Cheung, W.W.L, T.J. Pitcher and D. Pauly Using an Expert System To

Chapter 2

USING AN EXPERT SYSTEM TO EVALUATE VULNERABILITIES AND CONSERVATION RISK OF MARINE FISHES FROM FISHING

William W. L. Cheung,* Tony J. Pitcher andDaniel Pauly Fisheries Centre, The University ofBritish Columbia, Canada

ABSTRACT

Today, fishing is the only large-scale wildlife hunting activity. Unsurprisingly, it has contributed to the depletion and extirpation of numerous marine fish populations. We develop a fuzzy logic expert system to estimate the risk of depletion of marine fish populations from fishing. The frrst component of this expert system predicts the intrinsic vulnerability of marine fishes (i.e., fishes' inherent ability to withstand fishing mortality) from simple parameters of life history and ecology. The second component infers exploitation status of fishes from temporal features of their catch time-series. Combining the predicted intrinsic vulnerability with the inferred exploitation status, the expert system predicts the relative depletion risk of marine fishes. Heuristic rules relating the various input attributes with depletion risk are derived from published literature and expert opinions. Using published datasets and simulated data from numerical population dynamic models, we show that the expert system provides robust proxies of intrinsic vulnerability and relative depletion risk from fishing. Based on the validated system, we estimate relative depletion risk for 460 exploited marine fish species for which long catch time-series exist. Extrapolating our findings to all marine fishes, we estimate that a large proportion of marine fishes are facing high depletion risk from fishing. Our work suggests that marine fish may have a similar conservation risk from large-scale human activities to those of other vertebrates, and thus are candidates for a similar level of

orresponding author: Fisheries Centre Aquatic Ecosystems Research Laboratory (AERL), 2202 Main Mall, The University of Briti h Columbia, Vancouver, B, anada V6f 1Z4; Tel: +1 (604) 822-2731; fax: +1 (604) 822-8934; Email: w.cheunglsheries.ubc.ca 26 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly

conservation attention. The tudy highlights the potential applications ofan expert system in the study ofthe conservation biology of fishes.

I TRODUCTIO

Fi bing as a Conservation Threat to Marine Fi he

Fishing exploitation of the oceans increased rapidly in recent decades (Pauly e/ al. 2002). Based on fishery statistics collated by the United ations Food and Agriculture Organization (FAO), total reported landings (catches landed in ports) from the sea increased from less than 20 to over 82 million tonnes from 1950 to the 2000s. Ifdiscards (catches that are thrown back to the sea) and illegal, unreported and unregulated catches are included, estimated global marine catches peaked at almost 150 million tonnes in the late 1980s, after which they declined slowly (Pauly e/ at. 2002, Zeller & Pauly 2005). In 2003, about one-quarter of the stocks monitored by FAO were said to be underexploited or moderately exploited (3 percent and 21 percent, respectively), 52 percent were fully-exploited (production close to their maximum sustainable limits), while approximately one-quarter were overexploited, depleted or recovering from depletion (16 percent, 7 percent and I percent, respectively). These represented an increase in the proportion of overexploited and depleted stocks from about 10 percent in the mid-1970s to close to 25 percent in early 2003 (FAO 2004). Collapses of major fishery stocks and the endangerment of a number of marine fishes suggest that marine species are vulnerable to extreme depletions, or even extinction, resulting directly or indirectly from fishing (Roberts & Hawkins 1999, Powles e/ al. 2000, Dulvy e/ al. 2003, Sadovy & Cheung 2003). While the majority of the world's fisheries resources have been fully- to over-exploited (Pauly e/ al. 2002, Hilborn e/ al. 2004) fishing is considered to be a major conservation threat to marine fishes (Reynolds e/ al. 200 I, Sadovy 200 I, Dulvy e/ al. 2003, Reynolds e/ al. 2005). Parallel to the increa ing scale of fishing, the abundance of many marine fishes has declined greatly throughout the world over the past five decades. In the North Atlantic, predatory fishes have declined by two thirds since the 1950s (Christensen et al. 2003) while the top fish predators such as large sharks and tuna have declined by over 90% globally (Myers & Worm 2003). Over the past 50 years, breeding populations of 98 populations of marine fishes from around the world declined by a median of 65%, with over 28 populations declining by more than 80% (Hutchings & Reynolds 2004, Reynolds e/ al. 2005). Although it appears that fish stocks recovered after fishing pressure had been eased (Russ & Alcala 1996, Myers & Worm 2005), many stocks have shown little or no sign of recovery after up to 15 year, while those that have recovered are mainly c1upeoid fishes which are suggested to be intrinsically more resilient (Hutchings 2000, Hutchings & Reynolds 2004). The northern cod is a classic case of the lack of recovery after severe depletion (Shelton & Healey 1999, Hutchings & Reynolds 2004). Commercially-important species can be fished down to a vulnerable level because of their economic value, e.g., Chinese bahaba (Bahaba /aipingensis, Sciaenidae) (Sadovy & Cheung 2003) Southern bluefm tuna (Thunnus maccoyii, Scombridae) (Hayes 1997). However, species with little or no commercial value are not safe from the threats of fishing, since non-targeted species may be threatened through bycatch, e.g. Common skate, Raja balis, Rajiidae, (Brander 1981).

• Using an Expert System to Evaluate Vulnerabilities and Conservation Risk... 27

Moreover, fishing activities can create large disturbances and damage to benthic habitats (Jennings et al. 200 I, Kaiser et al. 2002, Kaiser et al. 2003). Declines and extinctions can be associated with loss of essential habitats critical to complete the life cycle of the species (McDowall 1992, Watling & Norse 1998). Fishing may also cause loss of biodiversity (Law 2000) which directly or indirectly affect the functioning of the ecosystem (Loreau et al. 200 I, Worm & Duffy 2003, Worm et al. 2006).

Approaches to Assessing Conservation Status of Fishes

Despite the wide range of impacts from fishing on the marine ecosystems and the potential vulnerability of marine fishes to fishing, our understanding of the conservation status of marine fishes - the largest group of vertebrates in the sea - lags behind the increasing rate of their utilization. Compared to other vertebrate groups, the proportion of fish species that have been assessed with the IUCN Red List criteria is very low (Figure I). The IUCN Red List criteria are widely accepted as the authority for determining extinction risk of animals and plants (Rodrigues et al. 2006), although their validity for marine fishes had been questioned (Powles et al. 2000, Punt 2000, Reynolds et al. 2005). If we consider marine fishes only, less than 7% of the 15,723 extant species have been assessed using the Red List criteria (Baillie et al. 2004). Among this 7%, over 35% of the assessed species were considered 'data deficient', i.e., at the time of the assessment, there were not enough data to determine the status ofthe species. If the current rate of Red List assessment is extrapolated, only about 20% of extant marine fishes would have been assessed by year 2020 (Figure 2). To complete the assessment of half of the marine fishes, the current rate of IUC Red List assessment would have to be tripled. However, the Convention on Biological Diversity has set a "2010 Biodiversity Target" which has a mission statement: "to achieve by 2010 a significant reduction of the current rate ofbiodiversity loss at the global. regional and national level as a contribution to poverty alleviation and to the benefit ofall life on earth. ' (Decision VII26, the Convention on Biological Diversity). To achieve such a target, species that are threatened or likely to be threatened should be identified. Given the current rate of Red List assessment for marine fishes, this target seems overly ambitious. The IUCN and its Species Specialist Commission realized the pressing need to increase assessment coverage in fishes and were devising strategies to increase their rate of assessments (Sadovy, J., chair of the IUCN Specialist Group of Groupers and Wrasses, pers. comm.). The marine fishes that need to be assessed are numerous, while population data for the majority are lacking. Data limitations restrict the application of conventional assessment approaches to the full spectrum of species, which require quantitative understanding of population dynamics (Dulvy et al. 2004). Quantitative data on fisheries and population status of exploited species are costly to collect (Reynolds et al. 2002; Dulvy et al. 2003). These data are available only for a small number of marine fishes, mainly commercially-targeted species in developed countries. The problem of data-limitation is particularly serious in tropical, developing country fisheries where specie diversity is high but resources for monitoring are low (Jennings & Polunin 1996, Johannes 1998). 28 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly

..... 100 ~ ~ "0 Q) III 80 III Q) III III III 60 III Q) 'u Q) Co 40 ..III C ~ 20 w 0 Mammals Birds Amphibians Reptiles Fishes

Figure I. Proportion ofextant vertebrates that have been assessed under the IUC Red List of ndangered pecies (Baillie el aI. 2004). The black and grey bars represent marine and freshwater fishes, re pectively.

