1 NORTH PACIFIC RESEARCH BOARD 2 3 BERING SEA INTEGRATED ECOSYSTEM RESEARCH PROGRAM 4 5 6 FINAL REPORT 7 8 9 10 Spatial Economic Models of Pollock and Cod 11 12 13 NPRB BSIERP Project B72 Final Report 14 15 16 17 18 Alan C. Haynie1, Lisa Pfeiffer2, and Jordan T. Watson3 19 20 21 22 1 National Oceanic & Atmospheric Administration, National Marine Fisheries 23 Service, Alaska Fisheries Science Center, Resource Ecology and Fisheries Management Div., 7600 Sand 24 Point Way NE, Seattle, WA 98115. (206) 526-4253, [email protected]. 25 26 2 National Oceanic & Atmospheric Administration, National Marine Fisheries 27 Service, Northwest Fisheries Science Center, 226 South Bldg., 2725 Montlake Blvd. East, Seattle, WA 28 98112-2097, [email protected]. 29 30 3 Pacific States Marine Fisheries Commission, 205 SE Spokane, Portland, OR 97202 and University of 31 Alaska Fairbanks, School of Fisheries & Ocean Sciences, 17109 Pt. Lena Loop Rd., Juneau, AK 98101, 32 [email protected]. 33 34 35 36 Abstract

37 38 This project focused on modeling the economic behavior of the Bering Sea pollock and Pacific cod 39 fisheries. These fisheries have several sectors, but here the majority of attention was given to the 40 vessels that are most mobile and reliant on the pollock and Pacific cod resources, namely the catcher 41 processor fisheries. By integrating economic, fishing, vessel characteristic, and environmental data, we 42 were able to examine how vessels have adjusted their spatial behavior across time in light of changing 43 conditions. Several primary conclusions emerge. First, the sea ice and the cold pool have a significant 44 impact on fishing behavior across the Bering Sea. A large cold pool will actually concentrate Pacific 45 cod in the north, so that fishing has been relative better in the north in cold years and fishing costs have 46 been higher in warmer years. Second, vessels have many ways to adapt to different climate (e.g., 47 choosing different locations, longer soak times, more travel), but this is costly. Third, because fish 48 abundance and climate are both impacting fish populations, there is little information to predict how the 49 pollock fishery will react in possible predicted future climate conditions with low abundance and a 50 small cold pool. Finally, changing institutions are affecting fishing behavior in new ways, so the 51 relationship between climate and fishing behavior is impacted by the transition of different fisheries to 52 catch shares. This allows greater flexibility within the fisheries, but minimal flexibility to alter species 53 targets for many vessels that currently target pollock and Pacific cod. 54 55 56 Key Words

57 Bering Sea, North Pacific, fisheries, CPUE, climate change, fleet behavior, North Pacific, range shifts, 58 modeling 59 60 Citation

61 Haynie, A.C., L. Pfeiffer, and J. Watson. 2014. “Spatial Economic Models of Pollock and Cod.” NPRB 62 BSIERP Project B72 Final Report, 128 pp.

Page 2 of 127 63 Contents 64

65 Study Chronology ...... 4 66 Introduction to the B72 Project ...... 5 67 Overall Objectives ...... 8 68 Chapter 1: Why economics matters for understanding the effects of climate change on fisheries ...... 12 69 Chapter 2: Climatic and economic drivers of the Bering Sea walleye pollock (Theragra chalcogramma) 70 fishery: implications for the future...... 21 71 Chapter 3: The effect of decreasing seasonal sea-ice cover on the winter Bering Sea pollock fishery ...... 35 72 Chapter 4: Climate fluctuations and fishing behavior in the Pacific cod (Gadus macrocephalus) longline 73 fishery ...... 48 74 Chapter 5: The effects of catch share management on rent generation through targeting and production 75 choices in the Pacific cod longline catcher processor fishery ...... 77 76 BSIERP B72 Project Conclusions ...... 100 77 PI Meetings and Other scientific interactions ...... 102 78 Past, current, and future collaborations ...... 103 79 Working with other modelers ...... 104 80 Interaction/connection with the LTK (local and traditional knowledge) component of BSIERP ...... 105 81 Next Steps ...... 105 82 Management and policy implications ...... 107 83 Publications ...... 108 84 BSIERP oral and poster presentations at scientific conferences or seminars ...... 109 85 Oral Presentations ...... 109 86 Poster Presentations ...... 111 87 Outreach / Workshops ...... 111 88 Acknowledgements ...... 113 89 Literature cited ...... 114 90 Appendix 1: Working paper: Utilizing VMS data to estimate unobserved Pacific cod fishing effort in the 91 Bering Sea ...... 115 92 Appendix 2: Headline document...... 126 93

Page 3 of 127 94 Study Chronology 95 96 This project was first approved in 2007 as BSIERP project M49, “Spatial fishery choices.” In 2008, 97 the project was re-designated B72, “Spatially explicit integrated economic model of pollock and cod.” 98 The name of the project later evolved to “Spatial Economic Models of Pollock and Cod.” The B72 99 project was initially placed into two MOU agreements, with the first agreement ending in 2009 and 100 years 3-5 budgeted to run from 12/1/2009 until 9/30/2013. An MOU was signed in 2013 which 101 provided a no-cost extension to run through July 31, 2014. 102 103 We attempted to hire a post-doc in 2008, but did not locate a suitable candidate in our first hiring 104 effort. We hired Dr. Lisa Pfeiffer in summer 2009 and Lisa worked full-time on the BSIERP project 105 from August 31, 2009 – December 2012. During the project, NOAA Fisheries AFSC obtained 106 additional resources from NOAA Fisheries Headquarters to extend the duration of her project work, 107 which allowed for an extension of Dr. Pfeiffer’s work at AFSC beyond 3 years. Dr. Pfeiffer was hired 108 at NOAA Fisheries Northwest Fisheries Science Center in December 2012, which freed additional 109 project resources that had been allocated for Dr. Pfeiffer’s salary and enabled the current extension of 110 the project. Dr. Pfeiffer has been able to work with Dr. Haynie on project manuscripts since then on a 111 limited basis.

Page 4 of 127 112 Introduction to the B72 Project 113 114 Project B72, “Spatial economic models of pollock and cod,” had a hybrid role of being a parallel or 115 competing model that was built outside of the integrated FEAST model but which potentially could also 116 interact with FEAST. The slower pace of development of FEAST was such that it became clear in the 117 middle of the project that any integration would have to occur in the future and the current project has 118 focused on examining the direct means through which climate affects the Bering Sea pollock and Pacific 119 cod fisheries. The primary goal was thus to have stand-alone models of fleet dynamics that would 120 explore how the pollock and Pacific cod fleets responded to changing environmental conditions on the 121 fishing grounds. This is illuminating because it enables a careful exploration of many interactions of 122 effects that impact fishing behavior. 123 124 As was recognized with the inclusion of social science components in BSIERP, it is impossible to 125 understand an ecosystem without including the role of humans in the system. In the Bering Sea, the two 126 largest fisheries are the pollock and Pacific cod fisheries. Fishing vessels attempt to maximize their 127 profits by trading off the revenue from catching fish with the costs of catching fish. The economic 128 literature has explored the impact of a range of factors on fishing behavior (e.g., Haynie et al 2009, 129 Mistiaen and Strand 2000, Holland and Sutinen 2000). Several recent reviews provide an overview of this 130 literature (e.g., van Putten 2011, Conrad and Smith 2012). 131 132 In the context of this economic modeling literature, how do we identify the impacts that changing climate 133 has on current fishing behavior and how will this evolve or change in the future? Climate is varying 134 across years, but this is occurring in the context of myriad other factors that are also changing. In this 135 project, we pay careful attention to identify and control for these factors and consider the response of the 136 pollock and cod fisheries. 137 138 Our focus in this work has been on the American Fisheries Act (AFA) Bering Sea pollock catcher 139 processor fishery and the BSAI Pacific cod freezer longline catcher processor fishery. The different fleets 140 that catch pollock and Pacific cod fisheries have quite different dynamics. As discussed in the different 141 manuscripts, we employed a range of modeling frameworks in our data exploration before arriving on the 142 models that are presented in the paper. We initially focused on haul-level discrete choice models of how 143 vessels chose to fish in different areas. As we were more concerned with large-scale shifts in effort, we 144 transitioned to an aggregate model to examine how the climate impacts the fishery.

Page 5 of 127 145 146 A major change that has impacted both the pollock and Pacific cod fisheries is the creation of fishing 147 cooperatives. With pollock, the American Fisheries Act (AFA) changed the nature of fishing profoundly 148 beginning in 1999. With Pacific cod, this has been a more gradual process, as some formal cooperation 149 began in 2004 but 2009 legislation allowed for the creation of the Pacific cod longline cooperative in 150 summer 2010. Complicating this transition is that there were large increases in Pacific cod total 151 allowable catch (TAC) as well as steep declines in prices. 152 153 The core components of this report are laid out as followed. Chapter 1 is our paper that appeared in the 154 ICES Journal of Marine Science in 2012 that provides a conceptual framework or overview of the many 155 means through which climate change may impact fisheries. Chapter 2 includes the published output from 156 our modeling of the pollock catcher processor fishery, which was published in 2013 in the Canadian 157 Journal of Fisheries and Aquatic Sciences (CJFAS). Chapter 3 focuses on how the pollock fishery 158 responded to changing ice conditions, which also appeared in the ICES Journal of Marine Science in 159 2012. Chapter 4 and 5 include draft manuscripts on the Pacific cod longline fishery that will soon be 160 submitted to peer-reviewed journals. Appendix 1 describes our work to date utilizing vessel monitoring 161 system (VMS) data to better describe trip and fishing behavior for unobserved fishing and for years 162 before trip information was recorded by observers. 163 164 Chapter 1 includes a conceptual framework of how fisheries maybe impacted by climate change. In this 165 paper, “Why economics matters for understanding the effects of climate change on fisheries,” we 166 distinguish between direct means, such as ice cover, and indirect means, such as fish recruitment. We 167 also discuss how institutions and economic factors interact with climate factors. 168 169 Chapter 2 is entitled “The effect of decreasing seasonal sea-ice cover on the winter Bering Sea pollock 170 fishery.” This paper explores how ice cover directly affects fishing in the Bering Sea pollock fishery. 171 172 Chapter 3 is arguably the most important paper in this final report, entitled “Climatic and economic 173 drivers of the Bering Sea walleye pollock (Theragra chalcogramma) fishery: implications for the future.” 174 175 Chapter 4 contains a manuscript that will be submitted this year to a peer-reviewed publication. The 176 manuscript, “Climate fluctuations and fishing behavior in the Pacific cod (Gadus macrocephalus) 177 longline fishery,” evaluates how climate variation in the 1990s and 2000s has impacted the observed 178 fishing conditions of the Pacific cod longline fishery. This research was originally presented as a poster at

Page 6 of 127 179 the Alaska Marine Science Symposium in 2012. 180 181 Chapter 5 investigates the manner in which the Pacific cod longline fishery has changed the products that 182 it makes since the creation of cooperatives. The manuscript, “The effects of catch share management on 183 rent generation through targeting and production choices,” examines how fishing has changed since the 184 implementation of cooperatives in the Pacific cod longline fishery in 2010. The paper has been revised 185 after internal review at AFSC. 186 187 As Appendix 1, we present a summary of our efforts to utilize vessel monitoring system (VMS) data to 188 identify fishing in unobserved Pacific cod longline trips. We have worked extensively with Bering 189 pollock catcher vessel data to characterize unobserved fishing effort. This Appendix summarizes our 190 efforts to extend this type of analysis to the Pacific cod longline fishery. As a second appendix, we also 191 include the “Headline” document prepared to summarize our work.

192 Finally, we offer overall conclusions, a discussion of our interaction with other components of the 193 BSIERP project, the policy and management implications of our research, and a description of our 194 ongoing research that builds upon this BSIERP project.

Page 7 of 127 195 Overall Objectives 196 The original description of the project was described as follows: 197 198 “This project will model how fishing effort is likely to change in the Bering Sea cod and pollock 199 fisheries under changing environmental conditions. These models will be constructed both 200 directly, through the inclusion of spatially-explicit environmental data into existing models, and 201 indirectly, through the inclusion of spatial predictions of fish abundance from FEAST or other 202 BSIERP-related models This project will allow us to model how research conducted on the 203 Bering Sea ecosystem (as reflected in hypotheses 1-4) will translate into changes in fishing 204 effort, as discussed more directly in hypothesis 5.” 205 206 The relevant portion of Hypothesis 5 states: “Climate-ocean conditions will change and thus affect the 207 abundance and distribution of commercial and subsistence fisheries…. For commercial fishermen, these 208 changes will lead to: 1) a change in home ports and distribution of fishing vessel rents, 2) vessels 209 traveling further, incurring greater fuel costs and peril at sea and 3) greater burden on smaller vessels. “ 210 211 The modeling component initially focused on pollock and then extended to Pacific cod. Previous work by 212 Haynie and Layton (2010) examined the Eastern Bering Sea pollock catcher vessel fleet. In this project, 213 we focused upon the pollock catcher processor fleet, which takes longer fishing trips and can pursue fish 214 further northward, following fish that might be shifting their distribution due to climate change. We 215 utilized a number of different statistical models to examine how the pollock fishery has adjusted its 216 behavior in response to changing climate, among other factors. Our modeling approach evolved to focus 217 on separating how climate and other factors interacted to impact fleet behavior. Climate variation is 218 occurring in the context of changing economic and management conditions as well. In Chapter 1 of this 219 report, we develop a conceptual framework which discusses how these factors interact to determine how 220 fish harvesters make decisions. 221 222 This interaction between climate, abundance, and economic factors proved to be an important element of 223 our research, as discussed in Haynie and Pfeiffer (2013), which appears in Chapter 2 of this final report. 224 Pfeiffer and Haynie (2012), presented here at Chapter 3, also displays how the fishery has been 225 constrained by ice, but so far it appears that ice is not covering up significantly better fishing conditions 226 and low-ice years have not had large effort shifts to the north.

Page 8 of 127 227 228 As described in the statement of work, “The spatial fishing choice models to be developed in this project 229 are both retrospective and predictive in nature.” We use available data from 1991-2012, or in some cases 230 a shorter period, to evaluate past fishing behavior to predict how future climate change is likely to impact 231 the fisheries. After we identified how climate and fish abundance each impacted location choice in the 232 pollock fishery, we came upon a difficult problem: there have been no warm, low-abundance years since 233 the American Fisheries Act (AFA) was implemented in 1999. Ianelli et al (2011) predicts lower average 234 pollock abundance in future warmer years, but this involves out-of-sample prediction from what we have 235 examined to date. We therefore can only hypothesize about how observed fishing behavior will extend to 236 these conditions. 237 238 In Chapter 4, we examine how the Pacific cod longline fishery has been affected by changing climate. 239 One important event in the fishery was the creation of a harvesting cooperative in the middle of 2010. 240 This cooperative led to the allocation of the sector-wide total allowable catch (TAC) to individual vessels. 241 As discussed in Chapter 1, the creation of catch shares can significantly alter fishing behavior in a manner 242 that could misrepresent the impact of changing climate on the fishery. In Chapter 5, we evaluate changes 243 in the fishing and processing strategies of the BSAI Pacific Cod longline fleet due to the formation of a 244 fishing cooperative. We find the volume, value, and types of byproducts increased after the formation of 245 the cooperative, reversing a period of decreasing recovery rates and byproduct production that 246 corresponded with an escalating race-for-fish prior to cooperative formation. However, we do not find 247 enormous changes in production behavior, as occurred in the pollock fishery after the AFA or in the 248 halibut fishery after individual fishing quotas (IFQs) were created in 1995. 249 250 One element that we initially considered including in this project was the collection of cost data for the 251 fisheries, which is a very elaborate process. Early in the project (April 2009), the North Pacific Fishery 252 Management Council (NFPMC) took final action on Amendment 91 to reduce Chinook salmon bycatch 253 in the pollock fishery, and planned a data collection program, leading us to await the output of that 254 project. Following the 2012 fishing year, AFSC and Pacific States began collecting these data and we are 255 beginning to utilize them in analyses for the pollock fishery. Currently there is no data collection for the 256 longline fishery but we are developing estimates based on other fisheries.

257 Early in the project, we thoroughly discussed the possibility of integrating our analysis with FEAST. As 258 it became clear that the timing of output from FEAST was not going to available in time to integrate with 259 our analysis, we have worked on longer term issues of development of tools to include FEAST-type

Page 9 of 127 260 output in our models. Thus we utilized spatially explicit output from the Bering Sea groundfish survey to 261 proxy for biomass estimates, which only occurs in the summer but allows us to examine the changing 262 relationship between biomass and other factors that affect fishing. As discussed in the conclusion of this 263 Final Report, we are developing the spatial economics toolbox for fisheries (FishSET), which will allow 264 output from FEAST and other ecosystem models to be coupled with realistic fisheries models in the 265 future.

Page 10 of 127 266 Manuscripts

267 We include our three published papers as well as two manuscripts that will soon be submitted to a peer- 268 reviewed journal. As an Appendix, we include a working paper that summarizes our recent Pacific cod 269 vessel monitoring system (VMS) data analysis work.

Page 11 of 127 270 Chapter 1: Why economics matters for understanding the effects of climate change on fisheries 271

272 Alan C. Haynie and Lisa Pfeiffer

273 The PDF from our publication follows.

Page 12 of 127 ICES Journal of Marine Science

ICES Journal of Marine Science (2012), 69(7), 1160–1167. doi:10.1093/icesjms/fss021

Why economics matters for understanding the effects of climate change on fisheries

Alan C. Haynie and Lisa Pfeiffer* Downloaded from Economics and Social Sciences Research Program, REFM Division, Alaska Fisheries Science Center, NOAA National Marine Fisheries Service, 7600 Sand Point Way NE, Seattle, WA 98115, USA *Corresponding author: tel: +1 206 5264696; fax: +1 206 5266723; e-mail: [email protected]. Haynie, A. C., and Pfeiffer, L. 2012. Why economics matters for understanding the effects of climate change on fisheries. – ICES Journal of Marine Science, 69: 1160–1167. http://icesjms.oxfordjournals.org/ Received 15 September 2011; accepted 12 January 2012; advance access publication 27 February 2012.

