difference = -0.2619, ANC standard deviation of differences = 2.207; t value for pH = 0.2816, t value for ANC = -0.531...... 2.22 2.7-2 Comparison of U.S.E.P.A. Quality Assurance Sample Values and Sample Measurements by EAL...... 2.25 2.7-3 Comparison of U.S.E.P.A. pH Values for Acid Rain Certification Program with Sample Measurements by EAL...... 2.25 2.7-4 Means and Standard Deviations of Replicate Measures for EAL Quality Control Samples...... 2.25 2.7-5 Average difference between monthly mean observed and expected quality assurance values in ARM I and ARM II...... 2.30 2.7-6 Mean results of double blind quality assurance samples for all local labs, compared with results for only labs accepted following quality assurance testing and the expected values...... 2.31 2.7-7 Chronology of quality control procedures...... 2.36 2.7-8 Results of double blind analysis of 14 replicate EPA quality assurance solutions .. 2.36 3.1-1 Site mean averages for pH and ANC, ARM I and ARM II...... 3.1 3.1-2 Statewide distribution of surface water samples in each ARM ANC Category and mean/median pH and ANC for ARM I and ARM II combined...... 3.6 3.1-3 Number and Percent of Streams per Sensitivity Category ...... 3.8 3.1-4 Number and Percent of Lakes per Sensitivity Category ...... 3.9
xviii 3.1-5 The difference between the percentage of lakes in each sensitivity category and the percentage of streams in that category...... 3.10 3.1-6 Analysis of variance in ANC among 660 lakes between (A) GLM model of six factors and (B) ANOVA of region on ANC. October 1983 and 1984 ANC data...... 3.12 3.1-7 Comparison of ARM October data with EPA/ELS. Sample counts are actual numbers of water bodies sampled. ARM estimate of total extends sample count percentages to all named freshwater lakes and ponds. Estimate of total and upper confidence limit (U.C.L.) estimates of total for EPA/ELS are statistical estimations...... 3.17 3.1-8 Comparison of ANC and pH distribution of ELS Northeast Lakes and ARM lakes expressed as a percentage of the target population...... 3.19 3.1-9 Comparison of Adirondack Lake Survey Corporation Survey results with ARM lake results...... 3.19 3.1-10 Comparison of ANC from Canadian surveys reported in Baker et al. (1990) with ARM...... 3.20 3.1-11 Comparison of pH from Canadian surveys reported in Baker et al. (1990) with ARM...... 3.21 3.1-12 Comparison of large-scale survey results for pH from Sweden as reported in Baker et al. (1990) with ARM results...... 3.22 3.1-13 Comparison of large-scale survey results for pH from Norway as reported in Baker et al. (1990) with ARM results...... 3.22 3.1-14 pH distribution of surface water chemistry survey results (based on Haines and Akielaszek, 1983)...... 3.23 3.1-15 ANC distribution of surface water chemistry survey results (based on Haines and Akielaszek, 1983). ANC values are in µeq L-1 ...... 3.24 3.2-1 Median chemistry for major cations and anions (Oct. 1984), along with physical data for Massachusetts lakes, and data subdivided by region within state. Median nitrate was below the limit of detection...... 3.28 3.2-2 Median chemistry for major cations and anions (Oct. 1984), along with physical data for Massachusetts streams, and data subdivided by region within state. Median nitrate was below the limit of detection...... 3.34 3.2-3 Comparison of ARM data with areas of high and low deposition throughout
the world. N = Number of sites surveyed, CB = total base cations in µeq/L, SO4 = sulfate in µeq/L, A- = organic acids, and ANC as the percent < 0 µeq/L. .. 3.37 3.3-1 Descriptive Statistics and Multiple Regression Results...... 3.46 4.2-1 Serial correlation for 181 lakes after removing seasonal means...... 4.2 4.3-1 Median ANC trend for the 81 streams with a significant trend among the total of 330 streams...... 4.11 4.3-2 ÷2 (ANOCHIS) Table Showing an Analysis of Trend and Homogeneity for Residual ANC in 330 streams...... 4.13 4.3-3 Cation trends in 330 Massachusetts streams. Note: ÓBC=sum of base cations. In most cases the trends were not homogenous across sites and seasons as described in the text. All trends are significant at á<0.001 except as noted...... 4.16
xix 4.3-4 Anion trends in 330 Massachusetts streams. Note: AA=sum of acid anions. In most cases the trends were not homogenous across sites and seasons as described in the text. All trends are significant at á<0.001 except as noted...... 4.16 4.3-5 ÷2 (ANOCHIS) Table Showing an Analysis of Trend and Homogeneity for Residual ANC in 181 lakes...... 4.19 4.3-6 ÷2 (ANOCHIS) Table Showing a Seasonal Analysis of Trend and Homogeneity for Residual ANC in 181 lakes...... 4.21 4.3-7 Median ANC trend for the 50 lakes with a significant trend among the total of 181 lakes...... 4.22 4.3-8 Cation trends in 181 Massachusetts lakes. Note: ÓBC=sum of base cations. In most cases the trends were not homogenous across sites and seasons as described in the text. All trends are significant at á<0.001 except as noted...... 4.24 4.3-9 Anion trends in 181 Massachusetts lakes. Note: AA=sum of acid anions. In most cases the trends were not homogenous across sites and seasons as described in the text. All trends are significant at á<0.001 except as noted...... 4.24
xx 1.0 Introduction
In September of 1982 at a public information presentation on acid rain, the Director of the Water Resources Research Center, Dr. Paul J. Godfrey, asked the group of 30 or so attendees at a Massachusetts Audubon Sanctuary if they would be willing to help fill in a glaring gap in knowledge about Massachusetts' surface water sensitivity to acid deposition. Within a month, the rudiments of a volunteer network in the Connecticut River Valley was functional; in two months Berkshire County in extreme western Massachusetts had asked to join in. By February 1983, the entire state was organized to begin sampling in March.
In December 1992, Dr. Godfrey learned that funding for the project by the Massachusetts Division of Fisheries & Wildlife was no longer possible. Despite a desperate search for continuing funds, the project could not be saved past July 1993.
In the intervening years, there were more than a thousand volunteers, tens of thousands of samples collected, hundreds of thousands of analyses performed, involvement in the intense national acid rain debate, awards from local, state, national and international groups, thousands of media interviews, and a database of tens of thousands of records of most lakes and streams in Massachusetts.
From those years comes a story of an unusual mix of science and public service, data quality and citizen volunteers, public attention and individual involvement. It is the purpose of this report to tell that story. Some of the story appears as normal scientific results reporting, some as a guidebook to the creation of volunteer networks, and some as a description of the people and events that helped make the Acid Rain Monitoring (ARM) Project a memorable effort and a database for future generations.
1.1 Mission
The Acid Rain Monitoring Project had two primary missions. From 1983 to 1986, the first two phases of the project sought to develop a comprehensive picture of the sensitivity of Massachusetts surface waters to acid deposition. From 1986 to 1993, the mission was to determine the long-term trend in sensitivity with respect to acid deposition levels and to evaluate the effect of state and federal emission controls. During the entire project, ARM also sought to demonstrate the utility of using citizen volunteers in developing a scientifically valid database and to make this database useable to researchers, managers and policy makers
1.1 1.2 Budget History
FY Budget Source 19831 14,500 TU 19841 45,500 DF&W 19841 19,888 USGS 19852 281,878 DF&W 1986 $156,985 DF&W 1987 $183,671 DF&W 1988 $184,400 DF&W 1989 $192,579 DF&W 1990 $200,074 DF&W 1991 $200,002 DF&W 1992 $200,000 DF&W 1993 $200,000 DF&W 1994 $24,823 USGS 1 Phase I study of 1000 lakes and streams monthly for 14 months with limited chemical analysis; 2 Phase II study of 2500 lakes and streams in October and April; all other years are for phase III study of 750 sites, four times annually. Sources: TU = Mass./Rhode Island Chapter and National Trout Unlimited; DF&W = Massachusetts Division of Fisheries & Wildlife; USGS = U.S. Geological Survey Water Resources Institute Program
1.2 2.0. Methods 2.1. Network Organization
The primary division of responsibility for sampling and analysis was by county, rather than by membership in other organizations. For the ARM Project, each county was represented by a volunteer who served as county coordinator. Numerous environmental and sportsmens’ organizations assisted in identifying individuals who might be interested in becoming ARM coordinators. Initially, the county coordinator worked closely with the county regional planner and other individuals to further subdivide the county into working groups called districts and to identify potential laboratories in each district. Each county was comprised of four to ten districts, depending on the number of sites being sampled, number of sample collection volunteers, and the geopolitical structure of the county. Districts generally consisted of towns that normally work together in other areas of public activity. The county coordinator also arranged two public presentations. The first was a general meeting to provide an overview of the acid rain problem and to instruct volunteers in appropriate sample collection techniques. The second was a meeting of participating laboratories to define common analytical procedures and describe the nature of the quality control program in which they would participate.
Volunteers within each district were supervised by a district coordinator, who was also responsible for acting as a local liaison for the district's analytical laboratory or laboratories. In Phase I, the district group decided which streams, lakes and reservoirs were to be sampled, starting with a suggested list compiled from state records; they resolved the specific logistics of collection and analysis at the local laboratories. Each county had between 30 and 150 participating volunteers. The role of each participant was specified in a handout given to each early in the project (Table 2.1-1).
After the preliminary organizational meetings, the district coordinator became the key liaison person in the ARM Project, assuring that the volunteers collected and delivered samples as arranged and that the laboratories did their part. The county coordinator assumed the principal role of local media contact person. In phase III, the role of the county coordinator began to assume many of the characteristics of the previous district coordinator because there were fewer volunteers and labs. This fluidity based on need by the county coordinators provided much of the longevity and flexibility of the project.
The initial volunteer network was not changed throughout the three phases of the project. There was some change in individuals acting as county and district coordinators as a result of attrition In Phase II, additional volunteers were sought to help with the greatly increased sampling load. Many volunteers were not asked to continue in phase III only because the number of sites was much reduced. More than a few wished they could have continued. During Phase III, replacement volunteers were occasionally needed as people moved, changed their other responsibilities, or, in some cases, died. When a volunteer resigned, it was his/her responsibility to find a replacement. If this didn't happen, it was usually the district coordinator who found a replacement and did the collection in the interim. Occasionally the statewide coordinator had to get involved, primarily through calling county and district coordinators and other volunteers for potential names. Articles and advertisements were sometimes sent to local newspapers, radio stations and specialized circulars to assist in the search for volunteers..
2.1 Table 2.1-1. Responsibilities of Participants in the ARM Project.
Sampling Volunteers 1. Will sample on a monthly basis throughout the hydrologic year (March 1983 - March 1984) (semiannually from 1984-1986 and quarterly from 1986 to 1993). 2. Will collect samples using the technique demonstrated. 3. Will be responsible for transporting samples to analytical station. 4. Will be responsible for finding a substitute to do monthly sampling if unable to sample in any given month. 5. Will be responsible for contacting the district coordinator when problems arise. 6. Will attend all follow-up meetings. 7. Will notify district coordinator of terminating. 8. Will find a replacement volunteer.
District Coordinator 1. Will deliver samples to analytical station if necessary. 2. Will mail data to WRRC if necessary. 3. Will maintain sufficient supplies of clean sampling bottles for group members. 4. Will contact WRRC or county coordinator if problems arise. 5. Will facilitate maximum coverage of district waters by volunteer samplers. 6. Will help find volunteer and lab replacements.
County Coordinator 1. Will coordinate districts within county. 2. Will arrange a county meeting(s) site for the Acid Rain Monitoring Project. 3. Will assist district coordinator in resolving problems. 4. Will have supplies of clean sampling equipment for district coordinators. 5. Will ensure sampling of all priority waters in the county. 6. Will help find volunteer and lab replacements.
Water Resources Research Center 1. Will conduct the research project. 2 Will analyze and report on data generated from the project. 3. Will provide sampling equipment needed. 4. Will demonstrate sampling techniques to participants. 5. Will act as media contact and spokesperson for the project, produce information releases, advise state and national legislators on results, seek ongoing support and publish scientific reports. ______Items in italics are responsibilities added after the initiation of the ARM project.
Communication between the statewide coordinator, county and district coordinators and volunteers varied depending on the form of communication. We learned early in the program that newsletters, data results and other regularly scheduled communications should be sent directly to all volunteers. In the first month or so of the Project, we attempted to save money and time by distributing data results to coordinators and assumed that they would copy and distribute them to volunteers. For a variety of reasons, this often did not happen. The neglected volunteers perceived project mismanagement. The personal attention of sending results and newsletters promptly to volunteers helped keep their loyalty. All sampling dates, changes in collection routine and instructions were also distributed directly. However, there were times when rapid communication was needed between the statewide coordinator and the volunteers. In those few cases, we used the coordinators in a telephone tree. We could communicate with all volunteers, even when there were
2.2 more than one thousand by making approximately 13 phone calls and no coordinator had to call more than another 13 people.
Over seventy local laboratories participated in the first phase of the project. They included universities, colleges and high schools, wastewater treatment facilities, water supply facilities, federal and state agency laboratories, environmental education centers and private environmental and water quality laboratories (Section 8.1). None of these local laboratories received any remuneration for their services.
In Phase II, the number of local laboratories was reduced to 25 to reduce the logistical problems associated with maintaining the quality control program. There were approximately two labs per county. This number was reduced by attrition with minimal replacement in Phase III to approximately 17 to 21 over the eight years of that phase. The number of labs asking to leave the project at any time during the ten years was remarkably small. Lab turnover in the last 9 years was minimal (after the major reduction to less than 25 labs).
The statewide coordinator was in touch with the volunteer labs, at least quarterly, by sending them QA samples and paperwork. The labs were supposed to call in their results to the statewide coordinator at least a week prior to collection. If they didn't, the statewide coordinator called them. Ion sample bottles, supplies and paperwork were then sent to each lab After the collection, Project personnel would drive to each lab to collect the completed paperwork and ion sample bottles. Each visit provided an opportunity to check on problems and successes.
Sample bottles were kept in each district; there were usually two sets of bottles so that samplers could pick up a clean set when they dropped off their samples at the lab. In some cases, the district coordinator was responsible for distributing bottles before the collection day. District coordinators were also responsible for distributing paperwork (collector's sheets, site location sheets, etc.) to the volunteer collectors.
2.2 Sampling Site Selection 2.2.1 Phase I
The selection of lakes, streams and reservoirs to sample was left largely to the volunteers. They were given two general suggestions for site selection:
1. Headwater streams in all areas were of high priority. 2. For all lakes, streams, and reservoirs not covered by the above and not on the priority list, an experimental approach was suggested. Each month a group of water bodies might be selected for a sampling survey. Those that were found to have 20 mg L-1 of alkalinity or less should be considered top priority and sampled monthly. Those above 20 mg L-1 may be sampled only once or twice per year.
Further, volunteers were given specific suggestions regarding which water bodies to sample in their region. For each county, a priority list of surface waters was developed, based on data from the files of the Massachusetts Department of Environmental Quality Engineering and Division of
2.3 Fisheries and Wildlife. These lists included all surface waters with data less than 10 years old exhibiting alkalinities below 20 mg/l. Volunteers were encouraged to use the lists as a starting point in their selection process. They were further asked to make their selections on the basis of their own assessments of water bodies important to their communities. Given this guidance, the volunteers and their district coordinators made the final selection of sites to sample.
2.2.1.1 Testing for Bias in Phase I
We have evaluated whether this process of water body selection in Phase I produced a random selection or one biased toward those on the priority lists. Table 2.2-1 lists, by county, the number of "priority" waters sampled relative to the total priority waters listed and the total sampled for April, l983. It is clear that, in most counties, the priority lists were relied upon to select water bodies to sample. If random sampling had occurred, one would expect that the ratio of priority waters sampled to the total number sampled in a county would not differ significantly from the ratio of total priority waters on the provided lists to the total number of water bodies in the state. In total, 531 lakes, streams and reservoirs were on the priority lists and there are a total of 4702 water bodies in the state. The resulting ratio is 0.112. The total number of water bodies sampled in April l983 was 1795; 382 were on priority lists. This ratio is 0.213. As the chi-square analysis (Table 2.2-2) demonstrates, lakes and streams were not sampled randomly for the state as a whole. Lake and stream selection in Franklin, Hampden, Middlesex, and Norfolk counties was not significantly different from random. Berkshire and Essex came quite close to meeting the statistical criterion for random sampling. (Berkshire county is unique in that its non-random bias is against those waters on the priority list.)
Although there is the suggestion that most phase I sites were selected partly on the basis of the priority lists provided, this does not imply that the eventual sites selected were or were not representative of all lakes and streams in the state. There was no evidence that the priority lists were other than random selections themselves. In addition, while not known at the time, the percentage of lakes and streams meeting the initial sampling criteria (ANC < 20 mg/l) was so large as to preclude relatively few sites (<17%). The sample size was so large and the adherence to these guidelines sufficiently loose that non-randomness is not certain. The results of additional tests of target population bias for phases I and II (see section 2.7.1) suggest that the priority lists did not result in noticeable bias for phase I.
2.4 Table 2.2-1. Comparison of the number of priority waters sampled with the total priority waters listed and the total number of waters sampled (April, 1983).
Priority* Priority Priority Total County Expected Sampled Listed Sampled
Barnstable 13.44 44 68 120 Berkshire 21.39 7 23 191 Bristol 10.19 25 37 91 Essex 11.98 21 27 107 Franklin 20.27 29 33 181 Hampden 25.87 28 36 231 Hampshire 9.18 20 26 82 Middlesex 26.10 27 34 233 Norfolk 15.46 21 21 138 Plymouth 19.49 41 53 174 Worcester(N) 11.98 50 72 107 Worcester(S) 15.68 69 101 140 ______Total ______201.04 382 508 1795 * based on the proportion of the statewide total of "priority" lakes and streams (508) to the total number of lakes and streams (4702). The resulting ratio (0.112) assumes that the number of water bodies on the priority list for a given county is solely a function of the number of water bodies in the county and not a function of regional differences in chemical characteristics.
Table 2.2-2. Chi-square analysis of the frequency of sampling of priority water bodies. No significant difference indicates random sampling.
County Chi-square at the 0.05 level Significant difference
Barnstable 67.2 Yes Berkshire 9.0 Yes* Bristol 20.1 Yes Essex 6.1 Yes* Franklin 3.3 No Hampden 0.1 No Hampshire 11.6 Yes Middlesex 0.006 No Norfolk 1.6 No Plymouth 22.6 Yes Worcester(N) 117.5 Yes Worcester(S) 197.9 Yes State 162.0 Yes ______* No significant difference at the 0.01 level
2.2.2 Phase II
ARM I sites were checked against a listing of ponds and lakes in Massachusetts compiled by the Massachusetts Division of Water Pollution Control (Ackerman et al., 1984) and a listing of
2.5 streams and rivers compiled by the Divisions of Water Pollution Control and Fisheries and Wildlife (Halliwell et al., 1982).
All sites appearing on either list but not sampled in ARM I were selected for ARM II. Lists of ARM II sites in each ARM district were then given to district and county coordinators and local laboratory personnel in a series of meetings held in each county in August and September, 1984. Coordinators were told to instruct volunteers not to sample any sites which were salt water or known to be seriously polluted from other causes, or which had been sampled in ARM I. The choice of sampling site was made by the volunteers, as in phase I.
2.2.3 Phase III
It was not practical to continue to sample the complete group of lakes and streams included in phases I and II for the long-term; the demands on volunteer collectors, volunteer labs and the central lab were only tolerable for the duration of Phase II. For analysis of long-term trends, we needed to develop a subsample of water bodies representative of a variety of characteristics and sufficiently large to permit statistically valid extrapolation to the unsampled population. We conducted a statistical evaluation of various criteria that could be used to develop a stratified random sample. We began with the assumption that the six acidity categories originally developed to describe phase 1 and 2 results would be the principal stratification criteria. However, we wished to determine if there were other factors that might account for significant variation so that sufficient numbers of lakes and streams with those characteristics could be included in the sample.
The 4374 water bodies sampled in the first two phases were assigned a sensitivity category according to their average winter/spring pH and alkalinity (if sampled in phase II, April values were chosen; if sampled in phase I, the average of December, January, February, March, and April values was calculated). The six categories were:
Number Category Alkalinity 1 Acidified <0 mg/l and pH#5.0 2 Critical 0 - 2 mg/l 3 Endangered 2 - 5 mg/l 4 Highly Sensitive 5 - 10 mg/l 5 Sensitive 10 - 20 mg/l 6 Not Sensitive > 20 mg/l
The April 1985 pH and alkalinity data were evaluated in relation to the following factors: bedrock geology, soil type, maximum and mean lake depth, lake elevation, lake surface area, lake type (drainage or seepage), stream length, stream gradient, and county. For some of these characteristics, there were only limited data available.
Bedrock geology category was assigned on a town by town basis by laying a transparent map of Massachusetts towns over the map of bedrock geology sensitivity produced by Norton et al.
2.6 (1982), using Norton's four bedrock sensitivity categories. All surface waters within a given town were assigned a category (1 - 4, where 1 denotes the areas most sensitive to acidification and 4 the least sensitive areas), based on the predominant category found in that town from the Norton map. Soil type was derived in the same way, using a map of sensitivity of surface and ground waters in Massachusetts to acidification as predicted by soil types (Venemann, 1984), except that in this system 1 is least sensitive and 4 is most sensitive. The county factor provided a rough measure of the correlation of geographic location with water chemistry.
Bedrock geology type and soil type were the factors best correlated with water chemistry (Table 2.2-3). That the county factor correlates significantly with the three water chemistry parameters is not surprising, given that both bedrock geology type and soil type are significantly correlated with county. Lake depth (mean and maximum), lake surface area, lake elevation, lake type, stream length, and stream gradient were either not significantly correlated with pH or alkalinity or marginally so.
From this analysis, we concluded that with sufficient geographic representation, the phase 3 subsample would be statistically representative. Geographic variation is explored further in Section 3.2.1.
Using the six ARM sensitivity categories to stratify the database, 40 streams and 40 ponds were randomly chosen in each sensitivity category to be monitored quarterly for ten years. The selected sites were sorted according to the other physical factors and a visual inspection made to see if reasonable numbers of sites occurred in each, even though no correlation was found with sensitivity. Over time, some random sites had to be abandoned (usually because they had dried up or the property owner suddenly denied access); the total number of sites in each category, the final number of random sites in each category, and the statistical weight assigned to each category are given in Table 2.2-4. However, application of these statistical weights must have an additional correction if any sites were not sampled for a particular collection date.
Lists of the randomly chosen sites were sent to the Project's volunteers, who were asked for feedback on the adequacy of each water body (was it known to be salt or brackish water, polluted, periodically dry, or totally inaccessible?). Sampling locations were to be at the same spot as they had been in phases I and II; site location sheets were sent to volunteers to help find the sites if they weren't already familiar with them. For streams, the sampling site was usually at a road crossing; for ponds, the sampling site was either at the outlet or along the bank, according to site selection protocols developed for phase I (Table 2.3-1).
Recognizing that many lakes and streams of special interest to state agencies or participating organizations might not be included in the random selection, we asked those groups to recommend additional sites and included approximately three hundred of these in the final list of sampling sites. Special interest sites are not included in any analyses requiring a statistically representative population. The complete list of sites sampled in ARM III is shown in Appendix 8.10.
2.7 Table 2.2-3 Results of regression of various physical factors against pH, alkalinity and ARM alkalinity category. April, 1985 data were used for the regressions involving the pH and alkalinity parameters. Regressions involving the ARM alkalinity category parameter were done using data from both April, 1985 and the winter-spring months of ARM I (1983 and 1984). See text for a discussion of the derivation of the bedrock geology type and soil type factors.
Factor pH Alkalinity ARM Alkalinity Category Bedrock r = 0.35 r = 0.37 r = 0.39 Geology Type1 n = 2409 n = 2391 n = 3497 p = 0.000 p = 0.000 p = 0.000 Soil Type r = -0.16 r = -0.17 r = -0.15 n = 2409 n = 2391 n = 3497 p = 0.000 p = 0.000 p = 0.000 Mean Depth NS NS r = 0.10 (lakes) n = 479 p = 0.027 Maximum Depth NS NS NS Lake Elevation NS NS r = -0.09 n = 1757 p = 0.000 Lake Surface r = 0.06 NS NS Area n = 1429 p = 0.015 Lake Type NS NS NS Stream Length r = 0.09 r = 0.10 NS n = 944 n = 938 p = 0.002 p = 0.001 Stream Gradient NS r = -0.06 r = -0.10 n = 938 n = 1642 p = 0.025 p = 0.000 County r = -0.12 r = -0.19 r = -0.10 n = 2409 n = 2391 n = 3497 p = 0.000 p = 0.000 p = 0.000 r = correlation coefficient n = number of cases p = significance level; NS = correlation not significant (p > 0.05) 1 Based on four bedrock categories (from Norton et al., 1982), where 1 is the most sensitive to acidification and 4 is least sensitive. 2 Based on four soil categories (from Veneman, 1984), where 1 is least sensitive and 4 is most sensitive.
2.8 Table 2.2-4. ARM III Random Site Distribution by Category.
Category Streams Ponds Population Random Weight Population Random Weight 1 76 40 1.900 129 35 3.686 2 209 37 5.649 368 38 9.684 3 284 38 7.474 394 40 9.850 4 312 38 8.210 433 37 11.703 5 314 39 8.026 368 37 9.946 6 277 36 7.694 295 38 7.763
Total 1472 228 1987 225
2.2.4 Site Selection Summary
The lake list (Ackerman et al., 1984) includes 2925 lakes, most greater than 2 ha and some as small as 0.4 ha in area. Not all of the lakes on the list were sampled by the ARM Project, but an additional unlisted 267 lakes were sampled. The final database for all phases of the Project contains some data for 2544 lakes representing 87% of the those on the list by Ackerman et al. (1984); 796 were part of ARM I, 1720 were part of ARM II, 57 were in both, and 378 in ARM III (85 were unique to ARM III, 225 were randomly selected and 11 of the 153 special interest lakes were limed in recent years).
The streams list (Halliwell et al., 1982) contains 2436 named streams and rivers. The final database for all phases of the ARM Project contains some data on 1670 streams representing 69% of those on the Halliwell et al. list; 602 were part of ARM I, 1023 part of ARM II, 24 sampled in both ARM I and II, and 347 in ARM III (69 were unique to ARM III, 228 were randomly selected and 10 of the 119 special interest streams had marine influences).
In total, 79% of the 5361 named and listed water bodies in the state were studied in either phase I or phase II. Many additional lakes and streams were sampled, particularly in ARM I, but their data did not pass ARM Project quality control (see Section 2.7). In addition, the data for the included sites may not be complete. For example, sites sampled only in ARM I have only pH and alkalinity; in a small number of instances either pH or alkalinity may not have passed quality standards and has been omitted from the database. Thus, for any specific analysis, the number of lakes and streams reported will vary from the numbers reported above. The consequences of this are discussed in Section 2.8.2. The lakes and streams sampled only in ARM III totalled 154. Of these sampled only in ARM III, 119 (77%) were special interest sites chosen by the sponsoring agency and other interested groups.
2.9 2.3 Field Collection 2.3.1 Phase I
All sample collection for the ARM Project was done by volunteers. Most of them had no previous experience with limnological sampling nor access to sophisticated equipment. Most had no access to boats that would permit sampling the middle of a lake. The collection protocol recognized these limitations and an appropriate educational and sampling program was developed. The collection protocol was described in detail during public presentations to elicit volunteer help. In addition, written instructions were distributed to all volunteers, and district coordinators were encouraged to assist volunteers in learning the appropriate techniques. A copy of the sample collection instructions provided to each volunteer is shown in Table 2.3-1.
Samples were collected statewide on the third Sunday of each month during the period March, l983 to April, l984. Samples that could not be delivered to a laboratory within a few hours of collection were refrigerated for preservation, with a maximum holding time of 24 hours. Although no extensive monitoring or supervison of volunteers was conducted in phase I, it appeared that volunteers scrupulously followed ARM Project instructions. This impression was supported by reports of volunteers at work, including direct observations by Water Resources Research Center personnel, conversations with local coordinators, visuals on TV news, and descriptions by reporters for the print media. Volunteer performance was specifically evaluated in phase III.
Volunteers were requested to provide very detailed descriptions of their sampling sites. As a rule of thumb, volunteers were instructed to make their site descriptions detailed enough for someone to locate the sample site without benefit of any information but the site description. Initially, the volunteers provided good descriptions of the actual sampling site but poor to non-existent descriptions of the route taken to reach the site. Site description forms were revised to force volunteers to complete both parts of the site description, and volunteers were asked to redo their site descriptions. The revised form provided for this purpose is shown in Table 2.3-2. These site descriptions are on file at the Water Resources Research Center and are included in a memo field in the database included with this report..
An unexpected difficulty arose as a consequence of permitting volunteer selection of sites. In many cases, knowledge of the water body's name and the town in which it was located provided sufficiently unique information for unmistakable identification. However, a non-trivial number of water bodies were not sufficiently uniquely identified in this manner. In some cases, local names or names found on county or town maps differed from those on topographic maps. In other cases, several water bodies in the same town shared a common name, e.g. Mill River and Long Pond are commonly used names occurring in multiples in several towns. A unique numerical for lakes (PALIS) had been developed by Godfrey et al. (1979); a similar system (SARIS) was developed by streams (Halliwell et al., 1982). These unique PALSARIS numbers were assigned to each lake and stream based on using the site descriptions and maps to confirm the location of each. After assigning these numbers to the 796 lakes sampled in ARM phase 1, we learned that a complete revision of the PALIS numbering system had just been completed and was being published by the state (Ackerman et al. 1984). The old numbers had been discarded without development of a conversion table. The labor intensive process of assigning PALIS numbers had to be redone.
2.10 Table 2.3-1. Sampling Protocol
Massachusetts Acid Rain Monitoring Project
LAKE AND STREAM SAMPLING INSTRUCTIONS
Because many people are involved in a stream and lake survey covering a large area, it is important that uniformity of collection and storage practice be observed. Please carefully follow the instructions below.
SAMPLE COLLECTION:
1. Site Selection. In streams, one should choose a location that yields a sample which is as representative of the stream as possible. The sample should be taken as near the center of the stream as is practicable. Avoid stagnant or still places; sample the flowing water. If the sampling site is at a road crossing, the sample should be taken on the upstream side of the road.
For lakes, sampling site selection is more complex. If a boat is available, a mid-lake sample is preferable. Use three shoreline landmarks to triangulate the site so that it can be found again. If a boat is not available, the next best sampling site is at the outflow of the lake. If there is no outflow or it is not accessible, select a shoreline area free of weeds and indications of pollution such as discharge pipes.
2. Sample Collection. In all cases (lake or stream), the sample should be taken below the surface. Care should be taken to not disturb and collect any bottom sediment. In lakes, sample an arm's depth below the surface but sufficiently far above the bottom to avoid including any sediment.
The sample bottle should be rinsed three times with the water from the lake or stream before collecting the sample. The bottle cap should also be rinsed. Pour the rinse water downstream or away from the place where the sample will be collected. To collect a sample, lower the empty sample bottle into the water, upside down, to the appropriate depth, then turn the bottle right-side up and allow to fill completely and cap. Try to leave no air in the bottle.
3. Sampling Containers. The bottle provided for samples is a 500 ml (16 oz) Nalgene bottle. The container must be absolutely clean and thoroughly rinsed. Rinsing is done with distilled water at the lab. Rinsing with lake or stream water on-site will remove the residual distilled water. The bottle is prelabelled with necessary site identification information. Be sure that the lake or stream name on the label matches that of the sampling site.
SAMPLE STORAGE AND PRESERVATION
No chemical preservation can be used prior to pH measurement. The pH measurement should be made as soon after collection as possible. If the sample must be stored (overnight is the maximum storage time) it should be kept in the dark and either in a refrigerator or on ice.
2.11 Table 2.3-2. Sample Site Location Form
SITE LOCATION SHEET
COLLECTOR’S NAME:
ADDRESS: (street) (town) (state) (zip)
TELEPHONE NUMBER: (area code) (number)
COLLECTION DATE; COLLECTION TIME:
County District
Lake or Stream
Code Number Town
Nearest Major Highway
Road names and/or numbers connecting major highway to the site access road:
Specific directions from access road (named above) to exact location of sampling site:
2.12 Table 2.3-3. Data collection form.
MASSACHUSETTS ACID RAIN MONITORING PROJECT
COLLECTOR’S SHEET
Collector Name: ______Phone (if new): ______
Address (if new): ______
CO DI STREAM OR LAKE NAME ID DATE COLLECTED TIME COLLECTED
______
______
______
______
______
______
______
______
______
______
______
______
______
______
______
QUESTIONNAIRE:
Mileage per collection: ______miles
Hours worked per collection: ______hours
SIGNED: ______DATE:______
2.13 2.3.2 Phase II
Samples were collected from the selected sites on two occasions - October 14, 1984 and April 14, 1985.
Samples were collected by volunteers in new, pre-labeled 500 ml. polyethylene bottles with screw caps, approximately one foot below the water surface, as in ARM I. Each bottle label was computer printed and contained the water body name, its unique PALIS or SARIS code number, and the county and ARM district. The bottle labels matched the lists of sample sites previously given to county and district coordinators, as well as the pre-printed lab data sheets. This consistency helped minimize confusion, particularly at the local laboratories. Volunteers were asked to fill out a sample site location form for each site (Table 2.3-2) and turn the completed forms in to the local labs along with their water samples. The site location sheets are maintained on file at the Water Resources Research Center. Site descriptions were required only once, but a data collection form was required for each sampling period (Table 2.3-3). Details of sample collection and preservation for phases 2 and 3 are shown in Table 2.3-4.
2.3.3 Phase III
The same sampling protocol was followed in phase III as used in the prior phases. During phase III, all bottles were relabeled with waterproof, bar coded labels. Of little importance to the field collection, this change greatly reduced the time spent preparing bottles for the labs and eliminated data entry mistakes during analysis at the central lab. Data collection sheets were also preprinted with the collector’s name, county, district, water body name and PALSARIS code for each collector. This minimized the amount of handwriting to be interpreted during key entry and reminded the collector of the sites to be sampled. Collector’s were asked to record and sign their time and mileage so that the state funding agency could request federal matching funds.