8000

...Q,I .Q (a) Same § 2000 z ..' t~_·;:' _ ... - o J~~~~~:..:...-....--.------.-----.-- 1995 2000 2005 2010 2015 2020 Year

Figure 2. umber of marine fishes that have been as essed under the IUCN Red List since 1998 (solid line) and the projected number of as essed marine fishes to 2020 (dotted lines) assuming the rate of assessment (number of species per year) (a) remains the same as the average between 2002 and 2005, (b) is 2-times the average between 2002 and 2005, and (c) is 3-time the average between 2002 and 2005.

To rapidly assess conservation status and hort-list pnonty species for detailed assessment, 'rule-of-thumb' approaches were proposed (Fagan el at. 200 I, Reynolds et at. 200 I, Dulvy et af. 2003, Dulvy et af. 2004). Such approaches use easily-obtainable information to approximately identify vulnerable or "priority" species that are in need of immediate conservation attention. The approaches are especially useful if their applications are combined with large databases, for instance, FishBase (www.lishbase.org) - a database that contains biological information of mo t fishes (Froese & Pauly 2004) and the Sea Around Using an Expert System to Evaluate Vulnerabilities and Conservation Risk... 29

U. Project database (www.seaaroundus.org), which presents a wide range of fisheries data ranging from spatially disaggregated catch data to prices of fishery catches (Watson et al. 2004). Results can also help focus longer tenn research on the priority species so that data could be made available for more accurate extinction.risk assessments (Figure 3). As life history and ecology detennine, at least in part, how fish populations respond to exploitation, these attributes could be used to develop 'rule-of-thumb' proxies to evaluate the intrinsic vulnerability of marine fishes to fishing (Jennings el al. 1998, Jennings et al. 1999, Reynolds el af. 200 I). Here, vulnerability of fishes is dermed as a combination of intrinsic vulnerability and exposure to some external threatening factors. 1ntrinsic vulnerability to fishing is the inherent capacity to respond to fishing. It relates to the fish's maximum rate of population growth and strength of density dependence (Cheung el af. 2005). The intrinsic factors act synergistically with external threatening factors, such as fisheries exploitations, climate change or coastal development, to affect the susceptibility of species or populations to risk of population depletion. For instance, when species with high intrinsic vulnerability to exploitation are being intensively fished, they are likely to have high depletion risk. Proxies of intrinsic vulnerability and risk of population depletion resulted from their interactions with the external threatening factors could be detennined from easily obtainable infonnation through these 'rules-of-thumb'. ------, I IMMEDIATE I Easily TERM obtainable data

'Rules-of-thumb" D

Priority-list

I------u------LO------GER I I I I TERM I Detailed data collection and assessments

Figure 3. Schematic presentations of the proposed framework to identify conservation tatus of marine fishes. 30 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly

Combining 'Rules-of-thumb' with a Fuzzy Logic Expert Sy tern

A fuzzy logic expert system can be useful in combining the rules-of-thumb' to identify the relative risk of population depletion of marine fishes to flShing. Fuzzy set theory is particularly useful because vagueness is a crucial aspect of our knowledge of fishes' biological characteristics, and their relationships and interactions with the depletion risk from fishing. For example, we know that large fishes tend to be associated with higher depletion risk. However, it is difficult to provide a clear cut definition of what a 'large fish' is, i.e., to separate large and small body size, and thus high and low intrinsic vulnerability to fishing. Moreover, other biological characteristics may confer low risk on a species despite large size. A fuzzy logic expert system should be useful in representing such vagueness and uncertainty.

STAGEl STAGE 2 Biological data: -Life history -Ecology

Heuristic rules Heuristic rules

Heuristic rules

Figure 4. Schematic diagram of the tructure of a fuzzy expert sy tem to predict depletion ri k of marine lishe to Ii hing.

Fuzzy expert system has been proposed and applied to study fisheries and conservation biology (Saila 1996). The applications range from assessing stock-recruitment relationships Using an Expert System to Evaluate Vulnerabilities and Conservation Risk... 31

(Mackin on el of. 1999 Chen 2001), predicting fish shoaling behaviour (Mackinson 2000) and identifying stock structure of fishes (Zhang 1994). It has also been applied to develop an analytical tool to assess conservation threats (Todd & Burgman 1998, Regan & Colyvan 2000) and assist the ]UCN Red List's sp~cies assessment (Akyakaya et 01. 2000). Fuzzy.logic was proposed to be used to assess extinction risks of different Pacific salmon stocks (Tinch 2000). Here, a fuzzy expert system was constructed to predict the risk of population depletion of marine fishes to fishing (Figure 4). The expert system is composed of two stages: the first stage infers intrinsic vulnerability to fishing while the second stage predicts depletion risk from the intrinsic vulnerability (from first stage) and time-series catch data. Heuristic rules were developed from published literature and expert knowledge. Assuming that depletion risk is a proxy to determine conservation status of marine fishes, the expert system is validated by comparing the predicted depletion risk with other widely used extinction risk classification system. The structure of the first stage (prediction of intrinsic vulnerability) and its validation has been detailed elsewhere (Cheung et at. 2005). Therefore, we only provide a brief summary ofthis component in the following section.

STRUCTURE OF THE FUZZY EXPERT SYSTEM

Stage 1 - Predicting Intrinsic Vulnerability to Fisbing from Life History

]n the first component of the fuzzy expert system, intrinsic vulnerabilities of marine fishes were inferred from easily-obtainable life history and ecological characteristics i.e., feature available from FishBase (Froese and Pauly 2004; www.fishbase.org). The input variables include maximum length, age at first maturity, longevity, von Bertalanffy growth parameter K, natural mortality rate, fecundity, strength of spatial behaviour, and geographic range. These life-history traits affect the responses of fish population to exploitation (Adams 1980, Roff 1984, Kirkwood et at. 1994). Correlations between life history, population regulation and thus vulnerability to fishing are supported by empirical evidence. For instance, meta-analysis using 54 stock-recruitment time-series showed that large-sized, late-maturing fishes bad strong density-dependence when they were at low abundance (thus smaller maximum spawner per spawner), but high equilibrium spawner-per-recruit when they were without exploitation (Goodwin et at. 2006). Analysis including data from other vertebrate groups produced similar conclusions, suggesting that the correlations between life history and population dynamics may be applicable to most vertebrates (Fagan el of. 1999). Also, empirical studie u ing historical abundance data of exploited fish populations find significant correlations between the rate of population declines (a proxy of vulnerability to fishing) and life history parameters such as maximum body size and age at maturity, but not fecundity (Jennings et at. 1998, Jennings et af. 1999). Current evidence suggests that body size is an important factor in determining vulnerability to hunting (Jennings et al. 1998, Jennings et of. 1999, Cardillo & Dromham 200 I, Reynolds et of. 200 I, Gaston & Blackburn 2003, Reynolds et of. 2005). Ecological and behavioural characteristics may also affect fishes' vulnerability to fishing. For instance, species forming large aggregations can be easily targeted by fishers. Thus high 32 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly catches can be sustained under low resource abundance (Hutchings 1996). [n particular, species which form spatially and temporally predictable spawning aggregations are e pecially vulnerable. Fishe may not form such spawning aggregations when adult population is small. Thus, depletion of spawning aggregations may permanently prevent reproduction in these populations (Sala et at. 2003, adovy & Domeier 2005). At the saine time, species with certain reproductive strategie such a hermaproditism or a high level of parental care may also be particularly prone to the effects of fishing (Rowe & Hutchings 2003, Hutchings & Reynolds 2004). Based on the known relationships between the biological attributes and intrinsic vulnerability to fishing, fishes are classified into verbal categories describing the input variables. The degrees of membership to these categories are based on the input fuzzy sets (Figure 5). The outputs include four levels of intrinsic vulnerability: (1) very high vulnerability, (2) high vulnerability, (3) moderate vulnerability and (4) low vulnerability. We assumed the simplest forms of fuzzy membership function: trapezoid and triangular membership functions: which are given by:

Membership =0 ifxSa eq. la

x-a Membership =-- if a

Membership =1 ifbSX9; eq. Ic

d-x Membership =-- if c

where x is the independent variable, and in this case, represents the ratio of annual catch to the maximum catch of the time-series. AII x values between a and d are in the particular fuzzy set; band c are the independent variables with maximum membership. For triangular membership function, band c are equal. The heuristic relationships linking life-history and ecological characteristics to intrinsic vulnerability were developed from published literature, excluding those overwhelmingly disproved by empirical data. For instance, high fecundity had been suggested to be associated with low vulnerability. However, both theoretical and empirical studies lately do not support such relationship (see Cheung et al. 2005). Thus the rules relating high fecundity and low vulnerability are excluded from the system. We assumed equal level of belief (50%) to all rules. That is:

Membershipconc/U'''on = Membership prenllse • CF eq. 2

where CF represents the weighting factor. Membership of the conclusion from each rule was combined using the knowledge accumulation method in Buchanan and Shortliffe (1984):

Membershipe = Membershipe_l + Membership, • (1- Membershipe_l) eg.3 Using an Expert System to Evaluate Vulnerabilities and Conservation Risk... 33

I.V

Q. 0.8 / Q. 0.8 :i: :c f 0.6 e 0.6 CI> \tI .8 s \It. .0 E 0.4 E 0.4 CI> CI> ~ 02 ~ 0.2

0.0 0.0 0 50 100 150 200 0 2 3 4 5 6 7 8 9 /9t at firs maturlty (year) a) MmTunIef9h (all b)

1.0\ll...w 1.0

Q. 0.8 Q. 0.8 :i: :i: l!! 0.6 f 0.6 Ql H Ql H ..Q ..Q E 0.4 E 0.4 Ql Ql ~ 0.2 ~ 0.2

0.0 r- I 0.0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0 0.2 0.4 0.6 0.8 VBGF parameter K(yea(1) Nitural rmrtality rate (}oea(1) c) d) Lw 1.0 1.0

Q. 0.8 Q. 0.8 :c :i: l!! 0.6 r! 0.6 Ql .8 VH ..Q E 0.4 E 0.4 Ql Ql ~ 0.2 ~ 0.2

0.0 0.0 0 20 40 60 0 2(0) 4000 Em> !OOJ Maxirrun age (year) Geographic range (km"~ e) f) 1.0 VI...w

Q. 0.8 :i: f 0.6 Ql ~ 0.4 Ql ~ 0.2

0.0 f-1---''I--\-----.------, o 100 200 ::m 1 g) Fe

Figure 5. Fuzzy sets defining the input life hi tory and ecological characteristics: (a) maximum body length, (b) age at first maturity (Tm), (c) von Bertalanffy growth parameter K, (d) nalural mortality rate 2 (M), (e) maximum age (Tmax), (I) geographic range (km ) (g) annual fecundity (egg or pup female-I year-I), (h) strength of aggregation behaviour (see Appendix 2.1). VLw - very low, Lw - low, NLw not low, M -medium/moderate, H - high, VII - very high, L- large, VL - very large, R- restricted, VR - very restricted, RV - not re tricted, S- mall. A fish species with maximum body length of 68 cm corre ponds to' medium body size" and "Iarge body size" with membership of 0.7 and 0.3 respectively (thre hold value = 0.2). This figure is a redrawn from Cheung et of. (2005). 34 WiJliam W. L. Cheung, Tony J. Pitcher and Daniel Pauly

1.0

0.8 Q. S; ...fIj 0.6 (l) Low Very high .0 E 0.4 Q) :E 0.2

0.0 1 20 40 60 80 100

Intrinsic vulnerability

Figure 6. Output fuzzy sets for the intrinsic vulnerability of marine fishes. The "Low" and "Very high" vulnerabilities are defined by trapezoid membership functions while the "Moderate" and "High' vulnerabilities are defined by triangle member hip functions. Intrinsic vulnerability was scaled arbitrary from I to 100. This figure is a redrawn from Cheung et al. (2005).

where Membership~ is the degree of membership of the conclusion after combining the conclusions from e pieces of rules, and Membership, is the degree of membership of the conclusion of rule i. Trapezoid membership functions were used for the conclusions: very high vulnerability' and the 'low vulnerability' categories, while symmetric triangular membership functions were used for the other two conclusion categories (Figure 6). The intrinsic vulnerability of fishes predicted from the fuzzy expert system was shown to be a valid indicator of vulnerability to fishing and robust to its assumptions (Cheung et al. 2005). The predicted intrinsic vulnerabilities correlate significantly with empirically estimated rate of population declines from a range of different fish assemblages. The goodness-of-fit to empirical data is also highest among other previously suggested indicators of intrinsic vulnerabilities. Moreover, the fuzzy expert system is robust to the assumed confidence factors to the rules.

Stage 2 - Predicting Depletion Risk

This component predicts the relative depletion ri k of exploited marine fish to fishing based on the more readily available catch and life history data. Depletion risk from fishing is defined here as the possibility of a population declines to reach the point of conservation concern because of fishing and the intrinsic properties of the species. Catch time-series are readily available in most fisheries, and are useful in understanding the overall status of a population (Grainger & Garcia 1996, Fiorentini et al. 1997, Caddy 2004, Worm et al. 2006). By definition, a population is under-exploited when its fishery is developing; in such case, catch increases as fishing effort increases (Hilborn & Walters 1992). A fishing effort approaches or exceeds maximum productivity, the population Using an Expert System to Evaluate Vulnerabilities and Conservation Ri k ... 35 becomes over-exploited, and the catch declines, and eventually collapses. A recovery phase may follow if fishing is reduced to a low level (Figure 7). The relationship between catch and population status becomes less tight when there are confounding ecological, environmental, economic and management effects. Firstly, catc.hes can be maintained by spatial changes in fishing effort and targeted sub-population. In such cases, catch may increase as fishing expands spatially (serial depletion), or catchability increases when a population reduces its rang . Also, catches can be reduced by the implementation of more stringent fisheries management policies, while population size may remain roughly constant. Moreover, change in market demand, catch value and costs of fishing may affect the operation of fleets without strong change in population abundance. On the other hand, given that these ecological, environmental and economic data are not available for many targeted fishes and fisheries, we have to rely on catch time-series as an approximate indicator to reflect population status at large spatial scale (Caddy 2004) to identify potentially over-exploited or depleted populations for more detailed analysis.

.c CJ ('J o-

Time

Figure 7. chematic diagram showing the classification of exploitation status of a population based on a catch time-series. Stage I- under-exploited, stage 2 - fully exploited, stage 3 - over-exploited, stage 4 - depleted, stage 5 - recovering.

To reduce the variability of the catch time-series that is resulted from non-fishing factors, the catch time-series were smoothed based on the life-history of the fishes. Firstly, since catch fluctuations caused by environmental variations might mask any trends due to fishing, catch time-series were smoothed with a running average. As populations of smaller-bodied species tend to response faster and more strongly to environmental variability than larger species ( pencer & Collie 1997, Goodwin elof. 2006), the running means were scaled by the inverse of the maximum length (maximum length :'S 30 cm: 9 years running means, maximum length = 30-90 cm: 5 years running mean, maximum length ~ 90: 3 years running mean). Based on the smoothed catcb time-series, each population at each year were categorized into different exploitation status. Firstly, the smoothed catch time-series were re-expressed as the ratio of each year's annual catch to the maximum catch in the time-series. Each data point in the time-series was also classified by the position relative to the maximum catcb in the data-series (i.e., before or after the maximum catch is reached). Based on the relative position in the time-series and the ratio to the maximum catch (Table I), each data point was then 36 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly

e pi ilalion lalu ad pled b the nil d tion d and gri ullure : (I) under-e ploited, (2) full -e ploited, ( 0 er- pi ited ( depl t d, Garcia 1996 (Figure 7). a b data pint ould b long to e eral ric. a h ~ ith an a ociated degree of membe hip calculat d fr m pre-defined memb hip funclion for the categories (Table I. The impl t form f memb hip functi n ' lrap l idal and Iriangular function ere u e

bl 1 at orizali tch tim ri uDder thr I (minimi uod r-

Fuzzy m mbcrship functi nJ

nefer pi il d Full 0.7 -1(1) 0.5-1(0.75) 0.15-1 (0.5-0.75) Deli re m imum T Ida! e pi II Full O. -I I) 0.75-1 (I) II rm imum Triangular e pi ilcd cr- O. --I (0.7 -I) Ii rem imum T z ida! pi iled o er- 0.1-0.7 (0.5) 0.2 -I (0.75) 0.7 -I (I II rm imum riangular e pi iled Depleled 0-0.25 (0.1) 0-0.5 (0.25) 0-0.7 (0.5) All rm imum Trapez ida!