Research attempting to predict the effect of climate change on fisheries often neglects to consider how harvesters respond to changing economic, institutional, and environmental conditions, which leads to the overly simplistic prediction of “fisheries follow fish”. However, climate effects on fisheries can be complex because they arise through physical, biological, and economic mechanisms that interact or may not be well understood. Although most researchers find it obvious to include physical and biological factors in predicting the effects of climate change on fisheries, the behaviour of fish harvesters also matters for these predictions. A

general but succinct conceptual framework for investigating the effects of climate change on fisheries that incorporates the biological at NW Fisheries Science Center on January 16, 2013 and economic factors that determine how fisheries operate is presented. The use of this framework will result in more complete, re- liable, and relevant investigations of the effects of climate change on fisheries. The uncertainty surrounding long-term projections, however, is inherent in the complexity of the system. Keywords: climate change, conceptual models, economic behaviour, fisheries management.

Introduction policy-makers may be interested in the potential effects of Changes in climate conditions have the potential to profoundly climate change on the feedback effect of changes in harvest pat- affect marine ecosystems, of which fisheries are an integral part. terns into the population dynamics and spatial distribution of The commercial fishing industry removes millions of tonnes of the stock, on the flexibility of policy to climate change, on the fish and shellfish from the marine environment annually, and rec- cost of fishing, or on socio-economic indicators such as income, reational and subsistence harvests can be locally significant to exports, the vulnerability of communities, or food security. Each marine ecosystems. Research on the effects of climate change on of these research objectives requires an understanding of how marine ecosystems is generally focused on the biological system the drivers of fishing fleet behaviour can affect the range of poten- rather than fish harvesters. Inferences from these biology-focused tial outcomes (Fulton et al., 2011). models have been used to predict the effect of climate change on Some of the factors that drive harvester behaviour are directly fisheries (Klyashtorin, 1998; Ko¨ster et al., 2003; Perry et al., related to the biological system, such as the spatial distribution 2005, Lehodey et al., 2006; Brander, 2007; Cheung et al., 2010; of the target species, or the total allowable catch (TAC) of a Ianelli et al., 2011b), but they often neglect to incorporate how in- fishery, which is a function of total abundance. Other factors dividual harvesters respond to a variety of changing economic, in- may be influenced by biological characteristics; for example, stitutional, and environmental conditions. These responses can ocean temperature patterns may separate large and small fish, significantly affect retrospective analyses and predictions. making harvested fish from certain locations more valuable, and Researchers, industry participants, and policy-makers are inter- the value of locations may shift with changing ocean conditions. ested in how the distribution of fishing effort may change in rela- Still other factors, such as fuel prices, may profoundly affect fleet tion to climate change. Potential changes include spatial and/or behaviour but are not affected by the ecological system. To temporal changes in the distribution of fishing effort, the targeting ignore how climate affects all these factors that influence a harvest- of particular species, the location where harvests are landed and er’s decision is to ignore the effects of basic ecosystem interactions processed, and the composition of fishing fleets. For example, (Davis et al., 1998).

# United States Government, Department of Commerce 2012. Published by Oxford University Press.

Page 13 of 127 Why economics matters for understanding the effects of climate change on fisheries 1161

The study of social–ecological systems can benefit greatly from where and when to fish. The diagram illustrates the mechanisms structural frameworks that outline the relationship between specific linking environmental, biological, and economic characteristics variables and the mechanisms that affect the complex whole of many fisheries and their ecosystems. (Ostrom, 2009; Reid et al., 2010). In this study, we develop and de- The solid lines connecting elements in Figure 1 represent scribe a general but succinct conceptual framework for the investi- mechanisms with relatively contemporaneous effects. For gation of the effects of climate change on fisheries. The framework example, the extent of winter ice cover is likely to affect the distri- recognizes that climate can affect fisheries through a variety of bution of water temperatures contemporaneously and throughout complex mechanisms and incorporates how climate is likely to the rest of the annual cycle. The dotted lines represent mechanisms impact the biological and the economic factors that determine that are likely to arise non-contemporaneously or have both year how fisheries operate. Characterizing both the degree of uncertainty of and lagged effects. Lagged effects are much more difficult to surrounding these mechanisms and the range of possible variation identify in empirical data because the correct lag structure is in climate and biological conditions is an essential first step in gen- often not known, and the interaction of contemporaneous and erating predictions and analysing management scenarios. non-contemporaneous factors may obscure identification. For example, fishing may affect total abundance through both the con-

Conceptual model temporaneous removal of fish from the ecosystem and through Downloaded from Figure 1 is an illustrative diagram of the mechanisms through effects on recruitment. The effects through recruitment may take which climate factors can affect the distribution of fishing effort. several years to become evident in abundance measures. The arrows represent the direction of causality. This work is focused on high-latitude regions, where the effects Climate change is predicted to affect many of the environmen- of climate change are expected to be the most dramatic and earliest tal characteristics of ecosystems, but with large regional and tem- (IPCC, 2007). However, the conceptual framework applies to tem- http://icesjms.oxfordjournals.org/ poral variability (IPCC, 2007). For example, climate change may perate, tropical, and subtropical regions as well. To capture the dy- affect atmospheric circulation patterns such as the Arctic namics through which climate change affects fisheries, each link Oscillation and the North Atlantic Oscillation (Overland and displayed in Figure 1 needs to be considered carefully, along Wang, 2005; Hurrell and Deser, 2010), which help to determine with the possibility of additional links, when adapting Figure 1 the extent of winter sea ice in the Arctic, and in turn affect for a particular fishery. water temperatures for the entire annual cycle (Thompson and The remainder of this study is organized as follows. We first Wallace, 1998). Environmental characteristics are likely to affect present a simple economic model of fishing behaviour, versions the biology of a fishery’s target species, including recruitment, sur- of which have been empirically supported in a large number of vival, and spatial distribution. Finally, the biological characteristics fisheries. In the subsequent subsection, we describe how environ-

of the target species affect the fishery through the variable compo- mental characteristics affect the biological characteristics of a fish at NW Fisheries Science Center on January 16, 2013 nents of a harvester’s decision-making process, the most general of harvester’s target species. Then, we explain how these biological which include expected catch (i.e. catch per unit effort, cpue), characteristics affect fishing decisions. Finally, we discuss the expected value of catch, the cost of fishing, and the TAC. implications of using this framework to guide research on the Harvesters consider these factors when making decisions about effect of climate change on fisheries. The use of this framework

Figure 1. Conceptual model of how the environment affects the distribution of fishing effort. Arrows represent the direction of causality, and dotted lines represent the mechanisms that may arise on a non-contemporaneous time-scale.

Page 14 of 127 1162 A. C. Haynie and L. Pfeiffer avoids the simplistic conclusion of “fisheries will follow fish” all other alternatives. This can be rewritten as the probability which is often professed without considering the complexity of that the harvester chooses alternative j: economic drivers of fisheries and their constraints. P jt = Pr(E(p jt . E(pkt)) ∀j = k. (2) Harvester behaviour The foundation of the study of fish harvester behaviour is the in- A harvester is less likely to travel to a more distant location because dividual harvester’s decision-making process. Harvesters have a a greater travel cost would be incurred, holding other factors equal. knowledge of the potential benefits and costs of decisions about However, an increase in E(Qjt) and/or E(pricejt) would increase how, when, and where to fish. They weigh the potential costs the probability of travelling to location j to fish, all other things and benefits of their options and must make trade-offs between, being equal. for example, fishing in one location vs. another, or fishing more The harvester’s optimization problem is often subject to a intensively at the beginning than at the end of a season. variety of constraints defined by the management system in Economists have put a great deal of effort into modelling fishing place for the fishery. For example, a harvester’s decision-making fleet behaviour (e.g. Bockstael and Opaluch, 1983; Eales and process is different in a rationalized fishery (a fishery with individ-

Wilen, 1986; Holland and Sutinen, 2000; Smith and Wilen, ual allocations of quota) vs. a non-rationalized fishery (no individ- Downloaded from 2003; Haynie and Layton, 2010). (A “fleet” can be defined as the ual allocation of quota, although often entry into the fishery is group of harvesters exploiting the resource. A group of subsistence limited and there is a TAC). Catch shares, fisheries cooperatives, harvesters, for example, can be considered a fleet.) Location choice individual fishing quotas, and individual tradable quotas are models are particularly relevant because of recent policy emphasis examples of rationalized fishery management systems. on spatial fishing restrictions such as spatial closures, marine pro- Rationalized management alters the harvester’s decision-making http://icesjms.oxfordjournals.org/ tected areas, and area-based fishing rights (Smith and Wilen, 2003; process into one of choosing areas in which to fish in each Wilen, 2004; Sanchirico et al., 2006), and because stock dynamics period so as to maximize the profit over the entire time available, may be influenced by changes in the distribution of fishing effort. such as a season, subject to an individual quota constraint (TACi). Fish harvesters maximize utility (which is often assumed to be That is, a harvester chooses a location to fish in each period t ¼ 1, “profit”, especially for commercial fisheries, because it may be ..., T, where T is the end of the season, so as to maximize the total more easily measured than “utility”) by considering the character- profit over the season, subject to the constraint that the total quan- istics of potential fishing locations, such as the quantity, quality, tity of fish caught must be equal to the vessel’s (denoted i) share of T ≤ and species of fish they expect to catch, and the cost of travelling the TAC, or t Qit TACi. to different areas at different times. Choices along other dimen- The constraints provided by the management system are im-

sions, such as the timing of fishing (Brown, 1974; Kellogg et al., portant because they may lead harvesters to place differing value at NW Fisheries Science Center on January 16, 2013 1988) and entry and exit decisions (Ward and Sutinen, 1994; on prices, catch rates, travel costs, or on other factors that are Pradhan and Leung, 2004) have also been modelled. not included in the model presented here. For example, in a non- A simple conceptual fishing location choice model involves a rationalized fishery, harvesters place more value on quantity, harvester choosing an area to fish that will maximize the profit E(Qj), than value, E(pricej), than they would if (or when) the from fishing. For example, consider a harvester who makes a fishery were rationalized. This is because in a non-rationalized choice between j fishing locations, j ¼ 1, ..., J. The harvester will fishery, there is an incentive to “race” for fish, i.e. to catch the receive revenues from the quantity of fish caught and the price quota before others do and the fishery is closed (Levhari and received for the catch from the chosen location. For a recreational Mirman, 1980). The profit-maximizing result is often the targeting or subsistence harvester, a non-market valuation or other utility of the highest catch rates. In a rationalized system, however, the in- measure (e.g. site quality) could be substituted for cash revenue. dividual allocation of quota allows vessels to consider the add- Perfect information about the quality and quantity of fish that itional profit that may be gained by making trade-offs between will be encountered is never available, so it is assumed that the fishing in areas with higher value, less abundant fish and areas harvester makes a choice based on an expectation, E(.), of revenues with lower value, more abundant fish, because fish not caught and the cost that will be incurred by choosing to fish in location now can be caught later in the season. Additionally, in a rationa- j. The harvester must incur the cost per kilometre (cj) of travelling lized fishery, changes in the constraints (such as TAC) or alterna- to the chosen location, which is distance Distj from port tive fishing opportunities may affect the harvester’s decisions or from his/her previous fishing location. Travel costs vary de- about location choice or the timing of fishing. For example, an in- pending on the type of fishery and location chosen and could crease in TAC, holding other factors equal, may increase the rela- include factors such as fuel, the time cost of travel, or safety. The tive importance of E(Qj)toE(pricej) and travel costs. As the TAC harvester’s expected profit, E(pjt), from fishing in location j at increases, a harvester will be more constrained by the number of period t is periods in the season because he or she needs to catch more fish in the same amount of time; sacrificing higher quality fish for E(p jt )=E(Q jt )×E( price jt )−c jtDist jt , (1) volume may lead to greater profit. If harvesters can participate in other fisheries, it may be optimal to fish in the alternative where E(.) is the expectations operator, Qjt the quantity of catch in fishery during its peak, then return to the rationalized fishery. area j in time t, and pricejt the unit value of fish caught in area j in Expected cpue, expected prices, and travel costs are the main time t. Qjt is a function of cpue and the amount of effort applied. variable components in a harvester’s decision-making process, The harvester maximizes the profit in each period t by choosing whereas TAC, management systems, and other factors constrain among the J location alternatives. A harvester will choose location fishing. These variables can be affected by factors that are not j in time t if the expected profit from that area at that time impacted by climate variation, such as demand for the processed (expected revenues minus travel costs) is greater than that from products, fuel prices, and politics; these are shown as external

Page 15 of 127 Why economics matters for understanding the effects of climate change on fisheries 1163 factors in Figure 1. Demand for the products being produced from Figure 1 illustrates them on the most basic level. For extreme lati- the catch affects the prices harvesters expect to receive and depends tudes, winds and atmospheric conditions create an environment on economic and market conditions. Fuel prices affect costs. Other characterized by a distribution of ice cover, and air and water tem- costs, such as crew salaries, the cost of maintenance, or the price of peratures that vary at many scales. leased quota, could affect the profitability of fishing and alter the The distribution of water temperatures is an important deter- level of effort. For recreational and subsistence harvest, factors minant of marine species’ locations; many species have “climate such as non-fishing employment opportunities, non-fishing recre- envelopes”—bands of temperature that they prefer as a result of ational activities, or the price of alternative food sources could physiological factors and the location of important food sources have a similar effect. Politics can affect the development and evo- (Box, 1981; Guisan and Zimmermann, 2000; Parmesan and lution of the management and system of regulation and may play a Yohe, 2003). Changes in the distribution of temperature bands role in the determination of TAC, fishing days, closures, or other may result in changes in the distribution of species. Many constraints. studies of the effect of climate change on species’ ranges rely on When attempting to identify the effects of climate factors using this climate envelope approach. For example, Richardson and retrospective data, it is important to consider how these other Schoeman (2004) find that sea-surface warming in the Northeast factors (those not directly related to climate variation) may have Atlantic is associated with changes in phytoplankton abundance Downloaded from affected what happened. It is also important to consider future that propagates up the foodweb. Atkinson et al. (2004) find that variation when making predictions. For example, a major increase sea-ice algae are an important food source for Antarctic krill, in fuel prices could cause a change in the spatial distribution of and their recruitment and survival are positively correlated with fishing effort by making travel more expensive. If the fuel price in- the extent of winter sea ice. Larger species such as penguins, alba- crease happened to be correlated with a change in climate condi- trosses, seals, and whales depend on Antarctic krill. Antarctic salps http://icesjms.oxfordjournals.org/ tions, the change in effort could be incorrectly attributed to occupy the warmer and lower productivity regions of the Southern climate. In addition, species targeting in a multispecies fishery is Ocean, and increases in their abundance are related to decreases in affected by the value of each species caught. For example, in a winter sea ice. Perry et al. (2005) find that the distributions of fishery where two abalone stocks occur in the same area near many North Sea fish have shifted north in response to increases Japan, harvesters selectively target the more-valuable species at in sea temperatures. the opening of the season. As the cpue of that species declines Other climate factors may also affect characteristics of target through fishing, harvesters begin to target the other species species that are important to the fishery (Walther et al., 2002); (Matsumiya and Matsuishi, 1989). An assumption of simultan- for example, temperatures may affect the phenology of marine eous exploitation could lead to erroneous conclusions about the species (Beaugrand et al., 2003; Edwards and Richardson, 2004;

dynamics of the fleet, which, if related to climate in some way, Yoneda and Wright, 2005; Bellier et al., 2007), as well as food avail- at NW Fisheries Science Center on January 16, 2013 could result in erroneous climate-related projections. Similarly, ability and the probability of overwinter survival (Beaugrand et al., using retrospective data from before a significant management 2003; Hunt et al., 2011; Mueter et al., 2011). Many other species change to predict a distribution of effort after the management and ecosystem-specific examples are possible (several are reviewed change could be seriously misleading, because constraints and in Brander, 2010). trade-offs of harvesters’ decisions are likely to have been affected Whereas the environmental characteristics of an ecosystem by the management change (Homans and Wilen, 2005). affect the harvester’s decision-making process through biological Incorporating a model of harvesters’ decisions is an important and ecosystem effects on the target species, Figure 1 shows how step in testing the statistical relationship between climate change some environmental variables, such as ice cover, for example, and observed variation in a fishery. can directly affect the harvester’s decision-making process. Variables such as ice cover or the frequency of storms can affect Environmental characteristics of an ecosystem affect harvester costs. These factors may make travel to some areas diffi- biological characteristics cult or impossible, and climate-related changes in weather or ice Climate-related changes in the environment such as increases in conditions can open or close areas to fishing seasonally. temperature, decreases in snow and ice-extent, and changes in hydrological systems have been observed to the greatest degree Biological characteristics of a species affect the fishery in northern latitudes (IPCC, 2007). Global climate models Figure 1 illustrates several mechanisms by which the environmen- predict continued changes in the long-term means and variability tal characteristics of an ecosystem are likely to affect the biological of atmospheric conditions and the environmental characteristics characteristics of the target species, and how those biological char- they affect. For example, changes in the Arctic Oscillation and acteristics are likely to affect the factors that matter to fish weaker northerly winds over the Pacific Arctic have caused harvesters. decreases in seasonal sea-ice cover and warmer ocean temperatures Non-contemporaneous factors such as recruitment, survival, or (Thompson and Wallace, 1998; Overland and Wang, 2005, 2007). the effect of fishing on the spatial distribution of the target species These changes are projected to continue, affecting many environ- in subsequent years (those represented by dotted lines connecting mental characteristics of marine ecosystems that, in turn, affect the elements) are likely to be more difficult to identify in empirical biological processes (Overpeck et al., 1997; Marshall et al., 2001; data. For example, the survival rate of juvenile fish affects the total Attrill and Power, 2002; Hunt et al., 2002; Beaugrand, 2004; biomass, but will not affect the fishable biomass until they have Grebmeier et al., 2006). recruited into the fishery, which can be many years later. Ocean In general, the environmental characteristics of an ecosystem temperature and current patterns can affect the degree of spatial interact in complex ways to affect the biological characteristics of overlap between juvenile and adult Alaska pollock; adults may the target species of the fishery. Evaluating the precise mechanisms cannibalize young pollock, resulting in a contemporaneous of these interactions is well beyond the scope of this paper, but benefit for adult pollock, but a lagged cost on future biomass