2.4 Sampling Dates
Times of collection varied from one phase to another. In phase I, sampling was conducted monthly for 15 months. In phase II, sampling was done only in October and April. In phase III, sampling was done quarterly (Table 2.4-1). The choice of sampling frequency was governed largely by the logistics of the survey while the choice of time was governed by what we knew about the behavior of the surface waters. In phase I, there was little knowledge of the seasonal change in water chemistry in Massachusetts lakes and streams, and there were only two parameters, pH and ANC to be measured. Sampling was done monthly, generally on the third Sunday.
The ARM I data provided clear evidence of significant seasonal variability in pH and alkalinity for surface waters in Massachusetts (Godfrey et al., 1985). It was therefore deemed necessary to consider this variability in planning the ARM II sample collection scheme. Because of the large number of sample locations and the desire to maintain synoptic sampling throughout the state, it was decided to conduct sample collections on two dates, one to represent the summer high
2.14 pH/alkalinity period, and one to represent the low pH/alkalinity period during spring. October was chosen for the high pH/alkalinity date for the following reasons:
1. Statewide mean pH and alkalinity for October, 1983 were statistically similar to the June - September, 1983 means during ARM I. The ARM I June - October months were classified as a "summer" period (Godfrey et al., 1985). 2. The October sample collection was intended to coincide roughly with Phase I of the eastern portion of EPA's National Lake Survey (NLS), permitting comparison of data for lakes sampled in both ARM and NLS. 3. The logistics and timing of project funding and organization were well suited to an October start.
Table 2.3-4. Sample collection and preservation protocols.
pH & alkalinity cations anions & color Sample Matrix natural surface water surface water + natural surface water HNO3 Total # of Samples 800/collection 800/collection 800/collection Sample Volume 500 ml 60 ml 60 ml Sample Container high density high density high density polyethylene bottles polyethylene bottles polyethylene bottles Preservation refrigeration acidified refrigeration Max Holding Time 24 hours 6 months 28 days
April was chosen for the spring low pH/alkalinity date based on the low mean statewide pH and alkalinity found in both April 1983 and 1984 in ARM I (Godfrey et al.,1985), and because of the likelihood of spring snowmelt runoff during that month. The selection of April also provided an increased margin of safety for volunteer sample collectors, because ice-out is nearly complete in most of the state's lakes and ponds by then. Prior to the April, 1985 sample collection, volunteers were requested to collect samples where possible from sites which had been dry in October, 1984.
In phase III, we sought to continue the sampling scheme initiated in phase II but added two additional sampling periods: January and July. January was added to provide some redundancy for the April sample in the event that weather conditions might make sampling difficult or unduly bias results. July was added at the request of the Massachusetts Division of Fisheries & Wildlife. Since their data were primarily collected during summer, July samples would be best for comparison. Quarterly sampling provided sufficient time for the central lab analyses to be completed and sample bottles readied for the next collection but kept the lab in continuous operation. Quarterly sampling also minimized the need to refamiliarize volunteers with procedures and recondition equipment at the field labs.
2.15 Table 2.4-1. Schedule of Collections.
Phase I - 1983 - 1984
Month/Day M A M J J A S O N D J F M A 20 17 25 19 17 21 18 16 20 18 15 26 18 15
Phase II - 1984 - 1985 October, April, 1984 1985 14 14
Phase III - 1985 - 1993
Year/Month Jan Apr Jul Oct 1986 12 6 13 19 1987 18 12 12 25 1988 24 10 24 16 1989 8 9 16 15 1990 21 8 15 14 1991 27 7 14 20 1992 26 5 12 18 1993 unfunded 4 25 unfunded
2.5. Local Laboratory Analysis
In 1983, 73 local labs were trained in county-wide regional meetings scheduled shortly after the county presentations to enlist volunteer help. Labs had already been selected by county and district coordinators. Dr. O.T. Zajicek reviewed the analytical protocols with local lab personnel and discussed several potential problems that might occur. He carefully reviewed the process of cleaning and calibrating pH electrodes. All lab personnel were provided with detailed procedures for measuring pH and alkalinity and maintaining equipment (Appendix 8.2. and 8.3). The first sampling in March 1983 served as a trial run to further resolve laboratory problems. Throughout the first year of the project, several project personnel were available to respond to questions and problems prior to sampling and on the day of analysis. In some cases, emergency deliveries of electrodes were
2.16 necessary on the sampling day. With improvements in the QA/QC program, the gained experience of the local labs and their reduced number in phases I and II, the need for sampling day expertise and troubleshooting diminished markedly after the first year.
Samples were delivered to local cooperating laboratories where pH and ANC were determined as soon as possible (within 24 hours). ANC was calculated from a double end point (two point Gran) titration, typically between pH 4.5 and 4.2 (method 310.1-7.2 EPA, 1983). In rare cases (<1%) where lake pH was less than 4.5 and titrations were not completed we estimated ANC from the negative of hydrogen ion (i.e. mineral acidity) (Stumm and Morgan, 1981). Only data which passed the respective pH and ANC quality control checks were used (Walk et al., 1992).
Sullivan et al. (1989) reported DOC can influence Gran ANC titrations. We compared our two-point method against the more typical multi-point Gran method (titrations extending below pH 4.0) on thirteen selected lake samples from April 1991, and 37 randomly selected samples from April of 1992. The data were similar for the two years and we have combined the results here. The two point titration underestimated the low pH Gran method by an average of 10.2 µeq L-1 (Gran ANC µeq L-1 = 4.7 + 1.02*[two point ANC]; n=50, r2>0.999). The differences between the titration methods were small, so we made no adjustments to our data.
To insure the use of standard operating procedures for coordinating the distribution of sample bottles and sampling forms, QA samples, and reporting and input of results, a written description of these details was developed (Appendix 8.4).
2.6. Central Laboratory Analysis
Soon after the ARM project began, an effort was made to expand the water chemistry analyses. With limited funds, basic wet chemistry for several major cations and anions was attempted. Local labs were asked to freeze aliquots in Whirl-pak bags at quarterly intervals from the summer of 1983 to spring of 1984. Difficulties were encountered in collecting these frozen samples from the 73 labs and maintaining them in frozen condition during the collection. Whirl-paks that had been stored for several months often leaked their contents when thawed. Analytical work was conducted by student volunteers. These data have been examined for data quality, but they do not pass Project quality requirements. They have not been included in any data analyses in this or other published reports. The effort did, however, help to justify the expense of automated equipment and the creation of the Environmental Analysis Laboratory used in the second and third phase of the project.
Beginning in October 1984, chemistry subsamples were analyzed at the Environmental Analysis Laboratory at the University of Massachusetts at Amherst as follows (Table 2.6-1). The equipment is listed in Appendix 8.5. Subsamples for anions were refrigerated and subsamples for cations and silica were acidified with nitric acid to pH <2. Ionic chemistry was analyzed by ICP
emission spectrophotometer for metals and SiO2 (EPA, 1983) and by single-column ion chromatography (Wescan) for sulfate, nitrate and chloride. Since the samples were unfiltered, the results for trace metals will be referred to as acid soluble, rather than as dissolved concentrations.
2.17 Table 2.6-1. Analytical procedures.
Parameter Method
Alkalinity Potentiometric titration, EPA 310.1
Aluminum, total ICP
Arsenic ICP
Cadmium ICP
Calcium ICP
Chloride Ion Chromatography
Chromium ICP
Color Modified EPA 110.2. Colorimetric, absorbance measured at 425 nm and compared to standards made with A.P.H.A. platinum-cobalt color standards
Copper ICP
Iron ICP
Magnesium ICP
Manganese ICP
Nickel ICP
Nitrate Ion Chromatography
pH Electrometric, EPA 150.1
Potassium ICP
Selenium ICP
Silica ICP
Sodium ICP
Sulfate Ion Chromatography
Titanium ICP
Vanadium ICP
Zinc ICP
ICP = Inductively Coupled Plasma Atomic Emission Spectrometric Method for Trace Element Analysis of Water and Wastes, Method 200.7, USEPA doc. EPA 600/4-79-020
2.18 Color was measured spectrophotometrically at 425nm on filtered (0.45 µm) samples against platinum color standards and reported as platinum color units (PCU). We chose this method over the Hach color disks because the spectrophotometric method was automated and is reported to be more precise (Cuthbert and del Giorgio, 1992). We also found that two Hach color disks both gave readings 10-20% low on tests of PCU color standards supplied by two different manufacturers.
We eliminated brackish lakes from the inorganic chemistry data set by excluding those with sodium or chloride exceeding 3480 or 4225 µeq L-1 (80 or 150 mg L-1 of sodium or chloride, respectively) which was roughly comparable to the conductivity limit of 1500 µS/cm used by the EPA/ELS to eliminate non-freshwater sites.
We tested and found 805 of 1025 lakes with non-missing major cations and anions passed the 15% ionic balance test of Hillman et al. (1986). Lakes that failed the ion balance test were highly colored (median PCU=60). We used the anion deficit in the charge balance as an estimate of organic anions (R-). Some of the failed data had unexplained high or low sodium to chloride ratios, suggesting a problem in the analysis of one of these ions. Upon further evaluation, the problem seemed related to high chloride levels in excess of the upper detection limit. Only the data from the 805 lakes that passed the ion balance test are presented here. Overall quality control of ionic chemistry determination in our lab was subsequently tested by the use of double blind EPA standard solutions disguised as 14 samples. All ions showed good agreement with expected values except chloride which was biased high by three percent.
2.7. Quality Control
2.7.1 Target Population of Lakes and Streams
Since our survey was an incomplete census rather than a statistical sample, we conducted additional tests to check for bias in the sampling design. We tested if the lakes sampled in phase I and phase II differed in pH or ANC from the 36% of those lakes not sampled. In April of 1992 we randomly selected and examined 60 lakes from the population of lakes that had not been sampled in phase I or II. Twenty four of the sixty lakes (40%) were not sampled for the following reasons: 16 were inaccessible (private, posted or on off-shore islands), 5 were saltwater, 3 were incorrectly identified or could not be located in the field. We sampled the remaining thirty six (60%) accessible, freshwater lakes and compared them to a concurrently collected stratified (by ANC) random sample of 213 lakes selected from the phase I and II population. A t-test on means revealed no significant differences in pH or ANC between the populations (á>0.05), and we concluded the phases I and II census provided a good representation of the accessible freshwater lakes in the state.
We also tested for differences in pH or ANC between October 1983 (phase I) and October 1984 (phase II) surveys. A nonparametric Wilcoxon test found no significant difference in the ranks of either pH (á=0.76) or ANC (á=0.34) between years for lakes. Similarly, there was no significant difference in the ranks of either pH (á=0.95) or ANC (á=0.66) between years for streams. In addition, for 13 lakes sampled in both years, paired t-tests revealed no significant differences in pH (n = 12, average difference = 0.028, standard deviation = 0.478) or ANC (n = 10, average difference
2.19 = 1.73 µeq/L, standard deviation = 164). For the 12 streams sampled in both years, paired t-tests showed no significant difference for either pH (n = 10, average difference = 0.071, standard deviation = 0.41) or ANC (n = 9, average difference = -0.87, standard deviation = 4.63). We concluded that no significant bias would be introduced by combining the data. For the 13 lakes and 12 streams sampled in both years we used the later (1984) data.
Because the results of previous surveys showed differences related to the sizes of lakes sampled we examined our sampling procedures for bias in lake size. The distribution of the surface areas of lakes sampled by the ARM survey is nearly identical to the distribution of lake areas in the PALIS list (Figure 2.7-1). In addition. a Chi-square test revealed no significant differences (á>0.05) in the frequencies of large vs. small lakes for the October samples (median lake size 4.1 ha) as compared to the PALIS list (median size 4.4 ha). As mentioned above, we could find no significant differences in pH or ANC between the population of lakes sampled in ARM (phase I plus phase II) and the lakes not sampled; we thus conclude that our survey provides a good representation of the acidity status of freshwater lakes in the state.
We also examined how different screening criteria may result in differences between our survey and that of the EPA/Eastern Lakes Survey (Baker et al., 1990). The ARM project's October 1984 survey sampled 958 (62%) of the 1545 lakes on the PALIS list with area >4 ha. Based on our random survey of 60 previously unsampled lakes we conclude most of the lakes not included in the phase I and II sampling could have passed our chemical screening criteria and would not have changed the results presented here if they had been included. These lakes were not omitted because of a selection bias, rather they were omitted because they were difficult to access and we lacked the time and personnel to complete the census.
The EPA/Eastern Lakes Survey (ELS) screening criteria were much more stringent with respect to the types of lakes selected. For example, in the ELS Southern New England subregion 1D, approximately 20 percent were excluded from consideration for reasons which include: urban/- industrial/agricultural influence; marsh/swamp; and very shallow lakes. These lakes would likely show higher concentrations of ions than the selected lakes and the resulting dataset may not accurately reflect the degree of human impacts on lakes in the region. Certainly the EPA/ELS data are most appropriate to the study of atmospheric acid deposition effects, but analysts should discuss the influence of screening criteria when reporting general conditions of water quality in a given region (see Section 3.1.7 for further discussion).
2.7.2. Sampling Sites
Throughout our studies, volunteers who sampled lakes followed instructions to avoid sites which appeared to be saltwater or seriously polluted. Samples were collected one foot below the surface at the shore at the outlet of the lake or from a clear shoreline away from weeds or inlets. We were concerned that a shore sample might not be representative of typical lake conditions. Normally the middle of the lake or the site of the deepest part of the lake would be chosen for sampling. In October, 1990, we tested the efficacy of shoreline sampling for the ARM population of phase III lakes using a randomly selected subset of 25 lakes representing all ANC categories. We chose
2.20 Figure 2.7-1. Relative size frequency of ponds and lakes in the combined ARM phases I and II versus all named ponds and lakes. October as a time when the differences between shore and mid-lake samples would theoretically be greatest. ARM staff collected shoreline samples at the site described by the volunteer and midlake samples. The statistical test of this sampling procedure using paired samples (center vs. shore), as shown in Table 2.7-1, found no significant difference in either pH or ANC between samples collected at the volunteers' shore sites vs. the centers of lakes (n=25 pairs, á>.05).
For streams, the volunteers usually collected from the upstream side of a convenient access road. We tested this site selection sampling procedure by sending professional staff to 21 randomly selected streams with instructions to sample from three locations within a one hour time period. The three sites included the volunteers' selected site (volunteer site), a site one hundred meters upstream from the volunteers' site (upstream site), and a randomly selected site on the stream (random site). The randomly selected sites were chosen by finding the stream on a topographic map and proceeding a random percent (0-100) of the stream length up from the mouth of the stream. Three of the upstream sites could not be sampled due to inaccessible conditions (e.g. presence of swamp or other restricted access). Examining the difference of the volunteer site results minus the upstream site revealed that the proximity to roads had no significant effect on pH (t-test á>0.05, average difference in pH=-0.03, SD=0.25). The difference for alkalinity was also not significant (t-test á>0.05, average difference in alkalinity=-0.51 mg/L, SD=2.43). Examining the difference of the volunteer site minus the random site also showed no significant effects on either pH (t-test á>0.05, average difference in pH=0.06, SD=0.53) or alkalinity (average difference in alkalinity=2.95 mg/L, SD=10.35). The results of the test indicate that streams are highly variable in pH and alkalinity along their length, thus
2.21 the choice of sampling location is important. However, the sites chosen by our volunteers showed no significant bias toward either more or less acidic conditions than those found at random site locations.
Table 2.7-1. Test of difference between lake shore sites used by volunteers and lake center. All samples collected by Project staff on October, 1990. N=21, pH mean difference = 0.0067, pH standard deviation of differences = 0.1059, ANC mean difference = -0.2619, ANC standard deviation of differences = 2.207; t value for pH = 0.2816, t value for ANC = -0.531.
Shore Center Size ANC ANC PALSITE (ha) pH µeq L-1 pH µeq L-1 11002 103.9 7.85 554 8.04 547 21040 102.1 7.46 226 7.41 230.5 21062 45.2 8.05 368 8.16 361 21102 5 7.43 102.5 7.36 105 31044 20.1 7.19 77 6.94 79.5 32031 1.8 5.16 32 5.11 31 33013 0.4 7.15 298 7.05 275 34018 11.7 6.67 144 6.81 152 34097 1.2 6.84 79 6.75 79.5 34103 50.9 6.17 48 6.17 48 35026 5.1 5.19 38 5.21 36 35035 16.6 6.34 51 6.35 52 35085 0.4 6.05 64 6.00 63 35095 7.4 6.64 44.5 6.73 43.5 36085 33.3 5.95 50 5.96 49 36092 44.7 6.69 89 6.74 89 36093 8.5 7.01 133 6.85 142 36128 1.7 7.33 185 7.29 171 32009 26.4 6.97 55 6.87 56.5 32012 9.3 6.71 125 6.87 186 81157 41.6 5.65 20 5.69 20
2.7.3. Volunteer Collection
We tested the performance of volunteers in collecting lake and stream samples by comparing the results of their collection to those of our staff sampling the volunteers' sites on the same day as, but unknown to, the volunteer. The professionally collected sample was then delivered to the lab for analysis and comparison with the volunteer's sample. A total of 135 volunteers were tested once during ARM Phase 3 in this manner and the results are shown in Figures 2.7-2 and 2.7-3. There was a significant difference from the expected one-to-one line for pH, but as Figure 2.7-2 shows, the agreement was still very good with volunteer samples showing only slightly higher pH at low pH
2.22 values and slightly lower pH at high pH values. There was no significant difference in alkalinity between professional and volunteer collections and the linear relationship is highly significant with an r2 of 0.986 (Figure 2.7-3). Overall we judged the differences to be minor and concluded that volunteers can collect samples and meet high quality control standards.
Figure 2.7-2. Comparison of pH results from volunteer collections and near-simultaneous (but unknown to the volunteers) professional staff collections. The regression is Vol = 0.45 + 0.93*Pro where N = 135, R2 = 0.949 and the root mean square error = 0.17.
Figure 2.7-3. Comparison of Alkalinity (ANC) results from volunteer collections and near-simultaneous (but unknown to volunteers) professional staff collections. The regression is: Vol = 0.21 + 1.00*Pro where N = 135, R2 = 0.986 and the root mean square error = 1.94
2.23 Examination of the stream reaches sampled shows there may have been a bias toward higher order reaches, probably because sampling was done at road crossings.
2.7.4. Local Labs -- pH and ANC
The three Project phases were conducted somewhat differently from each other, and the quality assurance plan evolved and improved with each phase. In phase I, quality assurance checks were confined to volunteer laboratory performance. A single quality assurance check sample was sent to each lab for each sample date. Quality checks of volunteer collectors' work was limited to sending all collectors a description of their responsibilities, the site selection criteria, and the proper way to collect samples. In phase II, quality assurance checks were increased to include one diagnostic, two audit samples per sampling period and one or more double blind audit samples. Phase III continued the phase II audit process and improved the stability of the audit samples, formalized the criteria for acceptance of results, and developed quality control documentation for each meter rather than each lab. The protocol in place for the last years of the project is shown in Appendix 8.6.
2.7.4.1 Phase I
The Phase I Quality Assurance program was designed to evaluate the performance of individual local laboratories, as well as their aggregate performance. At approximately monthly intervals, participating labs were sent "blind" quality assurance samples and asked to analyze these samples for pH, EPA double endpoint alkalinity (DET), and single endpoint alkalinity (SET) along with their regular monthly surface water samples. Through the course of the Project, an effort was made to reflect in the quality assurance samples the range of pH and alkalinity values typically encountered by local labs in their monthly surface water samples. The reported values of each measure were evaluated by comparison with the known or "true" value of the measure for the particular quality assurance sample. Statistical methods then were employed for the purpose of deleting data from labs which did not meet specified criteria.
During the startup period in March, 1983, the Center also assisted many labs in resolving such quality control problems as cleaning electrodes and calculating alkalinities (especially for the relatively unknown low level technique). A pH "check sample" provided by Corning assisted in determining the initial adequacy of local laboratory pH measurements. This sample was used by local labs as a known check for pH accuracy.
Two of the blind samples (March and November, 1983) sent to participating labs were EPA quality assurance samples. The remaining quality assurance samples were produced and evaluated by the Project's Environmental Analysis Laboratory (EAL). Since the basis of evaluating laboratory performance largely relies on these results, two separate means have been used to document the accuracy of EAL analyses. First, results on the two EPA quality assurance samples used in the ARM quality assurance program are presented (Table 2.7-2). Second, EAL results on the three EPA quality control samples provided for certification in the national acid rain program are compared with the "true" pH values listed by EPA (Table 2.7-3).
2.24 Table 2.7-2. Comparison of U.S. EPA Quality Assurance Sample Values and Sample Measurements by EAL.
U.S. EPA EAL DIFFERENCE MARCH pH 7.4 7.61 +0.21 alkalinity(mg/l) 16.0 16.9 +0.9
NOVEMBER pH 6.53 6.54 +0.01 alkalinity(mg/l) 3.7 3.9 +0.2
Table 2.7-3. Comparison of U.S. EPA pH Values for Acid Rain Certification Program with Sample Measurements by EAL.
SERIES U.S. EPA EAL DIFFERENCE 1218 4.45 4.45 0 2235 3.72 3.66 -0.06 3837 3.49 3.46 -0.03
With the sole exception of pH for the EPA quality assurance sample used in March, EAL results are quite close to the true values. EAL precision was further validated by investigating the ability to replicate results. Table 2.7-4 displays means and standard deviations reported for replicate analyses of project quality assurance samples by EAL. As demonstrated in this table, EAL's ability to replicate observations was consistently good. Standard deviations were, in all cases, less than or equal to 0.025 pH units and 0.5 mg L-1 of alkalinity. Therefore, for the purposes of evaluating the performance of participating ARM local laboratories, it is assumed that EAL results are both accurate (measure the true value) and precise (consistently reproduce the same value).
Table 2.7-4. Means and Standard Deviations of Replicate Measures for EAL Quality Control Samples.
MONTH pH DOUBLE ENDPOINT SINGLE ENDPOINT ALKALINITY ALKALINITY Mean s.d. Mean s.d. Mean s.d. May '83 6.31 0.025 8.3 0.18 9.52 0.10 June 6.14 0.017 17.4 0.2 19.4 0.2 August 7.07 0.02 15.2 0.2 16.9 0.2 September 6.38 0.006 28.8 0.4 31.3 0.2 October 5.95 0.01 12.2 0.5 14.65 0.4 February '84 5.74 0.004 9.1 0.16 12.1 0.18 March 4.94 0.004 - - - - April 7.073 0.001 47.0 0.3 49.2 0.3
The only questionable result produced by EAL in Tables 2.7-2 - 2.7-4 was the value of pH for the March EPA quality assurance sample. However, after repeated analysis of this sample with
2.25 several different meters and electrodes, and following initial statistical analysis of the pH quality assurance data set, the EPA's March value was shown to be highly suspect. Further discussion with EPA's Cincinnati laboratory revealed that EPA had problems with the stability of pH in mixed mineral samples, and had later begun sending pH quality assurance samples separately from those used for other anions and cations. The second EPA quality assurance sample used (November) was of this type. On the basis of this information, it was decided that the EAL's value would be used as the "true" pH value of the March quality assurance sample.
Evaluating the overall ability of each local lab to produce precise and accurate pH and alkalinity data was a primary objective of the quality assurance program. Least squares regression analysis was determined to be the most appropriate method for investigating the relationship between true and observed values for each measure within individual laboratories. If a lab's observed values of a particular measure matched perfectly with the true values, then a regression line would be drawn with slope equal to one and intercept equal to zero. From analysis of the regression lines, inferences were made concerning the accuracy and precision of a particular lab's data for pH, DET, and SET.
Based on the initial regression analysis performed on each measure, suspect observations in the plots of true versus observed values for each lab were tested for possible rejection as outliers. In such cases, a new regression analysis was performed on the data without the suspect observation. The observation was then treated as a new case, given the second regression line, and the probability of observing a deviation as great as that of the possible outlier from the new line was calculated. If the probability of this occurrence was less than .05, the observation was deleted from the QA data. Otherwise, the outlier was retained and the original regression line was considered valid for that particular lab and measure. It is important to note that this method for detecting outliers is designed to properly assess a specific lab's variability. This technique is only valid for within lab observations and ignores the possibility of outliers in the overall data set. Initial assessment of a lab's performance on a given measure involved use of the F test to evaluate the significance of the linear association between true and observed values. If the F test was not significant at the .05 level, then the lab's data for that measure were considered inaccurate for the purposes of this study. Labs which showed a significant linear relationship based on the F test were analyzed further in order to check for precision and accuracy of results.
Since it was possible for a lab's regression line to pass the above test for linearity but still deviate significantly from the hypothetical regression, an additional accuracy criterion was developed to limit the amount of deviation allowed from the expected regression. The accuracy of a given lab's data was evaluated by measuring the mean distance of the calculated regression line from the hypothetical 1:1 line within the range of "true" quality assurance values encountered for each measure. A lab with very accurate results on a particular measure would produce a regression line very close to this ideal line. In contrast, a lab with less accurate quality assurance data would produce, in most cases, a line which deviated further from the ideal line. The amount of deviation from the 1:1 line allowed for any lab was determined from the ARM quality assurance results and by evaluating results from previous studies which required pH and alkalinity measurements.
2.26 An indication of a lab's analytical precision for a particular measure was obtained from the Mean Square Error (MSE) term produced by regression analysis. If a lab was accurate in each monthly evaluation of a particular measure, relatively little scatter of points would result around the calculated regression line, producing a low MSE and indicating good precision. If a lab reported less accurate values for a measure and therefore exhibited more scatter about the calculated line, regression analysis would produce a larger MSE, indicating lower precision. The relative level of analytical precision required of local labs was set by selecting a maximum allowable MSE, based on evaluations of the ARM quality assurance data and other related databases.
With the above criteria in place for analysis of the quality assurance data, each lab's performance was evaluated as either acceptable or unacceptable for pH, DET and SET. Since it was hypothesized that a lab could do well on one measure and not on the others, it was decided to evaluate labs by each measure separately. Data from labs which produced unacceptable quality assurance results on a particular measure were then excluded from further analysis of the ARM data base.
Regression analysis was also used to assess the aggregate performance of the local labs for the entire Project, in conjunction with standard descriptive statistical methods which describe the general, statewide accuracy of pH and alkalinity measurements for each month.
2.7.4.2. Phase II
Local laboratories were required to analyze two types of quality assurance (QA) samples: those which were sent as unknowns to the labs but clearly labeled as quality assurance samples, and those which were sent to the labs disguised as regular surface water samples ("double blinds"). Approximately two weeks prior to the sample collection date, an initial QA sample was sent to each local lab, along with a postage paid return postcard for reporting of results. The local labs were asked to analyze this sample in duplicate for pH and alkalinity, and promptly report the results via the postcard. The results of this initial QA sample were then evaluated at WRRC. Any labs reporting data significantly different from the expected values were notified and requested to take corrective steps, such as rejuvenating electrodes. All local labs were subsequently sent two additional quality assurance samples to be analyzed during the day of analysis of the surface water samples. The labs were asked to analyze one of the samples in duplicate at the beginning of the analytical day, and the second in duplicate at the end of the analytical day. The two double blind QA samples were delivered by the volunteer sample collectors along with their other surface water samples and analyzed by the local labs as regular surface water samples.
The local lab QA samples were produced in the ARM central laboratory at UMass/Amherst and designed to cover a representative range of pH and alkalinity values. Expected values were derived from a minimum of five replicate analyses of the samples by EAL.
To determine the accuracy of the local laboratories, each lab's reported quality assurance results were regressed against the aggregate expected results from all sample dates in ARM I, and separately for each collection in ARM II. The reason for the different treatments in the two phases
2.27 is that in ARM I, at most one QA sample per month for 12 months was available, whereas in ARM II, five QA samples were available per lab each collection. Each regression plot was initially evaluated for the presence of possible within-lab outliers. When a possible outlier was found, the lab's least squares regression equation was recalculated without that data point, and the probability of finding the possible outlier within the new data set was calculated. Data points were rejected as outliers if this probability was less than 0.05, and all subsequent analyses were conducted excluding any such data points. Figure 2.7-4 compares volunteer lab QA sample pH with expected pH for all labs and all quality assurance samples for Phase II, and Figure 2.7-5 does the same for ANC. Each regression equation was evaluated for statistical significance using the F test at the 0.05 level of significance. Labs were then evaluated on the basis of the average distance of a lab's calculated regression line from the hypothetical 1:1 line, within the ranges of expected quality assurance values
2.28 for each measure. The range of expected pH quality control values was 4.76 to 7.67. The range of quality control values for ANC was 13.2 to 940 µeq L-1.
Given available information from other reports and quality control data sets, and the structure of this quality control data set, the limits for the average distance from the hypothetical 1:1 line were set at 0.25 pH units and 40 µeq L-1 for ANC in ARM I, and 0.15 pH units for pH and 40 µeq L-1 for ANC in ARM II, calculated across the above ranges for each measure. In ARM II, except for labs which previously were rejected by the F-test, all labs were shown to be within acceptable ranges on
2.29 this test. Final assessment of a lab's performance was conducted by evaluation of the mean square error term obtained by regression. Using these criteria, 56 labs out of 73 were accepted for the measurement of pH and 49 for the measurement of ANC in ARM I, and 23 labs out of 25 were accepted for both pH and ANC in ARM II. As a result, approximately 79% and 97% of the surface water sample data were considered valid for pH and 74% and 97% for alkalinity in ARM I and ARM II, respectively.
Table 2.7-5 shows the average monthly mean difference between observed (averaged within labs monthly) and expected QA values for pH and ANC in both phases of the project. Data for labs failing to meet QA/QC criteria have not been used in subsequent data analyses.
Results of the double-blind samples did not differ significantly from results of the other quality assurance samples for accepted labs, either in terms of differences from expected values or the magnitude of standard deviations (see Table 2.7-6).
Table 2.7-5. Average difference between monthly mean observed and expected quality assurance values in ARM I and ARM II.
ARM I ARM II
pH N = 468 N = 299
Mean difference for all months (pH units) -0.09 -0.03
Range in monthly mean difference (pH units) -0.21 - -0.08 - 0.00 0.07
ANC N = 365 N = 274
Mean difference (µeq L-1) -7 -4
Range in difference (µeq L-1) -28.8 - 12 -17.2 - 8.8
2.30 Table 2.7-6. Mean results of double blind quality assurance samples for all local labs, compared with results for only labs accepted following quality assurance testing and the expected values.
pH ANC (µeq L-1)
QA Sample N Mean SD Mean diff. N Mean SD Mean diff. Observed vs. Observed vs. Expected Expected
QA4 -all labs 20 5.05 0.17 19 218 46
-accepted 19 5.09 0.07 0.03 17 230 31 -4
-expected 5.06 234
QA5 -all labs 23 5.21 0.16 22 232 51
-accepted 22 5.24 0.09 -0.01 19 245 32 -15
-expected 5.25 260
QA9 -all labs 20 5.77 0.13 20 257 30
-accepted 19 5.8 0.07 -0.02 19 262 20 -6
-expected 5.82 268
QA10 -all labs 22 7.62 0.14 22 457 24
-accepted 21 7.65 0.08 0.07 21 457 24 -17
-expected 7.58 474
SD = Standard Deviation
2.7.4.3 Phase III
We used a network of 17-21 (usually 18) field laboratories across the state to analyze water samples for pH and alkalinity within a day of collection. Two weeks prior to each collection we sent out a diagnostic quality control sample to test and optimize each laboratory's performance. Just prior to the scheduled sample date the EAL prepared two regular QA samples and sent them to the field labs for analysis in duplicate at the same time as the field samples were analyzed. Neither the true pH nor the true alkalinity of the QA samples were known to the staff of the field labs.
The number and composition of the quality assurance samples varied during the course of the study. Prior to July, 1992 we used two QA solutions composed of tap water amended with acid, sodium bicarbonate, or dilute phosphate buffers, with potassium chloride occasionally added to increase ionic strength. These samples were each analyzed in duplicate by the field lab along with the other samples. Beginning in July 1991, we changed from two to three QA solutions per collection,
2.31 each being analyzed once per meter. These were made from air equilibrated sodium carbonate standard solutions with potassium chloride added. In October of 1988 we added a semi-annual double blind QA sample to our quality assurance program. This QA sample was disguised as a field sample, given a false name and code number on the label, and was delivered to the lab along with the other field samples. We added a special pH buffer QA beginning in October of 1991. The detailed procedure for quality control followed by Project staff is described in the Quality Control Plan used as standard operating procedure (Section 8.6) by the Project.
Our test for rejection of laboratory pH and/or alkalinity results consisted of the following procedure (see also Section 8.6). First, all QA results were entered into a computer database along with code numbers for the lab. Replicate analyses were then averaged for each lab, but after July 1992 the QA samples were analyzed once per meter (some labs had more than one meter in operation at a time). For each QA sample we calculated a mean and standard deviation for pH and alkalinity after removing the respective highest and lowest results. For pH, a meter failed a QA test if the reading differed by more than 0.3 pH units from the mean. A meter passed a QA test if it was within 0.2 pH units of the mean. Differences between 0.2 and 0.3 would pass if the meter fell within the 99 percent one-tailed confidence interval (Grubbs test for outliers). For alkalinity, a meter failed a QA test if the QA difference from the mean was greater than 3 mg L-1. A meter passed if the difference was within 1 mg L-1 or if the difference was less than 15 percent of the mean; otherwise the Grubbs test was applied, as above. During any given collection, an analyst had to fail more than one QA before the data for either pH or alkalinity from that meter were rejected. In 29 collections and for approximately 18 laboratories we rejected pH 14 times while alkalinity was rejected only once. Of the 19,715 samples collected for pH during ARM Phase 3 we rejected 453 samples for pH and 132 samples for alkalinity. The rejected data also include some individual samples that were failed for other reasons such as data incorrectly analyzed or improperly recorded. The data and statistics discussed below include only data that passed the tests discussed above.