Rec eong atch remained Ich remained After m imum and rapezoidal table/increasing lableJin reasing after c nditi os Ii r for at least 10 'er-e pi ited' r ears years 'coli p ed' cur

U Dom in of a set repre enl it all possible valucs of an independent variable f a fun Ii n. Values in p renlhe e repre enl Ihe value ( r range) of an independ nl ariabl \ ilh full member 'hip 10 Ihe el' hE limllied fr m lh rali f alch al year / 10 the maximum 81Ch (using calch lime- erie mo Ihed by tumlin a era e); < Po ilion ofdato-p int in tbe catch time-series (after running average) relali e 10 Ihe m imum attained at h in the dato- eric :. d T p fmember hip fun Ii n sumcd in the mod 1. a h of e pi ilali n t tu \ iLh degree of member hip 10 the el( d termined b membe hip un Ii n (Irapezoidal and triangular memb rship function ).

Ba d n the et f heuristic rules the predicted intrinsi vulnerabilities and pi ilalion taw re combined to infer depletion risk. The heuristic rule v ere d el p d from the as umption th I depl lion ri k of e plaited marine fishe in" population become full pI it doer- xploited and depleted (Table 2 . Depletion ri a categ rized into four I el: I ,mod rate, high, ery high - each repr enling a t of relali d pi tion risk inde that ran d fT m I to 10 ilb increasing risk. Heuri ti rul that d termin d the leIs f d pleli n ri ere e. pre sed in IF-THE clau in hich e pi italion tatu and 37 inlrin i ulnerabilit \ ere Ihe premi eSt while Ie el of dcpleti n ri k ere lh c nclu ions ( abl 2). r in Ian e.

"I inITin i ulnerabili ery high and populalion i d pI t d. TH depletion ri i hi h'.

11 mali rued to te t the n iii il to and Iidi fth rule (Table 2).

I 2. H pi itation tatll

Lo L Lo Lo ad. Lo L mod. od od od high High Lo Lo 00. od rate Lo Lo Lo mod. LO' Lo od. od. od. high High High . high V. high Low Lo mod. od. High Lo Lo ow & mod. Lo & mod. od. od. high Mod. & high High High &. high High & v. bigh V. high V. high Low & mod. Mod. Mod. & high Very high Low Low ow& mod. Mod. & high High High & v. high High High &. high V. high High & v. high V. high V. high Mod. & high High High & v. high n the ralionale Ihal deplclion ri kinde. ed a a p pulalion

wer

f memb hip 10 each depletion ri k Ie el in ea h year of a catch time- eries re determined ba d n th minimum m mb hip of th ir pr mi e i.e. intrinsic ulnerability and e ploitation tatu) (Table 2 . Wh n differ nl rul r ull in th arne onclu i n, m mber hip to the conclusion ere a cumul ted u ing equatioD 3 (Buchanan honliffe J ninde. of deplelion risk as estimaled from th inde. alue f each d pletion ri catego ighted b their degrees f memb rship. 38 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly

Applying the pert y tern to Predict Con ervation Risk

Spatially and taxonomically disaggregated catch time-series data that were produced from the FA 0 fishery statistics are available from the Sea Around Us Project (www.seaaroundus.org). We analyzed 460 species of marine fish' that had at least'10 years of catch time-series data, and catches of at least 100 tonnes. Catch time-series were aggregated by 19 FAO statistical areas (Figure 8), which yielded aggregates that used as proxy for identifying the stocks of a species. Thus the data for the 460 species compri e of I 313 catch time-series aggregated by the FAO areas. Parameter values for the life history and ecological characteristics were obtained from FishBase (www.fishbase.org).

SWP8Clk (11""'1

Figure 8. The 19 statistical areas delineated by FAO ( ource: ww~ .seaaroundus.org). The number of fish species by FAO area included in the analysi are shown in the parenthe es.

I/) 1.0 Q) U Q) 0.8 0.. 1970 I/) 0 0.6 -c 0 0.4 ~ 0 0.. 0.2 ..0 a. 0.0 5 15 25 35 45 55 65 75 85 Depletion risk index

Figure 9. Proportion ofthe 460 pecies of marine fishes with different classes ofcalculated depletion risk index in 1970 (open bars) and 2001 (gray bars). The values in the x-axis are the mid-point of the classes. Using an Expert ystem to Evaluate Vulnerabilities and Conservation Risk... 39

To obtain an approximate estimate of depletion risk of exploited marine fishes globally, we extrapolated the results from our analyse on the 460 species to all exploited marine fishes. To correct for any biases in OUT sample towards more vulnerable and targeted species, we grouped species by types (pelagic bony fish, demersal bony fish and elasmobranchs) and fishery importance (highly commercial commercial, minor commercial) using infonnation available from FishBase (Froese and Pauly 2004) - and excluded classes with low sample size (sample to global pecies number ratio :s 5%) before we extrapolated the predicted threatened status in each class (Table 3).

Table 3. Extrapolation from fish pecies with catch data to all exploited marine fish by fishery importance and fish type

Group Number of species Fishery With catch World Percent Extrapolation' ImportanceO data totalO represented Highly Pelagics 64 64 100 Included commercial Derner als 73 138 54 Included Elasmobranchs 3 7 43 Included Commercial Pelagics 53 155 34 Included Demersals 152 1272 12 Included Elasmobranchs 9 117 8 Included Minor Pelagics 25 185 14 Included commercial Demersals 61 1038 6 Included Elasmobranchs 13 165 8 Included Others Pelagics 1 146 1 Excluded Demersals 4 267 I Excluded Elasmobranchs I 44 2 Excluded

U Ba ed on Fi hBa e (www./i hba e.org)' b Pelagics and demersals include bony fish (teleosts) only;

C Groups that have 5% of less of the pecies with catch time-series data are excluded in extrapolating the number ofworld's threatened marine fish.

Our results showed that depletion risk of the 460 exploited fishes increased rapidly over the past three decades (Figure 9). In 200 I, about 24% of the evaluated species were associated with a very high depletion risk level (depletion risk index ~ 70), compared to 0% in the mid 1950s and 4% in 1970. Average depletion risk index of all species in 200 I is about 44 (highest depletion risk'" 100). Fishes that are in the very high depletion risk category include cod and haddocks, groupers, rays and skates grunts, tbreadfms. Comparing the predicted depletion risk index in 2001 between different fish groups, large demer al fish (maximum length ~ 30 cm) and elasmobranchs had a depletion risk index significantly higher than the average for all marine fishes, including small demersal and pelagic fishes, at 95% confidence level (logistic regression, P '" 0.003 and 0.022 respectively) (Figure 10). Among the 14-28%,20-56% and 8-24% of the 460 species that were in the 'very high', 'high and 'moderate' categories in 2001, elasmobranchs (sharks and rays) had the highest proportion in the 'moderate or higher categories (73%), followed by large demersal (61 %), large pelagic (48%) and then small pelagic bony fish (36%). 40 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly

>< Gl "0 50 c:: ~ III 'i: c:: 40 ..0 3:! c. Gl 30 C

20 +----'------'---....----'------'---~----'------'----r_----'------'----,._--'----'-__, Small pelagic Medium-large Small demersal Large demersal E1asmobranchs fishes pelagic fishes fishes fishes

Figure 10. Average depletion risk index of different fish group. tandard errors are indicated by the error bars.