Page 16 of 127 1164 A. C. Haynie and L. Pfeiffer

(Wespestad et al., 2000). These effects can be extremely difficult to The total abundance of the target species also affects many identify, especially with spatially and temporally limited biological characteristics of the fish harvester’s economic trade-offs. data, because of multiple and overlapping lags. As our focus here is Abundance affects travel costs because when total fishable on providing a framework to recognize the relationships that may biomass is larger, less time and travel are expended on the characterize how climate change may affect fisheries, not the search for acceptable quantities of fish. In other words, the travel details involved in identifying these relationships, we focus the dis- distance between fishing hauls is likely to be smaller. Abundance cussion on contemporaneous factors, although non- affects the TAC constraint because the estimated abundance is contemporaneous factors are also likely to impact the fishery. used to set the annual TAC. Abundance also affects expectations Figure 1 portrays a number of contemporaneous biological about the quantity of catch and cpue. Although generally increased factors that have the potential to affect the fishery. The spatial dis- abundance is expected to increase the expected catch, it may not tribution of the target species affects harvesters’ expectation of the increase uniformly, which is why there is a link in Figure 1 from quantity of fish they are likely to catch, E(Q), in each potential abundance back to the spatial distribution of the target species. fishing location. Additionally, the spatial distribution of the popu- For example, if increases in abundance result in an expansion of lation can affect the expected value of the catch, E(price), if it a species’ range, then the resulting increase in cpue may be varies by fish size, age class, and other factors that affect the greater on the expansionary edges of the range than in the areas Downloaded from quality of the fish and the value of the harvest, or if it affects the where the species is always present (Holt et al., 1997). uniformity of fish size or species composition in schools, which Vessel size, horsepower, the ability to process fish on board, or may increase the efficiency of harvest or post-harvest processing. harvester preferences may lead to different harvesters making dif- Figure 2 provides an example. It shows the average processed ferent decisions under changing conditions. For example, Loomis value of 1 t of Alaska pollock catch from the eastern Bering Sea and Ditton (1987) find that recreational anglers who fish in tour- http://icesjms.oxfordjournals.org/ catcher-processor fleet in 2006 by 0.5 × 18 grid. Clearly, the naments have very different preferences from those who fish for average value of pollock caught in the southern regions of the sport. Tournament anglers prefer trips where they catch larger fishing grounds is significantly higher than the average value in fish, whereas sport fishers prefer to catch a greater number of the north. This is because in winter, high-value roe is removed fish. Differentiating these two types of recreational fishers would from prespawning female pollock, processed, and sold to Asian be important in predicting changes in effort resulting from, for markets. The rest of the fish is processed into fillets and other pro- example, climate-induced changes in growth rates or biomass. ducts, so the harvest of roe represents a 50–200% bonus on regular fish prices. Spawning is concentrated in the southern region of the fishery, causing the spatial value differences shown in Figure 2. Discussion and conclusions

However, roe-harvesting is done almost exclusively in winter, so This paper has presented a general conceptual framework for the at NW Fisheries Science Center on January 16, 2013 this spatial price premium is absent in summer. Ignoring the study of the effects of climate change on fishing and argued why spatial and temporal differences in the value of pollock and focus- it is essential to include economics in this framework. The frame- ing only on cpue as an indication of how the fishery will change work applies to all types of fishing: commercial, subsistence, or would lead to serious inaccuracies in the prediction of the distri- recreational. Disregarding the economic behaviour of fish harvest- bution of fishing effort. ers can lead to serious mistakes in retrospective analyses or when making predictions about the effects of climate change on the eco- system and the spatial distribution of fisheries, just as ignoring any important interaction could make the resulting predictions unre- liable, incorrect, or irrelevant (Davis et al., 1998). For example, a simple error is to assume that one size fits all, merely focusing on abundance and ignoring how harvesters may target certain sizes or species of fish with different values, as illustrated by the examples in the previous section. The conceptual framework of Figure 1 illustrates that the effect of a change in a climate variable on the distribution of fishing occurs through many physical, biological, and economic mechan- isms. For example, a decrease in winter ice cover in a polar region may be expected to decrease the cost of fishing (harvesters have the opportunity to fish in previously ice-covered waters), increasing the expected profit in areas affected by the change, and potentially changing the spatial distribution of fishing effort. However, ice cover may also affect recruitment, which will affect the total abun- dance in future years. Total abundance affects the TAC constraint, the cost of fishing, and expected cpue. The extent of winter ice cover may also affect the distribution of water temperature, affect- ing the spatial distribution of a harvester’s target species, which Figure 2. Estimated average value of processed product per tonne of pollock caught by the catcher processor sector of the eastern Bering can affect expected cpue in different locations and various compo- Sea pollock fishery, by 0.58 × 18 grid, for the 2006 fishing year (as an nents of expected value. By following the links in the diagram, it is example). Grids with three or fewer vessels fishing in them are clear that each variable affects many other variables in the system, censored for confidentiality. Data are from the National Marine and ultimately has the potential to affect the distribution of fishing Fisheries Service Observer Program Database. in many ways.

Page 17 of 127 Why economics matters for understanding the effects of climate change on fisheries 1165

It is essential that the researcher examines the details of the en- reason. Human participants in fisheries respond to a variety of vironment that are unique to individual fisheries, including the economic incentives, but each incentive is affected by ecological characteristics of the ocean dynamics, region, species, harvesting mechanisms that are not completely understood. In each mechan- fleet, and management structure to determine which of these ism represented in Figure 1, there are varying degrees of uncer- mechanisms are likely to have the most important effects. The re- tainty. When an economic model of fish harvester behaviour is searcher must also carefully consider mechanisms that may occur added to other uncertain models or predictions, the resulting un- simultaneously and may have compounding or negating effects. A certainty is potentially compounded. Models to forecast fuel decrease in winter ice cover is likely to be simultaneous with an in- prices, consumer demand for the fish, or labour costs, for crease in winter water temperature, for example. Although their example, may be relevant. These would add multiple additional main effects may be different (ice cover directly affects costs, dimensions to the modelling process and, rather than narrowing whereas winter temperatures directly affect the maturation rate the possible outcomes or the surrounding uncertainty, expand of roe and therefore the expected value), their effects on the them. This is still preferable to ignoring or only superficially con- spatial distribution of effort will be confounded. For example, sidering the effects of human behaviour. When a large number of are the pollock harvesters depicted in Figure 2 travelling to factors affect behaviour with a large degree of uncertainty (which newly opened areas to fish because they can (less ice) or because characterizes much of the research on climate effects on marine Downloaded from they expect to find high-value roe-bearing fish (higher expected ecosystems), expecting a precise prediction of the effect of a par- price)? Alternatively, consider a change in the distribution of ticular climate factor, or climate change as a whole, on aggregate fish biomass resulting in more fish being located farther from a fishing effort can be unrealistic. In fact, recognizing when there main port that happened simultaneously with an increase in fuel is too much uncertainty surrounding the mechanisms or drivers prices. This may result in no change in the distribution of effort, of the system to generate reliable predictions and characterizing http://icesjms.oxfordjournals.org/ although holding either variable constant, a change in the distribu- the causes of that uncertainty are extremely valuable contributions tion of effort would have been predicted. The researcher must con- to the state of knowledge. They can serve to direct research towards sider carefully how to separate these effects in empirical data and the areas of uncertainty, spur data collection and analysis efforts, must be extremely cautious in drawing conclusions from short clue managers in to possible unintended effects of policy, and gen- time-series where there is little information available for under- erate entirely new topics of research. More modelling that includes standing the causal mechanisms. Therefore, establishing correla- environmental, biological, and economic elements across a range tions between fish populations and climate variables while of fisheries and ecosystems will help improve our understanding omitting economic factors can generate completely erroneous of how climate change can impact fisheries. explanations of causal relationships. Beyond the role that this conceptual framework has in helping

Even in fisheries and ecological systems with a relatively large researchers understand the effects of climate change on fisheries, at NW Fisheries Science Center on January 16, 2013 amount of observational data, separately identifying the effects an important contribution of this study is in providing fisheries of various climate factors can be challenging. Environmental con- managers with additional insight about the implications of ditions that are determined to be important drivers, and are climate change for management. Management can be designed expected to change, increase the data demands required to identify to be resilient to changing ecosystems, as well as impart upon separate effects. For example, in the eastern Bering Sea Alaska harvesters the flexibility to adapt to changing conditions. A pollock fishery, both the spatial distributions of the stock and careful examination of how economic, biological, and environ- the total biomass are expected to change as a result of climate mental characteristics of fisheries interact with management change (Wyllie-Echeverria and Wooster, 1998; Mueter et al., institutions under a changing climate is an important topic for 2011). However, since 1990, all years characterized by warm further research. ocean conditions in the eastern Bering Sea have also had a relative- ly high total pollock biomass due to large lagged recruitment Acknowledgements events. Cold years have been observed with both high and low Funding for this research was provided by the North Pacific abundances (Ianelli et al., 2011a). Therefore, there have been no Research Board (NPRB publication number 326) as part of the recent observations of warm, low abundance years in the Bering Bering Sea Integrated Ecosystem Research Program (BEST– Sea with which to empirically identify fishery behaviour under BSIERP project number 33). conditions that we are most concerned about predicting. Predictions under warm, low-abundance conditions would be References “out of sample” and are subject to significantly greater uncertainty. Atkinson, A., Siegel, V., Pakhomov, E., and Rothery, P. 2004. The researcher must also recognize that there are likely to be Long-term decline in krill stock and increase in salps within the uncertain, ignored, inestimable, or unknown elements in the eco- Southern Ocean. Nature, 432: 100. system and that these may influence findings and predictions Attrill, M. J., and Power, M. 2002. Climatic influence on a marine fish about the impact of climate change on fisheries. Long-term projec- assemblage. Nature, 417: 275–278. tions are often desired because they will potentially allow for better Beaugrand, G. 2004. The North Sea regime shift: evidence, causes, management adaptation to a changing climate. However, long- mechanisms and consequences. Progress in Oceanography, 60: term projections require strong assumptions about the degree to 245–262. which these other factors are held constant. Beaugrand, G., Brander, K. M., Lindley, J. A., Souissi, S., and Reid, P. C. 2003. Plankton effect on cod recruitment in the North Sea. Finally, the degree to which the scientific community has infor- Nature, 339: 556–559. mation about the magnitude, certainty, or even the sign of each Bellier, E., Planque, B., and Petitgas, P. 2007. Historical fluctuations in mechanistic link in Figure 1 varies widely. It has been noted that spawning location of anchovy (Engraulis encrasicolus) and sardine human behaviour is a key source of uncertainty in fisheries man- (Sardina pilchardus) in the Bay of Biscay during 1967–73 and agement (Fulton et al., 2011). Figure 1 illustrates part of the 2000–2004. Fisheries Oceanography, 16: 1–15.

Page 18 of 127 1166 A. C. Haynie and L. Pfeiffer

Bockstael, N. E., and Opaluch, J. J. 1983. Discrete modelling of supply Sea walleye pollock (Theragra chalcogramma) in a changing envir- response under uncertainty: the case of the fishery. Journal of onment. ICES Journal of Marine Science, 68: 1297–1304. Environmental Economics and Management, 10: 125–137. IPCC. 2007. Contribution of Working Groups I, II, and III to the Box, E. O. 1981. Macroclimate and Plant Forms: an Introduction to Fourth Assessment Report of the Intergovernmental Panel on Predictive Modelling in Phytogeography. W. Junk, The Hague. Climate Change. In Climate Change 2007: Synthesis Report, Brander, K. 2007. Global fish production and climate change. p. 104. Ed. by R. K. Pachauri, and A. Reisinger. IPCC, Geneva, Proceedings of the National Academy of Sciences of the USA, Switzerland. 104: 19709. Kellogg, R. L., Easley, J. E., Jr, and Johnson, T. 1988. Optimal timing of Brander, K. 2010. Impacts of climate change on fisheries. Journal of harvest for the North Carolina bay scallop fishery. American Marine Systems, 79: 389–402. Journal of Agricultural Economics, 70: 50–62. Brown, G. 1974. An optimal program for managing common property Klyashtorin, L. 1998. Long-term climate change and main commercial resources with congestion externalities. Journal of Political fish production in the Atlantic and Pacific. Fisheries Research, 37: Economy, 82: 163–173. 115–125. Cheung, W. W. L., Lam, V. W. Y., Sarmiento, J. L., Kearney, K., Ko¨ster, F., Hinrichsen, H-H., Schnack, D., St John, M. A., Mackenzie, Watson, R., Zeller, D., and Pauly, D. 2010. Large-scale redistribu- B. R., Tomkiewicz, J., Mo¨llmann, C., et al. 2003. Recruitment of

tion of maximum fisheries catch potential in the global ocean Baltic cod and sprat stocks: identification of critical life stages Downloaded from under climate change. Global Change Biology, 16: 24–35. and incorporation of environmental variability into stock–recruit- Davis, A. J., Jenkinson, L. S., Lawton, J. H., Shorrocks, B., and Wood, ment relationships. Scientia Marina, 67(Suppl. 1): 129–154. S. 1998. Making mistakes when predicting shifts in species range in Lehodey, P., Alheit, J., Barange, M., Baumgartner, T., Beaugrand, G., response to global warming. Nature, 391: 783–786. Drinkwater, K., Fromentin, J-M., et al. 2006. Climate variability, Eales, J., and Wilen, J. 1986. An examination of fishing location choice fish, and fisheries. Journal of Climate, 19: 5009–5030.

in the pink shrimp fishery. Marine Resource Economics, 2: Levhari, D., and Mirman, L. J. 1980. The Great Fish War: an example http://icesjms.oxfordjournals.org/ 331–351. using a dynamic Cournot–Nash solution. The Bell Journal of Edwards, M., and Richardson, A. J. 2004. Impact of climate change on Economics, 11: 322–334. marine pelagic phenology and trophic mismatch. Nature, 430: Loomis, D. K., and Ditton, R. B. 1987. Analysis of motive and partici- 881–884. pation differences between saltwater sport and tournament fisher- Fulton, E. A., Smith, A. D. M., Smith, D. C., and van Putten, I. E. 2011. men. North American Journal of Fisheries Management, 7: Human behaviour: the key source of uncertainty in fisheries man- 482–487. agement. Fish and Fisheries, 12: 2–17. Marshall, J., Kushnir, Y., Battisti, D., Chang, P., Czaja, A., Dickson, R., Grebmeier, J. M., Overland, J. E., Moore, S. E., Farley, E. V., Carmack, Hurrell, J., et al. 2001. North Atlantic climate variability: phenom- E. C., Cooper, L. W., Frey, K. E., et al. 2006. A major ecosystem shift ena, impacts and mechanisms. International Journal of in the northern Bering Sea. Science, 311: 1461–1464. Climatology, 21: 1863–1898. at NW Fisheries Science Center on January 16, 2013 Guisan, A., and Zimmermann, N. E. 2000. Predictive habitat distribu- Matsumiya, Y., and Matsuishi, T. 1989. Estimation of population para- tion models in ecology. Ecological Modelling, 135: 147–186. meters by optimizing catch effort allocation. Researches on Population Ecology, 31: 305–310. Haynie, A., and Layton, D. 2010. An expected profit model for mon- etizing fishing location choices. Journal of Environmental Mueter, F. J., Bond, N. A., Ianelli, J. N., and Hollowed, A. B. 2011. Economics and Management, 59: 165–176. Expected declines in recruitment of walleye pollock (Theragra chal- cogramma) in the eastern Bering Sea under future climate change. Holland, D., and Sutinen, J. 2000. Location choice in New ICES Journal of Marine Science, 68: 1284–1296. trawl fisheries: old habits die hard. Land Economics, 76: 133–149. Ostrom, E. 2009. A general framework for analyzing sustainability of Holt, R., Lawton, J., Gaston, K., and Blackburn, T. 1997. On the rela- social–ecological systems. Science, 325: 419–422. tionship between range size and local abundance: back to basics. Oikos, 78: 183–190. Overland, J. E., and Wang, M. 2005. The third Arctic climate pattern: 1930s and early 2000s. Geophysical Research Letters, 32: Homans, F., and Wilen, J. 2005. Markets and rent dissipation in regu- L23808. lated open access fisheries. Journal of Environmental Economics and Management, 49: 381–404. Overland, J. E., and Wang, M. 2007. Future regional Arctic sea ice declines. Geophysical Research Letters, 34: L17705. Hunt, G. L., Coyle, K. O., Eisner, L., Farley, E. V., Heintz, R., Mueter, F., Napp, J. M., et al. 2011. Climate impacts on eastern Bering Sea Overpeck, J., Hughen, K., Hardy, D., Bradley, R., Case, R., Douglas, food webs: a synthesis of new data and an assessment of the M., Finney, B., et al. 1997. Arctic environmental change of the Oscillating Control Hypothesis. ICES Journal of Marine Science, last four centuries. Science, 278: 1251–1256. 68: 1230–1243. Parmesan, C., and Yohe, G. 2003. A globally coherent fingerprint of Hunt, G. L., Stabeno, P., Walters, G., Sinclair, E., Brodeur, R. D., Napp, climate change impacts across natural systems. Nature, 421: 37–42. J. M., and Bond, N. A. 2002. Climate change and control of the Perry, A. L., Low, P. J., Ellis, J. R., and Reynolds, J. D. 2005. Climate southeastern Bering Sea pelagic ecosystem. Deep Sea Research II: change and distribution shifts in marine fishes. Science, 308: Topical Studies in Oceanography, 49: 5821–5853. 1912–1915. Hurrell, J. W., and Deser, C. 2010. North Atlantic climate variability: Pradhan, N. C., and Leung, P. 2004. Modeling entry, stay, and exit the role of the North Atlantic Oscillation. Journal of Marine decisions of the longline fishers in Hawaii. Marine Policy, 28: Systems, 79: 231–244. 311–324. Ianelli, J. N., Barbeaux, S., Honkalehto, T., Kotwicki, S., Aydin, K., and Reid, W. V., Chen, D., Goldfarb, L., Hackmann, H., Lee, Y. T., Williamson, N. 2011a. Assessment of the walleye pollock stock in Mokhele, K., Ostrom, E., et al. 2010. Earth system science for the Eastern Bering Sea. In 2011 North Pacific Groundfish Stock global sustainability: grand challenges. Science, 330: 916–917. Assessment and Fishery Evaluation Reports, pp. 51–168. Alaska Richardson, A. J., and Schoeman, D. S. 2004. Climate impact on Fisheries Science Center, National Marine Fisheries Service. http:// plankton ecosystems in the Northeast Atlantic. Science, 305: www.afsc.noaa.gov/REFM/docs/2011/EBSpollock.pdf (last 1609–1612. accessed 4 January 2012). Sanchirico, J., Malvadkar, U., Hastings, A., and Wilen, J. 2006. When Ianelli, J. N., Hollowed, A. B., Haynie, A. C., Mueter, F. J., and Bond, are no-take zones an economically optimal fishery management N. A. 2011b. Evaluating management strategies for eastern Bering strategy? Ecological Applications, 16: 1643–1659.