Figure 2.7-6. shows the standard deviation of alkalinity for the two regular QA samples over the course of the ARM Phase 3 program. The standard deviations varied somewhat with an overall average standard deviation of 1.14 mg L-1 which is equivalent to a coefficient of variation of only 9.7 percent in our data. We suspect the precision varied over time due to differences in the composition and alkalinity concentrations in the QA samples. Figure 2.7-7 shows a large reduction in the standard deviations of pH with time. Initially the standard deviations were quite variable, occasionally ranging higher than 0.3 and as low as 0.05 pH units. After 1991 (when air equilibrated carbonate solutions were first introduced) the standard deviations stabilized. The overall standard deviation was 0.14 pH units which we found to be acceptable.
2.32 Figure 2.7-6. Standard deviation of alkalinity (ANC) for the two regular samples over the course of the ARM Phase 3 program. The average standard deviation is 1.14 mg L-1, equivalent to a coefficient of variation of 9.7%.
Figure 2.7-7. Standard deviation of pH for the two regular samples over the course of the ARM Phase 3 program. The average standard deviation is 0.14 pH units.
2.33 Even though the true values of the regular QA samples were unknown to each analyst, we thought the analysts might analyze the labeled QA samples with extra care. If so, we would expect the double blind QA samples to have higher standard deviations because they were disguised as field samples. Comparing 1064 regular QA samples to 154 double blind samples, we do in fact see a small but significant (F=1.36, á<0.025) increase from 0.144 to 0.168 in the average standard deviation for pH. The average standard deviation for alkalinity on the other hand was not significantly different between the regular QAs and the double blind QAs (1.14 vs 1.07 mg L-1 respectively). From these tests we conclude that the volunteer analysts perform slightly better pH analyses on samples that are known to be QA samples. Thus, the standard deviation of the double blinds may be a more appropriate estimate of the true precision among labs for pH.
To test for bias in the analyses, we carefully prepared three identical solutions of known alkalinity (14.2 mg L-1) from sodium carbonate and potassium chloride salts which we used as the QA samples for the April 1993 collection. We established equilibration with atmospheric partial pressure of carbon dioxide by bubbling the samples with an air pump overnight. Statistics for the three solutions are combined in the discussion below. The pH and alkalinity of the resulting solution were measured at the EAL and found to be slightly lower than expected from theory (7.65 vs 7.77 and 13.85 vs 14.2 mg L-1 for pH and alkalinity, respectively), suggesting that theoretical air equilibration pH was not attained. The results of the volunteer laboratories were significantly lower than expected for both pH and alkalinity (pH= 7.38 STD=0.11 n=87; alk=13.53 STD=0.60 n=96). Three of the 32 meters in use at the time failed the QA test for pH and those pH results are not included here. These results imply an absolute error of -0.39 in pH from theoretical air equilibrium pH and -0.67 mg L-1 for alkalinity. Within lab errors (standard deviation within meter) were small (0.07 for pH, 0.44 mg L-1 for alk). We also tested the accuracy of the labs with a double blind standard solution of low pH and alkalinity (pH=4.35, alk=-2.25 mg/L) and found an average error of only +.01 pH and -0.2 mg L-1 for alkalinity. Overall, we considered these results acceptable because we believe most of the reported pH error may be attributed to changes in the relatively dilute QA solutions during shipment to the labs.
As a further test for errors in pH we used non color-coded commercial pH buffers (usually pH=6.86 or 6.42) as special pH QA solutions during 7 collections (n=164). As would be expected, the absolute errors in reading these buffers were quite low, as were the standard deviations. The results show the average error was -0.006 with an average standard deviation of 0.054. The buffer QA solutions were read with excellent accuracy and precision indicating that the meters were calibrated properly.
Throughout all phases of the ARM Project, we used double endpoint titrations for ANC determination. Other recent surveys (U.S. EPA National Surface Water Survey and Adirondack Lake Survey) have used Gran plot titrations. Gran plot titrations were not practical for our use with between 17 and 73 local volunteer laboratories, but we were concerned with the comparability of ARM data with these other surveys. Also, Sullivan et al. (1989) reported DOC can influence Gran ANC titrations. We compared our two-point method against the more typical multi-point Gran method (titrations extending below pH 4.0) on thirteen selected lake samples from April 1991 and 37 randomly selected samples from April 1992, representing a wide range of color variation (cf.
2.34 section 2.7.5. for data on color). The data were similar for the two years, and we have combined the results here. As shown in Figure 2.7-8, the two point titration underestimated the low pH Gran method by an average of 10.2 µeq L-1 (Gran ANC µeq L-1 = 4.7 + 1.02 * [two point ANC]; n = 50, r2 = 0.999). The differences between the titration methods were small, so we made no adjustments to our data.
5
4
3
2 Thousands
DET ANC ueq/L 1
0
-1 -1 0 1 2 3 4 5 Thousands Gran ANC ueq/L
Figure 2.7-8. Comparison of Gran plot ANC titration vs. double end point titration results on thirteen selected lakes from April 1991 and 37 randomly selected samples from April 1992. The regression is: Gran ANC µeq L-1 = 4.7 + 1.02 * (double end point ANC), n = 50, r2 >0.999.
2.7.5. Central Lab -- Inorganic Chemistry
The EAL quality control for inorganic chemistry included check solutions, blanks, duplicates, and spiked samples in each batch (as described in Table 2.7-7). Quality control procedures evolved over the ten years of the project. Initially, EAL participated in the EPA water quality and water supply audits to check performance on anion and cation analysis. These results are reported in Appendix 8.7. With very few exceptions, results on audit samples for the major ions were deemed acceptable by EPA. EAL also participated in audit program of the Acid Precipitation Mitigation Program (APMP) of the U.S. Fish & Wildlife Service from 1986 to 1991 and the audit program of the Living Lake program from 1988 to 1991. Batch quality control procedures also changed throughout the program. The chronology of procedural changes is shown in Table 2.7-7.
Double blind audits were not run until 1989. These were prepared from EPA certified standard solutions and disguised as samples. The results are presented in Table 2.7-8. The results show that the lab achieved a high degree of accuracy and precision for the major cations and anions
2.35 tested. Only chloride and sulfate showed a significant bias from expected concentrations and those errors were only +17 µeq L-1 and +7.0 µeq L-1, respectively. Since these errors are very small compared to average concentrations (see Mattson et al., 1992), we judged the data to be acceptable.
Table 2.7-7. Chronology of quality control procedures.
ICP Fall 1984 to April 1988 Calibration check run every 15 samples April 1988 to July 1988 Calibration check run every 15 samples 1 duplicate and 1 spike per 43 samples July 1988 to April 1989 Calibration check run every 15 samples 1 duplicate and 1 spike per 43 samples 1 outside QCCS every 22 samples April 1989 to July 1990 Calibration check run every 15 samples 1 duplicate and 1 spike per 43 samples 1 outside QCCS every 22 samples 1 blank every 22 samples July 1990 to July 1993 Calibration check run every 15 samples 1 duplicate and 1 spike per 43 samples 1 outside QCCS every 22 samples 1 blank every 22 samples 1 detection limit check every 22 samples
IC Fall 1984 to July 1993 1 duplicate and 1 spike every 11 samples and 1 or 2 EPA checks per 66 samples
Color Fall 1984 to July 1987 run on Beckman Model 20 spectrophotometer July 1987 to July 1993 run on Lachat dual channel flow injection auto analyzer
Table 2.7-8. Results of double blind analysis of 14 replicate EPA quality assurance solutions1. ______Cl- SO4= Na+ Ca++ Mg++ K+ Al Fe Mn ------µeq L-1------µg L-1--- ______Expected 535 417 1070 129 97 36 150 330 130 Mean 552* 424* 1077 129 98 34 140 330 130 STD2 11 8 26 4 2 4 20 10 10 ------LOD3 7 5 3. 0.5 0.8 2.6 20 10 10 ______*Significantly different at á<0.05. 1Nitrate was not available. 2STD= Standard deviation of QA solutions.
2.36 3LOD, limit of detection calculated from LOD=3xSTD of low standard solutions. The practical limit of detection for anions is about 15 µeq L-1.
Because we optimized our anion analyses for relatively high chloride and sulfate concentrations, nitrate was below the level of detection in most samples. Furthermore the level of detection fluctuated with the age of the chromatographic column. We screened the data for the 'worst case' detection limit and then replaced all values below this concentration (0.2 mg L-1 nitrate- N) with zero mg L-1. Although some data information was lost by this screening method, we felt it was the most conservative corrective measure we could apply to maintain equivalent data quality throughout the study.
We used the charge balance of anions and cations as a final test of data quality. With minor modifications, we computed the percent ion difference described by Hillman et al. (1986) as used by the U.S. EPA Eastern Lakes Survey, as follows:
% ion difference = 100*(Ócations - Óanions - Alk)/TI
where: Ócations = [Na+] + [K+] + [Ca2+] + [Mg2+] - 2- - Óanions = [Cl ] + [SO4 ] + [NO3 ] TI = Total ionic strength = [Alk] + Ócations + Óanions + 2*[H+] Alk = titration alkalinity and all concentrations are in units of µeq L-1.
For this test, samples passed if the absolute value of the percent ion difference was equal to or less than 15%. Of the 19,715 samples taken in phase III, 16,638 had complete chemistry and ion balances computed. Of these, 15,851 (over 95%) passed the 15% ion balance test. Due to time and money limitations the data failing this test were not rerun, but were flagged in the final data set.
Color in platinum-cobalt units was used as a surrogate for dissolved organic carbon measurement. When the project began, the capability for DOC measurements did not exist in the central lab. Even though the capability was subsequently added, the volume of samples precluded conducting DOC on all ARM samples. We tested how well our color measurements correlated with DOC concentrations (Figure 2.7-9), as determined by automated persulfate digestion + IR spectrophotometry, on the same 50 samples used for the Gran titration tests noted above. Our color method predicted dissolved organic carbon very well (DOC mg L-1 = 1.8 + 0.057 * PCU, n = 50, r2 = 0.94.
2.8 Conventions
2.8.1 Analytical Units
This report is a compilation of ten years of research on the Acid Rain Monitoring Project. The project has produced reports for volunteers, state agencies, journal publication and other forms
2.37 of information dissemination. The evolution of the project and the different audiences have had only two consequences for a final reporting. ANC (acid neutralizing capacity) and alkalinity are used interchangeably, but in all cases they are determined by double endpoint titration. First, for ANC and alkalinity, we report values as either mg L-1 or µeq L-1. To convert ANC or alkalinity from mg L-1 to µeq L-1, multiply by 20. Other conversions are provided in Appendix 8.9. Second, other chemical constituents are expressed in µeq L-1 except in the tables of QA/QC results where they are expressed in either mg L-1 or ug L-1 depending on the units preferred by the U.S. EPA audit programs.
35
30
25
20
15 DOC (mg/L)
10
5
0 0 100 200 300 400 500 Color (PCU)
Figure 2.7-9. Comparison of spectrophotometrically determined color at 425 nm vs. dissolved organic carbon for 13 selected lakes from April 1991 and 37 randomly selected sites from April 1992. The regression is: DOC mg L-1 = 1.8 + 0.057 * PCU, n = 50, r2 = 0.94.
2.8.2 Number of Sites
Throughout this report, the reader will find that the number of sites sampled varies from one analysis to another. This is either the consequence of missing data or quality control. Our rule of thumb has been to use all data that pass data quality criteria relevant to the analysis at hand, rather than restrict our analyses to only those surface waters for which we have complete sets of acceptable data. Therefore, there may appear to be one number of sites sampled for ANC, another for pH and a third for calcium. Analyses requiring all three will result in a fourth sample size resulting from the combination of acceptable data. This means that there may not be strict comparability between different analyses because the population of surface waters changes with the suite of parameters under consideration. For two reasons, we believe this sample size problem does not affect our results in any significant way. First, the ARM sample size is very large and not easily affected by the replacement or elimination of a small percentage of sites. Second, replacement or elimination of sites is independent of qualities relevant to water chemistry because it is caused by a missed volunteer
2.38 collection or error in the lab. Neither of these persisted very long without remedy through the quality control program. Therefore, these uncontrollable influences are unlikely to have introduced significant statistical or purposeful bias.
2.9 Data Management and Analysis
The scope of the ARM Project made data management a major task. In the very beginning, the Project had no computers (PCs were relatively new). With more than 1000 sites being sampled each month of phase I, there was an almost instantaneously overwhelming need for computerized data entry. Initial data analyses and summaries were completely manual and primitive. Throughout the project, ways were sought to speed the flow of data through data entry to reporting of data results in ways that minimized error. As we eventually caught up with the backlog, we began to seek ways to automate quality checking so that we could implement quality control changes in timely fashion, improve the structure of the database for later analysis, develop useable ways to export data to other software, and automate data output from analytical instruments in computer readable form. Mention of product names does not imply our endorsement of those products nor the exclusive need for their features.
We used Ashton-Tate's dBase program from the early days of computerization, migrating from a CPM version to DOS and then from dBase II to dBase IV. Partway through, we completely changed the structure of the database, changing fields from sampling date to analytical parameter and row from parameter to sampling date, and converted the prior file to the new structure. The final structure is shown in Section 8.8.1
In the lab, we convinced Perkin-Elmer to provide us with a program to convert their proprietary data file structure for the ICP to a DOS readable ASCII file that could be read into dBase. We discovered after approximately 7 years that the program was not "bug-free." At low levels, the program switched to scientific notation and then dropped the exponent with the effect of creating results many orders of magnitude higher than the true values. Fixing the error necessitated 6 months work in reconstructing the database. This bug affected only trace element analyses and did not affect any data reported prior to the time of discovery and correction. We attempted to develop automated ways to output data from the ion chromatograph, but found that all such programs available for the instrument could not accurately assess the baseline value, and, consequently, could not accurately calculate peaks. These were measured manually, but entered into a Lotus spreadsheet by the lab and imported into dBase.
A significant improvement in efficiency and accuracy was made when we added waterproof bar codes to all sample bottles and placed a bar code reader in the lab. An early problem with correct entry of the PALSARIS number for sample identification was solved and tracking and assembly of data produced by several instruments was greatly simplified.
A variety of codes were developed to describe limitations on the use of the database. These are described in Section 8.8.2. Site location includes the town location of the site. These are coded as shown in Section 8.8.3.
2.39 Programs were developed in dBase to screen data, separate QA/QC results from sample results, and rearrange data in forms acceptable by the prime database were developed. The primary screening program is listed in Section 8.8.4.
Statistical and graphical analyses were performed using dBase, Lotus, and PC-SAS. As the program matured and the database grew in size, PC-SAS became the preferred software. Additional analyses of a geographic nature, were analyzed using PC-ARC/INFO and data files provided by UMass and the Massachusetts Executive Office of Environmental Affairs MassGIS.
2.10. Volunteer Motivation and Reward
Management of a long-term, large scale, citizen monitoring project has many additional components not common to most research efforts. Included are initial volunteer involvement (section 2.1), unique aspects of quality control (section 2.7), scheduling (section 2.4), and data management (section 2.9). In this section, we discuss the long-term involvement of volunteers through a planned system of motivation and reward.
To create and maintain volunteer involvement, volunteers must perceive value to them and to a broader cause in return for their efforts. The broader cause is relatively easy to define; for ARM, it is described in the overall objectives (section 1.0). For the individual volunteer, the reward is not so easily discerned. All volunteers seem to need a clear, understandable, statement of the project objectives, scrupulous honesty and fairness on the part of the project organizers (i.e. promises are never broken), and rapid response to volunteer information needs. It is also important to realize that as a member of a group of volunteers, the individual will seek both group recognition and individual recognition. A workable system of appropriate rewards and motivation must keep both in mind.
In some cases, we came by this lesson the hard way. The project nearly collapsed after the appearance of a very critical newspaper article written by a reporter/volunteer who had not received any results for three months. We had sent results to his coordinator and asked that copies be sent to all volunteers. From then on, we sent results directly to all volunteers. Nearly at the same time, the project was criticized for prematurely releasing data with only two month's results and no peer review (the release was our return of results to volunteers with a brief summary), but the criticism was deflected because ARM results were compared to data from state agencies that were from one sampling and not peer reviewed. Nevertheless, the choice had to be made between rapid release of reasonably quality checked results to the volunteers versus sequestering results until all quality aspects were certain and manuscripts were peer reviewed; we chose rapid release. Ten years later, we do not regret the choice.
Rapid data release was not enough to ensure long-term commitment of volunteers. Many other forms of communication and expression of appreciation were developed. These took the forms of media reports (planned and requested), newsletters, presentations to groups, research and general publications, agency and governor press conferences, award and appreciation ceremonies, gifts of T- shirts, buttons, appreciation certificates, letters of support/recommendation to work supervisors,
2.40 calendars, and decals, design contests and presentation at scientific and other meetings. Below we present some examples of these various forms.
2.10.1 Rock Concerts
Very early in the project, prior to state funding, we were fairly desperate for funding. The Connecticut River Watershed Council helped initiate contact with a local band to do a benefit concert. The concert occurred, visibility of the project was improved but the monetary benefit to the project did not occur. It is the only time when this author has ever spoken to a group of partially deafened teens during the intermission of a rock concert, a memory which sometimes recurred as my kids became teenagers. In truth, with a son close to the local rock music scene, these concerts barely pay for themselves. Intentions were honorable and perhaps a few young people became aware of the environment.
2.10.2 Media Reports
The Acid Rain Monitoring Project received a staggering amount of media attention in the beginning. As time progressed, attention recurred, sometimes initiated by a press release on our part, other times by action at the national level, and, occasionally, by a reporter's search for a story when news was slow. There also developed a sense that reporters had circular rolodexes and that, at least annually, a significant newsmaker in the past would come around for a follow-up interview. We clipped all relevant articles from local newspapers, sought out major articles in state-wide newspapers and missed most of the other local newspaper reports except a few sent by our volunteers. Nevertheless, our records show that the Project was cited in at least 167 articles during its ten years. The Boston Globe carried 16 of these, the Springfield Union carried 41 stories, the Northampton Gazette carried 30 articles. At least thirty-six newspapers within the state carried articles. Nationally, stories were carried by the New York Times and NPR's All Things Considered. The Project was cited twice in the Congressional Record, most Figure 2.10-1. Daily Hampshire Gazette especially during the opening debate of the Clean Air announcement of ARM benefit Act. concert, November 21, 1982.
2.41 .
Figure 2.10-2. Ed Driscoll presenting Senator Edward Kennedy with an ARM t-shirt on Mt. Greylock with Ernie LeClair and Gene Chague (left to right) looking on.
2.10.3 Identification of Volunteers
As means of group identification and as a reward for service, the ARM Project provided T- shirts to volunteers. They provided opportunities to thank various officials for their efforts on behalf of acid rain control (Figure 2.10-2). The first time, we asked for reimbursement of costs; but after that, and with approval of the funding agency, we provided them free to all ARM volunteers who would tell us their size. Staff designed the first few and a statewide contest provided two other designs. Three of the four designs are shown in Figures 2.10-3 to 2.10-5.
2.42 Figure 2.10-3. 1987 T-shirt design, color on white background. Towns with median sensitivities in acidified or critical category - red: in endangered to sensitive category - orange; in not sensitive category - green (see Fig. 3.1-7 and Table 3.1-2), Design by ARM staff
Figure 2.10-4. 1988 T-shirt design, red on yellow background. Winner of statewide school contest. Design by Mariah Peelle, Shutesbury Elementary School.
2.43 Figure 2.10-5. 1990 T-shirt design, purple on light green. Original art by Scott Landry, University of Massachusetts student.
2.44 2.10.4 Buttons and Decals
Figure 2.10-6 Pin-on Button (blue on white) Figure 2.10-7 Stick-on decal (blue on white)
By the second year, we realized that easy recognition of an ARM volunteer was desireable to avoid problems during sampling and for awareness the rest of the time. Our first efforts focused on T-shirts, but T-shirts were not always the dress de rigeur. As a result, we produced both a button (Figure 2.10-6) and a stick-on decal (Figure 2.10-7) that were circular with blue printing on a white background. There is no scientific survey of their effectiveness, but a chance meeting of a very frustrated air traveler and ARM volunteer, bumped from her flight, who saw my button, may have made her day and justified the utility of the buttons. It is my impression that the decals were less useful than the buttons. Both were primarily effective in establishing rapid recognition between volunteers and landowners.
2.10.5 Newsletters
A newsletter was the primary vehicle to return collection results to volunteers. The newsletter had its own letterhead so that participating organizations would not feel co-opted. Contents of the newsletter focused on a brief summary of statewide results for that sampling period, attached data reports for that county, news on the national front, and sampling and analysis tips. The volunteer profile, described in Section 5, provided interesting anecdotes and amusing stories from volunteers that were shared via the newsletters.
2.45 2.10.6 Presentations to Volunteer Groups
Many organized groups helped develop and maintain the ARM network of volunteers. In return, no request for attendance at an organization’s meeting was ever denied. These requests included special presentations on project results, award ceremonies and annual meetings.
2.10.7 Publications, Presentations, Testimony Public Debate, and Awards
PUBLICATIONS INTERNATIONAL SCIENCE JOURNALS
1996 Evidence of Recovery from Acidification in Massachusetts Streams. Mark D. Mattson, Paul J. Godfrey, Marie-Françoise Walk, Peter A. Kerr, and O. Thomas Zajicek. in press. Water, Air and Soil Pollution. 1993 Road Salt Contamination of Streams in Massachusetts. M.D. Mattson and P.J. Godfrey. Environmental Management 18(5):767-773. 1992 Regional Chemistry of Lakes in Massachusetts. M.D. Mattson, P.J. Godfrey, M.F. Walk and O.T. Zajicek. Water Resources Bulletin 28(6): 1045-1056. 1992 Acidity Status of Surface Waters in Massachusetts. M.F. Walk, P.J. Godfrey, A. Ruby III, O.T. Zajicek, and M. Mattson. Water, Air, and Soil. Pollution 63 (3-4): 237-252.
POPULAR AUDIENCE
1991 Acid Rain: The Scientific Challenge. Paul J. Godfrey. Science Probe, July 1991: 71-80. 1991 Acid Rain Questions: Response. Paul J. Godfrey. Science Probe, October 1991: 8, 12, 117. 1988 Acid Rain in Massachusetts: The Massachusetts Acid Rain Research Program in Action. Paul J. Godfrey. Water Resources Research Center. More than 3000 copies distributed mostly to citizens and students in Massachusetts 1985 A Report on Massachusetts' Surface Waters. Paul J. Godfrey. Massachusetts Wildlife, 36(1), May-June 1985.
TECHNICAL AUDIENCE
1990 Effects of Acid Deposition on Surface Waters. M.-F. Walk and P.J. Godfrey. Jour. New England Water Works 104(4): 248-251. 1988 The Massachusetts Acid Rain Monitoring Project A.R.M.: Phase II. Armand Ruby III, Paul J. Godfrey, and O.T. Zajicek. Water Resources Research Center. 1988 Sensitivity of Surface Waters in Massachusetts: A Listing of the Sensitivity of All Surface Waters Sampled by the Acid Rain Monitoring Project. Nicholas Layzer, M-F. Walk. and Paul J. Godfrey. Water Resources Research Center, No. 157, University of Massachusetts, Amherst. 1985 The Massachusetts Acid Rain Monitoring Project A.R.M.: Phase I. Paul J. Godfrey, Armand Ruby III, and O. T. Zajicek. Water Resources Research Center, No. 147, University of Massachusetts, Amherst.
2.46 PROFESSIONAL REPORTS, PAPERS and CONFERENCES
Reassessing and Expanding Current Projects: How to Breathe New Life into Existing Citizen Monitoring Programs; Walk, M.F., S. Handley, K. Ellett. Third National Citizens' Volunteer Water Monitoring Conference, Annapolis, March/April 1992.
Use of GIS in Water Resources: PALIS, ARM, and SARIS. M. Mattson, Research & Education Opportunities Using GIS, April 22 1992, UMass.
Acid Rain and the Acid Rain Monitoring project; M.F. Walk invited speaker. Help Save our Environment Conference, Endicott College, Beverly, MA. June 12, 1992.
"The ARM Example." Massachusetts Congress of Lakes and Ponds Association. P.J. Godfrey. Lay Water Quality Monitoring Workshop. March 31, 1990. Worcester, MA.
"A Survey of Massachusetts Surface Waters Sensitivity to Acidification." P.J. Godfrey. Presentation at the annual meeting of the American society of Limnology & Oceanography, June 11-15, 1990.
"The Massachusetts Acid Rain Monitoring Project: Citizens Environmental Monitoring with an Emphasis on Quality Control." P.J. Godfrey. Presentation at the Second National Citizens Volunteer Monitoring Conference held December 1989, New Orleans, LA. (Abstract to be published).
"The Massachusetts Acid Rain Monitoring Project: Focus on Quality Control." P.J. Godfrey. Presentation at The Role of Citizen Volunteers in Environmental Monitoring Workshop held May 23-25, 1988, University of Rhode Island, Narragansett, RI. (Abstract published).
Conference on "Equitable Solutions to Acid Rain." P.J. Godfrey. Moderator and member of Steering Committee. Mt. Holyoke College and Oberlin College. April 29-30, 1988 at Mt. Holyoke College, S. Hadley, MA.
"Update on the Acid Rain Monitoring Project." P.J. Godfrey. presented at the Northeast Fisheries Symposium, Boston, MA, May 6, 1987
"Seasonal and Geographical Trends in pH and Alkalinity for Surface Waters in Massachusetts" P.J. Godfrey. presented at the annual meeting of the American Society of Limnology and Oceanography, Kingston, R.I., June 23-26, 1986.
"The Acid Rain Monitoring Project: The Characterization of Massachusetts Surface Water Sensitivity to Acid Deposition" P.J. Godfrey. presented at the Northeast Fisheries Symposium, Hershey, PA, April 29, 1986.
"Recent results on the Acid Rain Monitoring Project" P.J. Godfrey. presented during Acid Rain Awareness Week special ceremonies. 1985.
2.47 "Cooperative Project on Acid Rain Monitoring: University--Citizen Interaction" P.J. Godfrey. presented at 1985 University Council on Water Resources annual meeting.
"The Acid Rain Monitoring Project, Phase II." P.J. Godfrey. presentation at the Northeast Fisheries Conference Boston, MA. May 4, 1987. "Seasonal and Geographic Variations in pH and Alkalinity for Massachusetts' Surface Waters During Phase I of the Massachusetts Acid Rain Monitoring Project." Northeast States Acid Precipitation Symposium.
Moderator (P.J. Godfrey) of a panel discussion for American Society for Public Administration with panelists Lt. Governor John Kerry, George Rejohn from the Canadian Embassy, and the Director of the Illinois Environmental Protection Agency. 1984.
Panel discussion on Acid Rain at Dean Jr. College Environmental Forum. P.J. Godfrey with William Brown, Hudson Institute and George Rejohn, Canadian Embassy.
Panel discussion (P.J. Godfrey) at Western New England Law School.
Soil Conservation Association Annual Meeting presentation. P.J. Godfrey with K. Rahn, University of Rhode Island and C. Frink, New Haven Experiment Station.
Presentation (P.J. Godfrey) at Berkshire Museum public lecture series.
Presentation (P.J. Godfrey) at workshop sponsored by Massachusetts Department of Environmental Management.
"Update on the Acid Rain Monitoring Project" P.J. Godfrey. presented to the Governor's Acid Rain Working Group.
"The New England Perspective on Acid Rain" P.J. Godfrey. presented at the Indiana Energy Symposium, Purdue, Indiana, October 3-4, 1983.
"The Impact of Acid Rain on Massachusetts Water Resources" P.J. Godfrey. presented to the Governor's Acid Rain Working Group. Boston, Massachusetts
TESTIMONY
Testimony by P.J. Godfrey before Representatives Conti and Boland on the acid rain problem. Springfield, Massachusetts.
Testimony by P.J. Godfrey on Massachusetts impacts of acid rain and research needs before the Massachusetts Joint Committee on Agriculture and Natural Resources. State House, Boston, Massachusetts.
2.48 PUBLIC DEBATE
Public debate by P.J. Godfrey of the Massachusetts Acid Rain Emissions Reduction Referendum question on Public TV, Channel 57 - Springfield and Northampton Cable Access TV, 1986
AWARDS
1983 Certificate of Recognition from Governor Michael Dukakis presented at a press conference in the State House.
1984 ARM Volunteer Laboratory Recognition Awards and Ceremony presented by Lieutenant Governor John Kerry.
1984 Trout Unlimited Silver Trout Award.
1984 Recognition of volunteer effort by the University of Massachusetts and State Senator John Olver at two ceremonies.
1985 Secretary's Commendation, Secretary of Environmental Affairs, James Hoyte, presented at a press conference in the State House.
1985 Gulf Oil Conservation Award
1987 Certificate of Appreciation to each volunteer from the ARM Project
1990 Search for Success National Environmental Achievement Award, Category: Air Pollution Reduction. Washington, D.C., April 1990.
Friends of the United Nations Environmental Figure 2.10-8. President George Bush and Programme (FUNEP). FUNEP 500 Winner for Dr. Paul Godfrey at the Rose Garden reception for World Environment Day, April 19, 1990. National Environmental Achievement Award 1990 Recognition by President George Bush in a Rose Winners. Garden recepton for Search for Success awardees, April, 1990.
2.49 Figure 2.10-9. Reception at the Norwegian Embassy, Washington, D.C. for National Environmental Achievement Award Winners.
Figure 2.10-10. Rose Garden reception for National Environmental Achievement Award Winners
2.50 2.10.8 Appreciation Parchment
In 1986, we recognized the efforts of the volunteers by awarding certificates of appreciation to the volunteer labs and the volunteer sample collectors. The volunteer labs received a certificate from the Governor of Massachusetts, presented by Lieutenant Governor John Kerry. The volunteer collectors received a certificate from the project, each hand calligraphed and individually signed.
Figure 2.10-11. Example of the Certificate of Appreciation presented to volunteer sample collectors.
2.51 2.10.9 Calendar
We developed a calendar for two reasons. Primarily, we were looking for a mechanism to remind volunteers of the sampling dates as they were committing calendar dates to other things. We also sought something that would keep the ARM Project in their conciousness throughout the year. The calendar seemed fairly effective at accomplishing both purposes.
Figure 2.10-12. ARM Calendar
2.52 3.0 Comprehensive Survey -- Phases I &II 3.1 pH and ANC1 3.1.1 General Results
Sixteen hundred and ten sites for pH and 1515 sites for ANC met the quality assurance standards detailed in Section 2.7.4 and were accepted in ARM I. An average of 900 sites for pH and 829 sites for ANC consistently had acceptable analyses every month in ARM I. Close to twenty-five hundred sites were accepted in each of the two sample dates in ARM II. PH and ANC were averaged arithmetically for all 14 months of ARM I and the two collections of ARM II for each site, and these site means were then averaged for all sites, resulting in the statistics shown in Table 3.1-1. Also tabulated are the summer/fall (June through October) and winter/spring (December through April) averages as well as the October and April averages in ARM II.
Table 3.1-1. Site mean averages for pH and ANC, ARM I and ARM II.
pH ANC (µeq L-1)
Number Average Number Average
ARM I
Summer/Fall 1341 6.68 1262 370
Winter/Spring 1473 6.34 1366 260
Annual 1610 6.51 1515 330
ARM II
October 2482 6.56 2452 391
April 2450 6.50 2439 258
Overall 2764 6.53 2744 341
Throughout the course of ARM I and II, 382 sites or 9% of those sampled exhibited at least one measured occurrence of pH below 5.0; 2952 sites or 67% never exhibited a measurement of pH below 6.0; 1214 sites or 28% had an ANC below 50 µeq L-1 at least once, and the ANC of 1638 sites or 37% never dropped below 200 µeq L-1.
For the October samples (n=2001) we found 118 (5.9%) had ANCs less than zero, and 21.5% and 52% having ANCs less than 50 and 200 µeq L-1, respectively. The median pH was 6.58 and the median ANC was 184 µeq L-1. Acidic lakes were more frequently dark in color, with 46% having color values greater than 30 PCU. In comparison, the overall population has a frequency of 36.7 percent of lakes with PCU>30.
1 Expanded from Walk et al., 1992
3.1 9
· · · · · · · · 8 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 7 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ···· ·· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ····· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ······ · ········ · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·· ··· ·········· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·········· · ·· ····· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · pH · · · ······· ·········· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·· ·························· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ············ ······ ······ · · · · · · · · · · · · · ·········· ··· · · · · · · · · · · · · · · · · · · ················ ······ ··· · · · · · · · · · · · · · 6 ···· ··········· · ·· ······ · · · · · · · · · · · · · ···················· · ········ · · · · · · · · · ······· ······· ··· ···· ··· · · · · · · ·············· ·· · ·· ··· · · · ········· ·· · ·· · · ·· ·············· ····· · · · ··········· · · ·· · ··· ············· · · · ···· ···· ·· ·· · · · ······ ··· · ·· · ········· · · 5 ·········· ········ · · ·········· ··· ······ · · ·········· · · ·· ··· · ·· ·
4 -100 0 100 200 300 400 500 600 700 800 ANC (ueq/L) Figure 3.1-1. Lake pH vs. ANC for Lakes with ANC Less Than 800 ueq L-1. The upper curved line represents the relationship expected for lakes with carbonate ANC in equilibrium -3.5 with atmospheric carbon dioxide (pCO2 = 10 atm.). The lower line represents the -2.55 curve of best fit to the pH and ANC data (pCO2 = 10 )
9
· · 8 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 7 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ···· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ··· · ·· ·· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ··· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · pH ·· ···· · · · · · · · · · · · · · · · · · · · · · · · · · · ·· · · · · · · · · · · · · · · · · · · · · · · · · · · ····· ···· · · · · · · · · · · · · · · · · · · · · · · · · · · · ······ · · · · · · · · · · · · · · · · · · · · ·· ·· · ·· · · · · · · · · · · 6 ···· ·· · ····· ·· · · · · · · · · ······· · · · · · · · · · · · ··· ···· ···· ····· · · · · · ··········· · · ·· · · · · · · · · · · ·· · · · · ·· · · · · · · · ·· · · · · · · · ·· ··· · ····· · · 5 · ···· · · · · · · · · ··· ··· · · ··· ··· ··· · · ·· · ··· · 4 -100 0 100 200 300 400 500 600 700 800 ANC (ueq/L) Figure 3.1-2. Stream pH vs. ANC for streams with ANC Less than 800 ueq L-1. The upper curved line represents the relationship expected for streams with carbonate ANC in -3.5 equilibrium with atmospheric carbon dioxide (pCO2=10 atm.). The lower line -2.4 represents the curve of the best fit to the pH and ANC data (pCO2=10 atm.).