~30 ell .~ U QI ~ 20 ~ QI .;:,Q u Critically :: 10 endangered o

o Mammals Birds Amphibia Marine fish

Figure II. Proportion of de cribed mammals, birds, amphibians categorized as critically endangered (white) endangered (grey) and vulnerable (black). Status of mammals, birds and amphibians are obtained from the IU Red ist (Baillie el al. 2004), while status of marine fish are inferred from our rule-based model. The error bars are the upper and lower limit .

By extrapolating to all exploited marine fish, we found that the proportion of marine fishes that have moderate to very high depletion risk (depletion risk index over 55, maximum = 100) was considerable. Of the 3,503 species of marine fish that FishBase (Froese and Pauly 2004) classifies as being commercially exploited, 500-957 (3-6% of all marine fish), 218-730 (2-3%) and 641-1,763 (5-11%) were categorized as having very high, high and moderate depletion risk, respectively. The comparisons between the depletion risk index and the IUCN Red Li t categories sugge t that the 'very high' 'high' and 'moderate' depletion ri k may be Using an xpert System to Evaluate Vulnerabilities and Conservation Risk... 41 used a proxies to indicate a species being in the 'critically endangered', 'endangered' and 'vulnerable' categories respectively. Thus our results suggest that 3-6%, 2-3% and 5-11 % of all marine fishes may be critically endangered endangered, and vulnerable. This i in the same order of magnitude as for other vertebrates (mammals, birds and amphibians) for which, however, a much higher number has been evaluated under the lUC Red List procedure (Baillie et al. 2004) (Figure II). Th estimated threat ned status of marine fish resulted from fishing was consistent with other, mainly terrestrial, vertebrates. This differs from the frequently expressed view that marine fish populations should be inherently more resilient than other vertebrates (Hutchings 200 Ia). In fact, our findings are supported by abundant empirical evidence. For instance, population parameters such as intrinsic rate of population increase, and population variability in marine fish were shown to be similar to value for other vertebrates (Fagan et at. 200 I, Hutchings 200Ib). Maximum reproductive potential, a population parameter that reflects the ability to withstand fishery-induced losses, is similar among fish groups, and similar to values among mammals of the same sizes (Myers et al. 1999), while the geographic range of many fishes need not render them less vulnerable, given their propensity for range collapse when abundance declines (pitcher 1997, Jennings et al. 1998, Sadovy 2001, Dulvy et al. 2003). Studies on fish stock recovery after depletions showed that, except clupeids, recovery rate was generally low (Hutchings & Reynolds 2004). Although known contemporary marine extinctions are rare, this situation might partly be a result of poor detection ability (Dulvy et al.2003). The rapid increase in our estimated depletion risk index coincides with the dramatic expansion ofglobal fisheries (Pauly et al. 2002). Over the past few decades, fish had lost their natural refuges (in the form of inaccessible habitats) owing to improved technology e.g. GPS, sea-floor mapping, echo-acoustic; (Pauly et al. 2004). Bio-economic factors (e.g. diminishing return from depleted stocks) might not prevent extirpation (or even extinction) as, in some case market value increased with resource rarity (Sadovy & Cheung 2003). These factors can lead to depletion of populations across their geographic range. Vulnerable species such as elasmobranchs and other large predatory fishes, particularly demersal fish should be prioritized for monitoring and conservation. The life history of elasmobranchs (large-size, late maturation) renders them highly vulnerable to fishing (Stevens et at. 2000, Dulvy & Reynolds 2002, Baum e/ al. 2003), while large predatory fish are traditionally targeted by fishing (Pauly et al. 1998, Myers & Worm 2003) which contribute to the higher depletion risk in these groups (Hutchings & Reynolds 2004). Our results are also consistent with predictions from simulation modelling that suggested 20-50% of bony fish and 40-100% of sharks might be driven to extinction under a typical fishery removal rate i.e., 40% of the population size removed per year (Myers & Worm 2005).

Validation of the Fuzzy Expert System

Using simulated data from a dynamic population model, we compared the predicted depletion risk with the IUCN Red List threatened categories. Based on available stock recruitment database (Myers et al. 1999), 21 species of marine fishes with wide range of intrinsic vulnerability were chosen for this analysis. An age-structured population model (H ilbom & Walters 1992), with assumed variability in recruitment, fishing intensity, density 42 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly dependent change in catchability to fishing (Mackin on el af. 1997) were employed to imulate population dynamic for each of the 21 species:

-(F+M) N a+l.y+1 = Na,y' e eq.4

where Na.)' is number of age a individual at year y, F and Mare fi hing and natural mortality rates. Recruitment at time I (R,) wa specified by a Beverton and Holt function:

R = a· S, . e&(o.<7) eq.5 , (l + S, ·13)

where R, is expressed as a function of the egg production or weight of pawners, (X is the maximum annual recruitment per spawner and j3 determines the degree of density dependence, and S, is the spawning stock size. Variations of annual recruitment were assumed to be log-normally distributed (mean = 0 and standard deviation = 0.5). Population was in equilibrium without fishing mortality initially (year 0). Fishing mortality rate (F) then increased at a constant rate at each time-step (year). The rate of increase in F was randomly chosen for each simulation. Selectivity was assumed to be age dependent and follow a logistic function:

eq. 6

where Va is the probability of capture at age a, Ie is the age at 50% capture and P is a constant determining the slope of the selectivity curve (P = 5 in this analysis). Time-series of catch (C) and catch-per-unit-effort (CPUE) were generated from each simulation (Hilborn & Walter 1992). In each simulation run, catch and CPUE were calculated from:

F·v C = a.N .[1- e-(F-v•.q+M)]. w eq. 7 La (F· Va + M) a.y a

CPUE= C·q' eq. 8 F where Wa is the weight-at-age, q is the actual catchability coefficient while q' was the assumed catchability coefficient used by the observation model (q '=0.3). Density-dependent change in catchability was modelled by (Mackinson el af. J997)

o q =q,b " if q<0.8 eq.9

q =0.8 ifq>O.8 Using an Expert System to Evaluate Vulnerabilities and Conservation Risk... 43