Page 19 of 127 Why economics matters for understanding the effects of climate change on fisheries 1167

Smith, M., and Wilen, J. 2003. Economic impacts of marine reserves: Wespestad, V. G., Fritz, L. W., Ingraham, W. J., and Megrey, B. A. 2000. the importance of spatial behavior. Journal of Environmental On relationships between cannibalism, climate variability, physical Economics and Management, 46: 183–206. transport, and recruitment success of Bering Sea walleye pollock Thompson, D. W. J., and Wallace, J. M. 1998. The Arctic Oscillation (Theragra chalcogramma). ICES Journal of Marine Science, 57: signature in the wintertime geopotential height and temperature 272–278. fields. Geophysical Research Letters, 25: 1297–1300. Wilen, J. 2004. Spatial management of fisheries. Marine Resource Economics, 19: 7–20. Walther, G-R., Post, E., Convey, P., Menzel, A., Parmesan, C., Beebee, T. J. C., Fromentin, J-M., et al. 2002. Ecological responses to recent Wyllie-Echeverria, T., and Wooster, W. S. 1998. Year-to-year variations climate change. Nature, 416: 389–395. in Bering Sea ice cover and some consequences for fish distribu- tions. Fisheries Oceanography, 7: 159–170. Ward, J. M., and Sutinen, J. G. 1994. Vessel entry–exit behavior in the Yoneda, M., and Wright, P. 2005. Effects of varying temperature and Gulf of Mexico shrimp fishery. American Journal of Agricultural food availability on growth and reproduction in first time spawn- Economics, 76: 916–923. ing female Atlantic cod. Journal of Fish Biology, 67: 1225–1241.

Handling editor: Audrey Geffen Downloaded from http://icesjms.oxfordjournals.org/ at NW Fisheries Science Center on January 16, 2013

Page 20 of 127 282 Chapter 2: Climatic and economic drivers of the Bering Sea walleye pollock (Theragra 283 chalcogramma) fishery: implications for the future 284

285 Alan C. Haynie and Lisa Pfeiffer

286 The PDF from our publication follows.

Page 21 of 127 841 ARTICLE

Climatic and economic drivers of the Bering Sea walleye pollock (Theragra chalcogramma) fishery: implications for the future

Alan C. Haynie and Lisa Pfeiffer

Abstract: This paper illustrates how climate, management, and economic drivers of a fishery interact to affect fishing. Retro- spective data from the Bering Sea walleye pollock (Theragra chalcogramma) catcher–processer fishery were used to model the impact of climate on spatial and temporal variation in catch and fishing locations and make inferences about harvester behavior in a warmer climate. Models based on Intergovernmental Panel on Climate Change scenarios predict a 40% decrease in sea ice by 2050, resulting in warmer Bering Sea temperatures. We find that differences in the value of catch result in disparate behavior between winter and summer seasons. In winter, warm temperatures and high abundances drive intensive effort early in the season to harvest earlier-maturing roe. In summer, warmer ocean temperatures were associated with lower catch rates and approximately 4% less fishing in the northern fishing grounds, contrary to expectations derived from climate-envelope-type models that suggest fisheries will follow fish poleward. Production-related spatial price differences affected the effort distribu- tion by a similar magnitude. However, warm, low-abundance years have not been historically observed, increasing uncertainty about future fishing conditions. Overall, annual variation in ocean temperatures and economic factors has thus far been more significant than long-term climate change-related shifts in the fishery's distribution of effort. Résumé : Le présent article illustre comment les interactions du climat, de la gestion et des déterminants économiques d'une pêche influencent l'activité de pêche. Des données rétrospectives sur la pêche au goberge (Theragra chalcogramma) au navire-usine de la mer de Behring ont été utilisées pour modéliser l'impact du climat sur les variations spatiales et temporelles des prises et des lieux de pêche, et pour tirer des inférences sur le comportement des exploitants en cas de réchauffement climatique. Des modèles basés sur les scénarios du Groupe d'experts intergouvernemental sur les changements climatiques prédisent une diminution de 40 % de la glace marine d'ici 2050, ce qui entraînera une hausse des températures de la mer de Behring. Nous constatons que des variations de la valeur des prises se traduisent par des comportements différents selon la saison de pêche (hiver ou été). En hiver, des températures chaudes et de fortes abondances suscitent un effort intense au début de la saison afin de récolter les individus issus d'œufs a` maturation précoce. En été, des températures océaniques accrues sont associées a` des taux de prise plus faibles et a` une réduction d'environ4%del'activité de pêche dans les lieux de pêche plus nordiques, contrairement aux prévisions de modèles de type enveloppe climatique qui suggèrent que les pêcheurs suivront les poissons vers les pôles. Les

For personal use only. variations spatiales des prix associées a` la production ont une incidence semblable, en termes de magnitude, sur la répartition de l'effort. Cependant, le fait que des années plus chaudes de faible abondance n'aient pas encore été observées se traduit par une incertitude accrue concernant les conditions de pêches futures. Dans l'ensemble, les variations annuelles des températures océaniques et des facteurs économiques ont, a` ce jour, eu un effet plus important que les changements a` la répartition de l'effort de pêche associés aux changements climatiques a` long terme. [Traduit par la Rédaction]

Introduction Lehodey et al. 2003; Perry et al. 2005). This has led to predic- The economic decision-making process makes predicting the tions of fisheries following fish abundances toward the poles effects of climate change on fisheries more complicated than and potential increases in revenue due to higher productivity the already complex task of predicting the effects of climate in high latitude regions. change on fish populations. In many regions, fish harvesters Cheung et al. (2010) argue that global-scale projections of the are important components of ecosystem dynamics. The fishing effects of climate change on fisheries are useful for developing decisions of harvesters can be effectively modeled as a function policy scenarios and contributing to assessments of marine-related of a variety of economic factors, including the spatial distribu- socioeconomic issues. We argue, however, that the economic drivers tion of expected catches, fish and fuel prices, travel costs, and of a fishery, as well as the management structure, are complex, vary variation among fishing vessel capabilities (Bockstael and on a local level, and are impacted by climate factors in many ways. Opaluch 1983; Eales and Wilen 1986; Smith and Wilen 2003). However, predictions of how fisheries (as opposed to fish Thus, when predicting how climate will affect a fishery, it is vital to stocks) will respond to climate change have not included these consider the characteristics of the ocean dynamics, region, species, economic components and have often come from large-scale harvesting fleet, and management structure that are unique to par- models that predict shifts in the climate envelopes of species' ticular fisheries. These characteristics determine the nature of inter- ranges and changes in primary productivity (Cheung et al. 2010; actions between harvesters and their target species. Ignoring or Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14

Received 7 June 2012. Accepted 5 April 2013. Paper handled by Associate Editor Marie-Joëlle Rochet. A.C. Haynie. Economics and Social Sciences Research Program, Resource Ecology and Fisheries Management Division, Alaska Fisheries Science Center, National Marine Fisheries Service, NOAA, Seattle, WA 98115, USA. L. Pfeiffer. Economics and Social Sciences Program, Fishery Resource Analysis and Monitoring Division, Northwest Fisheries Science Center, National Marine Fisheries Service, NOAA, Seattle, WA 98115, USA. Corresponding author: Lisa Pfeiffer (e-mail: [email protected]).

Can. J. Fish. Aquat. Sci. 70: 841–853 (2013) dx.doi.org/10.1139/cjfas-2012-0265Page 22 of 127 Published at www.nrcresearchpress.com/cjfas on 10 April 2013. 842 Can. J. Fish. Aquat. Sci. Vol. 70, 2013

Fig. 1. The Eastern Bering Sea and the fishing areas of the catcher–processor fleet. Points represent the catch-weighted mean center of the distribution of fishing hauls by season. The borders of zones 1 through 4, which are referred to in the text, are shown.

Longitude 177° 180° 177° 174° 171° 168° 165° 162° 159° 156°

60° Area of detail

North America Zone 4

58° 2008 Pacific Ocean 2007 Ber 2009 2006 2005 ing

Sea Alaska

shelf 20032004

ude 56° 1999 2000 Eastern Bering Sea

atit 2001

L Zone 3 2002

Zone 2 54° A-season mean center of effort B-season mean center of effort Zone 1 ! Ports Zones Bering Sea shelf

52° ! ! Akutan 013026065 km Dutch Harbor

abstracting from these details, as a global-scale model must do, may zero after 2005, while fishing in the northernmost region (zone 4 lead to inaccurate predictions of fishery adaptation. shown in Fig. 1) expanded (Fig. 2b). Fishing in the winter season In this paper, we explore the changes that have occurred in the underwent no such marked redistribution (Fig. 2a). timing and location of the catcher–processor sector of the Bering In this paper, we investigate the driving factors that have ac- Sea (Alaska) walleye pollock (Theragra chalcogramma) fishery over companied these changes, and whether they are related to cli- 1999–2009, a period in which there has been considerable varia- mate variation. The “Background” section provides a description

For personal use only. tion in ice cover, temperature, and total walleye pollock abun- of the fishery and its characteristics. We then introduce a simple dance. Polar and subpolar ecosystems are expected to continue to structural diagram that illustrates how climate may affect har- experience changes in seasonal sea ice cover and ocean properties vesters' decisions about when and where to fish and use it to guide because of climate change (Grebmeier et al. 2006; Overpeck et al. our analysis. We find that while climate factors have had a role in 1997). Analysis of fishing behavior during past warm climate re- the changes in the distribution of effort, these changes are incon- gimes may be informative for predicting future fishing patterns in sistent with the “northward march” prediction of climate enve- the walleye pollock fishery, as climate change is expected to de- lope models for this particular fishery, because the northward crease the extent of winter ice cover and increase the frequency of shift occurred in the coldest years of the time period and was years characterized by warmer ocean conditions. For commercial driven by oceanographic patterns and market conditions. We also walleye pollock harvesters, climate change may lead to vessels look at the effect of climate and fishing location on the average traveling farther and incurring greater fuel costs and a greater distance traveled and distance-per-tonne of catch in a trip, proxies burden being placed on small vessels that are unable to travel as for travel costs. Finally, we investigate whether predictions can be far. It may affect the abundance of fishable biomass and thus total made from the observed data about the future course of the tim- allowable catches (TACs) and fishery revenue (Criddle et al. 1998; ing and spatial distribution of the walleye pollock fishery as win- Ianelli et al. 2011b). Climate change may have implications for the ter ice cover is reduced and the Bering Sea warms. This study complex marine ecosystem of the Bering Sea, of which walleye integrates climatic, biological, economic, and institutional char- pollock is an integral part (Hunt et al. 2002; Springer 1992). acteristics of an economically important fishery to empirically The premise of this paper is that the investigation of the mech- investigate the drivers of change in the fishery. anisms driving fish harvester behavior is essential to understand- ing patterns and changes in the distribution of fishing. A Materials and methods summary of the changes in the spatial distribution of the pollock catcher–processor fishery is useful for setting up the analyses. Background Since 1999, there has been a substantial shift in the distribution of The Bering Sea walleye pollock fishery is the largest commercial

Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14 fishing toward the northern reaches of the Bering Sea shelf off the fishery in the United States and has an annual product value of coast of Alaska (Fig. 1), which at first glance corroborates predic- over $1 billion (Hiatt et al. 2011). The TAC of Bering Sea walleye tions from climate envelope models. The mean center of the dis- pollock is split among several fishing sectors; our focus is on the tribution of winter season fishing has remained stable, while the offshore catcher–processor sector. The catcher–processor sector distribution of summer season fishing is centered much farther to of the fishery consists of 16 large midwater–pelagic trawling ves- the north in 2005–2009 than in 1999–2004. This change resulted sels that catch approximately 50% of the annual Bering Sea wall- from a shift in summer season effort. Effort in the most southern eye pollock TAC. The American Fisheries Act ended a seasonal regions of the fishing grounds (zone 1 shown in Fig. 1) fell to nearly race for fish in the fishery by allowing the formation of coopera-

Page 23 of 127 Published by NRC Research Press Haynie and Pfeiffer 843

Fig. 2. Change in the distribution of fishing effort by season and zone, 1999–2009. The location of the zones is displayed in Fig. 1.

(a) A-season (winter season) (b) B-season (summer season) 100 100 90 90 80 80 70 70 Zone 4 60 60 Zone 3 50 50 Zone 2 40 Zone 1 40 30 30 20 20 10

Distribution of effort by zone (%) 10 0 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 0 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

tives and allocation of quota to individual vessels beginning in Fig. 3. Average annual prices for walleye pollock roe (circles), 1999. This transition to “catch share” management changed fish- surimi (squares), and fillet (triangles). Standard deviations (SDs) ing dramatically as vessels went from catching as large a share of shown are the SDs across vessels of the vessel-level average prices. the common pool TAC as possible to maximizing the value of each fish caught as part of the vessels allocated share of TAC (Felthoven 2002; Wilen and Richardson 2008). Product recovery rates (the 16 000 proportion of a fish that is processed into marketable products) increased and vessels slowed processing to a rate that optimized the flow of fish through the vessels' factories. The walleye pollock fishery consists of two seasons: the 12 000 A-season (winter season) begins on 20 January and the B-season (summer season) begins on 10 June. During the A-season, vessels target roe-bearing fish that have aggregated to spawn. The abun- −1 dance and quality of roe varies spatially and generally decreases in 8 000

value as the season advances and the roe becomes over-mature. SD ($)•t and The highest value roe occurs 3–4 weeks prior to spawning. Wall- For personal use only. eye pollock spawning in the Bering Sea has a seasonal spatial Price 4 000 pattern, with the earliest spawning commencing in February (Bacheler et al. 2010; Ciannelli et al. 2007; Francis and Bailey 1983). Vessels follow the maturing roe such that roe is produced over the entire A-season. Roe is a high-value product in the Japanese con- 0 sumer market and may sell for over $20 000·t−1 for extremely 2002 2003 2004 2005 2006 2007 2008 2009 −1 high-quality lots; prices from 2003 to 2009 averaged $12 000·t . Year After roe is removed, the remainder of the fish is processed into other products. Depending on prices, roe can be a 50%–200% “bo- nus” on A-season fishing. The value of roe relative to other prod- Fishery-independent data derived from an annual bottom trawl ucts has decreased over the time period considered (Fig. 3). On survey of the Eastern Bering Sea (conducted by the Alaska Fisher- average, 50% of A-season revenue comes from roe, and total reve- ies Science Center groundfish stock assessment program in June nue per metric tonne of TAC is nearly 50% higher in the A-season. and July; RACE 2011) is used as a basic measure of the productivity In the B-season, schools of fish disperse along the outer Bering of a fishing location or expected catch per unit effort (CPUE). The Sea shelf, where they feed and gain mass throughout the season. survey timing corresponds with the beginning of the summer Surimi (an intermediate fish paste product used for imitation crab fishing season. The survey data were trimmed to the fishing area and other products) and fillets of different sizes and qualities are shown in Fig. 1, resulting in a dataset of approximately 200 sam- produced in the B-season and on average comprise 35% and 49% of pled areas per year, weighted by sampling density, from 1990 seasonal revenue, respectively. Roe is encountered only in small through 2009. This longer time series is used for the analysis of quantities in B-season. empirical relationships among biological factors. CPUE by species (in t·ha−1 trawled) is measured at each survey station, allowing the Data construction of mean CPUE by spatial zone (Fig. 1). The distribution of fishing is derived from haul-level data from Fishery-dependent CPUE was not used in this analysis to repre-

Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14 the NOAA Fisheries' Alaska Fisheries Science Center's North Pa- sent expected catch for a variety of reasons. First, it is contingent cific Groundfish Observer Program. Information on the time, haul upon harvesters' decisions to fish in an area and could result in no duration, exact location, species composition, and total tonnage observations of CPUE in areas when little fishing occurred. In the of each haul is recorded by an onboard government-trained ob- catcher–processor fleet, there exists the additional complication server. Because the focus of this paper is the impact of climate on of the production process. Vessels do not necessarily maximize fishing behavior, we excluded fishery data before the incentive CPUE; according to interviews with vessel operators, vessels fish structure was transformed by the change to cooperative manage- along the edge of walleye pollock aggregations to catch a large but ment in the fishery in 1999. manageable volume of fish. This optimizes flow through the fac-

Page 24 of 127 Published by NRC Research Press 844 Can. J. Fish. Aquat. Sci. Vol. 70, 2013

Fig. 4. The extent of the cold pool in 2004 (a warm year) and 2009 (a cold year), as measured by the annual summer bottom trawl survey of the Eastern Bering Sea.