Most lakes had pH values lower than expected from the theoretical distribution of pH vs. ANC for solutions in equilibrium with atmospheric carbon dioxide (Fig. 3.1-1). Similar results have been reported in other surveys (Linthurst et al., 1986; Driscoll et al., 1989; Driscoll et al. 1990; Baker et al., 1990) and are probably due to the pH depression caused by excess dissolved carbon dioxide, and/or the presence of organic acids. Streams also showed lower pH values than the theoretical
(Figure 3.1-2). The best fit curve suggests that the partial pressure of CO2 was lower in streams than
3.2 lakes on average. Comparison with the Adirondack Lake Survey results indicates that the combination of partial pressure and organic acids resulted in a steeper slope for the Massachusetts lakes' data than for Adirondack lakes while the Eastern Lakes Survey (Baker et al., 1990) exhibits an even steeper slope than Massachusetts lakes' data suggesting that excess carbon dioxide and organic acids are higher in Adirondack lakes and lowest in the Eastern Lakes Survey lake population.
3.1.2. Seasonal Patterns of pH and Acid Neutralizing Capacity
Significant variation was observed in both pH and ANC. Lowest values were observed in winter and spring, and highest measurements occurred in summer and fall (Figures 3.1-3 and 3.1-4). For ARM I data, the months June through October were grouped into a "summer/fall" period and compared through analysis of variance to a "winter/spring" period: March and April 1983, and December 1983 through April 1984. May and November did not consistently fit in any period for pH nor ANC and were thus considered transition months. An analysis of variance showed that the winter/spring and summer/fall periods were significantly different for both pH and ANC (p<0.0001). Winter/spring samples contained on the average 45% more H+ (arithmetic mean pH 6.44 vs 6.60) and 32% less ANC (257 vs 376 µeq L -1) than summer/fall samples.
Median alkalinities (overall median of the sites' medians) were substantially lower than their respective means (78 µeq L -1 for winter/spring and 210 µeq L -1 for summer/fall) because of skewed frequency distributions. The ANC distribution displayed a relatively "long tail" to the right which is apparently responsible for the upward skewing of the monthly means relative to the medians. The median therefore seems a more descriptive measure of ANC for the population of Massachusetts' surface waters. Median pH 's were also lower than means, particularly in winter/spring (6.2 vs. 6.44), and somewhat so in the summer/fall (6.58 vs. 6.60). The pH frequency distribution, however, appeared relatively normal.
3.3 Cumulative frequencies are a convenient way of presenting information on pH and ANC levels for a large number of surface waters and used in presenting results from the U.S. EPA Eastern Lakes Survey (Linthurst et al., 1986). For comparison the cumulative frequencies for spring and fall pH (Figure 3.1-5) and ANC (Figure 3.1-6) results for the ARM Project. are shown.
These data emphasize the importance of considering seasonal variability when planning surface water chemistry surveys. The collection dates for ARM II, April and October, were in part chosen to represent the two extremes of the seasonal spectrum of fresh water chemistry.
3.1.3. Sensitivity of Massachusetts Surface Waters to Acidification
For the purpose of classifying ARM's large data base, six sensitivity categories were established based on ANC and pH . These category designations were selected to permit comparisons with surveys by Massachusetts State agencies, EPA and others. They are meant to reflect the susceptibility of surface waters to acidification. Each category was named in order to facilitate communication with the citizen volunteers. Category description and the distribution in each category of all the surface waters sampled in the ARM project are shown for both October and April collections in Table 3.1-2.
3.4 Figure 3.1-5. Cumulative frequency of ñH for ARM Phase II, April and October Sites.
100%
80%
60%
40%
20%
0% -5 0 5 10 15 20 25 30 35 40 45 50
ALKALINITY (mg/l) OCTOBER APRIL Figure 3.1-6. Cumulative Frequency of ANC for ARM Phase II, April and October sites.
In the spring, which is the critical season for most aquatic reproduction, five and one half percent of the surface waters in Massachusetts were found to be acidified and 63% exhibited vulnerability to episodic acidification (ANC#200 µeq L -1).
3.5 Table 3.1-2. Statewide distribution of surface water samples in each ARM ANC Category and mean/median pH and ANC for ARM I and ARM II combined.
Category ANC (µeq L-1) Combined ARM I & II October Combined ARM I & II April samplesa samplesb
Percent Number Percent Number
Acidified <0 pH < 5.0 4.5 148 5.5 189
Critical >0 - 40c 10.1 337 16.7 573
Endangered >40 - 100 13.4 444 19.5 668
Highly >100 - 200 17.2 571 21.2 725 Sensitive
Sensitive >200 - 400 24.2 801 20.4 697
Not Sensitive >400 30.6 1014 16.7 571
Total 100 3315 100 3423
Mean pH 6.56 6.46
Median pH 6.61 6.52
Mean ANC 384 248 (µeq L-1)
Median ANC 228 133 (µeq L-1)
a Includes values from October 1983 and October 1984
b Includes values from April 1983, 1984, and 1985. If any given water body was sampled more than one October or April, the values were averaged for that water body.
c This category also includes ANC > 0 and pH < 5.0.
3.1.4. Comparison of Lakes vs. Streams
As shown in section 3.1.1, lakes and streams differ slightly in their pH/ANC relationship. Figure 3.1-7 shows that the percentage of lakes in more sensitive categories is higher than for streams and is due primarily to the increased percentage of lakes with ANCs less than 40 ueq L-1 (Table 3.1-3 and 3.1-4). Four counties have above average percentages of acidified streams: Berkshire, Worcester North, Plymouth and Norfolk (Table 3.1-3) Plymouth and Worcester North also have high percentages of acidified lakes (Table 3.1-4); Berkshire and Norfolk streams sharply contrast with their lakes (Table 3.1-5). At least in Berkshire County, the difference may be attributable to the occurrence of most lakes in the lowland areas (the Stockbridge region described in Section 3.2)
3.6 where high alkalinity surface waters are common, whereas the streams are mostly found in the highlands (the Berkshire region) with low alkalinities. Barnstable County also exhibits sharp differences between lakes and streams. For the most part these coastal streams fall in the middle range of sensitivity, but the county's lakes tend to be extremely sensitive.
Percent 100%
80% Not Sensitive 60% Sensitive Highly Sensitive Endangered 40% Critical Acidified 20%
0% Streams Lakes Site Type
Figure 3.1-7. Comparison of Lotic vs. Lentic Sites for Phase I and II Combined.
Category % Streams % Lakes Acidified 4.9 6.4 Critical 14.5 18.3 Endangered 19.2 20.4 Highly Sensitive 21.3 21.6 Sensitive 21.8 18.3 Not Sensitive 18.5 14.9
3.7 Table 3.1-3. Number and Percent of Streams per Sensitivity Category
Number Category Highly Not Acidified Critical Endangered Sensitive Sensitive Sensitive Row Total County BA 0 3 8 14 5 1 31 BE 15 4 14 12 15 2 62 BR 2 21 31 33 36 66 189 DU 0 1 3 1 0 0 5 ES 2 6 4 11 17 21 61 FR 9 50 40 28 54 56 237 HN 3 15 42 44 20 26 150 HS 3 12 11 37 33 14 110 MI 0 7 24 31 62 50 174 NO 3 1 3 11 17 3 38 PL 11 17 24 25 17 7 101 WN 19 34 27 10 10 3 103 WS 3 35 42 46 24 14 164 Total 70 206 273 303 310 263 1425
Percent BA 0.0% 9.7% 25.8% 45.2% 16.1% 3.2% BE 24.2% 6.5% 22.6% 19.4% 24.2% 3.2% BR 1.1% 11.1% 16.4% 17.5% 19.0% 34.9% DU 0.0% 20.0% 60.0% 20.0% 0.0% 0.0% ES 3.3% 9.8% 6.6% 18.0% 27.9% 34.4% FR 3.8% 21.1% 16.9% 11.8% 22.8% 23.6% HN 2.0% 10.0% 28.0% 29.3% 13.3% 17.3% HS 2.7% 10.9% 10.0% 33.6% 30.0% 12.7% MI 0.0% 4.0% 13.8% 17.8% 35.6% 28.7% NO 7.9% 2.6% 7.9% 28.9% 44.7% 7.9% PL 10.9% 16.8% 23.8% 24.8% 16.8% 6.9% WN 18.4% 33.0% 26.2% 9.7% 9.7% 2.9% WS 1.8% 21.3% 25.6% 28.0% 14.6% 8.5% Total 4.9% 14.5% 19.2% 21.3% 21.8% 18.5%
3.8 Table 3.1-4. Number and Percent of Lakes per Sensitivity Category
Number Category
Acidified Critical Endangered Highly Sensitive Not Row Total County Sensitive Sensitive BA 32 77 49 45 12 3 218 BE 3 16 18 20 17 81 155 BR 12 12 26 28 30 11 119 DU 0 1 6 6 3 5 21 ES 3 12 20 34 47 50 166 FR 3 20 16 11 9 8 67 HN 4 18 26 33 20 31 132 HS 2 7 15 14 22 10 70 MI 3 19 29 51 93 58 253 NO 3 9 19 27 46 13 117 PL 38 80 57 43 14 5 237 WN 24 68 44 15 14 7 172 WS 7 42 99 121 54 28 351 Total 134 381 424 448 381 310 2078
Percent
BA 14.7% 35.3% 22.5% 20.6% 5.5% 1.4% BE 1.9% 10.3% 11.6% 12.9% 11.0% 52.3% BR 10.1% 10.1% 21.8% 23.5% 25.2% 9.2% DU 0.0% 4.8% 28.6% 28.6% 14.3% 23.8% ES 1.8% 7.2% 12.0% 20.5% 28.3% 30.1% FR 4.5% 29.9% 23.9% 16.4% 13.4% 11.9% HN 3.0% 13.6% 19.7% 25.0% 15.2% 23.5% HS 2.9% 10.0% 21.4% 20.0% 31.4% 14.3% MI 1.2% 7.5% 11.5% 20.2% 36.8% 22.9% NO 2.6% 7.7% 16.2% 23.1% 39.3% 11.1% PL 16.0% 33.8% 24.1% 18.1% 5.9% 2.1% WN 14.0% 39.5% 25.6% 8.7% 8.1% 4.1% WS 2.0% 12.0% 28.2% 34.5% 15.4% 8.0% Total 6.4% 18.3% 20.4% 21.6% 18.3% 14.9%
3.9 Table 3.1-5. The difference between the percentage of lakes in each sensitivity category and the percentage of streams in that category. Differences of greater than 10% are shaded.
Lakes % - Streams % County Acidified Critical Endangered Highly Sensitive Sensitive Not Sensitive BA 14.7% 25.6% -3.3% -24.5% -10.6% -1.8% BE -22.3% 3.9% -11.0% -6.5% -13.2% 49.0% BR 9.0% -1.0% 5.4% 6.1% 6.2% -25.7% DU 0.0% -15.2% -31.4% 8.6% 14.3% 23.8% ES -1.5% -2.6% 5.5% 2.4% 0.4% -4.3% FR 0.7% 8.8% 7.0% 4.6% -9.4% -11.7% HN 1.0% 3.6% -8.3% -4.3% 1.8% 6.2% HS 0.1% -0.9% 11.4% -13.6% 1.4% 1.6% MI 1.2% 3.5% -2.3% 2.3% 1.1% -5.8% NO -5.3% 5.1% 8.3% -5.9% -5.4% 3.2% PL 5.1% 16.9% 0.3% -6.6% -10.9% -4.8% WN -4.5% 6.5% -0.6% -1.0% -1.6% 1.2% WS 0.2% -9.4% 2.6% 6.4% 0.8% -0.6%
3.1.5 Geographic Distribution of Water Bodies According to their Sensitivity
A substantial degree of geographic variability was found in surface water chemistry for Massachusetts: north-central and southeastern coastal counties have the lowest average pH and ANC levels. The sensitivity of Massachusetts' surface waters is shown geographically in Figure 3.1-8. The water bodies with lowest pH and ANC are found in southeastern Massachusetts and northeast Worcester county, areas characterized by very poor soils and granitic and gneiss bedrock, both of which would impart very little acid neutralizing capacity to local surface waters. The county showing highest ANC has the only significant limestone deposits in the state (Berkshire county).
3.1.6 Statistical Analysis of Geographic Variation in Lake ANC
We investigated which factors best explain the variation in ANC among lakes in the state. A map of the distribution of lakes in various ANC classes shows that lakes tend to group together by ANC class (Fig. 3.1-9). The highest ANC lakes are in the west (Region 1) and the acidic lakes are clustered in the north central (Region 4) and on Cape Cod (Region 6). Lacking detailed information on processes controlling ANC within each lake system, we gathered available information on 660 lakes (October 1983, 1984 data) which have complete data for ANC, color, lake area, elevation, and silica. Silica can be used as a measure of bedrock weathering (Yuretich and Batchelder, 1988). We also defined a drainage variable, which is 1 if the lake is open and 0 if closed. Open is defined as having a stream outlet marked on a 7.5 minute USGS map. We used a map of hydrogen ion deposition in the northeast (Driscoll et al., 1991) to roughly estimate H+ deposition for each lake.
3.10 Figure 3.1-8. Mean Alkalinity by Town of Massachusetts Surface Waters
We used an analysis of variance (ANOVA) to analyze the class variable region and compared the results to a general linear models (GLM) analysis (SAS, 1987) of a combined model based on a class variable (drainage type) with five additional linear variables to examine how well these two different models can explain the variation in ANC. The combined six-variable GLM model with color, area, elevation, silica, drainage and deposition could explain only 4.9% of the variation in ANC (Table 3.1-6). In comparison, a simple ANOVA based solely on region could explain 50.9% of the variance in the same data set.
3.11 Table 3.1-6. Analysis of variance in ANC among 660 lakes between (A) GLM model of six factors and (B) ANOVA of region on ANC. October 1983 and 1984 ANC data. ______(A) General Linear Models Source df Type I SS F Prob.>F Drainage 1 29445 0.16 0.6897 Color 1 126631 0.69 0.4077 Area 1 783934 4.25* 0.0397 Elevation 1 90780 0.49 0.4833 Silica 1 5001974 27.11** 0.0001 H+ dep. 1 156393 0.85 0.3576 Error 653 120496270 R2=0.049 (B) ANOVA Source df Type I SS F Prob.>F Region 5 64436880 135.4** 0.0001 Error 654 62248549 R2=0.509 ______*Significant at 0.05 á level **Significant at 0.0001 á level
3.12 3.1.7. Comparison with other Large Scale Surveys
The most directly comparable large scale survey is the U.S. EPA Eastern Lakes Survey (ELS). The ELS surveyed 97 lakes in Massachusetts (parts of ELS regions 1C and 1D) in late October and early November, 1984. Many of the ELS surveyed lakes were included in the phase II ARM Project which sampled 1897 lakes on October 14, 1984. Nevertheless, the two surveys differ in several respects. While ARM was nearly comprehensive in scope, site selections in Phase I and II were not statistically based. The ELS sites were statistically selected but included only lakes. In ELS, a variety of sites were excluded from consideration. All lakes not visible on 1:250,000 scale topographic maps were not included. This excluded nearly all lakes less than 4 hectares in area and a number of others of larger size that were not on the topographic maps. As shown in Figure 2.7-1, there was a slight bias against lakes smaller than 4 hectares in the ARM sample population. Lakes with suspected coastal or urban influences were also excluded by the ELS. Based on the description of the effect of these exclusions in the ELS report (Baker et al., 1990; Linthurst et al., 1986), the effect of exclusions versus the named lakes included in the PALIS listing (Ackerman et al., 1984) is shown in Figure 3.1-10. ELS sampling sites were selected from the target population and are shown versus the ARM sample population in Figure 3.1-11.
BREAKDOWN OF LAKE POPULATION COMPARISON BETWEEN EPA AND PALIS
3000
2500
2000
1500 PALIS <4 ha PALIS >4 ha 1000 EXCL. >16 ha EXCL. 8-16 ha EXCL. 4-8 ha 500 EXCL. <4 ha FRAME POPULATION TARGET POPULATION 0 EPA PALIS SURVEY Figure 3.1-10. Assessment of EPA statistical site selection process on the development of the target population for potential sampling (based on Johnson et al., 1989) versus lakes in the PALIS listing (Ackerman et al., 1984) categorized by area.
3.13 PERCENT OF TOTAL NAMED LAKE POPULATIONS
PERCENT 100%
80%
60%
40%
NON-CANDIDATE 20% SITE EXCLUSION MAP EXCLUSION ESTIMATED SAMPLED 0% EPA ARM SURVEY Figure 3.1-11. Comparison of the ARM sampled and estimated lake population with that of the EPA ELS. Differences also existed in the methods used for measuring pH and ANC. The ELS used closed head space and air equilibrated pH (with preference for closed head space results); ARM attempted to minimize degassing by instructing the volunteers to completely fill the sample bottles and not open them until moments before analysis. Samples were not air equilibrated. The ELS used Gran plot ANC titrations and ARM used double end point ANC titrations. We were interested in determining the bias that might be introduced by these different methods. A preliminary determination was made by comparing ELS and ARM results for lakes sampled by both. Samples were not collected at the same time, in fact, intervals of up to three weeks occurred between the ARM sampling date and the date of ELS sampling. Despite this confounding factor, ANC measurements were in good agreement (Figure 3.1-12). An additional comparison of double endpoint vs. Gran plot titrations is provided in section 2.7.4 and Figure 2.7-8. PH measurements exhibited more scatter, as might be reasonably expected. The ARM pH method was most comparable to the EPA closed head space pH, exhibiting little bias (Figure 3.1-13). There was some bias toward higher pH values with the air equilibrated method when compared to either the ARM method (Figure 3.1-14) or closed head space method.
3.14 COMPARISON OF EPA-ELS VS. ARM EPA ALKALINITY VS. ARM ALKALINITY
ARM ALKALINITY (mg/l CaCO3) 100.0
ARM = .836(EPA) + 0.61 80.0 r square = 0.973 60.0
40.0
20.0
0.0
0.0 20.0 40.0 60.0 80.0 100.0 EPA ALKALINITY (mg/l CaCO3) Regression Line 1:1 Line Figure 3.1-12. Comparison of ARM double endpoint ANC vs. ELS Gran titration ANC for lakes sampled by both in the fall of 1984. Actual sampling dates may differ by as much as three weeks.
Despite the biases introduced by the various map and statistical procedures used by the ELS, our percentage of acidic lakes (5.9%) agrees well with the EPA/ELS estimate of 5.6% for the state of Massachusetts (Linthurst et al., 1986; Baker et al., 1990). To compare our results to the EPA/ELS, we calculated the mean and standard errors (Snedecor and Cochran, 1980) for pH and ANC for both our unweighted data, and for the EPA/ELS data sampled in Massachusetts (Kanciruk et al., 1986) with the sampling strata weights applied. The mean pH values (6.51 vs. 6.54) for our data and ARM EPA data were not significantly different (t=0.4, á>0.05). However, the ARM mean ANC (311 µeq L-1) was significantly higher (t=-2.1, á<0.05) than the mean EPA ANC (236 µeq L-1).
Since the EPA/ELS restricted sampling to lakes larger than 4 ha we reanalyzed the statistics with the ARM data restricted to large lakes and found neither pH nor ANC were significantly different between the ARM and EPA/ELS surveys. The population of small lakes in the ARM survey tended to have both a high frequency of acidic lakes, as well as a high frequency of high ANC lakes, resulting in an overall higher mean ANC for the ARM dataset when compared to the EPA/ELS. A similar relationship between size and lake ANC was reported by Sullivan et al. (1990b).
3.15 COMPARISON OF EPA-ELS VS. ARM EPA CLOSED SYSTEM pH VS. ARM pH
ARM DETERMINED pH
8.0 ARM = .987(EPA) + 0
r square = .689 7.0
6.0
5.0
4.0 4.0 5.0 6.0 7.0 8.0 EPA CLOSED SYSTEM pH Regression Line 1:1 Line
Forced Zero Intercept Figure 3.1-13. Comparison between ARM pH and ELS closed head space pH for lakes sampled by both in the fall of 1984. Actual sampling dates may differ by as much as three weeks.
COMPARISON OF EPA-ELS VS. ARM EPA AIR EQUILIBRATED pH VS. ARM pH
ARM DETERMINED pH
8.0 ARM = .907(EPA) + 0
7.0 r square = .638
6.0
5.0
4.0 4.0 5.0 6.0 7.0 8.0 EPA AIR EQUILIBRATED pH Regression Line 1:1 Line
Forced Zero Intercept Figure 3.1-14. Comparison of ARM pH vs. ELS air equilibrated pH for lakes sampled by both in the fall of 1984. Samples were collected as much as three weeks apart.
3.16 A much larger bias, in our opinion, potentially results from extrapolation of the percentages of lakes in various ANC categories to the number of lakes in those categories. This bias results from the exclusions producing the limited target population compared to the true population of lakes in the state (Figure 3.1-10). Comparing ELS estimates of the number of lakes having a pH below 5.0 and below 6.0 shows that ELS estimates are lower than the number actually measured in ARM and that the upper confidence interval for lakes with pH less than 6.0 approximates the number actually measured in ARM (Table 3.1-7 and Figure 3.1-15). For ANC, the ELS estimates of numbers of lakes approximates the number of lakes greater than 4 hectares in area actually measured by ARM, but the upper confidence interval seriously underestimates the probable number of lakes in each category (Table 3.1-7 and Figure 3.1-16). While Linthurst et al. (1986) and Baker et al. (1990) have been careful in circumscribing the limits of use of ELS results and we do not disagree that the target population was a reasonable choice for evaluating the effects of acid deposition exclusive of confounding cultural factors, subsequent users have and probably will continue to assume incorrectly that ELS results describe the chemistry of Massachusetts lakes and the impact of acid deposition on them. We caution that ELS results should be used to typify only lakes larger than 4 hectares that are neither influenced by urban impacts nor coastal factors. As Figure 3.1- 10 shows, this excludes nearly two-thirds of Massachusetts lakes.
Table 3.1-7. Comparison of ARM October data with EPA/ELS. Sample counts are actual numbers of water bodies sampled. ARM estimate of total extends sample count percentages to all named freshwater lakes and ponds. Estimate of total and upper confidence limit (U.C.L.) estimates of total for EPA/ELS are statistical estimations.
ARM EPA/ELS Lakes < 4 ha. Lakes > 4 ha. Sample Est. of Sample Est. of Total Sample Est. of U.C.L. Count Total Count Count Total Est. pH<5 57 84 53 81 (5%) 6 (6%) 54 93 pH<6 207 304 240 369 (23%) 22 (22%) 180 243 pH>6 627 921 823 1264 (77%) 76 (78%) ANC<0 64 94 65 102 (6%) 7 (7%) 52 83 ANC<50 162 241 255 401 (25%) 29 (30%) 239 311 ANC<200 41. 615 596 938 (57%) 68 (69%) 578 685 ANC>200 409 610 442 695 (43%) 30 (31%)
3.17 COMPARISON OF ARM VS. EPA-ELS pH NUMBERS 400 pH<6.0
300
200
100 pH<5.0
0 EPA ARM>4 ARM<4 EPA ARM>4 ARM<4 LAKES BY SURVEY AND SIZE ESTIMATE UCL COUNT ESTIMATE
Figure 3.1-15. Number of lakes and ponds in low pH categories from ARM (October, 1984) and EPA/ELS
COMPARISON OF ARM VS. EPA-ELS ANC (ueq/l) ANC<200.0 NUMBERS 1000
800
600 ANC<50.0
400
ANC<0.0 200
0 EPA ARM>4 ARM<4 EPA ARM>4 ARM<4 EPA ARM>4 ARM<4 LAKES BY SURVEY AND SIZE ESTIMATE UCL COUNT ESTIMATE
Figure 3.1-16. Numbers of Lakes and Ponds in Low ANC Categories from ARM (October) and EPA/ELS.
3.18 Table 3.1-8 compares ARM results with the five subregions surveyed by the ELS. The Adirondack region has the greatest percentage of low ANC and low ñH lakes, followed by Massachusetts, southern New England, the Poconos and Catskill region, central New England, and Maine, respectively. Although the ELS in its survey of 155 lakes, representing a target population of 1290 lakes, observed greatest sensitivity in the Adirondack region, the Adirondack Survey (Kretser et al., 1989) of 1469 lakes observed even greater sensitivity (Table 3.1-9). Table 3.1-9 compares ALSC results with ARM lake results.
Table 3.1-8. Comparison of ANC and pH distribution of ELS Northeast Lakes and ARM lakes expressed as a percentage of the target population.
ANC (ueq L-1) pH <0 <50 <100 <200 <5.0 <5.5 <6.0 <6.5 Adirondacks 14.0 37.8 49.8 73.0 10.0 19.9 26.6 43.5 Poconos/ 5.9 13.5 20.2 40.9 0.8 5.7 7.8 17.2 Catskills C. New 4.2 23.3 48.8 67.6 1.7 7.8 12.9 32.3 England S. New 6.5 22.6 31.8 57.3 5.0 10.0 14.6 28.2 England Maine 1.6 14.7 35.9 66.8 0.5 1.6 4.8 13.2 All 6.2 21.9 37.1 60.9 3.4 8.6 12.9 26.3
ARM 7.0 30.0 69.0 6.0 22.0
Table 3.1-9. Comparison of Adirondack Lake Survey Corporation Survey results with ARM lake results.
ANC (ueq L1) pH <0.0 >0<40.0 >40.0<200.0 >200.0 <5.0 >5.0<6.0 >6.00 ALSC 26.4 21.0 34.4 17.2 24.0 18.2 57.8 ARM 7.0 18.3 42.0 33.2 6.6 15.8 77.6
Baker et al. (1990) cite the results of other large scale surveys. ANCs and pHs for various regions of Canada are shown in Tables 3.1-10 and 3.1-11. Massachusetts had substantially higher percentage of lakes and streams with no ANC than all Canadian regions except the Sudbury/Noranda and South Atlantic regions
3.19 Table 3.1-10. Comparison of ANC from Canadian surveys reported in Baker et al. (1990) with ARM.
Numbers CANADA ANC (ueq/L) Region Number <400 <200 <100 <50 <0 of Lakes West Ontario 568 467 322 121 36 0 North Ontario 1382 471 226 113 54 8 Southeast Ontario 2086 1928 1768 1439 864 86 Ottawa Valley 236 145 118 83 43 1 Southwest Quebec 749 718 616 449 254 4 Laurentide 743 736 715 627 445 8 North Quebec 538 516 465 373 239 7 South Atlantic 764 660 623 552 450 90 North Atlantic 442 434 419 373 254 14 Sudbury/Noranda 684 632 568 483 402 163 All 8192 6707 5840 4613 3041 381
ARM (October) 1895 1500 1077 730 478 178 ARM (April)
Percentage CANADA ANC (ueq/L) Region Number <400 <200 <100 <50 <0 of Lakes West Ontario 568 82.2% 56.7% 21.3% 6.3% 0.0% North Ontario 1382 34.1% 16.4% 8.2% 3.9% 0.6% Southeast Ontario 2086 92.4% 84.8% 69.0% 41.4% 4.1% Ottawa Valley 236 61.4% 50.0% 35.2% 18.2% 0.4% Southwest Quebec 749 95.9% 82.2% 59.9% 33.9% 0.5% Laurentide 743 99.1% 96.2% 84.4% 59.9% 1.1% North Quebec 538 95.9% 86.4% 69.3% 44.4% 1.3% South Atlantic 764 86.4% 81.5% 72.3% 58.9% 11.8% North Atlantic 442 98.2% 94.8% 84.4% 57.5% 3.2% Sudbury/Noranda 684 92.4% 83.0% 70.6% 58.8% 23.8% All 8192 81.9% 71.3% 56.3% 37.1% 4.7%
ARM 1895 79.2% 56.8% 38.5% 25.2% 9.4%
3.20 Table 3.1-11. Comparison of pH from Canadian surveys reported in Baker et al. (1990) with ARM.
Numbers CANADA pH Region Number of <6.0 <5.5 <5.0 Lakes West Ontario 569 21 6 0 North Ontario 1395 56 25 12 Southeast Ontario 2121 652 227 47 Ottawa Valley 253 48 10 3 Southwest Quebec 752 214 70 16 Laurentide 744 436 140 33 North Quebec 540 188 54 13 South Atlantic 765 416 274 121 North Atlantic 426 173 55 12 Sudbury/Noranda 706 353 219 136 All 8271 2557 1080 393
ARM 1936 472 264 129
Percentage CANADA pH Region Number of <6.0 <5.5 <5.0 Lakes West Ontario 568 3.7% 1.1% 0.0% North Ontario 1382 4.1% 1.8% 0.9% Southeast Ontario 2086 31.3% 10.9% 2.3% Ottawa Valley 236 20.3% 4.2% 1.3% Southwest Quebec 749 28.6% 9.3% 2.1% Laurentide 743 58.7% 18.8% 4.4% North Quebec 538 34.9% 10.0% 2.4% South Atlantic 764 54.5% 35.9% 15.8% North Atlantic 442 39.1% 12.4% 2.7% Sudbury/Noranda 684 51.6% 32.0% 19.9% All 8192 31.2% 13.2% 4.8%
ARM 1936 24.4% 13.6% 6.7%
3.21 Similar comparisons based on Baker et al. (1990) are shown for Sweden and Norway in Tables 3.1-12 and 3.1-13.
Table 3.1-12. Comparison of large-scale survey results for pH from Sweden as reported in Baker et al. (1990) with ARM results.
pH <4.9 5.0-5.9 6.0-6.9 >7.0 SWEDEN 5.6% 35.1% 53.3% 6.1%
ARM 5.9% 15.2% 44.5% 24.5%
Table 3.1-13. Comparison of large-scale survey results for pH from Norway as reported in Baker et al. (1990) with ARM results.
NORWAY pH
County <5.0 5.0-5.5 5.5-6.0 >6.0 South 55.4% 21.9% 11.1% 11.6% 11 43.2% 29.5% 13.6% 13.6% to 6.1% 27.3% 42.4% 24.2% 15-19 0% 6.2% 20.9% 72.9% North 0% 2.9% 8.8% 88.2%
ARM 6.4% 7.2% 11.9% 76.8%
Haines and Akielaszek (1983) compiled the results of other, earlier surveys. As shown in Table 3.1-14., the ARM percentages for lake pH are very similar to those obtained from a survey of New England lakes (Haines and Akielaszek, 1983). A similar comparison is shown for ANC (Table 3.1-15). Again, the distribution of ARM project ANC values approximates that found in the New England survey (Haines and Akielaszek, 1983), and they are within the range reported from other areas receiving highly acidic precipitation. Although the surveys of surface water chemistry in areas receiving precipitation with a pH greater than 4.6 are more limited in scope and may not be statistically based or comprehensive, they all found no surface waters with pH less than 5.0 and few with pH less than 6.0; whereas in areas with precipitation less than 4.6 the percentage of surface waters with pH less than 5.0 ranged from 2% to 64%. The limited data on ANC suggests that ANCs of 0 are also unlikely in these areas, but in areas with lower precipitation pH, between 3% and 100% of surface waters had ANCs less than 20 ueq l-1.
3.22 Table 3.1-14. pH distribution of surface water chemistry survey results (based on Haines and Akielaszek, 1983).
Location Number Percent in pH range Reference
of Sites <5 5-6 >6 Areas where precipitation pH averages < 4.6 Massachusetts 4379 9 24 67 Godfrey et al., 1985 Massachusetts 97 5.8 U.S. EPA, 1986 Central New England 164 17 U.S. EPA,1986 Southern New England 127 5 U.S. EPA,1986 New England 226 8 21 71 Haines and Akielaszek, 1983
West Sweden 314 36 21 43 Almer et al., 1974 South Sweden 51 2 20 78 Malmer, 1975 South Norway 155 18 38 44 Wright, 1977 South Norway 719 64 33 3 Wright and Snekvik, 1978 Scotland 72 26 36 38 Wright et al., 1980 La Cloche Mountains 152 28 34 38 Beamish and Harvey, 1972 Sudbury 150 13 15 72 Conroy et al., 1976 Adirondack Mountains 849 25 30 45 Pfeiffer and Festa, 1980 Adirondack Mountains 1469 24 18 58 Kretser et al., 1989
Areas where precipitation pH averages > 4.6 North Norway 77 0 13 87 Wright and Gjessing, 1976 Northwest Wisconsin 265 0 6 94 Lillie and Mason, 1980 North Minnesota 85 0 0 100 Glass and Loucks, 1980
3.23 Table 3.1-15. ANC distribution of surface water chemistry survey results (based on Haines and Akielaszek, 1983). ANC values are in µeq L-1
Location Number of Percent in ANC range Reference Sites <20 21-100 101-201 >201 Areas where precipitation pH averages < 4.6 Massachusetts 4379 28a 34b 38c Godfrey et al., 1985 Massachusetts 97 26a 62d U.S. EPA, 1986 Central New England 164 18a 68d U.S. EPA,1986 Southern New England 127 22a 57d U.S. EPA,1986 New England 226 23 18 12 47 Haines and Akielaszek, 1983
South Norway 62 3 11 2 84 Wright, 1977 Denmark 14 86 14 0 0 Rebsdorf, 1980 Nova Scotia 21 71 24 0 5 Watt et al., 1979 Central Ontario 26 12 73 15 0 Scheider et al., 1979 Ontario 600 16a 32b 52 Zimmerman and Harvey, 1979 La Cloche Mountains 4 100 Beamish, 1976 Adirondack Mountains 692 41 25 18 16 Pfeiffer and Festa, 1980 Adirondack Mountains 1455 48e 83d 17 Kretser et al., 1989
Areas where precipitation pH averages > 4.6 Northwest Wisconsin 265 15c 17 68 Lillie and Mason, 1980 North Minnesota 85 0 22 26 52 Glass and Loucks, 1980 a <50 ueq L-1 b 51-200 ueq L-1 c <100 ueq L-1 d <200 ueq L-1 e <40 ueq L-1 Adapted from Haines and Akielaszek, 1983
3.24 3.2 Ion Chemistry2
3.2.1. Specific Methods
We analysed inorganic chemistry from October 1984 (N=805) to examine the geographic variability of the inorganic lake chemistry across six regions within the state. In our analysis, we simply grouped lakes into geographic regions of similar geology and chemistry rather than examining the mechanisms of chemical transport and reaction within the watersheds.