b is the biomass relative to the un exploited biomass and [; is unifonnly distributed error with values between 0 and 1. Time-series of catch and CPUE were recorded for eight generations of the tested species (approximated by eight-times the age ofsexual maturity). We detennined the extinction risk of each population in each simulation using the IUCN Red List criterion E (based ~n probability of extinction). For each species, the population dynamics described by the above model was run for 100 times. Probability of (quasi-) extinction was measured as the frequency of a population reaching 1/1000 of the unfished equilibrium biomass in the 100 simulations (Punt 2000), i.e., we assumed that population that have been reduced by 99.9% is not viable. Therefore, the population was classified as critically endangered, endangered or vulnerability if the probability of(quasi-) extinction is at least 50% in 10 years or three generations, at least 20% in 20 years or 5 generations and at least 10% in less than 100 years, respectively (IUC 200 I). Simultaneously, using the generated catch-per-unit-effort time-series, we estimated the threatened status as defined by the IUCN Red List criterion A- trends of index of abundance (IUCN 2001). The population was categorized as critically endangered, endangered and vulnerable if the CPUE declined by 80%, 50% and 30% in three generations or 10 years, whichever is longer. In each simulation, depletion risk of each population was also estimated using the expert system developed in this study. The intrinsic vulnerability of each test species was estimated based on the available life history parameters (Myers et al. 1999). At each time-step, the exploitation status was inferred from the catch time-series recorded from the simulation model. Depletion risk was then estimated from the predicted intrinsic vulnerability and exploitation status. Populations were classified as having moderate, high and very high depletion risk if the calculated depletion risk index is above 40, 55 and 70, respectively. The depletion risk calculated from the expert system was compared with the extinction risk identified by using the IUCN criteria E and A. We considered that the extinction risk identified based on the probability of quasi-extinction (IUCN criterion E) was accurate, while the IUCN criterion A is the most widely used criterion to assess extinction risk of marine fishes (Punt 2000). We compared the conservation risk categories and the IUC categories detennined based on criterion A with the threatened categories detennined by criterion E. Considering that the conservation risk categories of moderate, high and very high correspond to the IUCN categorie of vulnerable, endangered and critically endangered, respectively, we calculated the probability of under- and over- estimating threatened status (Type I and II errors) from predictions of our rule-based model using the simulated data from the above population model. To evaluate the sensitivity of the output to the structure of the expert system, this analysis was repeated using alternative sets of heuristic rules (Tables I, 2). Analyses using the imulated data suggested that the fuzzy expert system was able to predict extinction risk categories of fishes that were similar to those predicted from the IUCN Red List criterion A (Figure 12). Using threshold depletion risk index of 70, 55 and 40 to define 'critically endangered', 'endangered' and 'vulnerable' categories, the probabilities of categorizing a species to a category that is lower than the prediction using the IUCN criterion E (based on probability ofextinction from the population model) (Type I error) are 0.03, 0.14, 0.36 for the three threatened categories, respectively. The probabilities of assigning a higher threatened status of critically endangered', 'endangered' and 'vulnerable' than those predicted from the IUCN criterion E (Type II error) are 0.33, 0.2, and 0.05, respectively. Comparing with the predictions based on the IUCN criterion A, the probability of over- 44 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly estimating risk (Type II error) from prediction of the depletion risk index was significantly lower than the IUCN criterion A while probability of under-e timation (Type I error) appeared slightly higher. Results from the two extreme ets of criteria and rul~s that represented conservative and liberal interpretations of depletion risk showed that our assumed fuzzy sets and heuristic rules performed best (Figure 13). Type I errors did not differ significantly while the 'liberal scenario performed significantly poorer on Type Jl error suggesting that predictions from our moderate' scenario were robust to the assumed rules and criteria.

0.5 ... 0 ... 0.4 Q)

Q) ~ 0.3 ~ 0 -~ 0.2 :c ell ~ 0.1 2 'l. 0.0 CR EN or higher VU or higher

0.6 ...... 2 0.5 Q) - Q) 0.4 Q. >. ~ 0.3 -0 ~ 0.2 ~ ell ~ ...0 0.1 'l. 0.0 CR EN or higher VU or higher IUCN Red List category b)

Figure 12. Comparisons ofType I and II errors between threatened status predicted by the IUC Red Li t procedure (criterion E, solid bars) and the rule-based model (white bar). R- critically endangered, - endangered, VU - vulnerable. The error bars represent the 95% confidence limits, as uming that errors are binomially di tributed. Using an Expert ystem to Evaluate Vulnerabilitjes and Conservation Risk ... 45

0.6

~ 0 ~ 0.5 ~ CD CD 0.4 Q. ~ 0.3 -0 ~ 0.2 .0 llJ .0 0 ~ 0.1 Q. 0 CR EN or higher VU or higher a) 0.8 o Conservative ~ 0 0.7 ~ • Moderate ~ CD = 0.6 • Liberal 8. 0.5 ~ '0 0.4 ~ 0.3 :E llJ 0.2 .0 0 ~ Q. 0.1 0 CR EN or higher VU or higher IUCN category b)

Figure 13. The probability of (a) under-estimating (type I) and (b) over-estimating (type II) risk using the depletion ri k index predicted from conservative (open bar ), moderate (default, gray bars) and liberal (dark bars).

MODEL U CERTAI TY

In this chapter, we showed that the fuzzy expert system presented here can provide reasonable prediction on the relative depletion risk of exploited marine fishes to fishing. Using simulated data, we showed that the probability of over-estimating extinction risk from predictions of our expert system (type II error) is significantly lower than the widely used IUCN criterion A, while probability of under-estimation (type I error) is statistically the same. This shows that the fuzzy expert system can provide estimates comparable to the IUCN Red List. Moreover, the predications of relative depletion risk were generally robust to the two extreme sets of criteria and rules that represent conservative and liberal interpretations of depletion risk. Type I errors do not differ significantly while the 'liberal scenario' perfonns 46 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly significantly poorer on type II error, which shows that predictions from OUf 'moderate" scenario arc robust to the assumed rules and criteria. On the olher hand, inherent uncertainry exists in using the fuzzy expert system 10 predict the depiction risk of marine fishes 10 fishing. Firstly, becauK of the poor understanding on dynamics of fish in small population size. the depletion risk index may not represent the 'truc' risk ofextinction (Dulvy el aL 2004). Examination ofstock-recruitment data from wide range of fishes (Myers et 0/. 1999) using Bayesian statistics suggested, however, that the distribution of productivity parameter at low population size extended well into the range where depe.nsalion could occur (Liermann & l-lilbom 1997). Future studies on the dynamics of small population size att needed (pitcher 1998, Dulvy ~I al. 2003). Stcondly, the estimation was dependent on the assumed heuristic relationships and categorization criteria. Also. the structure ofcatch time-series might be confounded by non-fishery factors (i.e., error in the statistical recording system) and not accurntel)' depict fishing yields. Statistical data uncertainties are panicularly serious in U'opicat deyeloping regions and reef fisheries where fishery monitoring is less effective but threats to their high biodiversity are acute (Johannes 1998). Many ofthe exploited sp«ies att not reported explicitly in the catch statistics and thus are excluded from our analysis. Thus our predictions may underestimate the threatened status in these regions. Moreover, indirCCl effects of fishing. e.g., habitat destruction (Kaiser el aL 2003), ecosystem effects (Jackson el al. 2001), or genetic effects (Law 2000), are not considered in our estimates. Other threats to biodiversity such as climate change (Roessig el 01. 20(4) were also not accounted for. These factors would likely increase our predicted extinction risk.

CONCLUSION

Marine ecosystems and their component organisms are affected in parnllel and synergistically by environmental and anthropogenic factors; mechanisms behind these interactions are complex and non-linear but quantitative knowledge about them is relatively poor. Although a comprehensive understanding may take considerable time and resources, this may be cxpectcd to improyc predictions of futurc changes in the marine ecosystem. On the other hand, the scale of human impacts on the ocean is large and many componcnts and organisms of the marine ecosystem are being strongly affected by it. Actions to reduce or mitigate the impacls are urgently needed. Therefore. in the short term, we have to make some predictions about the components or organisms that are being critically affected by these impacts even when quantitative understanding on the underlying mechanisms is not totally clear. Here, we demonstrated that a fuzzy expert system could be used to collate qualitative knowledge or 'rules-of-thumb' to predict the level of impacts of fishing on different marine fishes. A fuzzy expert system has the advantage ofeasily-adaptable to new knowledge. While we are gaining more knowledge about the mechanisms of impacts (environmental and human). the expert system could be constantly updated to improve its predictions. There is also potential to combine the expert system presented in this chapter with other systems that predict the effects of different enyironmental and human changes (e.g. climate change, Using an Expcrt System to Evaluafe Vulncrabilities and Conservation Risk ... 47

pollution. coastal construction, socio-economic profile}. The resulting system can be used as a multi-disciplinary dccision support system. Specifically, this study suggests that the conservalion status of exploited marine fishes is potentially poor and they have a finile risk ofextinction. Given data limitalions and lhe urgent need to improve the assessment of extinction risks for marine fishes (Reynolds el 01. 2005), lhis study provides a way to evaluate the scope of the threats using currently available data with explicit uncertainty.

ACKNOWLEDGEMENTS

We lhank Reg Watson and Yvonne Sadovy for useful comments on the manuscript. We acknowledge funding supports from the Sir Robert Black Trust Fund Scholarship. the University Graduate Fellowship from the University of British Columbia, the Natural Sciences and Engineering Research Council ofCanada and the Pew Charitable Trusts through the Sea Around Us Project.