Longitude 180° 177° 174° 171° 168° 165° 162° 159° 156°

Area of Detail 60° North America

Pacific Ocean

58°

Alaska ude

it 56° at L Pribiloff Islands

Extent of Cold Pool 54° <1.5 C (2009) <1.5 C (2004) Unimak Island Bering Sea Shelf

Bogoslof Island Akutan 02120 40km 52° Dutch Harbor

tory, which prioritizes constant catch rates. The resulting CPUE 2006). Cold-year regimes are characterized by a very large cold may be “hyper-stable”, showing little variation in periods or areas pool that extends southward into the inner shelf toward the of high and low abundance (Hilborn and Walters 1992). Finally, Alaska Peninsula, while in warm years the cold pool is much vessels have consistently heterogeneous average levels of CPUE, smaller (Fig. 4). The years 1999 and 2007–2009 were exceptionally even for a given level of abundance because of their horsepower cold years, while 2002–2005 were warm years in the Bering Sea. or net size or their ability to travel to the best fishing areas, ne- The size of the cold pool is measured as the percentage of the

For personal use only. cessitating standardization (Maunder and Punt 2004). bottom trawl survey area <1.5 °C (as recorded by the annual bot- Total abundance of walleye pollock from 1990 to 2009 is ob- tom trawl survey). An ice cover index, which is the annual nor- tained from estimates of the age 3+ (fishable) biomass in the East- malized anomaly from the average ice concentration for ern Bering Sea modeled and estimated for the annual stock January–May, is obtained from the NOAA Bering Climate Project. assessment process using data from the annual bottom trawl sur- Monthly average sea surface temperatures for the Bering Sea were vey and a semi-annual acoustic survey (Ianelli et al. 2011a). obtained from the NOAA National Climate Data Center (NOAA Prices were obtained from the Alaska Department of Fish and 2012). Game's Commercial Operator's Annual Report. Prices are annual, vessel-specific averages of prices received for each product type Methods sold in that year, from 2002 to 2009. Weekly production data that An illustrative diagram of the environmental, biological, and include the amount of each product type produced are from economic mechanisms through which climate may affect the dis- NOAA Fisheries' Alaska Region. Vessel-level price data are very tribution of walleye pollock fishing effort is provided (Fig. 5, de- limited prior to 2002, so in analyses that include prices, parame- rived from a more detailed figure in Haynie and Pfeiffer 2012). The ters are checked for robustness to inclusion of the years prior to arrows represent the direction of causality, and the solid lines 2002. Analyses that do not include prices are checked for robust- represent mechanisms that are investigated in this paper. Each ness to the exclusion of the years prior to 2002. numbered link is referred to in the text. Dashed lines represent Winter conditions such as temperature, ice cover, and prevail- mechanisms that we make no attempt to model explicitly, either ing wind drive the characteristics of the ecosystem and persist because they are environmental or biological mechanisms or be- through much of the year. The Bering Sea is partially ice-covered cause the available data are insufficient to test them. Dotted lines from December through April, and the extent of this sea ice im- represent mechanisms that are likely to occur noncontemporane- pacts ocean conditions for the rest of the year. How far the sea ice ously or have both year-of and lagged effects. For example, TAC extends into lower latitudes is determined by atmospheric tem- may affect total abundance through both the contemporaneous peratures and wind generated by the Aleutian Low Pressure Sys- removal of fish from the ecosystem and through its effects on

Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14 tem (Overland and Pease 1982). During the spring, the upper recruitment, which may take several years to impact abundance. portion of the water column is warmed as air temperatures and We do not attempt to model the lagged effects here. solar insulation increase, but cold water remains at depth, creat- A simple model of the harvesters' fishing location choice deci- ing a pool of arctic water, called the cold pool. Walleye pollock are sion involves a harvester choosing an area to fish that will maxi- a subarctic, pelagic species that generally avoid the cold pool mize the profit from fishing (Fig. 5a). This model is useful for (Kotwicki et al. 2005; Wyllie-Echeverria and Wooster 1998). The focusing attention on how climate may directly impact the factors cold pool persists into the warm season; its size and persistence is that influence how walleye pollock harvesters make decisions determined by the extent of the winter ice cover (Grebmeier et al. about when and where to fish by considering the characteristics

Page 25 of 127 Published by NRC Research Press Haynie and Pfeiffer 845

Fig. 5. Structural diagram of the effects of environmental characteristics on the distribution of fishing effort. The arrows represent the direction of causality, and the solid numbered lines represent mechanisms that are investigated in this paper. Dashed and dotted lines represent mechanisms that we do not model explicitly, either because they are environmental or biological mechanisms or because the available data are insufficient to test them. Dotted lines represent mechanisms that are likely to occur noncontemporaneously.

-

-

of potential choices, such as the quantity and value of fish they important relative to E(Pricej) and travel costs because a harvester expect to catch and the cost of traveling to different areas. will be more constrained by season length. Any compensatory

Consider a harvester i who makes a choice between fishing effect of E(Qj) increases can be captured through the effect of total locations j, j =1,…,J. The harvester will receive revenues from the abundance. TAC may be related to prices if the change in TAC is quantity caught and the price received for the catch from the large enough to affect world prices (Herrmann et al. 1996). The chosen location. The harvester must incur the cost per kilometre sector supplies about 3% of the global whitefish fillet market but (costs) of traveling to the chosen location, which is distance Dj as much as 28% of the surimi market, so aggregate quantity from port. All costs that vary by location choice are assumed to be changes could have some impact on the supply (FAO 2009), but captured in the costs parameter. The harvester's expected profit individual catcher–processers are assumed to be price-takers. from fishing in location j is E(Profitijt)=E(Qijt)·E(Priceijt) – costs · Dijt, Given these assumptions, expected CPUE (which determines Q), where E(.) is the expectations operator, Qj is the quantity of catch expected prices, and travel costs are the main variable compo- in location j, which is a function of the CPUE at location j, and nents in a harvester's decision-making process, while TAC con- Pricej is the unit value of fish caught in j that can implicitly incor- strains fishing. porate the price of subproducts if production is spatially variable. We use a variety of models to link behavioral responses of wall-

For personal use only. In a fishery with catch shares such as the walleye pollock fishery, eye pollock harvesters to economic and climate factors. This anal- a harvester is constrained by his or her individual allocation of ysis is not a structural model of micro-behavior, such as a spatial TAC. Thus, a harvester must choose a location to fish in each time discrete choice model would be (Abbott and Wilen 2011; Eales and period t = 1,…, T, where T is the end of the season, to maximize Wilen 1986; Haynie and Layton 2010). Limited data and complex expected total profit subject to the constraint that the total quan- structural assumptions that are unreasonable (using current tity of fish caught must be no greater than the vessel's share of the methodology) for the walleye pollock catcher–processor fleet T TAC, or ͚tϭ1 Qit ≤ TACi. A harvester will choose location j in time limit our ability to estimate this type of model. Rather, we take a period t if the expected profit from j (expected revenues minus reduced-form approach to analyze the relationships between cli- travel costs) is greater than that from all other alternative loca- mate factors and the distribution of fishing in the walleye pollock tions. All other things being equal, a harvester is less likely to fishery through the mechanisms outlined in Fig. 5. Homoskedas- travel to a more distant location because of greater travel costs ticity is tested for, and heteroskedasticity-robust standard errors (which could include fuel, the time cost of travel, or safety). However, are used. The characteristics of the fishery, as well as the observed an increase in E(Qj) and (or) E(Pricej) would increase the probability of changes, give guidance about which mechanisms are important. traveling to location j to fish, all other things being equal. Because of the differences in the value of walleye pollock between The manner in which vessels trade off these factors is compli- the A- and the B-seasons, the effects of climate are expected to vary cated and variable across vessels and among years. The effect of by season. TAC on distance traveled is ambiguous, because it constrains fish- ing (a vessel's TAC must be fished within the season) but is often A-season correlated with abundance (because it is generally based on stock The difference in the value per tonne of harvested catch (prices) surveys and abundance projections), which may affect CPUE. We in areas where roe-bearing fish are located causes harvesters to exploit several characteristics of the walleye pollock fishery to focus their effort on aggregations of prespawning walleye pollock separately identify the effects of TAC from the effect of abundance in the A-season. The location of fishing depends on the timing, through CPUE. In this fishery, individually allocated TAC is aggre- location, and progression of spawning. Large-scale spatial shifts

Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14 gated between vessels owned by the same company. If TAC is low, are unlikely if spawning location depends on bathymetric and a company may choose to not fish one or several of its vessels. In other physical features (Bacheler et al. 2010), although some spa- addition, in the Bering Sea there is a 2 million t “ecosystem cap” tial trade-offs exist between areas of high-recovery, lower-value that has constrained walleye pollock TAC to 1.5 million t to allow roe and low-recovery, high-value roe. Processors produce roe in fishing for other species even when abundance would allow for a the A-season regardless of the prices of other products, though higher TAC. This results in variation of vessel-level TAC that is not harvesters will concentrate more effort on roe if roe prices are completely correlated with variation in total abundance. Thus, an high relative to other products (Paul et al. 2009). Rather, a more

increase in TAC, holding other factors equal, makes E(Qj) more important variant in A-season fishing is the timing of the peak of

Page 26 of 127 Published by NRC Research Press 846 Can. J. Fish. Aquat. Sci. Vol. 70, 2013

roe value (link 3 in Fig. 5). Shifts in the temporal distribution of extent in March, so cold, high-ice years may prompt harvesters to fishing may occur as harvesters choose when to begin fishing in start earlier if ice has the potential to interfere with fishing (link the A-season. We explore several possible explanations for this 4). In addition, they may be more likely to delay the start of the choice. Biological evidence suggests that walleye pollock spawn season if TAC is low (link 8). If TAC is high, vessels may expend earlier in warm years (Smart et al. 2012). However, ice advances more effort early in the season to have time to catch their entire from the Arctic into the Bering Sea from January through April allocation. We empirically estimated how these factors have in- and can restrict fishing in some areas. Ice is often at its greatest fluenced decisions about when to start fishing with a logit model:

1 (1) Pr (Early trip ϭ 1) ϭ it ϩ Ϫ ␤ ϩ ␤ ϩ ␤ Ϫ1 ϩ ␤ ϩ ␤ ϩ ␤ (1 exp͕ [ 0 1TACit 2(Priceroe · Pricesurimi )it 30Icet 4LaggedSSTt 4Abundance]͖)

−1 Early tripit is a dichotomous variable equal to 1 if vessel i started where the Zone 4 · Zone 1 CPUE Ratiot is the annual ratio of fishing within the first 5 days of the opening of the A-season in average bottom trawl survey CPUE from zones 4 and 1, and %Cold-

year t, TACit is the vessel-specific seasonal total catch of walleye poolt is the percentage of the survey area <1.5 °C. If fish are more pollock (the fleet catches nearly 100% of their TAC each season, so concentrated in the north compared with the south by the cold −1 ␤ total catch is similar to TAC), (Priceroe · Pricesurimi )it is the vessel- pool, 1 is expected to be positive. If fish are driven out of the northern fishing regions by the cold pool, ␤ is expected to be specific roe to surimi price ratio, Icet is the ice cover index, 1 negative. High abundance is expected to increase CPUE (link 6), LaggedSSTt is the average sea surface temperature for several but it may not increase in all areas uniformly. In particular, the months prior to the start of the season, and Abundancet is the estimated population of age 3+ walleye pollock. The environmental– range of walleye pollock may expand in response to increased ␤ biological mechanisms that lead to early walleye pollock spawn- abundances (Holt et al. 1997). If 2 is positive, increases in abun- ␤ ing are not well enough understood to predetermine which dance increase CPUE in the north relative to the south, and if 2 is months to include in the SST measure. Two seasonal temperature negative, increases in abundance increase CPUE in the south rel- ␤ averages are presented here: the previous year's average summer ative to the north. If 2 is zero, abundance increases CPUE uni- SST (June–August) and the previous year's fall SST (September– formly. December). Other combinations of the previous summer and fall B-season: value of the catch SSTs were used as well, with similar results. Prices affect harvester behavior through the choice of a type of fish to target to produce a product to maximize expected profit. B-season: effect of climate on the distribution of walleye Both fish size and uniformity affect value, with larger fish typi- pollock biomass and CPUE cally but not always more valuable (it depends on product prices The analysis of the B-season is focused on how spatial variation at a given time). Different sizes of fish are located in different in CPUE and prices, as well as total abundance, TAC, the timing of areas of the fishing grounds in the B-season (Bailey et al. 1999; fishing, and other factors, affected the shift in fishing from the Lynde et al. 1986). Haul-level size uniformity of fish allows the south to the north.

For personal use only. onboard factory to recover more flesh. In a dynamic and con- First, we focus on CPUE. The distribution of the target species stantly changing process that varies by vessel, vessels choose how affects spatial variation in expected CPUE, which is measured to trade off expected lower levels of high-value fish with higher using fishery-independent estimates of CPUE. Water temperature catch rates of lower-value fish. Thus, the prices that harvesters patterns (the size of the cold pool) are expected to affect the spa- face can affect their decisions about where to fish (link 1). tial distribution of walleye pollock because they avoid the cold We econometrically examine the role that spatial variation in pool (Barbeaux 2012; Francis and Bailey 1983; Kotwicki et al. 2005). fish value had on observed fishing locations. Intra-annual price The mechanisms relating the size of the cold pool to CPUE differ- data are not available, which limits the ability to estimate the ences are complicated; for example, the shape and timing of the relationship between a vessel's location choice and the value of cold pool may interact with the cold water avoidance behavior the catch from each location. It also makes it more difficult to and affect migration patterns, resulting in varying levels of con- estimate a direct link between environmental factors and spatial centration of fish along the Bering Sea shelf. The cold pool may variation in the value of the catch. Instead, we tested to determine push walleye pollock out of the most northern fishing areas, it if value varied depending on where it was caught. Using prices, may concentrate fish that are in the north into a narrower band production, and observer data, value per tonne ($·t−1) of catch in a along the Bering Sea shelf (see Fig. 4), or it may concentrate fish season (total amount of each product type, multiplied by the ves- over the entire fishing area. The spatial distribution of walleye sel average price of each product, summed and divided by total pollock determines spatial expectations of CPUE (link 5). Because catch) was regressed on the percentage of vessel i's catch from

the most drastic changes in the distribution of fishing occurred in zone 4 in year t (%Zone 4it), before and after 2005 (to model the the most southern (zone 1) and the most northern (zone 4) regions shift in fishing location): of the fishing grounds, we focus on the CPUE differences between Ϫ1 ϭ ␤ ϩ ␤ ϩ ␤ these zones that may have driven the changes in effort patterns (3) Value · tit i t 1 Before 2005 · %Zone 4it and the shift of the mean center of the distribution of fishing to ϩ ␤ ϩ␧ 2 After 2005 · %Zone 4it it the north. The zone 4 to zone 1 ratio of CPUE is relevant because

Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14 generally speaking, CPUE increases the farther north a vessel trav- Ϫ1 els, but costs increase as well. As the CPUE ratio increases, a vessel Value tit is in natural logarithms so the coefficients are the is more likely to incur the increased travel costs to fish in the proportionate change in value for a unit change in each indepen- north. We estimate the linear regression: dent variable. Before 2005 and After 2005 are 0–1 variables indi- cating the period. Year indicators allow the intercept to vary over Ϫ time and control for differences in value that may be driven by (2) Zone 4 · Zone 1 1CPUE Ratio ϭ ␤ ϩ ␤ %Coldpool t 0 1 t relative prices, a vessel's effects allow the intercept to vary by ϩ ␤ ϩ␧ 2Abundancet t owner, which controls for company-level product marketing de-

Page 27 of 127 Published by NRC Research Press Haynie and Pfeiffer 847

cisions. The model is estimated as a fixed-effects panel data model effect of prices on the distribution of effort, if we could eliminate ␤ ␤ (Cameron and Trivedi 2005). If either 1 or 2 is significantly dif- the effect of the cold pool? Second, what was the effect of the cold ferent from zero, it indicates a value difference resulting from pool on the distribution of effort, if we could eliminate the effect zone 4 fishing either before or after 2005. of prices? To estimate the first, we compare the predicted distri- The source of any value difference can be investigated. The bution of effort after 2005 using historical prices with the predic- location of fishing may affect production choices (i.e., the propor- tion in which the price ratio was set equal to 1, which corresponds tion of fish processed into surimi versus fillets), which may affect to eliminating any value differences between fish caught in zone 1 the value of the processed product. Three regressions are esti- and zone 4. Then, holding the price ratio equal to one, we com- mated with fixed-effects panel data models: pared the predicted effort distribution assuming all years after 2005 were “typical” cold, average, or warm years. To define the ϭ ␤ ϩ ␤ ϩ ␤ (4) ShareSurimiit i t 1 Before 2005 · %Zone 4it “typical” cold, average, and warm year, we substituted the average ϩ ␤ ϩ␧ 2 After 2005 · %Zone 4it it size of the cold pool in each category (44.4%, 24.1%, and 10.0%, respectively) for the historical size (the results were not signifi- Ϫ1 ϭ ␤ ϩ ␤ ϩ ␤ cantly different if we instead randomly drew from the distribu- (5) Value · tit i t 1 Before 2005 · ShareSurimiit tion of each category). The difference between the prediction of ϩ ␤ After 2005 · ShareSurimi ϩ␧ 2 it it zone 4 effort when all years are assumed “average” and when all (6) Product recovery rate ϭ ␤ ϩ ␤ ϩ ␤ Before 2005 years are assumed “cold” or “warm” represent the magnitude of it i t 1 the effect of the cold pool if prices had no effect on the spatial × %Zone 4 ϩ ␤ After 2005 · % Zone 4 ϩ␧ it 2 it it distribution of effort.

B-season: travel costs where ShareSurimiit is the share of vessel i's B-season production processed into surimi in year t. Equation 4 estimates how changes Finally, one of the principal concerns related to climate change in fishing location affect production choices, and eq. 5 tests if is that vessels will have to travel farther and incur the resulting production changes translate into changes in value. Year indica- higher costs (link 9). It may seem obvious that travel costs would tors control for differences in surimi production that may be be likely to increase as the proportion of fishing in the most driven by relative prices, and vessel fixed effects control for dif- northern zone increases, but vessels may adjust other aspects of ferences in vessels' production processes. Equation 6 estimates fishing to mitigate travel costs. For example, they could travel less the product recovery rate (mass of processed products · mass of between hauls while fishing in the north. We estimated the linear catch−1) as a function of fishing location. If zone 4 fish are more panel data fixed-effects model: uniform, recovery rates from zone 4 fishing could be higher rela- Ϫ1 ϭ ␤ ϩ ␤ ϩ ␤ tive to other zones, increasing value. The dependent variables in (8) Dist · Triprit i 1%Zone 4rit 2Coldpoolt ϩ ␤ ϩ ␤ ϩ ␤ ϩ␧ eqs. 4 and 6 are shares in the interval [0,1], so a fractional logit 3Num · Haulsrit 4Abundancet 5TACt rit model was used, where E(y|x) is modeled as a logistic function ␤ (Papke and Wooldridge 1993; Wooldridge 2002), E(y|x) = exp(x )/ Ϫ1 [1 + exp(x␤)], and estimated using maximum likelihood. The esti- where Dist · Triprit is the total distance (km) traveled in a trip, and mated coefficients were exponentiated to obtain odds ratios. The %Zone 4rit is the percentage of a trip fished in zone 4. A similar odds ratio minus 1 is the proportionate change in the probability model with distance per tonne of catch per trip as the dependent For personal use only. resulting from a unit change in an independent variable. variable was also estimated. Distance per tonne of catch per trip represents the average cost of travel. If harvesters can catch more B-season: the role of climate and prices in the B-season fish by traveling farther, then their travel cost per tonne may effort distribution shift remain constant. The estimated coefficients on total abundance After establishing the link between climate and spatial differ- and TAC demonstrate the effects of easier fishing (higher abun- ences in CPUE and that prices must be included in a model of dance, holding TAC constant) and the need to fish more inten- harvesters' decisions to avoid misspecification, we developed a sively (higher TAC, holding abundance constant). If there are reduced-form model of the share of effort that a vessel expends in more fish present, the total distance traveled in a trip and distance

zone 4, by trip, which is indexed by r (ShareZone 4rit): per tonne is lower, because harvesters need to search less for acceptable rates of CPUE. If TAC is higher, holding abundance ϭ ␤ ϩ ␤ ϩ (7) ShareZone 4rit i 1ColdPoolt constant, harvesters need to fish faster to fish all of their TAC Ϫ before the season ends. If the coefficient on %Zone 4 is positive ␤ PriceZone 4 · PriceZone1 1 ϩ ␤ Abundance ϩ rit 2 it 3 t and significant, then vessels incur greater travel costs (total and ␤ ϩ ␤ Ϫ1 ϩ ␤ ϩ␧ 4TACit 5Hauls · Triprit mMonthrit rit per tonne of catch) to fish in the north. The coefficient on Cold- poolt tests for an independent effect of the size of the cold pool on travel (independent from the effect of the cold pool on travel to The dependent variable is a share, so the fractional logit model the north, which was estimated using eq. 7 and embodied in eq. 8 was used and the estimates are presented as the odds ratio minus Ϫ through the inclusion of %Zone 4 ). A significant negative effect 1. The ratio PriceZone 4 · PriceZone 1 1estimates the effect of rit it would indicate that given their choice of location, vessels travel vessel-level differences in average fish value between the north less when the cold pool is larger, presumably because of the and the south, which depend on production choices and prices. Ϫ higher concentration of biomass caused by fish avoiding the cold The number of hauls in each trip (Hauls · Trip 1) is included be- rit pool. cause longer trips may be more likely to be fished in zone 4. In this

Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14 section of the paper, we treat prices and total abundance as exog- Results enous. This allows us to estimate the effect that annual variation in the size of the cold pool and total abundance had on the pro- A-season portion of vessels' effort expended in zone 4, assuming that their While there was no trend in the spatial distribution of A-season effects are through the ratio of north to south expected CPUE. fishing, there is evidence that the temporal distribution of effort The results of eq. 7 are used to disaggregate the effect of the size varied by temperature regime. Vessels choose when to fish to of the cold pool from price effects to compare their magnitudes. maximize their net revenues; if roe quality and quantity is ex- To do this, consider two thought experiments. First, what was the pected to peak later in the season it would provide an incentive to

Page 28 of 127 Published by NRC Research Press 848 Can. J. Fish. Aquat. Sci. Vol. 70, 2013

Table 1. Logit model of the effects of climate, total allowable catch Value of the catch (TAC), and prices on the probability of taking a fishing trip within the To examine the role that spatial variation in prices had in the first 5 days of the opening of the A-season. observed changes in the fishery, we tested to see if fish caught in (1) Probability of (2) Probability of zone 4 were more valuable (i.e., processed into products with a an early trip an early trip higher total value), which would help to explain the shift in the distribution of effort toward the north beginning in 2005 (Table 3, Lagged summer SST 2.190** (0.69) column 1). While there was no significant value premium for fish Lagged fall SST 1.147* (0.47) caught in zone 4 in the period after 2005, there was a small value TAC (vessel-specific, 0.348** (0.10) 0.344** (0.10) penalty for fish caught in zone 4 relative to other areas before thousand t) 2005 (Table 3). This 0.4% penalty (ϳ$3.2·t−1) for a 1% increase in Ratio of roe to surimi prices −0.175 (0.24) −0.562** (0.19) zone 4 fishing would have made it slightly less likely for vessels to (average of t and t–1) fish in the north before 2005. This difference disappeared after Total abundance −0.254 (0.19) −0.234 (0.17) 2005. (million t of age 3+) Ice cover index −0.050 (0.21) −0.581** (0.10) Columns 2–4 explore the source of the disappearance in the Constant −17.444** (6.25) −3.652 (3.14) value penalty for fish caught in zone 4. The location of fishing may affect production choices (the proportion of fish processed into Observations 176 176 surimi versus fillets), which may affect the value of the processed Pseudo-R2 0.444 0.415 product. Column 2 contains estimates of eq. 4, the effect of fishing Note: *, p < 0.05; **, p < 0.01. Standard errors are in parentheses and are in zone 4 on surimi production. A 1% increase in zone 4 fishing led clustered by vessel. General linear mode (GLM) yields very similar results. Years to a greater-than-proportionate (1.7%) increase in the share of included are 1999–2009. Columns contain alternative specifications of sea sur- catch processed into surimi, before 2005. This effect was absent face temperature (SST). after 2005. Column 3 displays estimates from eq. 5, the value difference from producing a higher proportion of surimi. Before begin fishing later in the season. The warmest years (2003–2005) 2005, producing 1% more surimi led to a 0.6% decrease in value per had consistently high effort at the season's start. Cold years were tonne of catch (approximately $4.8·t−1). This effect disappeared more heterogeneous; some vessels delayed the start of the season after 2005. Finally, we tested if there was a difference in the recov- by up to 2 weeks in some of the coldest years. ery rate from catch in zone 4 relative to other zones, before and The factors correlated with a vessel's decision to start a fishing after 2005, but the effect was not significant (Table 3, column 4). trip within 5 days of the opening of the A-season include SST, TAC, These results indicate that before 2005, fishing in zone 4 de- ice cover, and roe prices (Table 1). The probability of early fishing creased the value of catch because a greater percentage of fish was increased with higher lagged SST (which may promote early processed into surimi at a lower value. This would have decreased spawning), higher TAC (harvesters must start fishing earlier to the propensity to fish in the north, given constant CPUE. Neither have time to fish their allocation), and a higher contemporaneous effect was significant after 2005. Before 2005, the zone 4 penalty ice cover index (harvesters may be restricted from some fishing may have disappeared because of increasing surimi prices (Fig. 3) grounds as ice advances later in the spring). If roe prices are low or an increasing propensity to produce fillet from zone 4 fish, relative to other products, the temporal targeting of roe is less especially given high fillet prices in 2008 (Paul et al. 2009 esti- important. Total walleye pollock abundance does not signifi- mated a higher own-price elasticity for fillet than for surimi).

For personal use only. cantly affect the decision to start fishing within 5 days of the season opening. The role of climate and prices in the B-season effort distribution shift B-season The preceding sections have demonstrated that the distribution Effect of climate on the distribution of walleye pollock biomass of walleye pollock in the Bering Sea is related to variation in the and CPUE size of the cold pool, total abundance, and prices. Harvesters In contrast with the A-season, a dramatic change in the distri- make decisions about where to fish based on trade-offs among bution of effort has occurred in the B-season (Fig. 1). The correla- expected CPUE, expected prices, and travel costs, so the distribu- tion among the ratio of north to south CPUE, the size of the cold tion of effort in the fishery is related to climate, at a minimum, pool, and abundance appears to be strong, especially after 1999 through the size of the cold pool and its effect on CPUE. (Fig. 6). Using the entire time period, the relationships are empir- The share of effort that a vessel expends in zone 4, by trip, is ically significant (Table 2). As the size of the cold pool increases by expected to increase when the cold pool is larger, because of the 1% (from a mean of 31%), the ratio of CPUE in zone 4 to CPUE in resulting higher concentration of biomass in the north relative to zone 1 increases by 0.023 (at the mean ratio of 1.737) (Table 2, the south. More northern fishing is observed in years in which the column 2). This is evidence for a concentration of fish in the north cold pool was large, when the price ratio was high, and when in the region outside the cold pool on the Bering Sea shelf in cold abundance was low (Table 4). The effect of abundance occurs years when the cold pool is large. Correlates of the size of the cold through both the effect on the spatial distribution of the stock pool, such as the ice cover index, the timing of ice retreat, average (toward the south in high abundance years, decreasing the north/ bottom temperatures, and average sea surface temperatures were south CPUE ratio), and through overall CPUE. An increase in used as alternative climate indicators, with similar results (only vessel-specific TAC (holding abundance constant) has no statisti- the ice cover index is shown, in column 4). As biomass increases, cally significant effect on the probability of zone 4 fishing. the CPUE ratio decreases, evidence that walleye pollock expand We disaggregated the effects of the size of the cold pool and into the south during periods of high abundance (Table 2, price variation (Table 5). The magnitude of the overall change in

Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14 column 3). However, the size of the cold pool is collinear with effort distribution resulting from the size of the cold pool was biomass (r = –0.39), and with only 20 years of data we cannot similar to the impact of prices. To see this, consider the first identify each effect separately very well (Table 2, column 1). In column of Table 5, which predicts the percentage of effort in zone addition, there have been few low-abundance study years in our 4 using the historical (observed) cold pool size. Holding the price dataset (4 years: 2006–2009), and in particular, there were no ratio equal to one isolates the predicted percentage of effort due warm years with low walleye pollock abundance. This makes it to other factors in zone 4 after 2005 if prices had no effect on the impossible to predict the CPUE ratio in warm, low-abundance spatial distribution of effort. Predicted effort falls by 3.9%, from years without making very strong assumptions. 69.1% to 65.2%. This difference is similar if we assume that all years

Page 29 of 127 Published by NRC Research Press Haynie and Pfeiffer 849

Fig. 6. Time series of the ratio of average CPUE in fishing zones 4 (the northern region) and 1 (the southern region) (black line), the size of the cold pool (percentage of the area surveyed by bottom trawl survey with temperature <1.5 °C) (dashed line), and the abundance of age 3+ walleye pollock (grey line). Abundance is normalized to the year of minimum abundance (2008).

4 60 Zone 4/Zone 1 CPUE rao 50 3 Cold 40 pool 2 30

20 pool cold of Size 1 Abundance Zone 4/zone 1 CPUE Zone 4/zone ratio

10 area <1.5ºC) (% of surveyed Abundance (normalized Abundance (normalized to min) 0 0 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 Year

Table 2. Relationship between the ratio of average zone 4 to average zone 1 CPUE, the size of the cold pool (or in column 4, the ice cover index), and total abundance. (1) Zone 4/Zone 1 (2) Zone 4/Zone 1 (3) Zone 4/Zone 1 (4) Zone 4/Zone 1 CPUE ratio CPUE ratio CPUE ratio CPUE ratio Size of cold pool 0.013 (0.009) 0.023* (0.009) (% of surveyed area) Ice cover index 0.249* (0.103) Total abundance −0.188* (0.076) −0.230** (0.070) −0.145* (0.074) (million t of age 3+) Constant 3.076** (0.956) 0.953** (0.285) 3.862** (0.729) 3.025** (0.741) Observations 20 20 20 20 Adjusted R2 0.374 0.232 0.342 0.476 Note: *, p < 0.05; **, p < 0.01. Standard errors are in parentheses. Homoskedasticity cannot be rejected. Data is from the bottom trawl survey, 1990–2009. Columns present alternative specifications.

Table 3. Regressions of the effect of fishing in zone 4 on the value of catch, before and after 2005. (1) Value (2) Share of production (3) Value (4) Recovery rate (ln of $·t–1 catch) made into surimi (ln of $·t–1 catch) (t production/t catch)

For personal use only. % of catch in zone 4 (before 2005) −0.004** (0.001) 0.017** (0.003) −0.001 (0.001) % of catch in zone 4 (after 2005) 0.0008 (0.001) 0.002 (0.003) 0.003** (0.001) % of production surimi (before 2005) −0.006** (0.001) % of production surimi (after 2005) −0.002 (0.001) Vessel effects Included Included Owner effects Included Included Annual effects Included Included Included Included Constant 6.190* (0.052) −2.201** (0.392) 6.308** (0.066) −0.783** (0.067) N 117 159 117 159 R2 or pseudo-R2 0.88 0.14 0.83 0.01 Note: *, p < 0.05; **, p < 0.01. Robust standard errors are in parentheses. Homoskedasticity cannot be rejected. Columns (1) and (3) use data from 2002–2009 because of price data limitations before 2002 and are estimated as semi-logs so the coefficients can be interpreted as the proportionate change in value from a unit change in the independent variables. Columns (2) and (4) use data from 1999 to 2009, are estimated as a fractional logit model, R2 is pseudo-R2, and the coefficients represent the proportionate change in value from a unit change in the independent variables. Results do not change significantly if 1999–2001 are dropped to match the time period used for columns (1) and (3), which include prices.

after 2005 were cold, average, or warm. This can be compared tance traveled in a trip (Table 6, column 1) and the travel cost with the effect of the size of the cold pool by holding the price (distance) per tonne of catch (Table 6, column 2). Total distance ratio equal to one and comparing the predicted share of effort in per trip is a proxy for total cost per trip, while distance per tonne zone 4 had each of the years after 2005 been typical cold, average, per trip is a proxy for the average cost of a trip. A 1% increase in the or warm years. Predicted effort in zone 4 is 4.2% greater in cold percentage of fishing occurring in zone 4 increased the distance years compared with average (64.5%–60. 3%) and 3.1% lower in traveled in a trip by 0.4% (ϳ10 km) and the distance per tonne of warm years compared with average years (60.3%–57.2%). These catch by 0.3%. This implies that walleye pollock catcher–processor Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14 results indicate that both the increase in the price ratio and the harvesters incurred a higher cost in terms of total travel and increasing number of cold years after 2005 helped to create the distance traveled per tonne to fish in the north, but the effect on observed shift in the distribution of effort toward the north after total distance was very slightly mitigated by catching a larger 2005. quantity of fish per kilometre traveled. Holding the choice of the Travel costs percentage of a trip to spend in the north constant (which is The characteristics of a trip, including the proportion of a trip affected by the size of the cold pool as discussed in the previous spent fishing in the north, are empirically related with the dis- section), a 1% increase in the size of the cold pool decreased the

Page 30 of 127 Published by NRC Research Press 850 Can. J. Fish. Aquat. Sci. Vol. 70, 2013

Table 4. The role of climate, prices, abundance, and TAC in the share of a fishing trip spent in the northernmost region (zone 4). (1) Share of trip (2) AME: EY/EX† (3) Mean of indep. fished in zone 4 (95% CI) variable (SD) Size of cold pool (% of surveyed area) 0.012* (0.01) 0.18% (0.11, 0.34) 30.3 (20.1) Price ratio (zone 4 to avg. of zones 1 and 2) 0.894** (0.27) 0.51% (0.21, 0.81) 1.1 (0.3) Total abundance (million t of age 3+) −0.269** (0.04) −1.21% (−1.58, −0.85) 8.8 (2.3) TAC (vessel-specific, thousand t) 0.027 (0.01) 0.33% (−0.02, 0.67) 23.7 (7.7) No. of hauls·trip–1 0.007 (0.006) 44.0 (17.0) Monthly effects Included Vessel effects Included Constant 0.019** (0.017) N 1009 Pseudo-R2 0.25 Note: *, p < 0.05; **, p < 0.01. Robust standard errors are in parentheses, except for column (3), which shows standard deviation (SD). The model is estimated as a fractional logit, and the coefficients can be interpreted as the proportionate change in value from a unit change in the independent variables. †Average marginal effect (AME) is presented as EY/EX (the elasticity) at the sample means and can be interpreted as the percent change in the share of a trip fished in zone 4 resulting from a 1% increase in the dependent variable. Years included are 1999–2009.

Table 5. Prediction of the average percentage of trip effort directed to zone 4, after 2005, using the historical north to south price ratio and substituting the price ratio equal to 1 after 2005. Historical All years after All years after All years after cold pool 2005, cold 2005, average 2005, warm Prediction of percentage of effort in zone 4 69.1% 68.6% 64.6% 61.6% after 2005, historical prices Prediction of percentage of effort in zone 4 65.2% 64.5% 60.3% 57.2% after 2005, price ratio = 1

Table 6. Determinants of distance traveled in a trip and travel cost per tonne of catch. (1) Distance·trip–1 (2) Distance·t–1·trip–1 (3) Mean of indep. (log of km) (log of km·t–1) variable (SD) % of fishing in zone 4 0.004** (0.00) 0.003** (0.00) 48.9 (41.8) Size of cold pool (% of survey area) −0.003** (0.00) −0.002* (0.00) 30.3 (20.1) No. of hauls 0.010** (0.00) −0.013** (0.00) 44.0 (17.0) For personal use only. Total abundance (million t of age 3+) −0.039** (0.01) −0.048** (0.01) 8.8 (2.3) TAC (vessel-specific, thousand t) −0.001 (0.00) −0.012** (0.00) 23.8 (7.7) Constant 7.414** (0.11) 1.095** (0.15) Vessel fixed effects Included Included N 1009 1009 R2 0.421 0.391 Note: *, p < 0.05; **, p < 0.01. Robust standard errors are in parentheses, except for column (3), which shows standard deviation (SD). Models are in semi-logs so the coefficients represent the proportionate change in distance from a unit change in the independent variables from their means. Heteroskedasticity has been modeled using feasible generalized least squares; the % of fishing in zone 4 and the number of hauls were included in the weighting model because they caused a majority of the heteroskedasticity. Results from a GLM are similar.

distance traveled in a trip by 0.3% and distance per tonne per trip economic factors and that these economic factors vary with by 0.2%. This indicates that the concentration of fish caused by a changes in environmental conditions. The mechanisms that de- larger cold pool results in less travel, independent from fishing termine fishing patterns are specific to this ecosystem and the location choice. The regressions also include the effect of the total fishing fleet. For example, if one failed to account for distinct number of hauls in a trip, total abundance, and vessel-specific economic drivers that characterize the two seasons of the Bering TAC. Higher abundance decreased both total distance traveled Sea walleye pollock fishery, the annual mean center of the distri- and distance per tonne, because harvesters needed to search less bution of fishing would obscure the changes taking place in the for acceptable rates of CPUE when abundance was high. The coef- fishery. Without disaggregating the seasons, one would see a dis- ficients on TAC estimate the effect of the need to fish more inten- tribution of effort centered in the middle of the fishing grounds, sively. Higher TAC is associated with a lower distance per tonne, with a small degree of northward shift over 1999–2009. In reality,

Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14 but no change in total distance traveled per trip. This may be high prices from roe-bearing fish relatively close to port dominate because vessels took longer trips and caught more fish on each higher CPUE that might be found elsewhere in A-season fishing, trip when TAC was high, negating the distance-related costs of while B-season fishing is driven by the trade-offs vessels make fishing more intensively. among spatially and temporally variable CPUE, prices, and travel costs. Each of these factors can be affected by climate, and the Discussion resulting aggregate distribution of fishing is a function of individ- We have shown that the distribution of fishing in the walleye ual harvesters' location choices over the fishing seasons. However, pollock catcher–processor fishery is influenced by a variety of several complexities inherent in this system push the modeling

Page 31 of 127 Published by NRC Research Press Haynie and Pfeiffer 851

Fig. 7. Summary of the effects of the size of the cold pool and total from those that have been historically observed, the potential for walleye pollock abundance on the intensity of early A-season (winter surprises increases. season) effort, B-season (summer season) CPUE, B-season effort, and The scenarios for which data correspondence does occur pro- B-season travel costs. Years in the sample characterized by varying vide insight into how economic drivers of the fishery affect the abundance and cold pool levels are listed on the horizontal and distribution of effort and how these factors are related to climate. vertical axes. The identifiable effects of annual climate variation on A-season fishing are temporal rather than spatial. Biologically, walleye pol- Low lock are shown to mature earlier in the season in warm years Intensity of early A season (Smart et al. 2012), and we find that lagged SST affects the propor- effort: - tion of vessels actively fishing for walleye pollock at the beginning

North/South CPUE ratio: + X of the A-season. Paul et al. (2009) found complementary results. North/South effort ratio: + (no data) Harvesters consider the shift in the peak of the highest value roe, Travel costs: + 2006 2008 2007 2009 2006 2008 2007 but are also influenced by TAC and the timing of the maximum extent of winter ice. There is evidence that vessels need to start fishing earlier to catch their share of the quota when TAC is high;

00 00 however, warm years in the sample have also been high-TAC Intensity of early A season Intensity of early A season effort: -/+ effort: + years, so we cannot reliably separate their effects. The data and analysis from the A-season suggest that we are unlikely to see Total abundance Total North/South CPUE ratio: + North/South CPUE ratio: - major changes in the location of fishing unless there is a (cur- North/South effort ratio: + North/South effort ratio: - rently unforeseen) regime shift that greatly affects the spatial Travel costs: + Travel costs: - distribution of spawning. High In the B-season the shift in the spatial distribution of effort is 2002 2003 1999 2001 20 2002 2003 related to climate through the cold pool and its effect on the Large (Cold) Small (Warm) spatial distribution of fish. However, the effect can be enhanced 1999 2008 2009 2007 2001 2005 2004 2002 2003 or mitigated by the spatial variation in the value of fish. The Size of cold pool spatial distribution of fish is also dependent upon total abun- dance, which is affected by climate through a lagged process of recruitment and survival. Cold years with large cold pools had higher B-season CPUE in the north relative to the south. This problem beyond the state of the art in spatial discrete choice effect was amplified by relatively low survey abundances in recent modeling of fishing locations. Walleye pollock are an extremely cold years (2007–2009). The combined effects of a large cold pool mobile species, making the temporal stability of endogenously and low abundance caused a divergence between CPUE in the estimated expected CPUE used in discrete choice models problem- north and CPUE in the south and helped to explain the shift in the atic. In addition, catcher–processors make decisions over multi- distribution of fishing to the north. The change in the distribution week trips that involve trading off constant flow through the of value complemented this shift. Increasing and more variable factory with the possibility of higher catch rates at the expense of surimi prices drove the disappearance of a value penalty for fish- lost processing time. Finally, the environment impacts the fishing ing in zone 4 after 2005.