We eliminated brackish lakes from the inorganic chemistry data set by excluding those with sodium or chloride exceeding 3480 or 4225 µeq L-1 (80 or 150 mg L -1 of sodium or chloride, respectively) which was roughly comparable to the conductivity limit of 1500 µS/cm used by the EPA/ELS to eliminate non-freshwater sites.
We tested and found 805 of 1025 lakes with non-missing major cations and anions passed the 15% ionic balance test of Hillman et al. (1986). Lakes that failed the ion balance test were highly colored (median PCU=60). We used the anion deficit in the charge balance as an estimate of organic anions (R-). Some of the failed data had unexplained high or low sodium to chloride ratios (see page 2.19). Only the data from the 805 lakes that passed the ion balance test are presented here. Overall quality control of ionic chemistry determination in our lab was subsequently tested by the use of double blind EPA standard solutions disguised as 14 samples. All ions showed good agreement with expected values except chloride which was biased high by three percent. We will use the inorganic chemistry to describe regional patterns in chemistry, but we cannot extrapolate these results to other lakes in the state with any statistical certainty.
3.2.2 Delineation of Regions
Previous studies have reported on the influence of bedrock on surface water chemistry (Hendrey et al., 1980; Bricker and Rice, 1989). Because of the close correspondence between bedrock geology and surface water chemistry in Massachusetts (Lindhult et al., 1988), we used the state map of bedrock formations (Zen, 1983) to delineate the boundaries between six regions of similar bedrock geology and water quality (Figure 3.2-1). The regions we chose are similar to the geographic distribution of lake alkalinity as shown in Omernik and Powers (1983). They differ more from the recently developed ecoregions of Massachusetts (Griffith et al., 1994), particularly in the eastern part of the state, reflecting the total phosphorus concentrations used to partially delineate the ecoregions.
2 Expanded version of results published in Mattson et al. 1992.
3.25 3.26 3.2.3. Lakes and Ponds
3.2.3.1. Regional Chemistry, October 1984
Median ionic chemistry and other related data for the state are presented in Table 3.2-1. along with the same data subdivided by region within the state. We find our median concentrations are higher for nearly all ions than those reported for areas in the northeast by the EPA/ELS (Brakke et al., 1988). Median values, means and quartiles for various chemical species are shown for the six regions in Figure 3.2-2. The ionic composition of lakes exhibits strong regional patterns across the state. In many cases the median concentration of a given ion is outside the interquartile range of concentrations in adjacent regions (Fig. 3.2.2). Nonparametric median tests (SAS, 1987) show significant differences between regions in the medians of all major ions as well as for pH, ANC, color and estimated organic anions (á<0.0001).
We find a general decline in median pH from west to east. Region 1 (Stockbridge) has a median pH of 7.92 while Region 6 (Cape) had a median pH of only 6.02. The lakes in Region 6 were also the most variable in pH. The regions in the central parts of the state had intermediate pH values.
ANC was highly variable across the state and closely followed the calcium concentrations. ANC was high in Region 1 with a median ANC of 1700 µeq L-1 compared to a median of only 52 µeq L-1 in Region 6. Again, the four central regions showed intermediate ANCs, with Regions 3 and 5 (Connecticut Valley and Boston) having higher ANCs than Regions 2 and 4 (Berkshire and Central Highland).
The sulfate concentration in precipitation exhibits a slight declining trend from west to east with an average of about 48 µeq/L (see Fig. 3, Driscoll, 1991). Our data on lake sulfate does not exhibit any consistent trend from west to east but varied considerably between Regions (Fig. 3.2.2). An unexpected result is the unusual positive association between sulfate and ANC; the three regions with the highest sulfate also have the highest ANCs and a Spearman rank correlation on the raw data show a significant relationship (r=0.51, á<0.0001).
Chloride concentrations are high and variable between regions. The highest median concentration of 633 µeq L-1 is found in the Boston Region; this is tenfold greater than the median of 62 µeq L-1 found in the Berkshire Region in the west. The acid-soluble iron and manganese concentrations are highest in the Connecticut Valley and Central Regions of the state. The acid soluble aluminum concentrations are near or below our limit of detection. The median silicate concentrations are highly variable between regions, with the highest concentrations found in Region 3 and the lowest concentrations found in Region 6 (Table 3.2.1.).
3.27 Table 3. 2-1. Median chemistry for major cations and anions (Oct. 1984), along with physical data for Massachusetts lakes, and data subdivided by region within state. Median nitrate was below the limit of detection. Massachusetts Region
1 2 3 4 5 6
All Stockbridge Berkshire Conn. Central Boston Cape
# with complete 805 25 45 44 276 302 113 chemistry1
Cations (µeq L-1)
Ca++ 273 1270 228 626 226 418 73
Mg++ 136 899 130 266 94 174 119
Na+ 430 227 70 388 340 654 446
K+ 36 31 23 57 39 42 23
Anions (µeq L-1)
ANC 201 1704 240 532 148 306 36
= SO4 167 214 124 238 152 224 123
Cl- 423 285 62 472 302 633 423
RCOO-2 97 28 495 125 94 114 78
pH 6.62 7.88 6.91 6.94 6.52 6.73 6.00 Trace metals3 (µg L-1)
Fe 230 140 180 360 270 250 100
Mn 30 20 20 60 30 30 10
Al 20 BD BD BD 40 30 20
Medians for other parameters:
Color (PCU) 19 9 19 19 19 29 18
SiO2 (mg/L) 2.5 2.5 1.5 4.1 2.6 3.0 0.6
Area (ha) 4.1 3.4 5.7 3.3 4.4 3.4 3.2
Elev. (m) 82 321 433 66 206 49 9
Other Statistics: Percent frequency of:
Drainage4 74 68 90 56 84 40 33
1N may be less than indicated for trace metals and other parameters due to occasional missing data. 2Organic anions (R-) for lakes were calculated from individual sample ion balances, not from ion balances of median statistics. 3Acid soluble metals (see methods). 4Drainage lakes are defined as having a stream outlet on a 7.5 min. USGS map.
3.28 The average ionic composition of lakes within each region is shown in Figure 3.2.3. with the ions drawn to the same scale. From this figure it is clear that the Stockbridge Region has lakes with much greater ionic strength compared to other regions of the state, and the mean ionic strength is nearly five times higher than in the adjacent Berkshire Region. Calcium, magnesium and ANC (presumably as bicarbonate) dominate the ionic composition of the Stockbridge Region. At the opposite extreme, the lakes in the Cape Region are dominated by sodium and chloride with little else. The Cape Region is also unusual in that it is the only region where average magnesium concentrations are higher than the average calcium concentrations (see Fig. 3.2-3). In the eastern regions there appears to be a large influence of sea spray (or saltwater intrusion) near the coast, and the high magnesium to calcium ratios in lakes of the Cape Region can be accounted for by seawater inputs based on chloride concentrations. Lakes in the Berkshire Region have the lowest ionic strength and have little salt compared to the sodium and chloride levels observed in the other regions of the state. The remaining three regions, the Connecticut Region, the Central Region and the Boston Region have moderately high ionic strength with high levels of sodium and chloride. These three regions also have the highest levels of potassium in the state.
3.29 3.2.3.2. Other Factors in Determining Regional Chemistry
As stated above, we used a map of bedrock geology to delineate the regions on the assumption that the bedrock geology is a major influence on the surficial geology and surface water chemistry. Previous research has also shown a strong geologic influence on subsurface water quality (Trombley, 1992). This should not be taken to imply that bedrock geology itself is the sole, or even the major, determinant of lake chemistry in all regions. In Region 1 (Stockbridge) the marble bedrock is obviously the source of the high concentrations of calcium and magnesium and results in high ANC and high pH values. High ANCs are also found in association with sedimentary bedrock in Regions 3 and 5 (Connecticut and Boston), but the correspondence between ANC and bedrock is confounded by other factors such as lower elevations and higher human population densities compared to areas such as the Berkshire and Central Regions (Table 3.2-1). In some parts of the state the influence of the bedrock is probably irrelevant. In Region 6 (Cape) for example, the surface waters are almost certainly isolated from bedrock by glacial sedimentary deposits (Stone, 1982). The low concentrations of calcium and the high frequency of acidic lakes there are probably a result of the high frequency of
3500.0
3000.0
2500.0
2000.0
1500.0
1000.0
500.0
0.0 Stockbridge Berkshire Connecticut Central Boston Cape Region Na+ Ca++ Mg++ K+ Cl- SO4= HCO3- R-
Figure 3.2-3. Ionic balance for lakes within six regions (October 1984 mean data). The left and right side of each pair of bars represents cations and anions, respectively. Titration ANC is - - represented here as HCO3 ; R is calculated by difference in the charge balance and is assumed to be organic acid anions.
3.30 seepage lakes (Table 3.2-1), the presence of organic acids, and the slow weathering rates of the sandy soils.
We currently do not have the detailed data on soils, vegetation, and hydrologic flow-paths for each lake necessary to conduct an examination of the mechanisms underlying the differences in lake chemistry (Eilers et al., 1983; Driscoll and Newton, 1985; Gherini et al. 1985). Lacking such specific data, our study demonstrates that regional analysis is a simple, flexible, and powerful classification system which incorporates the combined effects of a variety of environmental factors such as climate, vegetation, soils, and bedrock as well as land use (Bailey, 1983).
3.2.3.3 Road-Salt Influence
We compared our chloride data to the sea-salt chloride: distance-from-coast relationship developed for the Northeastern region lakes (Appendix B of Baker et al., 1990) and found that over 98% of our lakes exceeded the predicted chloride levels. The higher than expected chlorides are probably due to both the peninsular nature of the coastal region and high levels of road salt applied in this state (Mass. D.P.W., 1989).
The six regions exhibit perfect rank correlation (á<0.01) between median chloride and population density (Table 3.2-1), and this positive relationship suggests that the high levels of sodium and chloride found in the inland lakes are probably due to anthropogenic inputs such as road salt. The amount of salt applied to roads varies depending on snowfall, but can be as high as 230,000 metric tons per year on state roads alone, and has had a measurable impact on public water supplies (Mass.D.P.W., 1989). Proximity to roads has been previously identified as an important factor associated with salt impacted lakes (Driscoll et al. 1991). Some researchers have suggested neutral sea salts can cause temporary, but not long-term acidification of freshwaters, since the soil solution is assumed to be at steady state with respect to cation exchange reactions (Sullivan et al., 1988; Wright et al., 1988). Road salts have been suggested as a possible acidifying agent in Massachusetts by assuming soils adjacent to roads are not in steady state with respect to base cation inputs (Sullivan, 1990). We do not believe long-term acidification by road salts is likely. It is logical to assume that if the cation exchange reactions are equilibrium reactions, then hydrogen should be re-exchanged for sodium after the event and thus no long term acidification should occur. This may explain experimental data of Wright et al. (1988) which show that a temporary drop in ANC from 20 to -2 µeq L-1 during a salt addition was reversed later and ANC temporarily increased to 27 µeq L-1. Additional experiments are needed to verify the effect of variation in salt input on ANC export from watersheds over a variety of time scales. Further analysis of road salt impacts may be found in section 3.3.
3.2.3.4 Sulfate Sources
The high concentrations of sulfate in the Stockbridge, Connecticut and Boston Regions are difficult to explain. Based on an evapoconcentration factor of two and average precipitation concentrations, we would expect to see sulfate concentrations of about 96 µeq/L in the lakes. These three regions exceed this estimate by more than a factor of two (see Table 3.2-1) which would imply either unusually large unmeasured dry deposition rates, a terrestrial source of sulfate, or much higher evapoconcentration factors.
Local variations in dry deposition of sulfur may account for some of the variation in sulfate concentrations between regions. Typically, SO2 accounts for 75% of dry deposited sulfur (Sisterson,
3.31 et al., 1990), and emissions of SO2 are concentrated in industrial and power generating localities (Placet et al., 1990). Unfortunately, little information is available on local variations in dry deposition.
Much of the interregional differences in chemistry appear to be simply differences in overall 3500.0
3000.0
2500.0
2000.0
1500.0
1000.0
500.0
0.0 Stockbridge Berkshire Connecticut Central Boston Cape Region Na+ Ca++ Mg++ K+ Cl- SO4= HCO3- R-
Figure 3.2-4. Ionic balance for streams within the six regions (October mean data). The left and right side of each pair of bars represents cations and - - anions, respectively. Titration ANC is represented here as HCO3 ; R is calculated by difference in the charge balance and is assumed to be organic acid anions. ionic strength, suggesting that differential evapoconcentration may occur between regions. One way to estimate evapoconcentration is to use non-marine chloride as a conservative tracer. This method is difficult to apply in Massachusetts because of the large amounts of road-salting (Kahl et al., 1991). Instead, we estimated evapoconcentration factors from maps of precipitation and runoff. The resulting factors did not agree with the pattern of ionic strength seen between regions. We suspect this method also does not work well in Massachusetts because of the high frequency of seepage hydrology, and the method also does not account for hydraulic flushing rates of individual lakes. A more detailed analysis of hydrology is required to explain the large differences in ionic strength of lakes across the state.
The oxidation of sulfidic schists to sulfates has been hypothesized as a possible acidifying agent in the state (Parnell, 1983). The Central Region of the state has many acidic lakes and this region has significant amounts of sulfidic schists in the bedrock, whereas the other regions have little sulfidic schists. However, the Central Region has relatively low sulfate concentrations, indicating that bedrock sulfide oxidation is not the cause of the regional lake acidification. The lakes in this region are acidic because of relatively low base cations, not because of abnormally high sulfate concentrations (Figure 3.2-3). The non-coastal acidic lakes in the state are generally located in the high elevations of the Berkshire and Central Regions. In these regions there is an abundance of slow weathering granite, quartz and gneiss in the bedrock which probably accounts for the low base cation concentrations and the lower median ANCs. Another possibility to explain the high sulfates would be the anthropogenic
3.32 input of sulfates from industry, agriculture and domestic wastewater. The differences in sulfate concentrations between regions are probably due to a variety of factors such as sulfate adsorption in the soils (Fuller et al., 1985) or differential evapoconcentration (Sullivan, 1990).
The correlation of higher sulfate concentration with high ANC does not rule out sulfuric acids as an acidifying agent. The lakes in the state have relatively high sulfate concentrations and the major acidifying agent in the state is probably acidic deposition of sulfuric acid. The fraction of lakes where organic anions exceeds sulfate concentrations is only 23%, about the same fraction found in acidic lakes (20%). All of the organic dominant acidic lakes were located in the Cape Region.
3.2.4. Streams and Rivers
While streams display essentially the same pattern as lakes, some differences are evident. Figure 3.2-4 displays mean values for the six regions and Table 3.2-2 shows median values. Total ionic concentration is somewhat higher in all streams than for lakes, with the exception of the Connecticut region where there is little difference. The Cape region streams do not follow the pattern of higher magnesium than calcium found for lakes in that region. More specific differences between lakes and streams statewide and within specific regions are shown for median values in Figures 3.2-5 to 3.2-7. Statewide, streams are higher than lakes in calcium, magnesium, potassium, sulfate, aluminum and silica dioxide and are lower in sodium, chloride, organic acids, iron, and manganese. However, in the Stockbridge, Berkshire, and Boston regions, chloride is higher in streams than lakes; but lakes are higher in iron. In the Connecticut region, calcium, magnesium, potassium are lower in
3000 Stockbridge 2500
2000 Connecticut Boston 1500 Berkshire All 1000 Central Cape
500
0 L S L S L S L S L S L S L S
Ca Mg Na K
Figure 3.2-5. Comparison of cations in lakes versus streams for the six regions (October 1984 median data).
3.33 Table 3. 2-2. Median chemistry for major cations and anions (Oct. 1984), along with physical data for Massachusetts streams, and data subdivided by region within state. Median nitrate was below the limit of detection. Massachusetts Region 1 2 3 4 5 6 All Stockbridge Berkshire Connecticut Central Boston Cape Cod # with complete chemistry1 448 16 99 40 171 104 18
Cations (µeq L-1) Ca++ 383 1488 318 616 267 569 181 Mg++ 168 768 137 240 120 251 142 Na+ 320 177 108 357 308 730 481 K+ 41 38 28 43 43 62 28
Anions (µeq L-1) ANC 291 2561 314 508 196 355 114
= SO4 180 184 143 271 171 295 169
- 2 2 2 2 2 2 2 NO3 BD BD BD BD BD BD BD Cl- 322 346 80 384 269 702 378 RCOO-2 88 2 59 106 88 176 120
pH 6.68 7.56 6.83 6.94 6.40 6.60 5.95
Trace Metals (µg L-1))3 Fe 200 40 70 150 220 335 555 Mn 20 10 10 20 30 40 50 Al 40 0 0 0 60 70 80
Other Parameters Color (PCU) 19 9 19 19 19 39 29
SiO2 (mg/L) 10.6 6.1 8.0 12.0 11.4 11.3 9.1
Population per km2 269 98 29 398 162 573 200 1N may be less than indicated for trace metals and other parameters due to occasional missing data. 2Organic anions (R-) for lakes were calculated from individual sample ion balances, not from ion balances of median statistics. 3Acid soluble metals (see methods).
3.34 800 Cape 700
600 Connecticut 500 Boston Berkshire 400 All Central 300 Stockbridge 200 100 0 L S L S L S L S L S L S L S
Fe Mn Al
Figure 3.2-6. Comparison of Iron, Manganese and Aluminum for lakes versus streams in the six regions (October 1984 median data).
50 14 Connecticut Central Boston 12 40 All Berkshire Cape 10 30 Stockbridge 8
20 6 4 10 2
0 0 L S L S L S L S L S L S L S
Color SiO2
Figure 3.2-7. Comparison of Color (solid) and SiO2 (hatched) for lakes versus streams in the six regions (October median data).
3.35 streams than lakes and, in the Stockbridge region, magnesium and sulfate are lower. The Stockbridge region also had lower sulfate levels in streams than lakes. The Boston region streams, unlike most of the rest of the state, had streams with higher sodium, chloride, iron, manganese and organic acids than lakes. In the Boston and Cape regions, color and iron are higher in streams than lakes. Color was higher in Boston and Cape streams than lakes. Silica was substantially higher in all streams compared to lakes.
3.2.5. Comparison With Other Surveys
Compared to international sites reported in Baker et al. (1990), as shown in Table 3.2-3, Massachusetts lakes and streams are comparable in sulfate to water bodies in eastern Canada, and high deposition areas of Finland and more than three times as high as low deposition sites in the U.S. and Europe (Figure 3.2-8). In similar fashion, base cations in Massachusetts lakes and streams are approximately 100% higher than other areas (Figure 3.2-9). Massachusetts lakes and streams are also relatively high in organic acids, comparable to water bodies in Minnesota, northern Ontario, high deposition areas of Finland and Tasmania (Figure 3.2-10). Water bodies in Norway are characterized by low base cations, low organic acids, and moderate sulfate levels. Regions of high deposition in Norway have nearly 80% of surface waters with no ANC. Massachusetts and other sites in North America have higher sulfate levels but also have higher base cations and organic acids. Massachusetts has one of the highest values for organic acids and the highest base cations of the average for sites reported by Baker et al. (1990), so the combination of these features, that can buffer waters against acid deposition, may account for the smaller percentage of surface waters that are acidified. In section 4, we will discuss the trends in sulfate and base cation levels from 1984 to 1993.
3.36 Table 3.2-3. Comparison of ARM data with areas of high and low deposition throughout the world. N = Number
of sites surveyed, CB = total base cations in µeq/L, SO4 = sulfate in µeq/L, A- = organic acids, and ANC as the percent < 0 µeq/L.
UNITED STATES N CB SO4 A- %ANC< 0 High Deposition Areas Lakes Massachusetts All 805 945 167 97 4.7 Stockbridge 25 2976 203 28 0 Berkshire 45 442 124 495 2.8 Connecticut River 44 1495 238 125 0 Central 276 690 152 944 4.0 Boston 302 1374 224 114 0.3 Cape 113 669 123 78 22.1 Adirondacks 124 238 118 37 17.2 New England 310 533 136 38 5.8 Interior SE 89 330 33 19 0.0 Streams Massachusetts All 448 987 180 88 2.0 Stockbridge 16 2687 184 2 0 Berkshire 99 608 143 59 0 Connecticut River 40 1335 271 106 0 Central 171 750 171 88 3.5 Boston 104 1709 295 176 1.0 Cape 18 780 169 120 11.1 Mid-Atlantic Coastal 57 619 136 41 11.8 Plain Interior Mid-Atlantic 105 548 208 11 5.3 Northeast 56 546 192 19 6.2 Low Deposition Areas Minnesota (2A+2B) 177 471 50 102 0 Sierra (4A) 149 68 6 10 0 Cascades (4B) 150 140 11 13 0 Northern Rockies (4C) 143 254 16 12 0
3.37 Central Rockies (4D) 129 154 25 12 <1 Southern Rockies (4E) 139 389 35 15 0 Kenai Peninsula 47 123 0 66 0
CANADA High Deposition N. Ontario 22 520 110 52 0 SE Ontario 966 192 150 39 5 Ottawa Valley 83 242 123 47 0 N. Quebec 24 132 82 54 0 SW Quebec 466 226 150 40 1 Laurentide 340 143 85 44 <1 S. Atlantic 32 616 124 56 0 Sudbury 162 198 205 31 28 Low Deposition West Ontario 262 269 69 49 0 N. Ontario 55 594 25 104 0 North Atlantic 456 83 32 54 3 N. Quebec 14 115 41 30 0
NORWAY COUNTIES High Deposition Areas Counties 8-10 397 47 67 17 79.8 Counties 11 88 31 53 6 77.3 Counties 20 34 123 79 17 2.9 Low Deposition Area Counties 15-19 129 44 20 13 14.7
FINLAND High Deposition 304 354 140 105 7.8 Low Deposition 120 186 37 52 3.2
TASMANIA 170 429 47 103 -- (Low Deposition)
3.38 Mass. Lakes Adirondack New England Interior SE Mass. Streams Mid-Atlantic Coastal Plain Interior Med-Atlantic Northeast
Minnesota Sierra Cascades Northern Rockies Central Rockies Southern Rockies Kenai Peninsula N. Ontario SE Ontario Ottawa Valley N. Quebec SW Quebec Laurentide S. Atlantic Sudbury West Ontario N. Ontario North Atlantic N. Quebec Norway 8-10 Norway 11 Norway 20 Norway 15-19 Finland - High Dep. Finland - Low Dep. Tasmania 0 50 100 150 200 250 ueq/L
Figure 3.2-8. Comparison of Massachusetts lakes and streams sulfate data versus other sites as reported in Baker et al. (1990).
Mass. Lakes Adirondack New England Interior SE Mass. Streams Mid-Atlantic Coastal Plain Interior Med-Atlantic Northeast
Minnesota Sierra Cascades Northern Rockies Central Rockies Southern Rockies Kenai Peninsula N. Ontario SE Ontario Ottawa Valley N. Quebec SW Quebec Laurentide S. Atlantic Sudbury West Ontario N. Ontario North Atlantic N. Quebec Norway 8-10 Norway 11 Norway 20 Norway 15-19 Finland - High Dep. Finland - Low Dep. Tasmania 0 200 400 600 800 1000 1200 ueq/L
Figure 3.2-9. Comparison of Massachusetts lakes and streams base cation data versus other sites as reported by Baker et al. (1990).
3.39 Mass. Lakes Adirondack New England Interior SE Mass. Streams Mid-Atlantic Coastal Plain Interior Med-Atlantic Northeast
Minnesota Sierra Cascades Northern Rockies Central Rockies Southern Rockies Kenai Peninsula N. Ontario SE Ontario Ottawa Valley N. Quebec SW Quebec Laurentide S. Atlantic Sudbury West Ontario N. Ontario North Atlantic N. Quebec Norway 8-10 Norway 11 Norway 20 Norway 15-19 Finland - High Dep. Finland - Low Dep. Tasmania 0 20 40 60 80 100 120 ueq/L Figure 3.2-10. Comparison of Massachusetts lakes and streams organic acid data versus other sites as reported by Baker et al. (1990).
3.40 3.3 Identification of Road Salt Contamination Using Multiple Regression and GIS3
3.3.1 Introduction
The annual use of common salt (sodium chloride or NaCl) as a deicing agent for roads in the northern sections of the country has increased dramatically since 1950 when less than one million tons of salt were applied in the United States. The application rate has stabilized at about 10 million tons per year since 1970 (NRC, 1991). Massachusetts, in particular, applies large amounts of salt to the state's roads. The application rate on state roads in recent years (1984-1988), averaged about 11300 kg per lane kilometer per year (20 tons per lane mile) or a total of 223 million kg per year (245,000 tons) on state roads alone (Mass. DPW, 1989). The application rate for town roads and streets is unknown because the local agencies do not report it.
After the salt dissolves on the road the brine enters the adjacent soils, groundwaters and streams. In general, salt is not very toxic to aquatic animals. Crowther and Hynes (1977) reported that the behavior of aquatic insects was not affected by chloride concentrations below 1000 mg/L. Invertebrate numbers and diversity were lower in streams receiving drainage from salted and sanded roads, although the results suggested that increased fine sediments from sanding may be more important than the salt application (Molles, 1980; Demers, 1992) Unlike animals, plants are more susceptible to salt toxicity, and road salt can cause severe damage to roadside vegetation (Gidley, 1990; Leiser and John, 1990). Other plants naturally found near coastal areas and salt marshes are salt tolerant and may replace the native vegetation of inland wetlands which receive road salt (see review in Mass. DPW, 1989). Another potential problem with salt contamination involves the chemical stratification of small lakes receiving high amounts of road salt (Judd, 1970; Hoffman et al., 1981). This could lead to extended stratification, anoxic conditions, and possibly to fish kills.
Calabrese and Tuthill (1980) reported that high levels of sodium correlate with hypertension in school children. While there is currently no regulation of sodium in drinking water, concentrations of less than 20 mg/L are desirable. In 1986 there were 73 public water supplies in Massachusetts exceeding this concentration (Pollock, 1988). Nationally about $10 million is spend to mitigate salt impacts on drinking water, with most of the effort concentrated in the Northeast and Midwestern states (NRC, 1991).
Previous researchers have suspected that road salt is the major source of salt to groundwaters in the state, and road salting has been implicated as the source of salt in some high salinity well waters (Huling and Hollocher, 1972; Pollock, 1988). Computerized GIS systems now provide access to large amounts of land use data for nonpoint pollution studies (Khan and Liang, 1989). A recent study suggested multiple regression as a tool for policy makers to identify pollution loading rates from various land uses (Harper et al., 1992).
This study examines the salt concentration of a number of randomly selected streams in the state and attempts to determine the relative magnitude of salt inputs from road deicing as compared to precipitation as a source of the salt to the streams. Our general approach is to use the ARC/INFO GIS to link the chemistry of the streams with stream-specific salt loading estimated from associated land-use data (e.g., length and type of roads) and precipitation inputs. The data are then analyzed 3 Expanded version of results published in Mattson et al., 1994.
3.41 by several techniques. First we examine the observed ratio of sodium to chloride in the samples and compare them to ratios expected from different sources and perform a simple regression for atmospheric deposition. Second we construct a conceptual mass-balance model based on a priori knowledge of salting rates to predict stream sodium based on length of roads within the watershed. And finally we use multiple regression techniques to obtain empirical estimates of the magnitude of several sources of sodium to the streams. We used the SAS statistical program (SAS, 1987) to analyze the model for problems commonly encountered in environmental studies: outliers, stability, lack of fit, heteroscedasticity and collinearity. Finally we discuss potential problems in model construction and analysis and how they relate to policy decision making.
3.3.2 Methods
The chemical analyses for the Acid Rain Monitoring Project have included the concentrations of both sodium and chloride. Stream samples are collected on the upstream side of roads when possible to avoid local road salt contamination. We restricted our study to 'freshwater' sites by limiting our data to samples with concentrations less than 80 mg L-1 or 150 mg/L for sodium and chloride, respectively. Sodium shows relatively little seasonal variation but do show some dilution effect with increased runoff. Data collected quarterly from 1986 to 1993 show a mean of 11.3 mg/L with a standard deviation between seasons of 1.8 mg/L. The samples used in this analysis were collected on April 5, 1992 from 162 randomly selected streams whose watersheds fall within the state's borders.
We used the computer program ARC/INFO to digitize watershed boundaries for those streams whose watersheds lie within the state's borders (Figure 3.3-1). We then determined the length of roads of each given road classification (class 1 = Interstate highways, class 2 = major state roads, classes 3 and 4 are small roads and streets) from the digital roads coverage. Digital data on roads were supplied by the Massachusetts Executive Office of Environmental Affairs. We also digitized existing maps of stream runoff (Krug et al., 1990) and overlaid the map on the watershed map to determine total annual stream runoff in liters.
Sea-spray can add salts to otherwise dilute rainfall. Salt input to each watershed from combined sea-spray and precipitation was determined from a logarithmic relationship between distance from the coast and sodium deposition rates at four National Atmospheric Deposition Program monitoring sites (Na kg/ha = exp(2.26-0.022*km) R^2=0.92).
3.42 3.3.3 Results
3.3.3.1 Sea-spray Precipitation Inputs
The sodium concentration in the streams showed strong regional patterns with high concentrations of sodium in the eastern half of the state (Figure 3.3-2), but the highest concentrations were generally found in urban areas. Sodium and chloride showed a strong linear relationship (Na = 0.57+0.57xCl N=162 R^2=0.99). The slope of the regression line (0.57) is closer to that expected from diluted seawater (0.56) rather than the mass ratio of Na to Cl in pure salt (0.65).
The results of a simple regression based on deposition (calculated from distance from the coast, as above) reveal a highly significant relationship (Na mg/L = 4.0 + 15.0*SEA, N=162, R^2=0.28). The slope of the simple regression was 15, but was expected to be unity if the model was correct. Based on mass balance, the average expected sodium concentration in the streams from sea-spray and precipitation (SEA) was only 0.42 mg/L compared to the observed average sodium concentration of 10.31 mg/L. Thus, our mass balance calculations indicate sea-spray and
3.43 precipitation can account for only about 4 percent of the sodium in the streams, and we can therefore reject seawater as a significant source of salt to the average stream in Massachusetts.
3.3.3.2 Inputs from Roads
How does road salt compare to sea-spray as a source of salt to streams? We calculated expected sodium concentrations assuming the reported salting rate of 11,300 kg per lane km (Mass. DPW 1989) was the sole source of sodium to the streams. We further assumed that interstate highways had 4 lanes salted, major state roads had 2 lanes salted, and the smaller town roads and streets had the equivalent of 1 lane salted. The observed vs predicted sodium concentrations based on road salt loading are shown in Figure 3.3-3 with the best fit regression line of Na mg L-1= -1.46 + 0.66* Predicted Na mg L-1. These results show that even a simplistic mass balance model can explain 63 percent of the variation in sodium in the state's streams. The observed data (Figure 3.3.3) tend to fall below the expected 1:1 line, indicating that our assumed loading estimates were high (by about 50 percent), but the calculations certainly support the idea that road salt can supply the mass of salt implied by the observed concentrations of sodium in streams.
3.44 3.3.3.3 Combined Multiple Regression Model
We then combined our data on sodium deposition and the length of the four classes of roads in each watershed in a multiple regression model to obtain simultaneous empirical estimates of loading rates from both types of sources. In this case we made no assumptions about salting rates or lanes salted. The model equation was:
Na mg/L = SEA + C1 + C2 + C3 + C4 with the parameters defined in Table 3.3-1.
The results of the multiple regression analysis show that the model was highly significant (á<0.0001), and that sea-spray as well as each of the four classes of roads were significant in the model (Table 3.3.1). The coefficient for SEA (5.4) is still higher than the expected value of 1. The coefficients listed in Table 1 for the four classes of roads represent the model estimates of the metric tons of sodium applied per linear kilometer of road. Converting the units to kilograms of salt (NaCl), we obtain 22500, 17700, 4380 and 7530 kilograms for each of the four classes of roads, (or 5625, 8850, 4380 and 7530 kg per assumed lane km per year, respectively). These multiple regression estimates range from 39 to 78 percent of the state reported average application rate of 11,300 kg per lane km per year.
3.45 3.3.4 Discussion 3.3.4.1. General
Our mass balance model suggests roads are the likely source of sodium in stream water and the results agree with mass balance calculations based on reported application rates. In addition, we can say the multiple regression analysis indicates that the class 1 and 2 roads are associated with heavy salt use, while classes 3 and 4 (typically small town roads and streets) have total salt loads three to four times lower than the others. These results are not unexpected since the major roads (class 1 and 2) are known to be heavily salted and have multiple lanes. The class 4 roads (city streets) are associated with urban areas, sidewalks, industry and parking lots which may also be significant sources of salt, and thus some of the salt we have attributed to streets is probably due to these associated sources. Due to the lower amount of traffic and the smaller financial budgets of towns, the average salt application on class three roads appears to be small. Although the vast majority of roads in the state are class 3 and 4, our results show that streams with very high sodium are associated with class 1 and class 2 roads. Of the 25 streams that exceed the recommended sodium concentration of 20 mg/L, all 25 are associated with either a class 1 or 2 road, a result which was highly significant (Chi square = 9.71, 1 df, á<0.005). Such a relationship was not observed for the more common class 3 and class 4 roads.
TABLE 3.3-1. Descriptive Statistics and Multiple Regression Results.
Descriptive Statistics Variable: Definition Percent >0 Median Min. Max. Na: Stream Na conc., mg/L 100 5.32 0.82 66.37 SEA: Pred. Na conc., mg/L 100 0.27 0.03 2.04 C1: Class 1 Loading, km/L*1E9 25 0 0 2.85 C2: Class 2 Loading, km/L*1E9 44 0 0 2.53 C3: Class 3 Loading, km/L*1E9 92 1.58 0 7.23 C4: Class 4 Loading, km/L*1E9 87 0.63 0 12.82 ______ANOVA Source df Sum of squares F Pr>F Model 5 15100. 68. 0.0001 Error 156 6950. C. total 161 22050. (R^2=0.68; Adj. R^2=0.67) ______Model Results Source Coefficient. SE T Pr>T Tol. Intercept -1.76 1.03 -1.7 0.0887 ----- SEA 5.44 1.43 3.8 0.0002 0.77 C1 8.85 1.50 5.9 0.0001 0.92 C2 6.94 1.21 5.7 0.0001 0.96 C3 1.72 0.45 3.8 0.0002 0.96 C4 2.96 0.29 10.1 0.0001 0.80 ______
3.46 Previous studies have also revealed significant multiple regressions of various types of land- use on salt contamination of drinking well water. Noss (1989) reported that the area of three land- use types, open land, industrial land and low density development land, could predict about 50 percent of the variation in sodium concentrations of well water. The conclusions of Noss (1989) are invalid however, because Noss (1989) did not regress the land use areas on sodium concentration, but rather on sodium concentration weighted by the recharge area of each well. Because his regression model used an area weighted variable on both sides of the equation, his regression results cannot be used to infer a relationship between land uses and sodium.