REFERENCES

Adams, P. B. (1980). Life history patterns in marine fishes and their consequences for fisheries management. FIshery Bulletin, 78( I}, 1-12. Ak~akaya. H. R., Ferson, S., Burgman, M. A., Keith, D. A., Mace, G. M. & Todd, C. R. (2000). Making consistentlUCN classifications under uncertainty. Cmuervation Biolagy, 14(4},1001·1013. Baillie, J. E. M., Hilton·Taylor, C. & Stuart, S. N. (2004). 1UCN Red List of Threatened Species· A Global Species Assessment. IUCN, Gland, Switzerland. Baum, J. K., Myers, R. A., Kehler, D. G., Worm, B., Harley, S. J. & Doherty, P. A. (2003). Collapse and conservmion of shark populations in the Northwesl Atlantic. Science. 299, 389-392. Brander, K. (1981). Disappearance of common skate Raja baris from lhe Irish Sea. Nature, 290,48·49. Buchanan, B. G. & Shortliffe, E. H. (1984). Rille-based Expert Systems - the MYCIN Experiment.f o/the Stanford Heuristic Programming Project California, USA: Addison Wesely. Caddy, J. F. (2004). Current usage of fisheries indicators and reference points, and their potential application to management of fisheries for marine invertebrales. Canadian Journal o/Fisheries and Aquatic: Science, 61, 1307·1324. Cardillo. M. & Dromham, L. (2001). Body size and risk of ex Ii net ion in Auslralian mammals. COnJervalion Biology, 15(5), 1435-1440. Chen, D. G. (2001). Detecting environmental regimes in fish stock-recruitment relationships by fuzzy logic. Canadian Journal 0/ Fisheries and Aquatic Science, 58.2139·2148. Cheung, W. W. L., Pitcher, T. J. & Pauly, D. (2005). A fuzzy logic expen system to estimate inlrinsic extinction vulnerability of marine fishes to fishing. Biological Conservation, 124,97·111. 48 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly

Christensen, V., Guenette, S., Heymans J. J., Walter, c., Wat on, R., Zeller D. & Pauly, D. (2003). Hundred-year decline of orth Atlantic predatory fishes. Fish and Fi heries. 4, 1467-2979. Dulvy, N. K. & Reynolds J. D. (2002). Predicting extinction vulnerability in kates. Conservation Biology, 16 (2), 440-450. Dulvy, . K., Sadovy, Y. & Reynolds, J. D. (2003). Extinction vuLnerability In marine populations. Fi h and Fisheries, 4,25-64. Dulvy, N. K., Elli , J. R., Goodwin, . B., Grant, A., Reynolds, J. D. & Jennings, . (2004). Methods of assessing extinction risk in marine fi he . Fish and Fisheries, 5 255-276. Fagan, W. F., Mieir, E. & Moore, J. L. (1999). Variation thre holds for extinction and their implications for conservation strategies. The American Naturalist, 154, 5\ 0-520. Fagan, W. F., Meir, E., Prendergast, J., Folarin, A. & Karieva, P. (2001). Characterizing population variability for 758 species. Ecology Leiters, 4, 132-\38. FAO (2004). The State of World Fisheries and Aquaculture. Rome: Food and Agriculture Organization of the United Nations, 153. Fiorentini, L., Caddy, J. F. & De-Leiva-Moreno J. I. (1997). Long- and short-term trends of Mediterraneanfishery resources. GFCM Stud. Rev. 0.69. 14. Froese R. & Pauly, D. (2004). FishBase. http://www.fishbase.org. Gaston, K. 1. & Blackburn, T. M. (2003). Birds, body size and the threat of extinction. Phi/. Trans. R. Soc. Lond. B, 347, 205-212. Goodwin, . B., Grant, A., Perry, A. L., Dulvy, . K. & Reynolds, J. D. (2006). Life history correlates of density-dependent recruitment in marine fishes. Canadian Journal of Fisheries andAquatic Science, 63, 494-509. Grainger, R. J. R. & Garcia S. M. (1996). Chronicles of marine fishery landings (1950 1994): Trends analysis and fisheries potential. FAO Fisheries Technical Paper. No. 359. Rome: FAO Hayes, E. (1997). A review ofthe southern bluefin luna fishery. Implications for ecologically sustainable management. Sydney, Australia: TRAFFIC Oceania, 34. Hilborn, R. & Walters, C. 1. (1992). Quantitalive fisheries stock assessment: choice, dynamics and uncertainty. New York: Chapman & Hall. Hilborn, R., Branch, T. A., Ernst, B., Magnusson, A., Minte-Vera, C. V., Scheuerell, M. D. & Valero, J. L. (2004). State of the world's fisheries. Annual review ofEnvironment and Resources, 28, 359-399. Hutchings, 1. A. (1996). Spatial and temporal variation in the density of northern cod and a review of hypotheses for the stock's collapse. Canadian Journal ofFisheries and Aquatic Science, 53,943-962. Hutchings, J. A. (2000). Collapse and recovery of marine fishes. ature, 406,882-885. Hutchings, 1. A. (200 Ia). Conservation biology of marine fishes: perceptions and caveats regarding assingment of extinction risk. Canadian Journal of Fisheries and Aquatic Science, 58, 108-121. Hutchings, J. A. (200 Ib). Influence of population decline, fishing, and spawner variability on the recovery of marine fishes. Journal ofFish Biology, 59A, 306-322. Hutchings, J. A. & Reynolds, J. D. (2004). Marine fish population collapses: consequences for recovery and extinction risk. BioScience, 54(4),297-309. IUCN (200 I). IUCN Red List Categories and Criteria: Version 3. J. IUCN pecies urvival ommission Gland, Switzerland and Cambridge, UK: IUCN. Using an Expert System to Evaluate Vulnerabilities and Conservation Risk ... 49

Jackson, J. B. C., Kirby, M. X., Berger, W. H., Bjomdal, K. A., Botsford, L. W., Bourque, B. J., Bradbury, R. H., Cooke, R., Erlandson, J., Estes, J. A., Hughes, T. P., Kidwell, S., Lange, C. B., Lenihan H. ., Pandolfi, J. M., Peterson, C. H., Steneck, R. S., Tegner, M. J. & Warner, R. R. (2001). Historical overfishing and the recent collapse of coastal ecosystems. Science, 293, 629-638. Jennings, S. & Polunin, N. V. C. (1996). Impacts of fishing on tropical reef ecosystems. Ambio, 25,44-49. Jennings, S., Reynolds, J. D. & Mills, S. C. (1998). Life history correlates of responses to fisheries exploitation. Proceedings ofthe Royal Society of London: Biological Science, 265,333-339. Jennings, S., Reynolds, 1. D. & Polunin, N. V. C. (1999). Predicting the vulnerability of tropical reef fishes to exploitation with phylogenies and life histories. Conservation Biology, 13(6), 1466-1475. Jennings, S., Pinnegar, J. K., Polunin, N. V. C. & Warr, K. J. (2001). Impacts of trawling disturbance on the trophic structure of benthic marine communities. Marine Ecology Progress Series. 213,127-142. Johannes, R. E. (1998). The case for data-less marine resource management: examples from tropical nearshore fmfisheries. Trends in Ecology and Evolution, 13(6), 243-246. Kaiser, M. J., Collie, S. J., Hall, S. J., Jennings, S. & Poiner, I. R. (2002). Modification of marine habitats by trawling activities: prognosis and solution. Fish and Fisheries, 3, 114 136. Kaiser, M. J., Collie, J. S., Hall, S. J., Jennings, S. & Poiner, I. R. (2003). Impacts of fishing gear on marine benthic habitats. In M. Sinclair, &G. Valdimarsson (Eds.), Responsible Fisheries in the Marine Ecosystem, pp. 197-217). Rome: FAO. Kirkwood, G. P., Beddington, J. R. & Rossouw, J. A. (1994). Harvesting species of different lifespans. In P. 1. Edwards, R. M. May, &N. R. Webb (Eds.), Large-Scale Ecology and Conservation Biology. pp. 199-227). Oxford: Blackwell Science Limited. Law, R. (2000). Fishing selection, and phenotypic evolution. ICES Journal of Marine Science, 57, 659-668. Liennann, M. & Hilborn, R. (1997). Depensation in fish stocks: a hierarchic Bayesian meta analysis. Canadian Journal ofFisheries andAquatic cience, 54, 1976-1984. Loreau, M., Naeem, S., Inchausti, P., Bengtsson, J., Grime, J. P., HEctor, A., Hooper, D. U., Huston, M. A., Raffaelli, D., Schmid, B., Tilman, D. & Wardle, D. A. (2001). Biodiversity and ecosystem functioning: current knowledge and future challenges. Nature, 294, 804-808. Mackinson, S. (2000). An adaptive fuzzy expert system for predicting structure, dynamics and distribution of herring shoals. Ecological Modelling, 126, 155-178. Mackinson, S., Sumaila, U. R. & Pitcher, T. J. (1997). Bioeconomics and catchability: fish and fi hers behaviour during stock collapse. Fisheries Research, 3 I, 11-17. Mackinson, S. Vasconcellos, M. & Newlands, . (1999). A new approach to the analysis of stock-recruitment relationships: "model-free estimation" using fuzzy logic. Canadian Journal ofFisheries andAquatic Science, 56 686-699. McDowall, R. M. (1992). Particular problems for the conservation of diadromous fishes. Aquatic Conservation: Marine and Freshwater Ecosystem, 2,351-355. Myers, R. A. & Wonn, B. (2003). Rapid worldwide depletion of predatory fish communities. Nature, 423,280-283. 50 William W. L. Cheung, Tony J. Pitcher and Daniel Pauly