For personal use only. process in a complex and temporally and (or) spatially lagged There has been substantial speculation as to whether climate manner, often at uncertain scales. While these issues are the sub- change will affect commercial fishing through increased travel ject of ongoing research, this paper presents a robust and infor- costs if fisheries shift north and farther from their traditional mative reduced-form analysis of economic drivers of the fishery ports. We find evidence of increased travel per trip and distance and how they are related to local climate conditions. Understand- traveled per tonne of fish caught that is associated with the shift ing these mechanisms is an important foundation for future inte- to zone 4 fishing; however, increases in zone 4 fishing are contem- grated modeling efforts. poraneously associated with colder, low-abundance years. We This paper also identifies one of the difficulties in predicting the also find that TAC has an effect on travel beyond abundance. High effects of climate change in complicated systems with limited abundance reduces search travel, as expected, but high TAC, hold- historical data. Although 20 years of historical biological survey ing abundance constant, makes vessels less willing to expend data and 11 years of fishery data were used, separately identifying travel time for marginal improvements in CPUE or prices. In con- the effects of many important variables is difficult. For example, trast with more general climate-envelope models, our analysis of separate identification of TAC, ice cover, and lagged temperatures the economic drivers of the fishery with the available data suggest in the A-season, or identification of the CPUE ratio, total abun- that a warming trend will lead to a more southerly distribution of dance, and the size of the cold pool in the B-season, is problematic fishing effort by the catcher–processor sector if abundance re- because we simply have no data on warm, low-TAC and low- mains high. This will be driven by the smaller ratio of north to abundance years. Figure 7 illustrates this point by showing the south CPUE, which is driven by a smaller cold pool. In reality, years in our economic dataset in which total abundance was rel- although climate models forecast a warming trend, variation in atively high and low and years in which the cold pool was small annual temperatures is predicted to remain high (IPCC 2007). and large, along with the key results of the preceding analysis. Figure 7 makes it obvious that our findings are not adequate to There was no correspondence between low-abundance, small- make an informed prediction about the distribution of the fishery cold-pool (warm) years, and only 1 year of correspondence be- in warm, low-abundance years. tween high-abundance and large-cold-pool (cold) years. Because It is critical to recognize the importance of management insti-

Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14 biological evidence suggests that the predicted increased fre- tutions in shaping the behavior of harvesters and their response quency of warmer climate regimes may result in decreased to changes in the factors that drive them, including climate. The abundances (Hunt et al. 2011; Mueter et al. 2011), this lack of cor- creation of catch shares in the walleye pollock catcher–processor respondence represents a major limitation in predicting how the fishery in 1999 completely changed the fishery's incentives, so behavior of the catcher–processor walleye pollock fleet will be much so that it is impossible to use data from both periods to affected by climate change. Given the current state of knowledge analyze the effects of climate without accounting for the change. and available data, out-of-sample forecasting results would be Before this change in management, the race to fish caused vessels highly speculative. If a warmer future has characteristics different to fish at full intensity from the moment of season opening, mak-

Page 32 of 127 Published by NRC Research Press 852 Can. J. Fish. Aquat. Sci. Vol. 70, 2013

ing time spent traveling rather than fishing much more expen- under uncertainty: the case of the fishery. J. Environ. Econ. Manage. 10(2): sive. Catch share management allows vessels, companies, and 125–137. doi:10.1016/0095-0696(83)90021-9. Cameron, A.C., and Trivedi, P.K. 2005. Microeconometrics: methods and appli- cooperatives to make spatial and temporal trade-offs to maximize cation. Cambridge University Press, Cambridge. value. Ongoing institutional changes will continue to impact the Cheung, W.W.L., Lam, V.W.Y., Sarmiento, J.L., Kearney, K., Watson, R., Zeller, D., fishery. For example, the implementation of catch shares in the and Pauly, D. 2010. Large-scale redistribution of maximum fisheries catch west coast groundfish fishery enables vessels that fish in both potential in the global ocean under climate change. Global Change Biol. 16(1): 24–35. doi:10.1111/j.1365-2486.2009.01995.x. the walleye pollock fishery in the Bering Sea and the whiting Ciannelli, L., Bailey, K.M., Chan, K.-S., and Stenseth, N.Chr. 2007. Phenological fishery off of Washington and Oregon to coordinate efforts in and geographical patterns of walleye pollock (Theragra chalcogramma) spawn- both regions. Changes in salmon bycatch regulation will also con- ing in the western Gulf of Alaska. Can. J. Fish. Aquat. Sci. 64(4): 713–722. doi:10.1139/f07-049. tinue to impact the timing and location of fishing in the Bering Criddle, K.R., Herrmann, M., Greenberg, J.A., and Feller, E.M. 1998. Climate Sea. The design of management institutions to be flexible to the fluctuation and revenue maximization in the Eastern Bering Sea fishery for potential impacts of climate change is an important policy objec- walleye pollock. N. Am. J. Fish. Manage. 18(1): 1–10. doi:10.1577/1548- tive. For example, anticipating changes in the spatial distribution 8675(1998)018<0001:CFARMI>2.0.CO;2. Davis, A.J., Jenkinson, L.S., Lawton, J.H., Shorrocks, B., and Wood, S. 1998. Mak- and abundance of the stock (or changes in their variance) and ing mistakes when predicting shifts in species range in response to global incorporating them into stock assessments and management warming. Nature, 391: 783–786. doi:10.1038/35842. PMID:9486646. choices could help avoid unintended overfishing. If enough infor- Dichmont, C.M., Pascoe, S., Kompas, T., Punt, A.E., and Deng, R. 2010. On imple- mation on spatial stock dynamics and costs were available, the menting maximum economic yield in commercial fisheries. Proc. Natl. Acad. Sci. 107(1): 16–21. doi:10.1073/pnas.0912091107. PMID:20018676. stock could be managed (in terms of the determination of total Eales, J., and Wilen, J. 1986. An examination of fishing location choice in the annual TAC by the regulating body) to achieve maximum eco- pink shrimp fishery. Mar. Resour. Econ. 2(4): 331–351. nomic yield rather than the current policy of maximum sustain- FAO. 2009. Fisheries and aquaculture statistics. . able yield (Dichmont et al. 2010). Allowing flexibility in the timing Felthoven, R.G. 2002. Effects of the American Fisheries Act on capacity, utiliza- tion and technical efficiency. Mar. Resour. Econ. 17(3): 181–206. and location of harvesters choices (by reducing spatial and sea- Francis, R.C., and Bailey, K. 1983. Factors affecting recruitment of selected gad- sonal limitations, for example) can reduce unnecessary costs im- oids in the northeast Pacific and East Bering Sea. In From year to year. Edited posed by the limitations. Even if regulations were originally by W.S. Wooster. Washington Sea Grant, Seattle, Wash. pp. 35–60. Grebmeier, J.M., Overland, J.E., Moore, S.E., Farley, E.V., Carmack, E.C., efficiently designed, climate-related impacts on fish and fisheries Cooper, L.W., Frey, K.E., Helle, J.H., McLaughlin, F.A., and McNutt, S.L. 2006. can make existing regulations extremely costly. A major ecosystem shift in the northern Bering Sea. Science, 311(5766): 1461– Ultimately, the entire marine ecosystem, of which fisheries are 1464. doi:10.1126/science.1121365. PMID:16527980. an integral part, is being affected by changing climate. In this Haynie, A.C., and Layton, D.F. 2010. An expected profit model for monetizing fishing location choices. J. Environ. Econ. Manage. 59(2): 165–176. doi:10.1016/ research we show that there are a variety of causes of the observed j.jeem.2009.11.001. changes in walleye pollock fishing, and market conditions have Haynie, A.C., and Pfeiffer, L. 2012. Why economics matters for understanding larger impacts on the fishery distribution than trends in climate the effects of climate change on fisheries. ICES J. Mar. Sci. 69(7): 1160–1167. over the study period. Just as ignoring interactions between spe- doi:10.1093/icesjms/fss021. Herrmann, M., Criddle, K.R., Feller, E.M., and Greenberg, J.A. 1996. Estimated cies can lead to erroneous predictions about the effect of climate economic impacts of potential policy changes affecting the total allowable change on the range and abundance of species (Davis et al. 1998), catch for walleye pollock. N. Am. J. Fish. Manage. 16(4): 770–782. doi:10.1577/ ignoring the interaction between human apex “predators” and 1548-8675(1996)016<0770:EEIOPP>2.3.CO;2. their target species can lead to similarly inaccurate predictions. In Hiatt, T., Dalton, M., Felthoven, R., Fissel, B., Garber-Yonts, B., Haynie, A., Kasper- ski, S., Lew, D., Package, C., Sepez, J., and Seung, C. 2011. Stock assessment

For personal use only. some fisheries, excellent data concerning economic drivers of the and fishery evaluation report for the groundfish fisheries of the Gulf of fishery is available. Improving the quality and quantity of eco- Alaska and Bering Sea/Aleutian Islands area: Economic status of the ground- nomic data and using it to support ecosystem studies of fisheries fish fisheries off Alaska. Edited by N.P.F.M. Council, Anchorage, Alaska. will result in research that can better inform management in the Hilborn, R., and Walters, C.J. 1992. Quantitative fisheries stock assessment: Choice, dynamics and uncertainty. Chapman and Hall, New York. face of climate change. Holt, R.D., Lawton, J.H., Gaston, K.J., and Blackburn, T.M. 1997. On the relation- ship between range size and local abundance: back to basics. Oikos, 78(1): Acknowledgements 183–190. doi:10.2307/3545815. Funding for this research was provided by the North Pacific Hunt, G.L., Stabeno, P., Walters, G., Sinclair, E., Brodeur, R.D., Napp, J.M., and Bond, N.A. 2002. Climate change and control of the southeastern Bering Sea Research Board (NPRB publication No. 423) as part of the Bering pelagic ecosystem. Deep Sea Res. Part II: Top. Stud. Oceanogr. 49(26): 5821– Sea Integrated Ecosystem Research Program (BEST-BSIERP publi- 5853. doi:10.1016/S0967-0645(02)00321-1. cation No. 100). Angie Grieg assisted with the processing of cli- Hunt, G.L., Coyle, K.O., Eisner, L.B., Farley, E.V., Heintz, R., Mueter, F., Napp, J.M., Overland, J.E., Ressler, P.H., Salo, S., and Stabeno, P.J. 2011. Climate impacts mate data, and Terry Hiatt assisted with the organization of on eastern Bering Sea foodwebs: a synthesis of new data and an assessment of economic data. Franz Mueter provided a compilation of monthly the Oscillating Control Hypothesis. ICES J. Mar. Sci. 68(6): 1230–1243. doi:10. and annual climate data. Steve Barbeaux, Ron Felthoven, and 1093/icesjms/fsr036. three anonymous reviewers provided helpful comments. These Ianelli, J.N., Barbeaux, S., Honkalehto, T., Kotwicki, S., Aydin, K., and Williamson, N. 2011a. Assessment of the walleye pollock stock in the findings are those of the authors and do not necessarily represent Eastern Bering Sea. In 2011 North Pacific Groundfish Stock Assessment and the views of the National Marine Fisheries Service. Fishery Evaluation Reports North Pacific Fisheries Management Council, Anchorage, Alaska. pp. 51–168. References Ianelli, J.N., Hollowed, A.B., Haynie, A.C., Mueter, F.J., and Bond, N.A. 2011b. Evaluating management strategies for eastern Bering Sea walleye pollock Abbott, J.K., and Wilen, J.E. 2011. Dissecting the tragedy: a spatial model of (Theragra chalcogramma) in a changing environment. ICES J. Mar. Sci. 68(6): behavior in the commons. J. Environ. Econ. Manage. 62(3): 386–401. doi:10. 1297–1304. doi:10.1093/icesjms/fsr010. 1016/j.jeem.2011.07.001. IPCC. 2007. Contribution of Working Groups I, II, and III to the Fourth Assess- Bacheler, N.M., Ciannelli, L., Bailey, K.M., and Duffy-Anderson, J.T. 2010. Spatial ment Report of the Intergovernmental Panel on Climate Change. In Climate and temporal patterns of walleye pollock (Theragra chalcogramma) spawning Change 2007: Synthesis Report. Edited by R.K. Pachauri and A. Reisinger. IPCC, Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by NORTHWEST FISHERIES SCIENCE CENTER on 06/15/14 in the eastern Bering Sea inferred from egg and larval distributions. Fish. Geneva, Switzerland. Oceanogr. 19(2): 107–120. doi:10.1111/j.1365-2419.2009.00531.x. Kotwicki, S., Buckley, T.W., Honkalehto, T., and Walters, G. 2005. Variation in Bailey, K.M., Quinn, T.J., Bentzen, P., and Grant, W.S. 1999. Population structure the distribution of walleye pollock (Theragra chalcogramma) with temperature and dynamics of walleye pollock, Theragra chalcogramma. Adv. Mar. Biol. 37: and implications for seasonal migration. Fish. Bull. (NOAA), 103(4): 574–587. 179–255. doi:10.1016/S0065-2881(08)60429-0. Lehodey, P., Chai, F., and Hampton, J. 2003. Modelling climate-related variability Barbeaux, S.J. 2012. Scientific acoustic data from commercial fishing vessels: of tuna populations from a coupled ocean–biogeochemical–populations dy- Eastern Bering Sea walleye pollock (Theragra chalcogramma). Ph.D. thesis, namics model. Blackwell Science Ltd. pp. 483–494. School of Aquatic and Fisheries Sciences, University of Washington, Seattle. Lynde, C.M., Lynde, M.V.H., and Francis, R.C. 1986. Regional and temporal dif- Bockstael, N.E., and Opaluch, J.J. 1983. Discrete modelling of supply response ferences in growth of walleye pollock (Theragra chalcogramma) in the eastern

Page 33 of 127 Published by NRC Research Press 300 Chapter 3: The effect of decreasing seasonal sea-ice cover on the winter Bering Sea pollock fishery 301

302 Lisa Pfeiffer and Alan C. Haynie

303 The PDF from our publication follows.

Page 34 of 127 ICES Journal of Marine Science

ICES Journal of Marine Science (2012), 69(7), 1148–1159. doi:10.1093/icesjms/fss097

The effect of decreasing seasonal sea-ice cover on the winter Bering Sea pollock fishery

Lisa Pfeiffer and Alan C. Haynie* Downloaded from REFM Division, Alaska Fisheries Science Center, Economics and Social Sciences Research Program, NOAA National Marine Fisheries Service, 7600 Sand Point Way NE, Seattle, WA 98115, USA *Corresponding author: tel: +1 206 5264253; fax +1 206 5266723; e-mail: [email protected]. Pfeiffer, L., and Haynie, A. C. 2012. The effect of decreasing seasonal sea-ice cover on the winter Bering Sea pollock fishery. – ICES Journal of Marine Science, 69: 1148–1159. http://icesjms.oxfordjournals.org/ Received 15 September 2011; accepted 8 April 2012; advance access publication 8 June 2012.

The winter fishing season for eastern Bering Sea pollock (Theragra chalcogramma) is during the period of maximum seasonal sea-ice extent, but harvesters avoid fishing in ice-covered waters. Global climate models predict a 40% reduction in winter ice cover by 2050, with potential implications for the costs incurred by vessels travelling to and around their fishing grounds and the value of their catch. Additionally, it may open entirely new areas to fishing. Using retrospective data from 1999 to 2009, a period of extensive annual climate variation, the variation in important characteristics of the fishery is analysed. When ice is present, it restricts a portion of the fishing grounds, but in general, ice-restricted areas have lower expected profits at the time of restriction than the areas left at NW Fisheries Science Center on August 29, 2012 open. Some areas show a change in effort in warm years relative to cold, but the global redistribution of effort attributable to ice cover is small. This is largely because the winter fishery is driven by the pursuit of roe-bearing fish whose spawning location is stable in the southern part of the fishing grounds. Keywords: Bering sea, climate change, fisheries, pollock, sea ice.