The results of Harper et al. (1992) generally agree with ours. In their analysis they found three factors were significant in predicting sodium concentration in wells: the length of roads in the buffer zone, medium density residential area, and commercial land-use area. Six other factors in the model were not significant, but the model was able to explain 28.9 percent of the variation. Harper et al. (1992) estimated that adding a mile of road within a one-half mile radius buffer zone should increase sodium concentrations in the well by about 2.5 mg/L. Assuming sodium behaves conservatively and is diluted by an average of 60 cm of runoff water per year, their model implies that 7800 kg of salt were applied per km of road each year. This is a reasonable estimate based on reported salting rates and roughly agrees with our results discussed above. The portion of salts carried by surface vs. groundwaters are likely to vary across the state.
In our data analysis we have implicitly assumed that there are no other sources of sodium and chloride to streams. Data from an undisturbed forested watershed in New Hampshire indicate chloride outputs in stream water generally balance precipitation inputs. Sodium outputs exceed inputs, presumably due to weathering reactions, yet the loading rate was estimated at only 5.6 kg Na per hectare (Likens et al., 1977). Such rates of weathering could explain less than 10 percent of the observed sodium concentrations in our data.
Anthropogenic inputs such as sidewalk deicing, sewage, industry and other sources all contribute to the salt contamination of streams but these sources are likely to be small on average. Highway deicing is the single largest use of salt in the United States. Data provided by the salt industry show that deicing accounts for 42.7%, while the chemical industry, water conditioning and agriculture account for 16.7, 8.4 and 8.1 percent, respectively (Dickinson, 1983). To separate the effects of these other land-use factors from each other would likely require intensive site-specific studies and perhaps experimentation (e.g., limit road salting in an area for a given set of time and observe the result on stream sodium). The strong linear relationship between sodium and chloride suggest salt as the common source, and that other non-salt sources of sodium or chloride related to weathering of rocks, fertilizer application or various other land uses are much smaller in comparison.
The decision to restrict the use of salt must weigh the societal benefits against the potential for adverse health and environmental effects (Mass. DPW, 1989). In general, salt is not a very toxic substance to most aquatic organisms. And the levels that we observed in Massachusetts streams are probably not high enough to cause significant damage, although concentrations would probably be much higher along roadside ditches and ponds. In local areas where high levels of salt are found in the groundwater it has been found that the application of salt substitutes such as calcium magnesium acetate can reduce the sodium concentration in nearby wells (Pollock, 1990).
3.47 3.3.4.2 Model reliability
Although the model and all factors are significant, how sure are we that the model is reliable? After all, the coefficient for SEA has changed markedly from 15 in our first regression to 5.4 in the current model, and we expected the 'true' value to be one. This provides an example of an unstable regression coefficient due to collinearity between the regressors. While multiple regression can still obtain a good fit to the data with collinearity, the coefficients can be very imprecise and lose their meaning (Neter and Wasserman, 1974). This problem can be severe with percentile data if the land uses under study tend to sum to nearly 100 percent.
An examination of correlation coefficients between the regressors indicates significant positive correlations between SEA and roads of class 1, 2 and 4, as well as between roads of class 1 and 4. The correlations range from r=0.17 to 0.42. We examined the tolerance of each regressor to see how stable the model is to the effects of collinearity. The tolerance can be described as the fraction of 'new' information in the regressor after the combined effects of the other regressors are factored out. The regressors in our model have relatively high tolerance (Table 3.3-1) to collinearity and after further examination we felt the coefficients were stable enough to demonstrate the general effects of road salt on streams. A more serious problem with the multiple regression approach concerns model specification. As in any study of this type, we may have left out a significant source of sodium in our model, and this could change both the magnitude and significance level of the loading coefficients.
Other problems may not be as obvious. For example we checked residual plots and found no problems with lack of fit, but we did see some evidence of increased variance in the residuals as the predicted sodium increased. While such heteroscedasticity will ordinarily not bias our estimates of the coefficients, it can reduce the significance level and our confidence in the estimates of coefficients. Rather than use a data transformation to 'fix' the problem, we chose to keep the relationship linear as theory suggests, and we recognize that our confidence in the significance of the coefficients is somewhat less than indicated in Table 3.3-1.
We used a jackknife technique called PRESS, the predicted sums of squares, computed by sequentially omitting each observation from the data set. In terms of sodium mg/L, the average estimated error was 6.7 mg/L with the model compared to 7.8 mg/L for the average error predicted from PRESS, which indicates that 'outliers' in the data do not unduly influence the model estimates.
While the statistical tests discussed above are not intended to be inclusive, we feel they represent some of the more common problems encountered in the analysis of environmental data. Based on our results, we can not suggest multiple regression as a tool to set policy favoring one land use over another (Harper et al., 1992) unless rigorous validation of the models are performed and reported.
3.48 3.3.5 Conclusions
Our data indicate that most of the sodium and chloride in Massachusetts streams is due to road salt application. A simple mass balance model of road salt and stream runoff can account for more than half of the observed variation and all of the mass of sodium in streams, and conversely salt in precipitation can account for only about 4 percent of the observed stream concentrations. A multiple regression model of sodium inputs to streams indicates interstate and major state roads are large sources of salt to streams, but the accuracy of the estimated loading rates is still in question. Problems such as collinearity, heteroscedasticy and model specification may reduce our confidence in the coefficients estimated by the multiple regression approach.
Computer geographic information systems such as ARC/INFO can easily extract large amounts of land-use data for environmental modeling. While multiple regression can be a powerful tool in such models, caution should be used in the interpretation of the results. High significance levels are not sufficient; models must also be tested for stability and validated before they are used to interpret the effects of land-use on environmental quality or to set policy.
3.49 4.0. Trend Analysis - Phase III 4.1. Introduction
Several lines of evidence including paleolimnologic reconstruction, historic data and modeling studies have all indicated acidification of numerous lakes in the northeastern United States within the last century (Sullivan et al., 1990a; Asbury et al., 1989; Wright et al., 1986; NRC, 1986). Fewer studies have reported on historic trends in alkalinity (or Acid Neutralizing Capacity, ANC) of streams and rivers (Stoddard, 1991). With the passage of the amendments to the Clean Air Act in 1990 and the recently observed declines in sulfate, base cation, and hydrogen ion deposition (Hedin et al., 1987; Hedin et al., 1994; Likens, 1989), there is a great interest in detecting the presence of corresponding recent trends in the acid-base chemistry of surface waters. It is important to distinguish between “acidified systems” and the process of “acidification.” Note that the “acidification” process is not necessarily restricted to those systems with acid neutralizing capacity less than zero (ANC<0). Here, as in previous studies, we consider any reductions in ANC or pH by strong acid inputs to be evidence of acidification.
Analyses of recent acid-base trends in lakes have suffered from a lack of statistical power to detect trends in the short term data sets used (Driscoll and Van Dreason, 1993; Stoddard and Kellogg, 1993; Webster et al., 1993). Significant trends have been observed in stream surveys, but the studies often reported conflicting trends between measures of acid neutralizing capacity (ANC) and other parameters (Smith et al., 1987; Lettenmaier et al., 1991; Murdoch and Stoddard, 1993). We used meta-analysis to combine results of trend analyses from 330 Massachusetts' streams into a single, powerful test for trend and a similar trend analysis of 181 lakes in the state. Our results show strong evidence for improvements in the pH and ANC over the past ten years, and the rate of recovery in lakes and streams parallels the observed improvements in acid deposition inputs. Trends in pH and ANC in the lakes and streams are driven largely by significant declines in sulfate.
4.2. Methods 4.2.1 Sample Selection
Our analyses are based on 10 years of data collected by the Acid Rain Monitoring (ARM) Project (Mattson et al., 1992; Walk et al., 1992) using quarterly samples between April 1983 and July, 1993 from 330 streams and 181 lakes throughout the state of Massachusetts. Although more data were available, we limited our analysis to streams and lakes which had a minimum of 20 (average 29) non-missing values for both pH and ANC and non-significant serial correlations. To minimize bias in the data we used quality control samples and consistent methods and instrumentation throughout the 10 year study. All data used in the analysis passed our quality control checks as described in Section 2.0 and Mattson et al. (1994), Mattson et al. (1992) and Walk et al. (1992). The 330 streams include 219 streams selected from a stratified random design and 111 special interest streams with a significant sport fishery. The available data on 338 lakes were analyzed for serial correlation in raw and detrended data for pH, alkalinity, sulfate and base cations. Lakes with significant serial correlation were removed leaving 181 lakes which included 112 random lakes and 69 special interest lakes (lakes with significant fisheries or with active lake associations). Serial correlations were small in the final data set for both quarterly and yearly lagged lake data.
4.1 Only the yearly lagged serial correlations would effect the assumptions of the ANOCHIS procedure as used here. Table 4.2-1 shows that these correlations are small and are ignored in subsequent analyses. As both random and special interest groups showed significant and similar trends, the results are combined here. Both streams and lakes showed significant and similar trends.
Table 4.2-1. Serial correlation for 181 lakes after removing seasonal means.
Variable Lag Correlation
Raw ANC Yearly 0.010
Residual ANC Yearly 0.001
Raw ñH Yearly 0.042
Residual ñH Yearly -0.029
4.2.2 Statistical Analysis
The statistical methods consist of three parts: 1) detrending the raw data for hydrologic effects by nonlinear regression against runoff; 2) temporal trend detection for each chemical variable for both raw data and detrended data; and 3) meta-analysis to combine statistical results from many individual streams or lakes into state-wide trends. By analyzing trends in both raw data and detrended data, we have evaluated how changes in runoff over the ten years may have affected our conclusions.
4.2.2.1 Detrending for Runoff
To adjust the data for effects due to stream runoff, the raw data were detrended by regression against the estimated runoff on the day of collection, expressed in millimeters per year. The runoff data were from reported values and direct readings (U.S.G.S., 1985-1994; Mattson, unpublished data) at the nearest of seven selected U.S.G.S. gaging stations (station numbers 01171300, 01331500, 01162000, 01101000, 01110000, 01109070 and 01185500). We used a stream water chemistry model (Johnson et al., 1969) to relate ANC to flow with a bi-weight robust, non-linear regression (SAS, 1987). The residuals from this non-linear model (hereafter referred to as residual ANC) were used in the trend analysis described below. As an additional check on the detrending method, we compared the trend results of residual ANC data generated from the Johnson et al. (1969) model to trend results of residuals from simple linear regression on log transformed runoff. The trend results, slopes, and conclusions of the two methods were nearly identical, indicating the choice of the hydrochemical model used in the detrending process is not critical for this analysis. The trend results from the Johnson et al. model are presented here. An example of the detrending process for streams, including the raw data, the runoff data and the residual ANC is shown for Cronin Brook in Figure 4.2-1. In this case, the trend in ANC is obscured both by seasonal and year to year variability in runoff, while the residual data show a significant increase in ANC for this stream.
4.2 The detrending process for lakes is shown for Ramshorn Pond in Figure 4.2-2. In this case the seasonal variation in alkalinity is less than in typical streams and the relationship of alkalinity to runoff is not as strong. The two low outliers (38 µeq/L and 55 µeq/L) shown in Figure 4.2-2b are both samples taken in the month of January when the alkalinity of samples taken just under the ice may be very low and not strongly related to surface water runoff rates. The resulting residuals are shown in Figure 4.2-2c, but in this case there is no obvious improvement in the ability to detect a trend in the data.
For detrending other ions, we used a simple linear regression against the log of runoff; for pH we used linear regression vs. runoff. The residuals from these regressions were used in the trend analysis tests for each ion.
4.2.2.2 Temporal Trend Analysis
We used the nonparametric seasonal Kendall test (SKT) for trend detection on the raw data and on the residual data after detrending for any effects due to stream flow (Hirsch et al., 1982). The SKT is the preferred test for water quality data because the test is powerful and can accommodate non-normality, outliers, missing data, and some types of censored data (Hirsch et al., 1982; Loftis et al., 1989). We found that autocorrelation coefficients of residual data between quarters were small (average r=0.016, SE=0.01 for ANC and r=0.089, SE=0.014 for pH) so no corrections were made for serial correlation in the stream data. Lakes with significant serial correlations in raw or detrended pH, alkalinity, sulfate or base cations were removed, as described above, prior to trend analysis by SKT (Hirsch et al., 1982).
Based on previous work (van Belle and Hughes, 1984), and our own Monte Carlo simulations on both 5 and 10 year randomly generated data sets, we omitted the continuity correction in the SKT calculations to avoid overly conservative test results in the meta-analysis. Both the raw and the adjusted data showed the same general results and our discussion focuses on the flow adjusted (residual) trends because we are most interested in changes in acid base chemistry due to changes in acidic deposition rather than changes driven by stream runoff. The SKT can be calculated either as a single z statistic for each site or as four z statistics for each site, one for each of the four seasons (van Belle and Hughes, 1984; Yu et al., 1993).
4.3 Figure 4.2-1. Example of steps in the analysis of trends in ANC shown for Cronin Brook, MA. (A) Quarterly stream water ANC concentrations vs. time. (B) ANC vs. daily runoff from nearby gaging station on the Quinsigamond River in North Grafton, MA. The hydrology data based on Water Resources Data Massachusetts and Rhode Island Water Year 1982-1992 publications and direct gage readings. The line is from nonlinear regression fit to the data as described in the text. (C) The detrended residuals vs. time. The SKT test shows significant increases in ANC at this site (Z=2.59, á<0.01).
4.4 Figure 4.2-2. Example of steps in the analysis of trends in ANC for Ramshorn Pond, MA. (A) Quarterly stream water ANC concentrations vs. time. (B) ANC vs. daily runoff from a nearby gaging station on the Quinsigamond River in North Grafton, MA. The hydrology data based on Water Resources Data Massachusetts and Rhode Island Water Year 1982-1992 publications and direct gage readings. The line is from nonlinear regression fit to the data as described in the text. (C) The detrended residuals vs. time. The SKT test shows significant increases in ANC at this site (Z=2.7, á=0.1).
4.5 4.2.2.3 Meta-Analysis
We used meta-analysis to statistically combine the results of the SKT on 330 streams, regardless of their individual significance levels, into a single, much more powerful, test for the overall trend in streams (Wolf, 1986). The same was done to determine the overall trend in lakes. Basically, meta-analysis goes one step beyond the traditional statistical trend analysis of a single stream. Meta-analysis is a statistical test on a population of statistical test results, conducted to see if the group shows a significant trend overall, even if the individual tests do not (Wolf, 1986). Consider, for example, that 330 random data sets would result in an average of 16.5 significant trends (5%) at the 95% confidence interval based on chance alone (with as many increasing trends as decreasing trends). Real trends will result in deviations in the distribution of significance levels from that expected by chance. Meta-analysis answers the question “What is the probability that the observed distribution of the individual test results would have occurred by chance alone?”.
To use meta-analysis, we need some method of combining independent test results into a single test statistic. Two types of meta-statistics are considered here. The first method simply takes the overall z score from the SKT trend analyses of 330 individual streams. When these z’s are squared and added together, they have a ÷2 distribution with 330 degrees of freedom (van Belle and Hughes, 1984). Thus, the sum of the squares of the z scores can be tested against the expected ÷2 given in any statistics text. This example meta-analysis, however, assumes the trends are homogeneous.
The second method, here referred to as an analysis of ÷2s or ANOCHIS, is an extension of the simple sums of squared z statistics described above, except here we use the 1320 (330*4=1320) seasonal z statistics to examine both trend and homogeneity of trend simultaneously. The majority of our results and discussion will focus on the second method (ANOCHIS) of meta-analysis, as described by previous researchers (van Belle and Hughes, 1984; Yu et al., 1993), where the overall trend is tested while simultaneously testing the homogeneity of trends between sites and between seasons.
To test for homogeneity of trends between the four seasons as well as between the 330 stream sites, we calculated the Kendall test for each season on each site. The resulting 1320 z statistics are squared and the sums of squares partitioned into sources of variation in the ÷2 (ANOCHIS) procedure (see sections 4.3.1.4 and 4.3.2.4) for residual ANC, in a manner similar to the analysis of variance (ANOVA) (van Belle and Hughes, 1984; Yu et al., 1993). In fact, output from the typical ANOVA analysis (SAS, 1987) can be used to generate the desired sums of squares which are compared directly to ÷2 tables of significance. With this method, we can test for trends between sites and between seasons. The process is described in detail below for the results of the residual stream ANC trends and in Section 4.3.2 for lakes. Meta-analysis for trends in stream and lake pH and other chemistry were conducted in a similar manner for both raw and residual data.
4.6 4.3 Analysis and Interpretation of Trends 4.3.1 Streams and Rivers 4.3.1.1 Raw Data
Figures 4.3-1 through 4.3-4 show the median unadjusted (raw) data for ANC, pH, sulfate, and sum of base cations for each collection and illustrate the year to year and seasonal variation observed over the ten year period. Due to missing data in each collection, the data shown in Figures 4.3-1 through 4.3-4 should not be used for trend detection but are presented as visual reference for the discussion that follows.
4.3.1.2 Trends in Raw ANC and pH Data
Examining the results of the SKT, we find that 34 streams show significant increasing trends in the raw ANC data over the ten year period at the 95% confidence level (Z > 1.96). Nine other streams show a significant decline (Z < -1.96) and the remaining 287 show no significant trend (see Figure 4.3-5). Note that the distribution is shifted to the right of zero, indicating an increasing trend for the group. The overall median slope for raw ANC was an increase of 1.5 µeq/L/yr. This increase in ANC was also reflected in the trends in raw pH which show 71 streams significantly increasing, 4 decreasing and 225 with no significant trend. The median slope of pH was an increase of 0.014 pH/yr.
4.3.1.3 Trends in Residual ANC and pH Data
There was no significant trend (á>0.05) in the overall stream runoff during the 10 year period and, therefore, we did not expect to see large differences between raw and residual data. The residuals of the regressions on runoff were used regardless of significance level. In fact, significance is irrelevant in the robust weighted regression used here for ANC detrending. The average r2 for the ANC data of 330 streams was 0.52; where r2 is defined here as 1-(residual SS/Corrected Total SS) for weighted regressions. In comparison, the Cronin Brook ANC regression on runoff, shown in Figure 4.2-1, had an r2 of 0.58.
The trend results for the residual data were very similar to the trends seen in the raw data, although the significance level was often higher and the magnitude of the estimated slopes were often somewhat greater. We find that 70 of the 330 streams show statistically significant increases in residual ANC, while only 11 streams show significant decreasing trends at the á=0.05 level (Table 4.3-1). The remaining 249 streams show no significant trend, but show more positive than negative indications (see Figure 4.3-6). The median slope of residual ANC vs. time is an increase of 2.4 µeq/L/yr. Again these trends are supported by trends seen in residual pH. Ninety-six streams show significant increases in residual pH and only two show significant decreases with 232 showing no significant trend in either direction. The median slope was +0.021 pH/yr.
4.7 Figure 4.3-1. Median stream water ANC vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes= +2.4 µeq/L/year and +1.5 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only.
Figure 4.3-2. Median stream water pH vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = +0.021 pH units/year and +0.014 pH units/year), representing detrended and raw data trends respectively, are shown for visual reference only.
4.8 Figure 4.3-3. Median stream sulfate vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = -1.8 µeq/L/year and -1.6 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only.
Figure 4.3-4. Median sum of stream base cations vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = +1.0 µeq/L/year and -1.4 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only.
4.9 Figure 4.3-5. Frequency distribution of Z scores for streams from the seasonal Kendall test on raw ANC.
Figure 4.3-6 Frequency distribution of z scores for streams from the seasonal Kendall test on residual ANC after detrending for hydrology.
4.10 4.3.1.4 Meta-Analysis Results of Trends in ANC
Although many more streams show increasing trends in pH and ANC than show decreases, most of the streams show no significant trend at the á<0.05 level. The fact that most of the streams show no significant trend is not surprising given the relatively short period of the study and the natural variability of water quality. For simple meta-analysis, consider the 330 SKT Z scores for adjusted ANC trends. These normal Z statistics, when squared, sum to 838, which is highly significant (á<<0.001) when compared to the expected ÷2 of 421 with 330 degrees of freedom and indicate an increase in ANC in the population of streams as a whole.
Using the meta-analysis ANOCHIS procedure described above, we reanalyzed the residual ANC data based on the 1320 seasonal z statistics and produced tests of trend and homogeneity shown in Table 4.3-2. The overall trend is highly significant at the á<0.001 level (÷2@ 1 d.f.=250) and in agreement with the simple meta-analysis described above. The additional information in Table 4.2-1 shows that the trends are homogeneous across all seasons (÷2@3 d.f.=0), but the trends are not homogeneous across all sites (÷2@329 d.f.=578). The significant site effect indicates some of the sites have trends which differ significantly from trends at other sites. The site by season interaction effect was found to be not significant. Table 4.3-1. Median ANC trend for the 81 streams with a significant trend among the total of 330 streams. 1 Streams recovering from acidification (acidification=ANC<0); 2 streams becoming acidified. PALSARIS Median ANC Trend Code Name Town ueq/l/yr mg/l/yr POSITIVE TREND 1202250 Whitman Brook Hancock 91 4.6 8143925 James Brook Groton 60 3.0 2105150 Sykes Brook Pittsfield 58 2.9 1100625 Sweet Brook Williamstown 53 2.7 1101000 Hudson Brook1 North Adams 49 2.5 3313375 Stafford Brook Colrain 43 2.2 7239175 Clematis Brook Waltham 32 1.6 2105350 Barton Brook Dalton 31 1.6 3315425 Chickley River Hawley 20 1.0 7341350 York Brook Canton 19 1.0 3421500 Bottom Brook Northfield 18 0.9 3420100 Esther Brook Conway 17 0.9 8451650 Limit Brook Tyngsborough 17 0.9 7240225 Dix Brook Medway 14 0.7 7240075 Miller Brook Franklin 13 0.7 5131325 Spring Brook Mendon 13 0.7 5132625 Cronin Brook Grafton 13 0.7 3315800 Gulf Brook Savoy 13 0.7 3208950 Drake Brook Southwick 13 0.7
4.11 PALSARIS Median ANC Trend Code Name Town ueq/l/yr mg/l/yr 3313650 South River Ashfield 11 0.6 5131425 Aldrich Brook Millville 11 0.6 7240050 Cress Brook Norfolk 11 0.6 3420000 Ground Brook Conway 11 0.6 5131800 West River Grafton 10 0.5 3316050 Todd Brook Charlemont 10 0.5 3209700 Moose Meadow Brook Montgomery 8 0.4 9354725 Alewife Brook Essex 8 0.4 5131775 Laurel Brook Douglas 8 0.4 3419075 Unquomonk Brook Williamsburg 8 0.4 8143675 Gulf Brook Pepperell 8 0.4 3522400 Jacks Brook (Upper) Northfield 7 0.4 5233750 Bungay River North Attleborough 7 0.4 2105425 Anthony Brook Dalton 7 0.4 8246625 Pond Brook1 Westford 7 0.4 9662475 Hawes Run Yarmouth 6 0.3 3314650 Underwood Brook Heath 6 0.3 3209275 Ripley Brook1 Granville 6 0.3 5132850 Stone Brook Millbury 6 0.3 6237625 Black Brook Easton 5 0.3 5131300 Muddy Brook Mendon 5 0.3 9456425 Stony Brook Norwell 5 0.3 4129125 Cady Brook Charlton 5 0.3 8145625 Muschopauge Brook Rutland 5 0.3 6235750 Snake River Raynham 5 0.3 9661650 Herring River Wellfleet 5 0.3 3107575 Pond Brook Granville 5 0.3 3316550 Lord Brook Rowe 5 0.3 8145900 Babcock Brook Princeton 5 0.3 3420200 Cushman Brook Leverett 5 0.3 8145400 Governor Brook1 Princeton 4 0.2 9662775 Mashpee River Mashpee 4 0.2 3627400 Shattuck Brook1 Phillipston 4 0.2 3627000 Hop Brook New Salem 4 0.2 3524375 Bluefield Brook1 Ashburnham 3 0.2 3627150 West Branch Fever Petersham 3 0.2 Brook 3522450 Whetstone Brook1 Wendell 3 0.2 3523575 Thousand Acre Brook Phillipston 3 0.2 3524075 Mahoney Brook1 Gardner 3 0.2 5131700 Emerson Brook Uxbridge 3 0.2 3626475 Maynard Brook1 Oakham 3 0.2
4.12 PALSARIS Median ANC Trend Code Name Town ueq/l/yr mg/l/yr 3627850 Burrow Brook Oakham 2 0.1 3524175 Scott Brook1 Royalston 2 0.1 3524150 Priest Brook1 Winchendon 2 0.1 3106825 Fox Brook Granville 2 0.1 3419600 Dean Brook1 Shutesbury 2 0.1 5133125 Scott Brook1 Holden 2 0.1 3419675 Amethyst Brook Amherst 2 0.1 3524200 Towne Brook1 Royalston 2 0.1 3522425 Osgood Brook1 Wendell 1 0.1 3523750 Kenny Brook Royalston 1 0.1
NEGATIVE TREND 3421450 Mill Brook2 Northfield -1 -0.1 6235125 Rattlesnake Brook2 Freetown -3 -0.2 9560175 Shingle Island River2 Dartmouth -3 -0.2 8144125 Mason Brook Townsend -4 -0.2 9661550 Silver Spring Brook Wellfleet -5 -0.3 9559550 Spring Brook Lakeville -5 -0.3 8248425 Whitehall Brook Hopkinton -7 -0.4 3209950 Gibbs Brook Blandford -9 -0.5 7442425 Martin Brook Randolph -12 -0.6 5334150 Clear Run Brook Seekonk -30 -1.5 2103800 Dry Brook Sheffield -148 -7.4
Table 4.3-2. ÷2 (ANOCHIS) Table Showing an Analysis of Trend and Homogeneity for Residual ANC in 330 streams.
Source÷2 á leveldf Trend 1 250 <0.0001 Site 329 578 <0.005 Season 3 0 NS* Site*Season 987 934 NS* Total 1320 1762
* NS not significant at level á=0.05
4.13 4.3.1.5. Other Ion Trends in Streams
Meta-analysis of homogeneity and trends in other parameters were calculated as described above for ANC. Unlike ANC trends, the trends in the other parameters showed significant seasonal heterogeneity (Figure 4.3-7). The trends were in the same direction in all four seasons for both pH (increasing) as well as sulfate (decreasing). However, acid anions and base cations showed some differences in trend direction between seasons, and this demonstrates why one should not assume, as some have suggested (e.g. Stoddard and Kellogg, 1993), that the original SKT precludes the need to analyze seasons separately. Residual chloride, for example, shows a significant negative trend of -0.37 overall; yet, an examination of seasonal trends (Figure 4.3-7) shows large decreasing trends in residual chloride in the months of January, July, and October, but an increasing trend in April over the past 10 years. Results of the trend and meta-analyses for both raw data and hydrology adjusted data are summarized in Table 4.3-3 for cations and in Table 4.3-4 for anions. Significance levels for site, season and interaction effects are not reported here. Our discussion will focus on the adjusted trends.
The trends in pH and ANC appear to be driven primarily by reductions in acid anions (-2.3 µeq L-1 year-1), and most of this was due to declines in sulfate (-1.6 µeq L-1 year-1). Other researchers have found similar downward trends in sulfate, particularly in the northeast (Murdoch and Stoddard, 1993; Driscoll et al., 1989). The sum of base cations showed a significant decline in the raw data of -1.4 µeq L-1 year-1, but when the data were adjusted for hydrology the trend was no longer significant (Table 4.3-3).
However, we were severely constrained in analyzing surface water trends by our relatively high level of nitrate detection (14 µeq L-1). Trends that may or may not be occurring at lower concentrations are veiled. For those streams with nitrate levels higher than 14 µeq L-1, trend analysis shows that some of those streams exhibit significant upward trends (significant positive z scores) and others show a significant downward trend (negative z scores). The sum of z scores is negative suggesting that more high nitrate streams are exhibiting declining nitrates than increasing nitrates. When combined, however, the trends cancel each other to yield a median slope of zero.
4.3.2. Lakes and Ponds 4.3.2.1 Raw Data
Figures 4.3-8 through 4.3-11 show the median unadjusted (raw) data for ANC, pH, sulfate, and sum of base cations for each collection and illustrate the year to year and seasonal variation observed in the lakes over the ten year period. Due to missing data in each collection, the data shown in the figures should not be used for trend detection but are presented as a visual reference for the discussion that follows.
4.3.2.2 Trends in Raw ANC and pH Data
The results of the SKT on the 181 lakes show that 37 of the lakes show a significant upward trend in raw ANC, eight show a significant downward trend and the remaining 136 lakes show no significant trend (á<0.05). The median trend slope for all lakes is an upward trend of 1.5 µeq/L/yr. This increase was also reflected in the trends in raw pH where 30 lakes showed a significant upward trend, eight showed a significant downward trend and the remaining 143 lakes showed no significant trend (á<0.05). The overall median slope was an increase of 0.010 pH units/year.
4.14 Figure 4.3-7. Trends for streams in major ions and ANC by season with the overall yearly trend for comparison. The left and right side of each pair of bars represent the cations and anions, respectively. Titration ANC is shown as a line. Increasing trends are shown as positive values, decreasing trends as negative values, and nonsignificant values are not shown.
4.15 Table 4.3-3. Cation trends in 330 Massachusetts streams. Note: ÓBC=sum of base cations. In most cases the trends were not homogenous across sites and seasons as described in the text. All trends are significant at á<0.001 except as noted. ______Trend pH ÓBC Ca++ Mg++ Na+ K+ (pH/year) ------(µeq/L/year) ------______Raw data trends Slope +0.014 -1.4 -0.6 -0.8 -0.0 +0.3 (÷2) 196. 12. 14. 180. 0. 59. Significance1 ** ** ** ** NS **
Adjusted data trends Slope +0.021 +1.0 +0.1 -0.8 +0.6 +0.3 (÷2) 396. 0. 0. 112. 6. 73. Significance1 ** NS NS ** * ** ______1 Significance levels for á: ** < 0.001; * < 0.05; NS, not significant.
Table 4.3-4. Anion trends in 330 Massachusetts streams. Note: AA=sum of acid anions. In most cases the trends were not homogenous across sites and seasons as described in the text. All trends are significant at á<0.001 except as noted. ______= - - Trend ANC ÓAA SO4 Cl NO3 ------(µeq/L/year) ------______Raw data trends Slope +1.5 -3.4 -1.6 -1.0 -0.0 (÷2) 60. 119. 194. 21. 20. Significance1 ** ** ** ** **
Adjusted data trends Slope +2.4 -2.3 -1.8 -0.37 ND2 (÷2) 250. 74. 192. 11.2 ND2 Significance1 ** ** ** ** ND2 ______1 Significance levels for á: ** < 0.001; * < 0.05; NS, not significant. 2 ND, nitrate trends not determined on adjusted data due to high censoring (high LOD).
4.16 Figure 4.3-8. Median lake water ANC vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes +2.4 µeq/L/year and +1.5 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only.
Figure 4.3-9. Median lake water pH vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = +0.013 pH units/year and +0.010 pH units/year), representing detrended and raw data trends respectively, are shown for visual reference only.
4.17 Figure 4.3-10. Median lake sulfate vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = - 1.10 µeq/L/year and -1.39 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only.
Figure 4.3-11 Median sum of lake base cations vs. time. Missing collections are indicated by a dotted line connecting points. The solid and dashed lines (slopes = +1.51 µeq/L/year and +0.98 µeq/L/year), representing detrended and raw data trends respectively, are shown for visual reference only.
4.18 4.3.2.3 Trends in Residual ANC and pH Data
As noted previously there was no significant trend in overall stream runoff during the 10 year period (section 4.3.1.3) and we did not expect to see large differences in trends between raw and residual data. During the detrending process we observed that less of the variation in lake ANC was explained by runoff in the weighted non-linear regressions (average r2=0.36) than was observed for detrending of stream water quality data described previously. An example of detrending is shown for Ramshorn Pond in Figure 4.2-2. In this example, there is less variation in raw ANC data shown in Figure 4.2-2a than seen in Cronin Brook in Figure 4.2-1a, but the fit to the runoff (Fig. 4.2-2b) is not as good (r2=0.27), and there is little difference in trends between the raw data and the detrended data. The median trend slope for the raw and detrended data were both 4µeq/L/yr and both were significant (á<0.05).
The trend results for residual data were similar to the trends in the raw data. We find that 43 lakes showed significant upward trends in residual ANC, 7 showed significant decreases (Table 4.3-7). The remaining 131 lakes showed no significant trends (á<0.05). The median overall trend for residual ANC was +2.4µeq/L/yr, slightly higher than for the raw ANC trends, and the same as was found in streams discussed previously. Only 18 lakes showed significant upward trends in residual pH, 5 showed significant declines and the remaining 158 showed no significant trends (á<0.05). The median trend in residual pH was +0.013 pH units/yr, slightly higher than that observed for raw pH above.