M)crs. RA Ii Worm, D. (2005). Exline/ion. survival. or recow!ry oj/urge preddtoryfishes Philosophical Transactions oflhe Royal Society of London: B. 360. 13·20. Myers. R. A., Bowen, K. G. & Barrowman. N. J. (1999). Maximum reproductive rate of fish at low population sizes Canadian Journal ofFisheries and AqllOlic Science, 56. 2404 2419 Pauly. D. Watson, R. &. Alder. J. (2004). Globallfcnds in world fisheries: impacts on marine ecosystems and food security. Philosophical Transucliom of ,he Royal Sociery of London: B, 360, 5-12_ Pauly, D. Christensen. V.• Dalsgaard, J.• Froese, R. &. Torres Jr.• F. (1998). Fishing down marine food webs. Science. 279, 860·863. Pauly, D.• Christensen. V" Guenette. S., Pilcher. T. J., Sumaila. U. R.• Walters. C. J., Watson. R. &. Zdlu. O. (2002). Towards sustainability in world fisheries. Nature. 418. 689-695. Pitcher, T. J. (1997). Fish shoaling behaviour as a key factor in the resilience of fisheries: shoaling behaviour alone can generate range collapse in fisheries. Journal, 14)·148. Pitcher. T. J. (1998). A covery story: fisheries may drive stocks 10 extinction. R~l";i!WJ;n FiJh Biology and Fisherie.s. 8. )67·370. Powles. H.• Bradford. M. J.• Bradford. R. G., Doubleday. W. G., Innes. S. &:. Levings, C. O. (2000). Assessing and protecting endangered marine species. ICES Journal ofMadne Sdence. 57. 669--676. Punt. A. E. (2000). Extinction of marine renewable resources: a demographic analysis. Population Ecology, 42.19-27. Regan, I-I. M. &:. Colyvan. M. (2000). Fuzzy sets and lhreatened species classification. Conservation Biology. 14(4), 1197·1199. Reynolds, J. D., Jennings. S. &:. Dulvy, N. K. (2001). Life histories of fishes and population responses 10 exploil8lion. In J. D. Reynolds. G. M. Mace, K. H. Redford. &J. G. Robinson (Eds.). Conservation of exploited species. pp. 147-168). Cambridge: Combridge University Press. Reynolds. J. D., Dulvy. N. K., Goodwin. N. B. & HUlchings, J. A. (2005). Biology of extinction risk in marine fishes. Proceedings ofIhe Royal Society ofLondon: Biological Science. 272.2337·2344. Robens, C. M. &:. lIawkins, J. P. (1999). Extinction risk in the sea. Trends in Ecology and Evol/llioll, 14(6).241-246. Rodrigues. A. S. L., Pilgrim, J. D., Lamoreux, J. F.• Hoffmann, M. & Brooks. T. M. (2006). The value of the IUCN Red List for conservation. Trends in Ecology and Evolution. 2 t, 71-76. Rocssig. J. M.• Woodley. C. M., Cech, J. J. & Hansen. L. J. (2004). Effects of global climale change on marine and estuarine fishes and fisheries. Reviews in Fish Biology and FiJherie.s, 14,251-215. Roff, D. A. (1984). The evolulion of life history parameters in teleosts. Canadian JOllrnal of Fisheries andAquatic Science. 4 I, 989-1000. Rowe. S. &. Hutchings, J. A. (2003). Mating systems and the conservation of commercially exploited marine fish. Trends in Ecology and Evolution. 18(11),561·572. Russ. G. R. &:. Alcala. A. C. (1996). Marine reserves: tates and patterns of recovery and decline of large predatory fish. Ecological Applications. 6(3), 941·961. Sadovy, Y (2001). The threat of fishing to highly fecund flShes. Jounwl ofFish Biology. 59 (Supplemenl A). 90-108. Using an Expert System to Evaluate Vulnerabilities and Conservation Risk ... 51

Sadovy, Y. & Cheung, W. L. (2003). Near extinction of a highly fecund !ish: lhe one that nearly got away. Fish and Fisheries, 4, 86-99. Sadovy, Y. & Domeier, M. (2005). Reef !ish spawning aggregating need management: meeting the challenge. Coral Reefs. 24, 254-262. Saila, S. B. (1996). Guide to some computerised artificial intelligence methods. In B. A. Megrey, &E. Moksness (Eds.), Computers in Fisheries Research, pp. 8-37. London: Chapman and lIall. Sala, E., Aburto-Oropeza. 0., Paredes, G. & Thompson. G. (2003). Spawning aggregations and reproductive behaviour of reef fishes in the Gulf of California. Bulletin of Marine Science, 72( I), 103-121. Shelton, P. A. & Healey, B. P. (1999). Should depensation be dismissed as a possible explanation for the lack of recovery of thc northern cod (Gardus morhua) stock? Conadian Journal ofFisheries andAquatic Science, 26, 1521-1524. Spencer. P. D. & Collic, J. S. (1997). Patterns of population variability in marine fish stocks. Fisheries Oceonography, 6, 188-204. Stevens. J. D.• Bonlil. R., Dulvy, N. K. & Walker, P. A. (2000). The effects of fishing on sharks, rays, and chimaeras (chondrichthyans), and the implications for marine ecosystems. ICES Joun/al ofMarine Science, 57.476-494. Tinch, R. (2000). Assessing extinction risks: A novel approach usi"g fuz..:.~' logic. Norwich: University of East Angila, 26. Todd, C. R. & Burgman, M. A. (1998). Assessment of threat and conservation priorities under realistic levels ofuncertainty and reliability. Consen'otion Biology, 12(5),966-974. Watling, L. & Norse, E. A. (1998). Disturbance of the seabed by mobile fishing gear: a comparison to forest clearculting. Consen-'ation Biology, 12, 1180-1197. Watson, R.• Kitchingrnan, A., Gelchu, A. & Pauly, D. (2004). Mapping global fisheries: sharpening our focus. Fish and Fisheries, 5(2), 168-177. Wonn, B. & Duffy, J. E. (2003). Biodiversity, productivity and stability in real food webs. Trends in Ecolog)' and Evolution, 18(12),628-632. Wonn. B., Barbier, E. B., Beaumont, N., Duffy, J. E., Falke, C., Halpern, B. S., Jackson, J. B. C., Lotze, H. K., Micheli, F., Palumbi, S. R., Sala, E., Selkoc, K. A., Stachowicz, 1. J. & Walson, R. (2006). Impacts of biodiversity loss on ocean ecosystem services. Science, 314, 787-790. Zeller, D. & Ilauly. D. (2005). Good news, bad news: global !isheries discards are declining, but so are total catches. Fish a"d Fisheries, 6, 156-159. Zhang, C. (1994). A fuzzy distinguishability on similar degree of composition of spawning stock ofyellow croaker in the south Fulian fishing ground. Joumal ofFisheries ofChina, 18(4),335-339.