4.3.2.4 Meta-Analysis Results of Trends in ANC in Lakes
The simple meta-analysis based on the summed squares of 181 SKT Z scores for residual ANC in lakes shows that the result of 618.9 is highly significant, but assumes no heterogeneity of trends. The second meta-analysis (ANOCHIS) procedure is shown in Table 4.3-5 for the 181x4= 724 Z statistics from the seasonal SKT. Here the trend is still highly significant with a trend chi- square of 147.2. The site effect is also significant as was seen in stream trends, indicating that the differences in trends among sites were greater than expected due to chance. Unlike stream trends where differences in residual ANC trends between seasons was not significant, we did find significant seasonal heterogeneity in residual ANC trends in lakes. There was no significant interaction effect between site and season effects (á>0.05). Table 4.3-5. ÷2 (ANOCHIS) Table Showing an Analysis of Trend and Homogeneity for Residual ANC in 181 lakes. ÷Source2 á leveldf Trend 1 147.2 <0.0001 Site 180 451.1 <0.005 Season 3 37.6 <0.005 Site*Season 540 469.8 NS* Total 724 1105.7
*NS not significant at level of á=0.05
4.19 Figure 4.3-12. Frequency distribution of Z scores for lakes from the seasonal Kendall test on raw ANC
Figure 4.3-13. Frequency distribution of z scores for lakes from the seasonal Kendall test on residual ANC after detrending for hydrology
4.20 The significant season effect was further investigated by examining the trends for the four seasons individually as shown in the bottom of Table 4.3-6. The trends were significant in all four seasons and the median slope were all in the same direction (upward trends) for all four seasons.
Table 4.3-6. ÷2 (ANOCHIS) Table Showing a Seasonal Analysis of Trend and Homogeneity for Residual ANC in 181 lakes. Season df ÷2 á level January 1 5.70 <0.05 Site 180 198.9 NS Slope 1.2 µeq/L/yr
April 1 69.9 <0.0001 Site 180 231 <0.01 Slope 1.6 µeq/L/yr
July 1 14.2 <0.001 Site 180 233 <0.01 Slope 0.7 µeq/L/yr
October 1 94.9 <0.0001 Site 180 257 <0.005 Slope 3.8 µeq/L/yr
NS not significant at level á level=0.05
4.21 Table 4.3-7. Median ANC trend for the 50 lakes with a significant trend among the total of 181 lakes. 1 Lakes recovering from acidification (acidification =ANC<0); 2 lakes becoming acidified. PALSARIS Median ANC Trend Code Name Town ueq/l/ year mg/l/ year POSITIVE TREND 62147 Poquoy Pond Lakeville 49.2 2.5 51150 Silver Lake Bellingham 32.8 1.6 32078 Wright Pond Holyoke 21.0 1.1 32016 Chapin Pond Westfield 19.8 1.0 92025 Hood Pond Ipswich 14.4 0.7 96186 Lovers Lake Chatham 13.0 0.7 81100 Phoenix Pond Shirley 12.1 0.6 84033 Long Sought For Pond Westford 11.3 0.6 81056 Heald Pond Pepperell 11.3 0.6 94169 Weeks Pond Plymouth 9.4 0.5 72095 Pleasant Street Pond Franklin 9.3 0.5 51024 Coes Reservoir Worcester 9.3 0.5 95034 Curlew Pond Plymouth 8.7 0.4 42033 Low Pond Dudley 7.5 0.4 91010 Pentucket Pond Georgetown 7.2 0.4 92063 Stiles Pond Boxford 7.2 0.4 42004 Buffum Pond Charlton 7.1 0.4 36036 Cloverdale Street Pond Rutland 6.6 0.3 81041 Fitch Basin Sterling 6.4 0.3 62150 Prospect Road West Pond Plympton 6.2 0.3 51068 Hovey Pond Grafton 6.0 0.3 62121 Morse Pond Easton 5.9 0.3 51086 Lee Reservoir Uxbridge 5.3 0.3 42009 Cedar Meadow Pond Leicester 5.3 0.3 41014 East Brimfield Reservoir Brimfield 5.0 0.3 35095 Watatic Pond1 Ashburnham 4.9 0.2 42065 Wee Laddie Pond Charlton 4.9 0.2 95003 Atwoods Reservoir1 Carver 4.6 0.2 51159 Stevens Pond Sutton 4.4 0.2 96244 Peters Pond Sandwich 4.1 0.2 51126 Ramshorn Pond Sutton 4.0 0.2 36051 Demond Pond Rutland 3.7 0.2 93085 Wallace Pond Gloucester 3.1 0.2 41019 Hamilton Reservoir Holland 2.9 0.1 51063 Holden Reservoir Holden 2.9 0.1 41016 Lake George Wales 2.8 0.1 52013 Falls Pond North Attleborough 2.7 0.1 96226 Northeast Pond1 Wellfleet 2.7 0.1 36001 Adams Pond Oakham 2.6 0.1 96015 Big Sandy Pond1 Yarmouth 2.0 0.1
4.22 35017 Lake Denison Winchendon 1.6 0.1 93001 Babson Reservoir1 Gloucester 1.1 0.1 96068 Duck Pond1 Wellfleet 1.0 0.1
NEGATIVE TREND 94149 South Triangle Pond2 Plymouth -2.5 -0.1 96050 Crystal Lake Orleans -2.7 -0.1 96163 Kinnacum Pond2 Wellfleet -2.8 -0.1 95112 New Long Pond2 Plymouth -3.4 -0.2 73024 Hemenway Pond Milton -3.7 -0.2 95030 College Pond Plymouth -5.3 -0.3 92022 Four Mile Pond Boxford -5.6 -0.3
4.3.2.5 Other Ion Trends in Lakes
Meta-analysis was performed on the other ions in a manner similar to that described for ANC above. As shown in Tables 4.3-5 and 4.3-6, the trends were generally significant except for the sum of acid anions which showed insignificant trends in both raw and adjusted data. As with ANC, significant heterogeniety was observed for most trends for both site and season effects but not for the interaction effect. Unlike ANC, however, many of the adjusted ion trends were non-significant in some seasons, but the significant trends were all in the same direction for a given ion, as shown in Figure 4.3-12. The same was true for the trends in raw ion data except for the case of unadjusted magnesium data. Even though the typical SKT test would show a significant negative trend for raw magnesium data, it would be inappropriate to report a yearly trend where there are significant, yet opposing trends in some seasons.
As in streams, the upward trends in pH and ANC appear to be related to downward trends in sulfate (-1.1 µeq/L/yr for adjusted sulfate). Unlike streams however, the trends in base cations in lakes is clearly upward for both raw and adjusted data (+1.0 and +1.5 µeq/L/yr, respectively) which also would be expected to related to increases in pH and ANC.
The trend in the sum of acid anions was not significant in either the raw data or the adjusted data in lakes. Part of this may be due to the strong upward trend in chloride in lakes would tend to cancel the downward sulfate trend. The upward trend in chloride was associated with similar upward trends in sodium, especially in the month of April, and this may be related to spring runoff from road salting during the winter (see Figure 4.3-12). The raw nitrate trends showed a significant downward trend, but the median slope was zero. As in streams, analysis of adjusted nitrate trends was not performed due to the high number of censored data associated with the high limit of detection for nitrate (see section 4.3.1.5).
4.23 Table 4.3-8. Cation trends in 181 Massachusetts lakes. Note: ÓBC=sum of base cations. In most cases the trends were not homogenous across sites and seasons as described in the text. All trends are significant at á<0.001 except as noted. ______Trend pH ÓBC Ca++ Mg++ Na+ K+ (pH/year) ------(µeq/L/year) ------______Raw data trends Slope +0.010 +1.0 +1.0 -0.3 +0.8 +0.2 (÷2) 57.8 7.16 16.2 13.8 14.0 32.2 Significance1 ** * ** ** ** **
Adjusted data trends Slope +0.013 +1.5 +0.9 -0.2 +1.0 +0.2 (÷2) 34.7 9.04 21.8 15.2 13.4 35.0 Significance1 ** * ** ** ** ** ______1 Significance levels for á: ** < 0.001; * < 0.05; NS, not significant.
Table 4.3-9. Anion trends in 181 Massachusetts lakes. Note: AA=sum of acid anions. In most cases the trends were not homogenous across sites and seasons as described in the text. All trends are significant at á<0.001 except as noted. ______= - - Trend ANC ÓAA SO4 Cl NO3 ------(µeq/L/year) ------______Raw data trends Slope +1.5 +0.4 -1.4 +1.6 -0.0 (÷2) 77.2 0.03 43.1 10.9 4.38 Significance ** NS ** ** *
Adjusted data trends Slope +2.4 -0.1 -1.1 +1.4 ND2 (÷2) 147.2 0.47 39.8 10.8 ND2 Significance1 ** NS ** ** ND2 ______1 Significance levels for á: ** < 0.001; * < 0.05; NS, not significant. 2 ND, nitrate trends not determined on adjusted data due to high censoring (high LOD).
4.24 Figure 4.3-14. Trends for lakes in major ions and ANC by season with the overall yearly trend for comparison. The left and right side of each pair of bars represent the cations and anions, respectively. Titration ANC is shown as a line. Increasing trends are shown as positive values, decreasing trends as negative values, and nonsignificant values are not shown.
4.4 Discussion and Synthesis
Our trend analysis of ANC, pH, sulfate and base cations are generally consistent with declines in acid sulfate inputs. Our results showing an increase in residual ANC of 2.4 µeq/L/yr for streams and 2.4 µeq/L/yr for lakes are also supported by the observed significant (á<<0.0001) increase in residual pH of +0.021 pH units/yr for streams and +0.013 pH units/yr for lakes. For streams, the trends in pH and ANC appear to be driven primarily by reductions in residual acid anions, particularly residual sulfate (-2.3 µeq/L/yr and -1.8 µeq/L/yr, respectively) with no significant
4.25 change in residual base cations. Unlike streams, there was no significant trend for residual acid anions in lakes where declines in residual sulfate were balanced by an increase in residual chloride (-1.1 and 1.4 µeq/L/yr, respectively). In lakes, the increase in chloride appears to be related to sodium (+1.0 µeq/L/yr); both show strong increases in April, presumably due to snowmelt reunoff from road salting. Residual base cations in lakes did show a significant upward trend due mainly to trends in residual calcium (+1.5 µeq/L/yr and +0.9 µeq/L/yr, repectively). Unfortunately, we could not test for trends in residual nitrates due to a high frequency of censored data (nitrate concentrations were usually below our limit of detection of 14 µeq/L).
The decline in sulfate and increase in ANC and pH in the streams and lakes is correlated with the decline in atmospheric sulfate. The effect due to trends in base cations is not clear. A study of deposition trends in the northeast shows significant declines in sulfate of about -1 µeq/L/yr in precipitation, while corresponding trends in base cations have ranged from nonsignificant to -0.9 µeq/L/yr (Hedin et al.,, 1994). Why lakes would show an increase in residual base cations but streams do not is not clear. Much of the difference is due to increases in residual calcium in lakes which is not seen in streams. Increasing eutrophication trends may cause increased CO2 production in bottom sediments, decreasing ñH in the sediments. The increased hydrogen ions may exchange for calcium, thus explaining the increase in calcium in lakes but not streams. Alternatively, calcium increases may be due to liming or other human impacts in the watershed.
There is no reason to believe that the declines in lake and stream sulfate are due to increases in bacterial sulfate reduction, since atmospheric sulfate inputs are declining. As discussed previously (Mattson et al., 1992 and Chapter 3.2), variations in surface water sulfate concentrations of the state may be due to a variety of factors including geologic, local variation in dry deposition, and variations caused by differences in evapotranspiration. We do not have sufficient information to test if these sources, rather than atmospheric inputs are driving the observed downward trend in stream and lake sulfate. We do not have any reason to believe there are significant changes in bedrock weathering, evapotranspiration or dry deposition.
Previous studies have indicated little change in pH and ANC in stream water quality in the northeastern United States. For example, at the Hubbard Brook Experimental Forest in New Hampshire, declines in stream sulfate concentrations have been balanced by declines in base cations with the result that stream acidity has been relatively unchanged (Driscoll et al., 1989). Streams of the Catskill Mountains of New York have also shown declines in sulfate, but there the decline is balanced by increases in nitrate levels resulting in generally nonsignificant increases in ANC, and inexplicably, some evidence of a decline in pH (Murdock and Stoddard, 1993). The streams in the Catskill region show some evidence of declines in base cation concentrations, but there, as in our study, the trends were not clear. The National Stream Quality Accounting Network shows strong upward trends in pH and ANC, which support our results; however, this same study reported that sulfate concentrations were inexplicably increasing at many of these same stations (Lettenmaier et al., 1991). Some of the differences between the studies cited above may be due to differences in the time period under study.
4.26 4.4.1. Site Effects To examine the pattern of heterogeneity between sites, we plotted the approximate positions of the streams and lakes with the ANC trends as shown in Figures 4.4-1 and 4.4-2. While most of the upward trends are distributed evenly, there is a small cluster of downward trends in the southeastern portion of the state for both streams and lakes. Our previous work on surface water quality shows strong regional patterns in ANC of surface waters in the state (Chapter 3, this report; Walk et al., 1992; Mattson et al., 1992) and that the southeast is an area of low relief containing numerous wetlands and many low ANC and darkly colored waters. We hypothesized that trends in organic acids, particularly in the southeast part of the state, could account for the observed trends in ANC. To test this hypothesis, we examined our data on stream and lake water color (as a surrogate for organic acids) and found a significant (á<0.05) declining trend over time of -0.3 PCU/year for both streams and lakes. Converting this first to DOC (see page 2.38 and Mattson et al., 1992) and then to acid equivalents (Sullivan et al., 1989), we roughly estimate this could account for only 0.1 µeq/L/yr decrease in ANC. Thus, it does not appear that trends in organic acids are responsible for the observed trends (2.4 µeq/L/yr) in ANC. We did notice that average stream color was inversely related to ANC trends (Spearman rank correlationr = -0.12; á<0.025). That is, colored waters tend to have smaller increases in ANC than clear water systems, and this may account for some of the differences in ANC trends between sites.
Figure 4.4-1. Distribution of significant (á<0.05) trends in adjusted ANC for 330 streams in Massachusetts. The solid triangles indicate upward trends, the open triangles represent downward trends and the dots represent no trend.
4.27 Figure 4.4-2. Distribution of significant (á<0.05) trends in adjusted ANC for 181 lakes in Massachusetts. The solid triangles indicate upward trends, the open triangles represent downward trends and the dots represent no trend.
4.4.2 Relative Sensitivity of High vs. Low Alkalinity Waters
Researchers have often assumed that the water bodies most sensitive acid inputs would be those waters with low ANC (Schindler 1988; Herlihy et al. 1991; Newell 1993). If 'sensitivity' refers solely to sensitivity to changes in pH, then our results support the assumption that the low ANC systems show the greatest rates of pH change. If the definition of sensitivity refers to sensitivity to changes in ANC as a function of ANC, then our results do not agree with the common viewpoint. Instead, we find that streams with high ANC (greater than 100 µeq/L) show a faster rate of ANC increase than the low ANC streams (4.2 µeq/L/year vs 1.3 µeq/L/year, respectively); both groups showed significant trends (÷2 = 149 and 102, respectively). The pattern for lakes is similar with high ANC lakes (greater than 100 µeq/L) showing a faster rate of ANC increase than the low ANC lakes (4.5 µeq/L/year vs 0.4 µeq/L/year, respectively); both groups showed significant trends (÷2 = 164 and 13.3, respectively). A plot of residual ANC trend vs. mean ANC shows a significant relationship, with ANC increasing 2 percent per year in streams (Figure 4.4-3) and 1.4 percent per year in lakes (Figure 4.4-4). While this result is unexpected, it is not without precedent. A paleolimnologic study on Adirondack lake acidification suggest the magnitude of acidification, as measured by changes in pH, was greatest in low alkalinity lakes as expected (Cumming et al. 1992). But Figure 6 in the same study also suggests that the greatest magnitude of alkalinity changes
4.28 (increases in this case) were observed in lakes with pH higher than about 6.7 with current ANC greater than about 40 µeq/L (Cumming et al. 1992). Asbury et al. (1990) also found that the greatest acidification (difference between historic and modern alkalinity) in Adirondack lakes was weakly, but positively correlated with historic alkalinity. It could be argued that Asbury et al. (1990) results may be spurious because the analysis was a difference between two measurements, but such spurious correlation is not likely in our SKT analysis.
There may be many other mechanisms to explain the greater increases in ANC in high ANC systems. For example, long term studies on streams in the Catskill Mountains suggest factors such as changes in land use can have a strong effect on water quality trends (Stoddard, 1991). Since development is more likely to occur in lowland areas of high ANC waters, one might expect ANC increases in such areas. While we do see an increase in base cations in lakes, we do not see any significant increase in base cations in streams that would be expected from anthropogenic activities. Another mechanism to explain the relationship would be if the high ANC systems are also systems which tend to have low runoff to precipitation ratios. In such cases, the evapo-concentration factor (the inverse of the runoff to precipitation ratio) is large and any changes in the concentrations of ions in precipitation are magnified into greater changes in stream water concentrations. Unfortunately, the evidence to support this is ambigous. The runoff to precipitation ratio for most of Massachusetts is about 0.5 (see Figure 5b of Church et al. 1995). We find higher ratios in the low ANC Berkshire Region 2, which supports our hypothesis, but we also find lower ratios in the Southeastern Massachusetts where there are also low ANC systems which does not support the hypothesis. Perhaps local variations in evapoconcentration may explain the differences, but further studies are required to resolve the issue. Some of the observed heterogeneity in trends among lakes may be due to difference in hydrology. Of the 181 lakes studied, 141 are hydrologically open (drained by a stream) while the remaining 40 are closed (seepage lakes). The median trend in adjusted alkalinity in open lakes was 2.8 µeq/L compared with 0.7 µeq/L in closed lakes. Thus open lakes are recovering from acidification faster than closed lakes. It should be noted that most of the closed lakes are found in the southeast coastal and Cape Cod regions of the state along with unique factors of geology, relief, and open vs. closed. This correlation makes it difficult to separate effects and attribute cause and effect in this or any correlational analysis.
In terms of pH sensitivity, the high ANC streams show a more gradual increase in pH than the low ANC streams (0.018 pH units/year vs 0.026 pH units/year); the trends were significant in both cases (÷2= 223 and 173 respectively). This is to be expected due to the nonlinear relationship between pH and ANC and the lack of pH buffering in low ANC systems. Thus low ANC systems are sensitive in terms of change in pH, but are not as sensitive to changes in ANC when compared to the high ANC systems. Whether a stream is 'sensitive' depends on the definition used. In terms of biotic effects, which respond mainly to pH, the low ANC streams are most sensitive and are recovering most rapidly.
4.29 Figure 4.4-3. The median trend slope for each of 327 streams is shown vs. the mean ANC of the stream. The best fit regression is shown. Three high ANC streams are omitted for clarity.
Figure 4.4-4. The median trend slope for each of 181 lakes is shown vs. the mean ANC of the lake. The best fit regression line is shown
4.30 Figure 4.4-5. Comparison of adjusted trend regressions between lakes and streams for pH, ANC, sulfate, and base cations. Regression lines are centered on median values for each parameter. All regressions are significant at the 95% confidence level except stream base cations.
4.31 4.4.3. Comparison with other Trend Studies
Previous studies have indicated little change in pH and ANC in stream water quality in the Northeastern United States. For example, at the Hubbard Brook Experimental Forest in New Hampshire, declines in stream sulfate concentrations have been balanced by declines in base cations with the result that stream acidity has been relatively unchanged (Driscoll et al., 1989). Streams of the Catskill Mountains of New York have also shown declines in sulfate, but there the decline is balanced by increases in nitrate levels resulting in generally nonsignificant increases in ANC, and inexplicably, some evidence of a decline in pH (Murdoch and Stoddard, 1993). The streams in the Catskill region show some evidence of declines in base cation concentrations, but there, as in our study, the trends were not clear. The National Stream Quality Accounting Network shows strong upward trends in pH and ANC, which support our results; however, this same study reported that sulfate concentrations were inexplicably increasing at many of these same stations (Lettenmaier et al., 1991). Some of the differences between the studies cited above may be due to differences in the time period under study.
Detecting recent trends in water quality is inherently hampered by the lack of power because the statistical tests are applied to short term data sets. For example, lake surveys have reported relatively few significant trends or conflicting trends at different sites (Driscoll and Dreason, 1993; Stoddard and Kellogg, 1993; Kahl et al., 1993). In some cases researchers have set the significance level to á<0.10 to increase the power of trend detection, but this increases the type 1 error rate. The presence of serial correlation in data from lakes further reduces the power of trend detection in the studies cited above, and this effect can be severe in lakes with relatively long water residence times. Other in-lake processes such as sedimentation and sulfate reduction may add further variability to the trends in lakes, all of which suggests that streams may be more appropriate for the early detection of water quality trends, although our data on the typically small Massachusetts lakes suggest the trends for lakes and streams are similar. In both lakes and streams, meta-analysis can increase the power of trend detection and also allows testing for non-homogeneous trends across seasons and sites.
Our stream results are generally consistent with other recent studies of trends in stream water chemistry (Murdoch and Stoddard, 1993; Smith et al., 1987; Driscoll et al., 1989), although some regional differences are present. For example, at the Hubbard Brook Experimental Forest in New Hampshire, declines in stream sulfate concentrations have been balanced by declines in base cations with the result that stream acidity has been relatively unchanged (Driscoll et al., 1989). Streams of the Catskill Mountains of New York have also shown declines in sulfate, but there the decline is balanced by increases in nitrate levels resulting in generally nonsignificant increases in ANC, and some evidence of a decline in pH (Murdoch and Stoddard, 1993). The streams in the Catskill region may be declining in base cation concentrations, but there, as in our study, the trends in base cations were not clear.
4.32 4.4.4 Meta-analysis and SKT Power
Detecting recent trends in water quality is inherently hampered by the lack of power because the statistical tests are applied to short term data sets. For example, lake surveys have reported relatively few significant trends or conflicting trends at different sites (Driscoll and Van Dreason, 1993; Stoddard and Kellogg, 1993; Kahl et al., 1993). In some cases researchers have decreased the significance level (increasing á from á<0.05 to á<0.10) to increase the power of trend detection, but this obviously doubles the type 1 error rate of reporting false trends. Recently Driscoll et al. (1995) reported that lakes in the Adirondacks of New York show declines in sulfate of -1.81 µeq/L/yr, but these trends are balanced by similar declines in base cations with no systematic change in ñH and ANC. The median sum of cations for Adirondack lakes is only 184 µeq/l (Kretser et al., 1989) compared to 781 µeq/l in our study. The presence of serial correlation in data from lakes further reduces the power of trend detection in the studies cited above, and this effect can be severe in lakes with relatively long water residence times. Other in-lake processes such as sedimentation and sulfate reduction may add further variability to the trends in lakes, all of which suggests that streams may be more appropriate for the early detection of water quality trends.
In both lakes and streams, meta-analysis can increase the power of trend detection and also allows testing for non-homogeneous trends across seasons and sites. For these reasons we expect meta-analysis to be used with increasing frequency in the analysis of water quality data, but we should also offer the following cautions. For new applications of meta-analysis, Monte Carlo simulations should be conducted to verify the expected behavior of the statistical results. Even minor changes or bias in analytical methods or technique may be detected as a trend in the data and it is important to ask if the quality control is sufficient to detect small changes over time, above and beyond that due to bias (Alexander et al. 1993). For example, if the long term errors (bias) of chemical analyses during the course of the study is on the order of 10 µeq/L, then over a ten year survey the realistic limit of detection for trends is about 1 µeq/L/year. In our study, we would not put too much confidence in the results of any single trend shown in Table 2 or Table 3 with slopes less than +1 µeq/L/year. Finally, one should ask, is the trend consistent with other independent lines of evidence? In our case, independent measurements of pH, ANC, and acid anions are all consistent with the conclusion of recovery from acidification while base cations are ambiguous (Figures 3-6). Furthermore, our results are also generally consistent with deposition studies and other regional studies of surface water trends. As a final check on our analysis we reanalyzed our alkalinity data using Alley's (1988) detrending method. Because Alley's method adjusts the temporality of the data, it precludes the use of the seasonal application of the SKT. Nevertheless, the results of the Kendall tau test were significant and in agreement with our analysis, and the estimated Sen slope (median slope for all data) for residual ANC was only slightly higher than our original analysis (2.66 vs. 2.4).
4.5. Summary of Trend Analyses
Many streams and lakes in the northeastern U.S. are influenced by anthropogenic activities other than atmospheric deposition, and thus it is impossible to assign a direct cause and effect in this type of study. Long term studies on streams in the Catskill Mountains suggest factors such as changes in land use can have a strong effect on water quality trends (Stoddard, 1991). The agreement in direction and magnitude of the trends in precipitation and stream water chemistry in
4.33 our study do, however, provide some evidence in support of a causal link between trends in atmospheric inputs and stream and lake concentrations. A study of deposition trends in the northeast shows significant declines in sulfate of about -1 µeq L-1 year-1, while corresponding trends in base cations have ranged from nonsignificant to -0.9 µeq L-1 year-1 (Hedin et al., 1994). Trends in hydrogen ion are more difficult to detect, but the long term record at the Hubbard Brook site in New Hampshire does show a significant decline of -1.45 µeq L-1 year-1 since 1963 (Likens, 1989). Our estimates of trends in stream water adjusted sulfate (-1.8 µeq L-1year-1) and lake water adjusted sulfate (-1.1 µeq L-1year-1) combined with increases in ANC (+2.4 µeq L-1 year-1 for both streams and lakes), roughly agree with the precipitation trends if we assume a two fold increase in stream and lake trends due to the effects of evapoconcentration. The concentration of ions results from approximately 120 cm of annual precipitation (Climates of the States, 1959) being reduced to approximately 60 cm of annual runoff in the streams (Krug et al., 1990). We suggest that the observed declines in sulfate inputs are not completely balanced by equivalent declines in base cations and this results in increases in pH and ANC in the streams. Despite earlier skepticism of emission reductions producing benefits (Katzenstein, 1985), our results provide evidence that streams and lakes respond quickly to atmospheric inputs.
4.34 5.0. A Profile of the Acid Rain Monitoring Project Citizen Volunteer
More than 1000 volunteers have provided the manpower for the Acid Rain Monitoring Project. Most have been sample collectors but others have been professional lab technicians; all have been volunteers. Because similar volunteer environmental efforts hold substantial promise for meeting future monitoring needs at affordable cost, examination of the profile of the Acid Rain Monitoring Project volunteer may yield insight into the needs for attracting and motivating future volunteers for similar efforts.
The Acid Rain Monitoring Project began locally in October, 1982 but quickly spread statewide by March, 1983. In its first phase, which sampled more than 1000 surface waters monthly for 14 months, nearly 1000 volunteers were involved. Its second phase, which sampled 2500 surface waters semiannually, involved more than 1000 volunteers, most also involved in the earlier effort. Its third phase sampled 800 surface waters quarterly and required 300 volunteers. Of these past and present volunteers, the Project had mailing addresses for 693. During the winter of 1988-89, the Massachusetts Water Resources Research Center surveyed these volunteers associated with the Project to develop a profile of the "ARM volunteer." Questions and responses are shown in detail in Appendix 8.11
Forty percent (275) of the 693 questionnaires distributed were returned. Many volunteers from earlier days were no longer at their previous mailing addresses. Most of the respondents were then current volunteers (161 vs 106). Of the 231 volunteers responding to the question on years of involvement, 58 had been with the Project since its inception (Figure 5.1-1) and 139 had completed one or more phases (Figure 5.1-2). Volunteers who only collected samples from surface waters were most numerous, representing 96%. Laboratory volunteers represented 5% and volunteer coordinators represented 14%. Individuals who not only collected samples but also helped coordinate activity within towns or a county represent 13%, but relatively few lab volunteers
Figure 5.1-1. Years of involvement of ARM volunteers.
5.1 attempted to also collect samples (3%). A small number of volunteers (1.5%) did all aspects of the volunteer effort.
Figure 5.1-2. Phases of involvement of ARM volunteers for those who completed one or more phases
Of 268 respondents who provided information on their sex, 65% were male and 31% were female. Four percent responded that they worked as couples, even though we had not offered that response option on the questionnaire. Most were married (79%), but 12% were single, 3% divorced or separated, 3% widowed, and 3% living with a partner. Volunteers were mostly college graduates (88% had some college education): 70% had a college degree and 36% had advanced degrees.
ARM Project volunteers lived mostly in small towns (63%) and roughly equal numbers lived in city/urban environments (16%) and rural environments (20%). They primarily owned their own home (79%), but 12% rented and 5% lived in a home owned by a relative.
Most ARM volunteers were natives to Massachusetts (54%) or lived most of their life in Massachusetts (24%). Many had lived in other New England states (11%) while 10% were relative newcomers to New England. They came from places as far away as Europe, Brazil, western U.S., and, ironically, the Midwest (Figure 5.1-3).
5.2 Figure 5.1-3. Geographic distribution of volunteer's state of origin
The questionnaire provided 19 employment categories for the volunteers to use. With one exception, there was no clear occupation that defined an ARM volunteer (Figure 5.1- 4). The exception was the category of "retired," but among these could be seen all of the other occupations in earlier life. Household income tended to fall in the range of $20,000 to $75,000 (75%), with most (37%) in the $30,000 to $50,000 range (Figure 5.1-5). A few earned more (11%) and a few earned less (14%). The age distribution of volunteers reflected the population age distribution with most volunteers in middle age but with a pronounced Figure 5.1-4. Volunteer occupations number of retirement age (Figure 5.1-6).
5.3 ARM volunteers must have devoted considerable time and money to other volunteer organiza- tions. They listed memberships in 69 civic organizations, some requiring only donations to be a member but most requiring active participation. Volunteers averaged membership in 1.2 such organi- zations, in addition to their efforts for the ARM Project. The Massachusetts Audubon Society and Trout Unlimited were the most common organizations listed by Figure 5.1-5. Estimated household income of ARM volunteers. volunteers, a probable reflection of the way in which the volunteer network was originally created. In fact, 24 volunteers became involved through Trout Unlimited, 18 through newspaper stories, 9 through the Massachusetts Audubon Society and 17 were asked by a friend. Many volunteers chose not to list a religious affiliation, but those that did listed 20 different religions or religious beliefs. Few (21) chose to list their political affiliation: 52% were Democrats, 24% Republicans, 19% Independent and 5% Libertarian.
The most common hobbies for volunteers Figure 5.1-6. Age distribution of ARM volunteers were hiking (57%), camping (51%), wildlife identification (51%), fishing (48%), boating (39%), and photography (38%). Most other popular hobbies and sports were outdoor in nature. Overall, 76 hobbies, sports and interests were listed by the volunteers.
Clearly, the ARM volunteers were a diverse group, but they all had the aim of helping protect Massachusetts surface waters against acid rain. Most had invested many years in the effort. How did they feel about their involvement? Practically all seemed to be very pleased with the results of their effort; 94% considered the effort worthwhile, 0.3% considered it not worthwhile, and 6% were unsure.
The ARM Project was involved with participants in the acid rain controversy at all levels: federal, state, and industry. Volunteers were asked how satisfied they were with efforts at these levels to address the acid rain problem (Figure 5.1-7). For our own interest, we asked about their perception of our effort at UMASS. We asked them to use one of 5 categories ranging from very dissatisfied to very satisfied for each and then converted to the familiar A+ to F range used in schools. Volunteers were most distressed with the response of the federal government as of 1989 and gave it a grade of D-. Industry came next with a grade of D. Massachusetts government fared
5.4 better, but not as well as we expected with a grade of only D+. The volunteers were clearly quite critical, because even the ARM Project did not fare so well. Volunteers gave UMASS researchers only a B+ for their efforts. Subjectively, however, we detected an interesting phenomenon. The ARM Project had tried diligently since 1983 to downplay the organizations involved so that no organization would feel slighted for its efforts. We seem to have become a victim of our diligence. That is, many Figure 5.1-7. Volunteer satisfaction with effort by groups to volunteers did not realize that the solve acid rain problem. project was coordinated by UMASS researchers or chose to consider us in a different category, apart from other UMASS researchers. We based that hypothesis on the fact that 97% of the volunteers considered their effort for ARM as worthwhile, the many written comments commending ARM, and the many responses stating that ARM volunteers did not know what UMASS was doing in acid rain research. However, it was nearly as likely that volunteers thought the ARM project would singlehandedly solve the acid rain problem, perhaps not understanding that the Project was contributing a substantial part, but only part, of the needed information. A third possibility, and one of at least equal likelihood, was that, despite our efforts, we had not been perfect. The message for us was to keep trying but try harder. Volunteers did their own part in helping to dis- seminate information about the ARM Project. Seventy percent dis- cussed their efforts with their friends, 29% with environmental organiza- tions, 24% with government officials, 12% gave public talks on the subject, 10% talked to school groups, 9% wrote letters to the Figure 5.1-8. Volunteer promotion of ARM Project and solutions to acid rain editor, and 7% used problem. other methods (Figure 5.1.8).
5.5 6.0 Conclusions
The Acid Rain Monitoring Project was a statewide, nearly comprehensive volunteer effort to chemically sample the streams and lakes of Massachusetts to determine their sensitivity to acid deposition. It consisted of three phases: I (1983-84), II (1984-85) and III (1986-93). The first two provided a nearly complete picture of surface water sensitivity in the state and the seasonal variation in pH and ANC and used as many as 1000 citizen volunteers to collect and help analyze more than 40,000 samples from 2444 lakes and 1670 streams, respectively 87% and 69% of the named lakes and streams in the state. Phase III monitored a representative 453 randomly selected and 119 special interest lakes and streams for eight successive years (1985-1993) with approximately 300 volunteers. Interestingly, the Project was almost never limited by the willingness of volunteers to contribute time and effort. Usually, the limits were set by the capacity of the central lab and, infrequently, there were short-term problems in replacing local coordinators and local labs.
The Project’s network of volunteers was created with the assistance of existing environmental and sportsmen’s groups, notably Trout Unlimited and Massachusetts Audubon, but the volunteers represented a broad cross-section of the state with respect to age, sex, income and affiliation. The thousand or more volunteers participating in the project were profiled through a questionnaire distributed in 1989. Of the 40% responding, roughly 50% were original participants, 69% were male, 79% were married, 70% had a college degree, 63% lived in small towns, 54% were native to Massachusetts, and 75% had household incomes between $20,000 and $75,000. In total, they listed 76 hobbies, sports and interests. They also listed membership in 69 civic organizations and 20 different religions or religious beliefs. Ninety-seven percent considered their efforts on the project to be worthwhile.
The network organization was hierarchical for ease in communication; each volunteer had a local contact who had a county contact. The Water Resources Research Center at the University of Massachusetts provided central communication, quality control, chemical analyses, data management, and most media relations. Contributions of time and facilities always exceeded the contractual cost of the project. The role of the Center was critical in maintaining the project. When funding for the Center’s role in the project ceased in 1993, volunteer activity also ceased even though the volunteers had sufficient equipment and supplies to maintain pH and ANC sampling for several sampling periods and were encouraged to do so. There are few, if any, instances where large volunteer networks work without some level of paid staff support, but many assume that “volunteer” translates to “free.“ This assumption often undermines the creative harnessing of the potential of volunteers.
The contributed effort of volunteer sample collectors and laboratories combined with contributions by the University for central lab analyses and other facilities kept project costs modest. The ten years of the project contract cost $1,904,300 with contributions always exceeding annual contract costs by approximately 50%. For ARM, volunteers increased the effort possible by at least 150% (estimates of volunteer time contributions assume minimum wage).
The creation of the central laboratory for the ARM Project made it possible for Massachusetts to participate in a managed liming program for lakes with critically low ANC and for the Division
6.1 of Fisheries & Wildlife to be awarded a contract to experimentally test stream liming (for which most lab costs were provided as an additional in-kind contribution of the Project).
Quality control and quality assurance were major components of the project. Data quality compares favorably with other major surveys. A variety of procedures, ranging from training workshops to automated computer programs were established to maintain quality and improve efficiency. With a final tally of nearly 39,000 data records, data management became increasingly important throughout the project. Personal computers played a key role in the development of this database. ARM began with no computers and quickly became computer intensive. Future projects have the advantage of using PCS and the Internet to collect and distribute information in more timely fashion.
Results for the nearly comprehensive initial phases of the project show that 5.5% of lakes and streams in Massachusetts are acidified (pH < 5.0 and ANC < 0 µeq/l); 57.4% were sufficiently low in acid neutralizing capacity to be considered threatened by acid deposition (0
While results are more representative of Massachusetts lakes than the Eastern Lakes Survey because of planned exclusions in the latter’s site selections, the differences are relatively small. Comparisons between ARM results and those from other areas and countries suggest that Massachusetts surface waters are comparable to those from other areas of the world that are very sensitive to acid deposition.
Most of the differences in water chemistry can be related to the underlying geology. The western edge of the state is substantially higher in most cations and anions than average while southeastern, north-central and the Berkshire Mountain region are lower. In three of six regions, sulfate levels were twice as high as anticipated from deposition and evapotranspiration. Massachusetts lake and stream sulfate levels were comparable to eastern Canada and other high deposition sites. Streams were higher than lakes in calcium, magnesium, potassium, sulfate, aluminum and silicon dioxide and lower in sodium, chloride, organic acids, iron, and manganese. The combination of relatively high organic acids and base cations may account for the somewhat smaller percentage of acidified surface waters than in other high deposition areas.
The coastal region has a higher proportion of sodium and chloride resulting from sea salt influences. Sodium and chloride levels exceeded the amount expected solely from sea salt. A substudy of 162 randomly selected streams suggested that only 4% of the salt concentration of these streams could be attributed to sea salt. The remainder was highly correlated with the number of road lane miles in each stream’s watershed. Approximately 63% of the variance in stream sodium concentration were explained by the number of lane miles. Class 1 and 2 (interstate highways and major state roads) made the highest contribution with urban streets next. Rural roads contributed relatively little.
6.2 Analyses of long-term trends in pH, ANC and selected ions from 10 years of data on 330 streams and 181 lakes showed a significant but small increase in average pH and increase in ANC. For streams, the median slope after correction for hydrologic variation was +0.021 pH/yr and, for ANC, it was +2.4 µeq/l/yr. For lakes, the median slope (hydrologically corrected) was +0.016 pH/yr and +1.9 µeq/l/yr for ANC. Streams with higher ANCs show a faster rate of recovery than low ANC streams (4.15 µeq/L/yr vs 1.33 µeq/L/yr).
For pH, high ANC streams showed a more gradual increase than low ANC streams (0.018 ñH units/yr vs. 0.026 ñH units/yr. Open (with inlets and outlets) lakes increased alkalinity faster than close lakes (2.8 ueq/l vs. 0.7 ueq/l). Sulfate declined by 1.8 and 1.4 µeq/L/yr for streams and lakes, respectively. There was no trend for base cations in streams but a significant trend for base cations in lakes (+1.5 µeq/L/yr). The latter was counteracted by an increase in chloride of 1.4 µeq/L/yr.
Most lakes and streams exhibited no significant trend for the 10 years of the study. However, 70 of 330 streams showed statistically significant increases in ANC, 11 showed decreases, 43 of 181 lakes increased in ANC while 7 decreased. Most of the streams and all of the lakes exhibiting statistically significant declines occurred in the southeastern portion of the state. Only three lakes and three streams became acidic (dropped below 0.0 ANC and ñH 5.0) during the ten years of the study; five lakes and fifteen streams improved enough that they are no longer acidic.
ARM was an early demonstration of the important role that a carefully designed and managed citizen volunteer effort can have on the protection of the environment. The results of ARM provided impetus for a statewide referendum that led to enaction of the Massachusetts Clean Air Act in 1986 and for strong Massachusetts Congressional delegation support of the Federal Clean Air Act of 1990. With five year implementation periods for both, it is probable that only the Massachusetts Clean Air Act had much effect during the ARM Project. Review of the deposition data for Massachusetts suggests that acid deposition improvements began in 1990. It is reasonable to assume that the full effect of national legislation will continue the trend seen in the ARM Project, but all survey efforts to measure this trend have been discontinued. Unanswered will be the relative contributions of air pollution that locally-derived versus long-range transport, the importance of base cations and the eventual status of surface waters.
ARM also served as an early model for other volunteer water quality monitoring efforts throughout the nation. The degree of attention to quality control in ARM convinced many others that data quality was a fundamental part of the credibility of citizen monitoring.
The ARM project was not without flaws. Staffing and resources were always less than necessary. The relentless pace of sampling resulted in deferral or omission of other tasks. For example, reanalysis of samples in the event of minor analytical error in the central lab was not possible if the lab was to be ready for the next sample collection. As a result, later statistical analyses was often constrained by missing data. Publication of results in peer-reviewed journals was delayed because staff were fully occupied in managing the influx of data. Lack of an equipped central laboratory in the first year resulted in a lost opportunity for more comprehensive water chemistry in phase I. Similarly, the opportunity to also analyze nutrients was missed until the last year of the project. ARM also did not monitor organics or biota.
6.3 The size of the database required high levels of computerization at all stages. When the project began to enter the data analysis stage, the state was in the early stages of developing geographic information system (GIS) data layers critical to the project. Locating sampling sites in a GIS data layer was a critical part, but the level of effort required to review site descriptions, locate sites on topographic maps and digitize them could never be spared. Inexpensive geopositioning equipment did not become commonplace until after the project ended. Similarly, digitizing watershed boundaries for thousands of lakes and streams was not feasible, although several hundred were done. Other basic hydrologic and watershed data did not exist or existed in non-digital form, so in order to include lake type, stream flow, lake retention rate, topography, soils in a data analysis meant collecting or computerizing these data.
Finally, ARM was generally seen only as a means to determine acid deposition effects on surface waters. When that ceased to be a hot media and controversial issue because of the passage of the Federal Clean Air Act, determination of trends and monitoring of other important water quality parameters was not sufficiently critical to elicit legislative and administrative support for continued funding. However, the Division of Fisheries & Wildlife continued to fund the project until efforts to reduce government and cut budgets made the project’s end unavoidable. Ironically, throughout the existence of ARM, it had demonstrated a way to monitor surface waters at significantly less cost and with minimal government involvement.
With the demise of the ARM Project, there is no ongoing, large-scale, statistically representative assessment of surface water quality in Massachusetts. Stream monitoring is restricted to an intensive survey of major tributaries in each major river basin every seven years; lake monitoring is minimal. New volunteer initiatives are expanding but results cannot be considered statistically representative, and the array of water chemistry conducted is limited by the personal funds of the volunteers.
ARM is the largest baseline of surface water inorganic chemistry data ever developed in Massachusetts. It has been used in numerous, unrelated water quality projects conducted by state, federal, private and local organizations. While it was helpful in moving the state and nation toward controls that are reducing acid deposition impacts on the Commonwealth’s surface waters, its total value may continue to grow as it increasingly serves as a baseline for comparison. That and the remembered efforts and commitment of thousands of concerned citizens over ten years has helped us to complete this final report of the ARM Project.
6.4 Section 7.0. Literature Cited
Ackerman, M.T., R.A. Batiuk, and T.M. Beaudoin, 1984. Compilation of lakes, ponds, reservoirs and impoundments relative to the Massachusetts lake classification program. Massachusetts Division of Water Pollution Control #13786-216-30-8-84-CR, Westborough, MA., 204 pp.
Alexander, R.B., R.A. Smith and G.E. Schwartz. 1993. Correction of stream quality trends for the effects of laboratory measurement bias. Water Resources Research 29(11): 3821-3833.
Alley, W.M. 1988. Using exogenous variables in testing for monotonic trends in hydrologic time series. Water Resources Research 24: 1955-1961.
Almer, B., W. Dickson, C. Ekstrom, E. Hornstrom, and U. Miller. 1974. Effects of acidification on Swedish lakes. Ambio, 3(1): 30-36.
Asbury, C.E., F.A. Vertucci, M.D. Mattson, and G.E. Likens. 1989. Acidification of Adirondack lakes. Environmental Science & Technology, 23: 362-365.
Asbury, C.E., M.D. Mattson, F.A. Vertucci and G.E. Likens. 1990. Comment on “Acidification of Adirondack Lakes.” Environmental Science & Technology, 24(3): 387-390
Bailey, R.G., 1983. Delineation of ecosystem regions. Environmental Management, 7(4):365-373.
Baker, L.A., P.R. Kaufman, A.T. Herlihy and J.M. Eilers, 1990. Acidic Deposition: Current status of surface water acid-base chemistry. NAPAP SOS/T Report 9. National Acid Precipitation Assessment Program. Government Printing Office, Washington DC.
Beamish, R.J. 1976. Acidification of lakes in Canada by acid precipitation and the resulting effects on fishes. Water, Air & Soil Pollution, 6: 501-514.
Beamish, R. J. And H.H. Harvey. 1972. Acidification of the La Cloche Mountain lakes, Ontario and resulting fish mortalities. Journal of the Fisheries Research Board of Canada, 29:1131- 1143.
Brakke, D.F., D.H. Landers, and J.M. Eilers, 1988. Chemical and physical characteristics of lakes in the Northeastern United States, Environmental Science & Technology, 22:155-163.
Bricker, O.P. and K.C. Rice, 1989. Acidic deposition to streams. Environmental Science & Technology, 23:379-385.
Calabrese, E.J., and R.W. Tuthill. 1980. The influence of elevated levels of sodium in drinking water on elementary and high school students in Massachusetts. Journal of Environmental Pathology and Toxicology, 4-2,3:151-165.
7.1 Church, M.R., G.D. Bishop and D.L. Cassell. 1995. Maps of regional evapotranspiration and runoff/precipitation ratios in the northeast United States. J. Hydrology, 168: 283-298
Climates of the States, Massachusetts. 1959. U.S. Dept. Of Commerce, Weather Bureau
Conroy, N., K. Hawley, W. Keller, and C. LaFrance. 1976. Influences of the atmosphere on lakes in the Sudbury area. Journal of Great Lakes Research, 2 (Suppl. 1): 146-165.
Crowther, R.A. and H.B.N. Hynes. 1977. The effect of road deicing salt on the drift of stream benthos. Environmental Pollution, 14:113-126.
Cumming, B.F., J.P. Smol, J.C. Kingston, D.F. Charles, H.J.B. Birks, K.E. Camburn, S.S. Dixit, A.J. Uutala and A.R. Selle. 1992. How much acidification has occurred in Adirondack region lakes (New York, USA) since preindustrial times. Can. J. Fish. Aquat. Sci. 49: 128-141.
Cuthbert, I.D. and P. del Giorgio, 1992. Toward a standard method of measuring color in freshwater, Limnology and Oceanography, 37(6):1319-1326.
Demers, C.L. 1992. Effects of road deicing salt on aquatic invertebrates in four Adirondack streams. Pages 245-251. in: F.M. D'Itri (ed.) Chemical deicers and the environment. Lewis Publishers, Boca Raton.
Dickinson, W.E. 1983. Salt products in North America and outlook for the future. p 657-659. in: Sixth International Symposium on Salt. Vol. II. B.C. Schreiber and H.L. Harner (eds). Salt Institute, Alexandria, VA.
Driscoll, C.T., 1991. Northeast overview. in: Acidic Deposition and Aquatic Ecosystems, D.F. Charles, (ed.), 129-132, Springer-Verlag, New York.
Driscoll, C.T. and R.M. Newton, 1985. Chemical characteristics of Adirondack lakes, Environmental Science & Technology, 19:1018-1024.
Driscoll, C.T., R.D. Fuller and W.D. Schecher, 1989. The role of organic acids in the acidification of surface waters in the Eastern U.S. Water, Air, and Soil Pollution, 43:21-40.
Driscoll, C.T., G.E. Likens, L.O. Hedin, J.S. Eaton, and F.H. Bormann. 1989. Changes in the chemistry of surface waters. Environmental Science & Technology, 23: 137-143.
Driscoll, C.T., J. Gallagher, and R.K. Munson, 1990. Evaluation of the chemistry of ALSC lakes. in: Adirondack Lake Survey 1984-87. Adirondack Lake Survey Corp. Ray Brook, NY, 2-1,- 2-116.
Driscoll, C.T., R.M. Newton, C.P. Gubala, J.P. Baker, and S.W. Christensen, 1991. Adirondack mountains. in: Acidic Deposition and Aquatic Ecosystems, D.F. Charles, (ed.), 133-202, Springer-Verlag, New York.
7.2 Driscoll, C.T., and R. Van Dreason. 1993. Seasonal and long-term temporal patterns in the chemistry of Adirondack lakes. Water, Air, and Soil Pollution, 67: 319-344.
Driscoll, C.T., K.M. Postek, W. Kretser and D.J. Raynal. 1995 Long-term trends in the chemistry of precipitation and lake water in the Adirondack Region of New York, USA. Water, Air, and Soil Pollution, 85: 583-588.
Eilers, J.M., G.E. Glass, K.E. Webster, J.A. Rogalla, 1983. Hydrologic control of lake susceptibility to acidification. Canadian. Journal of Fisheries and Aquatic Science, 40:1896-1904.
EPA. 1983. Methods for Chemical Analysis of Water and Wastes., EPA-600/4-79-020., U.S.E.P.A, Cincinnati, OH.
EPA. 1986. Characteristics of Lakes in the Eastern United States. Volume I. Population Descriptions and Physico-Chemical Relationships, EPA/600/4-86/077a. June 1986.
Fuller, R.D., M.B. David, and C.T. Driscoll, 1985. Sulfate adsorption relationships in forested spodosols of the Northeastern USA, Soil Science Society of America Journal, 49:1034-1040.
Gherini, S.A., L. Mok, R.J.M. Hudson, C.W. Chen, and R.A. Goldstein, 1985. The ILWAS model: formulation and application, Water, Air and Soil Pollution, 26:425-459.
Gidley, J.L. 1990. The impact of deicing salts on roadside vegetation on two sites in California. Pages 20-48 in C.R. Goldman and G.J. Malyj (Eds.) The Environmental Impact of Highway Deicing. Institute of Ecology Publication No. 33, University of Calif. Davis, CA.
Glass, G. And O. Loucks. 1980. Impacts of air pollutants on wilderness areas of northern Minnesota. EPA-600/3-80-044. Environmental Research Laboratory, Duluth, Minnesota.
Godfrey, P.J., S. Joyner, E. Goldstein, L. Ross. 1979. The development of PALIS: A ponds and lakes information system for Massachusetts. Publ. No. 108. Massachusetts Water Resources Research Center, University of Massachusetts, Amherst, MA.
Godfrey, P.J., A. Ruby III, and O.T. Zajicek, 1985. The Massachusetts Acid Rain Monitoring Project. Phase I., 228pp. Water Resources Research Center, University of Massachusetts, Amherst, MA.
Griffith, G.E., J.M. Omernik, S.M. Pierson and C.W. Kiilsgaard. 1994. The Massachusetts Ecological Regions Project, Publ. No. 17587-74-70-6/94-DEP. Massachusetts Department of Environmental Protection, Boston, MA.
Haines, T.A. and J. Akielaszek. 1983. A regional survey of chemistry of headwater lakes and streams in New England: Vulnerability to acidification. U.S. Fish and Wildlife Service, Eastern Energy and Land Use Team, FWS/OBS-80/40.15, 141 pp.
7.3 Halliwell, D.B., W.A. Kimball, and A.J. Screpetis. 1982. Massachusetts stream classification program. Part I: Inventory of rivers and streams. Publ. No. 13002-138-72-10-82-CR Massachusetts Division of Water Pollution Control, Westboro, Mass.
Harper, C.R., W.J. Goetz, and C.E. Willis. 1992. Groundwater protection in mixed land-use aquifers. Environmental Management, 16(6):777-783.
Hedin, L.O., G.E. Likens, and F.H. Bormann. 1987. Decrease in precipitation acidity resulting from = decreased SO4 concentration. Nature, 325: 244-246.
Hedin, L.O., L. Granat, G.E. Likens, T.A. Buishand, J.N. Galloway, T.J. Butler, and H. Rodhe. 1994. Steep declines in atmospheric base cations in regions of Europe and North America. Nature, 367: 351-354.
Hendrey, G.R., J.N. Galloway, S.A. Norton and C.L. Schofield, 1980. Geological and hydrochemical sensitivity of the eastern United States to acid precipitation. EPA-600/3-80- 024, 110pp, U.S.E.P.A., Corvallis.
Herlihy, A.T., P.R. Kaufmann, and M.E. Mitch. 1991. Stream chemistry in the eastern United States 2. Current sources of acidity and low acid-neutralizing capacity streams. Water Resources Research, 27: 629-642.
Hillman, D.C., J.F. Potter, and S.J. Simon, 1986. National surface water survey eastern lake survey (phase I - synoptic chemistry) analytical methods manual., 219pp. EPA/600/4-86/009, Las Vegas, NV.
Hirsch, R.M., R.B. Alexander, and R.A. Smith. 1991. Selection of methods for the detection and estimation of trends in water quality. Water Resources Research, 27: 803.
Hirsch, R.M., J.R. Slack, and R.A. Smith. 1982. Techniques for trend analysis for monthly water quality data. Water Resources Research, 18: 107-121.
Hirsch, R.M. and J.R. Slack. 1984. A nonparametric trend test for seasonal data with serial dependence. Water Resources Research, 20: 727-732.
Hoffman, R.W., C.R. Goldman, S. Paulson, and G.R. Winters. 1981. Aquatic impacts of deicing salts in the central Sierra Nevada Mountains, California. Water Resources Bulletin, 17(2):280- 285.
Huling, E.E. and T.C. Hollocher. 1972. Groundwater contamination by road salt: Steady-state concentrations in East Central Massachusetts. Science, 176:288-290.
Johnson, N.M., G.E. Likens, F.H. Bormann, D.W. Fisher, and R.S. Pierce. 1969. A working model for the variation in stream water chemistry at the Hubbard Brook Experimental Forest, New Hampshire. Water Resources Research, 5: 1353-1363.
7.4 Johnson, C.B., T.J. Sullivan, and D.J. Blick, 1989. Defining regional populations of lakes for the assessment of surface water quality, Water Resources Bulletin, 25(3),565-572.
Judd, J.H. 1970. Lake stratification caused by runoff from street deicing. Water Research, 4:521- 532.
Kahl, J.S., S.A. Norton, C.S. Cronin, I.J. Fernandez, L.C. Bacon, and T.A. Haines, 1991. Maine. in: Acidic Deposition and Aquatic Ecosystems, D.F. Charles, (ed.), 203-235. Springer-Verlag, New York.
Kahl, J.S., T.A. Haines, S.A. Norton, and R.B. Davis. 1993. Recent trends in the acid-base status of surface waters in Maine, USA. Water, Air, and Soil Pollution, 67: 281-300.
Kanciruk, P., J.M. Eilers, R.A. McCord, D.H. Landers, D.F. Brakke and R.A. Linthurst, 1986. Characteristics of lakes in the eastern United States. Volume III. Data compendium of site characteristics and chemical variables. 439pp. EPA/600/4-86/007c, U.S.E.P.A., Washington,DC.
Katzenstein, A.W. 1985. Letters. Science, 228: 390-391.
Kretser, W., J. Gallagher, and J. Nicolette, 1989. Adirondacks lakes survey 1984-1987., 391pp. Adirondack Lakes Survey Corp., Ray Brook, NY.
Khan, M.A. and T. Liang. 1989. Mapping pesticide contamination potential. Environmental Management, 13(2):233-242.
Krug. W.R., W.A. Gebert, D.J. Graczyk, D.L. Stevens Jr., B.P. Rochelle, and M.R. Church. 1990. Map of mean annual runoff for the Northeastern, Southeastern, and Mid-Atlantic United States, water years 1951-80. U.S. Geological Survey, Water-Resources Investigations Report 88-4094.
Leiser, A.T., and S.A. John. 1990. Evaluation of the effects of calcium magnesium acetate on selected plant species. Pages 49-96 in: C.R. Goldman and G.J. Malyj (eds.) The environmental impact of highway deicing. Institute of Ecology Publication No. 33, University of Calif. Davis, CA.
Lettenmaier, D.P., E.R. Hooper, C. Wagoner, and K. Faris. 1991. Trends in Stream Quality in the Continental United States, 1978-1987. Water Resources Research, 27(3): 327-339.
Likens, G.E. 1989. Some aspects of air pollution effects on terrestrial ecosystems and prospects for the future. Ambio, 18(3): 172-178.
Likens, G.E., F.H. Bormann, R.S. Pierce, J.S. Eaton, N.M. Johnson. 1977. Biogeochemistry of a forested ecosystem. Springer-Verlag, New York. 146pp.
7.5 Lillie, R. And J. Mason. 1980. PH and alkalilinity of Wisconsin lakes -- a report to the acid deposition task force. Wisconsin Dept. of Natural Resources Report.
Linthurst, R.A., D.H. Landers, J.M. Eilers, D.F. Brakke, W.S. Overton, E.P. Meier, and R.E. Crowe, 1986. Characteristics of lakes in the eastern united states. Volume 1. Population descriptions and physico-chemical relationships. 136pp. EPA/600/4-86/007a, U.S.E.P.A., Washington,DC.
Loftis, J.C., R.C. Ward, R.D. Phillips, and C.H. Taylor. 1989. An evaluation of trend detectiion techniques for use in water quality monitoring programs. EPA/600/3-89/037.
Malmer, N. 1975. Inventering om sjoars forrsurning (Inventories of lake acidification). Statens Naturvardsverk, Solna, Sweden. PM 676. (In Swedish, English summary).
Mass. D.P.W., 1989. Draft Generic environmental impact report of the snow and ice control program., Massachusetts Dept. Public Works., Boston, MA.
Mattson, M.D., P.J. Godfrey, M.F. Walk, P.A. Kerr, and O.T. Zajicek. 1992. Regional chemistry of lakes in Massachusetts. Water Resources Bulletin, 28(6):1045-1056.
Mattson, M.D., M.-F. Walk, P.A. Kerr, A.M. Slepski, O.T. Zajicek, and P.J. Godfrey 1994. Quality assurance testing for a large scale volunteer monitoring program: the Acid Rain Monitoring Project. Lake and Reservoir Management, 9(1):10-13.
Molles, M.C. Jr. 1980. Effects of road salting on stream invertebrate communities. Eisehower Consortium Bulletin, U.S.D.A. Forest Service, Fort Collins, CO, 10:1-9.
Murdoch, P.S., and J.L. Stoddard. 1993. Chemical characteristics and temporal trends in eight streams of the Catskill Mountains, New York. Water, Air, and Soil Pollution, 67: 367-395.
Neter, J. and W. Wasserman. 1974. Applied linear statistical models. Richard D. Irwin, Inc., Homewood, IL. 842pp.
Newell, A.D. 1993. Inter-regional comparison of patterns and trends in surface water acidification across the United States. Water, Air, and Soil Pollution, 67: 257-280.
Norton, S.A. , J.J. Akielaszek, T.A. Haines, K.L. Stromborg, and J.R. Longcore. 1982. Bedrock geologic control of sensitivity of aquatic systems in the United States to acid deposition. Manuscript.
Noss, R.R. 1989. Recharge area land use and well water quality. The Environmental Institute publication number 89-2, University of Massachusetts, Amherst, MA. 61pp.
NRC, 1986. Acid Deposition: Long-term Trends. National Research Council, National Academy Press, Washington, D.C. 506 pp.
7.6 NRC, 1991. Highway Deicing. Comparing salt and calcium magnesium acetate. Transportation Research Board Special Report 235, National Research Council, National Academy Press, Washington, D.C. 170pp.
Omernik, J.M. and C.F. Powers, 1983. Total alkalinity of surface waters - a national map., Ann. Assoc. Am. Geog. 73:133-136.
Parnell, R.A., 1983. Weathering processes and pickeringite formation in a sulfidic schist: A consideration in acid precipitation neutralization studies, Environmental Geology, 4:209-215.
Pfeiffer, M.H. and P.J. Festa. 1980. Acidity status of lakes in the Adirondack region of New York in relation to fish resources. N.Y. State Dept. Environmental Conservation, FW-P168(10/80), 36 pp. + appendices.
Placet, M., R.E. Battye, F.C. Fedsenfeld and G.W. Bassett, 1990. NAPAP Report 1 Emissions involved in acidic deposition processes. National Acid Precipitation Assessment Program, Government Printing Office, Washington, DC.
Pollock, S.J. 1988. Highway deicing salt contamination problems and solutions in Massachusetts. Pages 353-370 in Proceedings of the focus conference on the eastern regional ground water issues. National Water Well Assoc., Dublin, OH.
Pollock, S.J. 1990. Mitigating highway deicing salt contamination of private water supplies in Massachusetts. Pages 157-170 in: C.R. Goldman and G.J. Malyj (eds.) The environmental impact of highway deicing. Institute of Ecology Publication No. 33, University of Calif. Davis, CA.
Rebsdorf, A. 1980. Acidification of Danish soft-water lakes. in: Ecological Impact of Acid Precipitation. Proc., Int. Conf. Ecol. Impact Acid Precip., D. Drablos and A. Tollan (eds.) Sandefjord, Norway, Maarch 11-14, 1980. Oslo-Aas, Norway. Pp. 238-239.
SAS. 1987. SAS/STAT Guide for Personal Computers, Version 6 Edition. Cary, NC, SAS Institute Inc. 1028pp.
Scheider, W., W. Snyder and B. Clark. 1979. Deposition of nutrients and major ions by precipitation in south-central Ontario. Water, Air & Soil Pollution, 12: 171-185.
Schindler, D.W. 1988. Effect of acid rain on freshwater ecosystems. Science, 239: 149-157.
Sisterson, D.L., V.C. Bowersox, A.R. Olsen, T.P. Meyers and R.J. Vong, 1990. NAPAP Report 6, Deposition monitoring: Methods and results. National Acid Precipitation Assessment Program, Government Printing Office, Washington, DC.
Smith, R.A., R.B. Alexander, and M.G. Wolman. 1987. Water-quality trends in the Nation's rivers. Science, 235: 1607-1615.
7.7 Snedecor, G.W. and W.G. Cochran, 1980. Statistical Methods, 7th Ed. Iowa State Univ. Press, Ames. 507p.
Stoddard, J.L. 1991. Trends in Catskill stream water quality: evidence from historical data. Water Resources Research, 27(11): 2855-2864.
Stoddard, J.L., and J.H. Kellogg. 1993. Trends and patterns in lake acidification in the state of Vermont: Evidence from the long-term monitoring project. Water, Air, and Soil Pollution, 67: 301-317.
Stone, B.D., 1982. The Massachusetts state surficial geologic map. in: Geotechnology in Massachusetts. O.C. Farquhar (ed.), pp. 11-27. Graduate School, University of Massachusetts, Amherst, MA.
Stumm, W. and J.J. Morgan, 1981. Aquatic Chemistry, 2nd. ed. John Wiley & Sons, New York. 780p.
Sullivan, T.J., 1990. Historical changes in surface water acid-base chemistry in response to acidic deposition. Report 11, in: NAPAP State of Science and Technology, National Acid Precipitation Assessment Program., 181pp + append. NAPAP, Washington, DC.
Sullivan, T.J., C.T. Driscoll, J.M. Eilers, and D.H. Landers, 1988. Evaluation of the role of sea salt inputs in the long-term acidification of coastal New England lakes. Environmental Science & Technolology, 22:185-190.
Sullivan, T.J., C.T. Driscoll, S.A. Gherini, R.K. Munson, R.B. Cook, D.F. Charles and C.P. Yatsko, 1989. Influence of aqueous aluminum and organic acids on measurement of acid neutralizing capacity in surface waters. Nature, 338:408-410.
Sullivan, T.J., D.F. Charles, J.P. Smol, B.F. Cumming, A.R. Selle, D.R. Thomas, J.A. Bernert, and S.S. Dixit. 1990a. Quantification of changes in lakewater chemistry in response to acidic deposition. Nature, 345: 54-58.
Sullivan, T.J., D.L. Kugler, M.J. Small, C.B. Johnson, D.H. Landers, B.J. Rosenbaum, W.S. Overton, W.A. Kretser and J. Gallagher, 1990b. Variation in Adirondack, New York, lakewater chemistry as function of surface area, Water Resources Bulletin, 26(1):167-176.
Trombley, T.J., 1992. Quality of water from public-supply wells in Massachusetts, 1975-86. USGS Water-Resources Investigation Report 91-4129, 63pp. Boston, MA.
U.S. Geological Survey. 1985. U.S. Geological Survey Water-Data Reports MA-RI-83-1.
U.S. Geological Survey. 1986. U.S. Geological Survey Water-Data Reports MA-RI-84-1.
U.S. Geological Survey. 1987. U.S. Geological Survey Water-Data Reports MA-RI-85-1.
7.8 U.S. Geological Survey. 1988. U.S. Geological Survey Water-Data Reports MA-RI-86-1.
U.S. Geological Survey. 1989. U.S. Geological Survey Water-Data Reports MA-RI-87-1.
U.S. Geological Survey. 1990. U.S. Geological Survey Water-Data Reports MA-RI-88-1.
U.S. Geological Survey. 1990. U.S. Geological Survey Water-Data Reports MA-RI-89-1.
U.S. Geological Survey. 1991. U.S. Geological Survey Water-Data Reports MA-RI-90-1.
U.S. Geological Survey. 1992. U.S. Geological Survey Water-Data Reports MA-RI-91-1.
U.S. Geological Survey. 1993. U.S. Geological Survey Water-Data Reports MA-RI-92-1.
U.S. Geological Survey. 1994. U.S. Geological Survey Water-Data Reports MA-RI-93-1. van Belle, G., and J.P. Hughes. 1984. Nonparametric tests for trend in water quality. Water Resources Research, 20: 127-136.
Veneman, P.L.M. 1984. Evaluation of the Buffering Capacity of Massachusetts Soils to Negotiate the Impact of Acid Deposition on Water Quality. Completion Report to the Massachusetts Division of Air Quality Control. December, 1984.
Walk, M.F., P.J. Godfrey, A.III Ruby, O.T. Zajicek, and M. Mattson, 1992. Acidity status of surface waters in Massachusetts, Water, Air, and Soil Pollution, 63(3-4):237-252.
Watt, W.D., D. Scott, and S. Ray. 1979. Acidification and other chemical changes in Halifax County lakes after 21 years. Limnology and Oceanography, 24: 1154-1161.
Webster, K.E., P.L. Brezonik, and B.J. Holdhusen. 1993. Temporal trends in low alkalinity lakes of the Upper Midwest. (1983-1989). Water, Air, and Soil Pollution, 67: 397-414.
Wolf, F.M. 1986. Meta-analysis. Quantititative methods of research synthesis. Sage University Paper. Series No. 59., Beverly Hills.
Wright, R.F. 1977. Historical changes in the pH of 128 lakes in southern Norway and 130 lakes in southern Sweden over the period 1923-1976. SNSF-project, TN34/77. 71 pp.
Wright, R.F. and E.T. Gjessing. 1976. Acid precipitation: Changes in the chemical composition of lakes. Ambio, 5:219-223.
Wright, R.F. and E. Snekvik. 1978. Acid precipitation chemistry and fish populations in 700 lakes in southernmost Norway. Ver. Internat. Verein. Limnol., 20:765-775.
7.9 Wright, R.F., N. Conroy, W.T. Dickson, R. Harriman, A. Henriksen, and C.L. Schofield. 1980. Acidified lake districts of the world: A comparison of water chemistry of lakes in southern Norway, souther Sweden, southwestern Scotland, the Adirondack Mountains of New York, and south eastern Ontario. in: Ecological Impact of Acid Precipitation. Proc., Int. Conf. Ecol. Impact Acid Precip., D. Drablos and A. Tollan, eds. Sandefjord, Norway, Maarch 11-14, 1980. Oslo-Aas, Norway. Pp. 377-379.
Wright, R.F., B.J. Cosby, G.M. Hornberger and J.N. Galloway. 1986. Comparison of paleolimnological with MAGIC model reconstruction of water acidification. Water, Air & Soil Pollution, 30: 367.
Wright, R.F., S.A. Norton, D.F. Brakke, and T. Frogner, 1988. Experimental verification of episodic acidification of freshwaters by sea salts. Nature, 334:422-424.
Yu, Y.S., S. Zou, and D. Whittemore. 1993. Non-parametric trend analysis of water quality data of rivers in Kansas Journal of Hydrology, 150: 61-80.
Yuretich, R.F. and G.L. Batchelder. 1988. Hydrogeochemical cycling and chemical denudation in the Fort River watershed, central Massachusetts: An appraisal of mass-balance studies. Water Resources Research, 24(1):105-114.
Zen, E., 1983. Bedrock geologic map of Massachusetts., 2 sheets, scale 1:250000. Massachusetts Department of Public Works, Boston, MA.
Zimmerman, A.P. and H.H. Harvey. 1979. Sensitivity to acidification of waters of Ontario and neighboring states. Final report to Ontario Hydro, University of Toronto, Ontario. Pp. 118.
7.10 7.11