Introduction areas with little or no ice cover. Vessels cannot physically Annual climate variation in the eastern Bering Sea off the coast of enter the areas where ice cover is thick, there may be increased Alaska is characterized by variation in the extent of winter sea-ice risk associated with fishing even where there are marginal cover. Sea ice begins to form in December and January and can amounts of ice cover, and pollock generally avoid the very cold persist through May. Ice coverage is greatest in February and water that may be present beneath the ice (Swartzman et al., March (Mysak and Manak, 1989; Chapman and Walsh, 1993). 1994; Wyllie-Echeverria and Wooster, 1998). Very little of the The extent to which the sea ice extends into lower latitudes (as winter season pollock fishing takes place in areas with .10% ice low as 588N), the timing of its advance and retreat, and the cover (Table 1). extent to which it persists into spring are determined by atmos- Because vessels that participate in the pollock fishery generally pheric temperatures and wind generated by the Aleutian Low avoid areas with ice cover, the distribution of winter fishing could Pressure System (Overland and Pease, 1982). Global climate shift as the ice cover in the Bering Sea declines with anticipated models predict warming of the Bering Sea and 40% reduction in climate change. This may have implications for the costs that winter ice cover by 2050 (Overland and Wang, 2007). vessels incur by travelling to and around their fishing grounds. The Bering Sea shelf is home to the walleye pollock (Theragra Additional gear interactions among vessels or between sectors chalcogramma) fishery, the largest commercial fishery in the could arise as vessels move out of the areas they fished historically. United States, generating .$1 billion in revenues annually Changes in the distribution of fishing could result in local deple- (Hiatt et al., 2009). Some 40% of the total allowable catch tion and potentially affect other species, such as marine (TAC) of the fishery is harvested in the winter (“A”) fishing mammals that forage on pollock. Finally, characterizations of the season, which opens on 20 January each year and generally lasts extent to which vessels avoid ice and predictions of the future dis- through April, the time of year when ice cover in the Bering Sea tribution of fishing are important in the development of spatially is at its greatest. However, most winter fishing takes place in explicit bioeconomic models (Aydin et al., 2010).

# United States Government, Department of Commerce 2012. Published by Oxford University Press.

Page 35 of 127 Decreasing seasonal sea-ice cover and the winter Bering Sea pollock fishery 1149

Table 1. Percentage of total observed hauls in statistical areas upper portion of the water column is warmed as air temperatures (after Kong et al., 2004), with ice cover at the time of the haul and solar insolation increase, but the water remains cold at depth, (1999–2009 for CPs, 2000–2009 for CVs). where it is insulated from warming. The temperature of the Percentage ice Percentage of hauls Percentage of bottom layer depends on the temperature of the water column cover in area (%) (CPs and motherships) hauls (CVs) at the time when this stratification is taking place, generally .0 22.2 32.8 around April or May (Stabeno et al., 2007, in press). “Cold” .2 13.8 18.8 years are characterized by extensive winter ice cover that persists .5 5.0 7.4 through March and April (Stabeno et al., in press). The ice restricts .10 3.6 5.2 the warming of the water column, resulting in a cold bottom layer. .20 1.8 2.0 This cold pool of water, which is commonly defined as water .30 0.7 0.5 ,28C(Wyllie-Echeverria and Wooster, 1998), persists into the .40 0.2 0.1 warm season, dividing the Bering Sea ecosystem into Arctic and .50 0.0 0.0 Subarctic waters and driving many ecosystem processes for the Source: NMFS observer data. The percentages in the table are inclusive, so entire year (Grebmeier et al., 2006). “Warm” years are character- “.X%” indicates the percentage of hauls in areas with .X% ice cover. ized by little ice cover or ice that retreats quickly, allowing more Downloaded from spring warming of the bottom layer of water (Stabeno et al., in press). Figure 1 shows the extensive variability in ice extent and In this paper, retrospective data are used to investigate changes timing even among years classified as “cold” (1999 and 2007– in the distribution of fishing that are correlated with variation in 2009) and “warm” (2001–2005; Stabeno et al., in press). The ice cover in the winter pollock fishery. Annual variation in seasonal IPCC models forecast a 40% decrease in average ice cover in the sea-ice cover is great (Chapman and Walsh, 1993), and global http://icesjms.oxfordjournals.org/ Bering Sea by 2050 (Overland and Wang, 2007). The forecasted warming scenarios predict more frequent occurrences of low ice warming will result in sea ice becoming less common in spring cover [Intergovernmental Panel on Climate Change (IPCC), over the southern Bering Sea shelf and warmer ocean temperatures 2007]. The distribution of fishing in past warm, low ice-cover (Schumacher et al., 2003). Years with less ice and warm tempera- years can provide insight into the future distribution of fishing tures such as those of 2001–2005 are predicted to become more when years with such conditions become more common in the Bering Sea. Here, we focus on the direct spatial changes caused frequent. by ice in the winter fishing season, which increase the cost (in The winter pollock fishery is driven largely by the pursuit of terms of travel and danger) of fishing in some areas. Although roe-bearing fish. Roe is removed from the fish and the rest of ice cover and the resulting cold water temperatures may cause the carcass processed into fillets or surimi (a fish-paste product) changes in the phenology or other biological mechanisms of and other products. Average roe prices range from 3 to 6 times at NW Fisheries Science Center on August 29, 2012 pollock, which may then affect the harvesters’ decisions on the average price of fillets, so, although harvesters’ production where to fish, these indirect effects are beyond the scope of this decisions are a function of a suite of product prices and paper, but they will be considered in future research. contractual obligations for specific products, roe amounts to a Changes in the distribution of fishing attributable to ice cover 50–200% bonus on fishing. Pollock spawning has a spatial and are investigated. If ice covers areas of the fishing grounds that are temporal progression throughout the winter season. Spawning relatively valuable, one would expect redistribution of fishing commences in February around Bogoslof Island (currently towards those areas in warm years. The effects of the climate closed to fishing), then progresses north of Unimak Island regime on catch per unit effort (cpue) are also analysed, because before continuing eastwards as the water warms (approximately there is some evidence that pollock avoid very cold water March through May). Finally, spawning on the grounds around (Swartzman et al., 1994; Wyllie-Echeverria and Wooster, 1998; the Pribilof Islands occurs in approximately April through June Kotwicki et al., 2005) and that cold periods are good for recruit- (Francis and Bailey, 1983; Ciannelli et al., 2007; Bacheler et al., ment (Hunt et al., 2011; Mueter et al., 2011). Increased concentra- 2010; Smart et al., in press). The location of these islands is tion of biomass as a consequence of cold-water avoidance could depicted in Figure 2. Roe is at its highest commercial value typic- affect fishery cpue, as could greater abundance. In addition, the ally 3–4 weeks before spawning. Vessels follow the maturation cycle such that roe is produced over the entire winter season. data are examined for evidence of a direct effect of ice on fishery There is only extremely limited spawning in the areas north of cpue. A period of extensive ice cover in 2008 forced harvesters the winter fishing grounds highlighted in Figure 2. The summer to fish in both ice-covered and ice-free waters, allowing a compari- pollock fishery takes place in the north region, as well as the son between the two. Finally, we analyse the data for evidence of winter and summer fishery for Pacific cod (Gadus macrocephalus; differences in roe recovery by the climate regime. Roe recovery is a fishery where the roe is low value and so not extensively tar- a significant determinant of the value of the fishery, so systematic geted), but pollock harvesters do not travel to the north in the changes in roe recovery (whether the effect is biological or caused winter. In summer, catcher-processors (CPs) are largely restricted by harvester behaviour) could provide information on the likely from fishing in the southernmost portion of Bering Sea by the future value of the fishery as warm years become more catcher-vessel operation area that preserves this area for CV and common. Given the results, the potential for changes in the distri- Community Development Quota fishing. Therefore, although bution of fishing as warmer, low-ice years become more frequent the north is often affected by ice cover descending from the in the Bering Sea are discussed. Arctic, it is effectively not in the choice set of pollock harvesters fishing in winter. Background Harvesters choose where to fish by making trade-offs between Seasonal sea-ice cover and water temperatures at depth vary dra- the value per tonne of harvest, cpue, the distance they must matically in the Bering Sea from year to year. During spring, the travel, and the safety and availability of different areas within the

Page 36 of 127 1150 L. Pfeiffer and A. C. Haynie Downloaded from http://icesjms.oxfordjournals.org/

Figure 1. Total ice cover in the fishing region (%) by day of the winter season for the period 1999–2009. A horizontal line at 20% total ice cover is added for reference. at NW Fisheries Science Center on August 29, 2012

Figure 2. Summer (light + dark grey area) and winter pollock fishing areas (dark grey area) of the Bering Sea.

fishing grounds (Bockstael and Opaluch, 1983; Eales and Wilen, recovery or value. Bycatch avoidance and regulations meant to 1986; Smith and Wilen, 2003). Because the Bering Sea pollock reduce bycatch may affect where harvesters choose to fish. fishery has been rationalized, individual vessels have shares of Several regulations regarding the bycatch of Chinook salmon the overall TAC, which allows harvesters to decide how most effi- (Oncorhynchus tshawytscha) affect the winter season pollock ciently to fish their share without concerns that slowing down will fishery. Full details regarding those programmes are available at lead to other harvesters catching a greater share of the common- the National Marine Fisheries Service Alaska Regional Office pool TAC. Roe is an important consideration in the value per website http://www.fakr.noaa.gov/sustainablefisheries/bycatch/ tonne, but the cpue is often higher in the areas of lower roe default.htm.

Page 37 of 127 Decreasing seasonal sea-ice cover and the winter Bering Sea pollock fishery 1151

Fuel costs and travel time make areas closer to port preferable, rationalized the fishery in 1998, vessels faced different incentives all else being equal. The presence of sea ice may increase the danger in their decisions on fishing location. Vessels competed to catch associated with travelling to some areas, or it may cut off access al- fish before the fleet-wide TAC was reached, so steaming to together. Other factors, such as the TAC and total abundance, distant locations was more costly because the time spent transiting affect harvester behaviour too. High total abundance may increase led to a reduction in the time spent fishing. This makes the data on the cpue overall, or it may expand the area in which the population the spatial distribution of the fleet before and after rationalization is located (Holt et al., 1997). TAC may also impact fishing; a high not directly comparable; post-AFA data are more representative of TAC may provide an incentive for vessels to fish more intensively the fishery as it is likely to be in future. The analysis is also early in the season, whereas a low TAC affords more time to restricted to the winter fishing season, January through May, identify the highest value fishing locations carefully. because in summer, sea ice retreats completely from the fishing grounds. The cold pool created by winter ice persists into Data and methods summer and is likely to affect summer fishing, but here we con- Haul-level data on catch location, quantity, and composition were sider only the direct effect of ice in the winter fishing season. taken from the Alaska Fisheries Science Center’s North Pacific All CPs and motherships in the Bering Sea must carry two

Observer Program Database for the three types of vessels observers for 100% of their days at sea. The observers collect infor- Downloaded from (sectors) that participate in the Bering Sea pollock fishery: CPs, mation on the time, duration, exact location, species composition, motherships, and inshore CVs. The American Fisheries Act and total tonnage of all hauls, among other information. CVs (AFA), passed in 1998, rationalized the fishery, ending the race ,125 ft are required to have observers on just 30% of their days for fish that arose through the 1990s. The AFA allocated the at sea, although from 2011 on, 100% of trips by CVs are being annual TAC among several sectors of the pollock fishery. In all, observed. Given the mix of small and large vessels, 50% of the http://icesjms.oxfordjournals.org/ 16 CPs harvest 50% of the annual catch, including the quota sector’s hauls are observed. Catch and general location informa- allocated to and leased from Community Development Quota tion about unobserved CV fishing is collected at landing, but it groups, as recorded by the National Marine Fisheries Service is not possible to disaggregate the landings data to the level of in- (NMFS) Alaska Regional Office. CPs are large vessels (60– dividual hauls. We assume that the data on unobserved trips can 114 m) with onboard factories that immediately process the be represented by the average from the observed trips. The catch into fillets, surimi, roe, and marketable by-products. CVs fishing region of the Bering Sea is divided into Alaska with exclusive delivery contracts with one of three motherships Department of Fish and Game (ADF&G) statistical areas of 1/ (large, mobile processing vessels that process fish at sea) harvest 2 × 18 (Kong et al., 2004, shown in Figure 3) and matched to 10% of the annual catch. The final 40% of the total catch is har- fishing haul location data. Data at the haul level were aggregated

vested by CVs that deliver their catch to shoreside processing to the daily observations of the number of vessels and hauls in at NW Fisheries Science Center on August 29, 2012 plants. We restrict our data to the years after rationalization: each statistical area to match daily climate measures. 1999–2009 for CPs and motherships, and 2000–2009 for CVs. Prices were obtained from the ADF&G’s Commercial In the race to fish operations that existed before the AFA Operator’s Annual Report (COAR). Prices are annual, vessel-

Figure 3. Ice quintiles and fishing zones of the eastern Bering Sea pollock fishery. Darker blue areas are more often affected by ice. The fishing zones, outlined with dashed lines, represent the spatial and temporal evolution of fishing as the season progresses.

Page 38 of 127 1152 L. Pfeiffer and A. C. Haynie specific averages of prices received for each product type sold, from The data from the fishery were analysed for the effect of ice on 2002 to 2009. The average annual prices are a poor representation the availability of fishing grounds, cpue, and value of the harvest. of the actual value of fish harvested from different areas at different Ice may affect the number of areas available for fishing (here, times of the year, principally because of the progression of the pro- defined as ,20% ice cover on a given day). Cut-off values of 10 duction of different values of roe over time and space. Weekly pro- and 30% ice cover on a given day were analysed too; although duction data from the National Marine Fisheries Service Alaska the number of areas affected changed, the conclusions from each Region were matched with observer programme catch location analysis remained the same. In reality, there is no strict cut-off data and used to estimate a spatial, seasonal price per tonne of value; vessels may fish in icy areas if they expect good returns, or catch that incorporates the average production ratios of surimi, if there are few alternatives. We compared the number of areas fillet, and roe produced in each statistical area in each season. available with the total percentage ice cover over the fishing Vessel-level production data are available from CPs and mother- area, by day. In addition, the areas affected by ice may be more ships only, but we assume that the estimated prices also apply to or less valuable than areas not affected by ice. If ice cuts off valu- the fishing decisions of CVs. Roe prices are generally lower for able regions of the fishing grounds during times when harvesters CVs because the fish are delivered to shoreside processing plants would wish to fish there, then a greater welfare loss would be in-

at the end of a trip, resulting in lower quality roe than if it is curred than if ice cuts off areas of the fishing grounds that yield Downloaded from removed immediately upon catch. Here, it is assumed that the a relatively low cpue, a low value per tonne, and/or are distant spatial differences in value are similar for CVs and CPs. from the processing ports of Dutch Harbor and Akutan. Daily sea-ice concentration interpolated from satellite sensors Quintiles of expected cpue and the proportion of effort in each from 1999 to 2009 was obtained from the National Climate Data statistical area per day, for each year, were constructed using Center. The surface data were averaged to the statistical area area–day combinations (area–days) for which ice cover was level and matched to the statistical area of the fishing haul location ,20% (non-ice times), as well as an additional category for the http://icesjms.oxfordjournals.org/ data. Total percentage ice cover is the daily sum of average ice expected cpue or proportion of effort equal to zero. These categor- cover over the entire pollock fishing area shown in Figure 2. ies were averaged over the 11 years of the period being analysed to Spatial indices of the location of fishing in cold, average, and obtain the average quintile of expected cpue and proportion of warm years summarize the global distribution of fishing. The effort, by day, for each statistical area during non-ice times. centre of gravity (CG) and standard distance (SD, or 1 s.d. of Similarly, quintiles of expected prices were created for each year, the distribution) are often used to measure shifts in the spatial dis- but because the estimates of prices do not vary over a season, we tributions of fish stocks (Woillez et al., 2007); they are used here to simply averaged the quintile of prices over the 11-year period to describe the spatial distribution of fishing locations. The CG of the obtain an average price quintile per statistical area. Quintiles

distribution of fishing is the mean centre of fishing locations: were used because the average cpue and prices may vary by year, at NW Fisheries Science Center on August 29, 2012 because of high abundance or a strong market for pollock, for   N x N y example. Averaging an area–day over several years would bias x = i=1 i , y = i=1 i . (1) n n the average towards the high-abundance or high-price year; aver- aging the quintiles removes this bias. Finally, quintiles of the dis- The SD represents the dispersal of the distribution of fishing tance from Dutch Harbor to the centroid of each statistical area, around its CG and is estimated from which are constant over time, were created. For area–days that were ice-restricted owing to their having .20% ice cover, the    mean non-ice quintile of expected cpue, prices, proximity to N (x − x)2n N (y − y)2n SD = i=1 i + i=1 i . (2) port, and proportion of effort were tabulated as a summary of n n the quality of ice-restricted areas. Ice (or a combination of the climatic conditions correlated with The CG and SD for each sector were compared by year to indicate variation in ice cover) may affect cpue. Realized cpue from obser- whether ice resulted in any significant global change in effort ver data were standardized with a linear regression that controlled distribution. for vessel-specific differences. Differences in standardized cpue Ice may also affect fishing on local scales. The fishing area, ex- were tested for between groups of years corresponding to warm cluding the northern region that pollock harvesters do not visit in and cold climate regimes by including interactions between the winter even in the warmest years, was split into regions based on month of the haul, the fishing zone in which it occurred, and two different definitions of “region”. First, the area was split into climate-regime indicators. In addition, differences in cpue “ice quintiles” of the frequency and percentage of ice cover to between hauls in icy areas and hauls in ice-free areas were tested show how ice affects some regions of the pollock fishing for; if pollock avoid the cold water that may be present beneath grounds more than others (Figure 3). Areas in the north and the ice, cpue may be lower in ice-covered areas. The data are east are most often affected by ice, whereas areas in the south limited to 2008 and 1 month and zone of 2007 (the only time in are affected least often. Second, the fishing region was split into the dataset when effort was similar in timing and locations in three fishing “zones” (Figure 3). These zones describe how the dis- both icy and ice-free areas). Included in the linear regression are tribution of fishing evolves spatially over the winter season and interactions between the month of the haul, the fishing zone in reflect the spatial and temporal maturation of roe and the trade- which it took place, and an indicator of whether the area contained offs harvesters make between expected value, expected catch, .20% ice at the time of the haul. and distance. Fishing in January and early February is almost ex- Similarly, the ice or climate regime may affect the value of har- clusively in zone 1, but in February, the fishing extends to zone vested fish. The largest drivers of value are roe recovery and roe 2 around the Pribilof Islands, then in March and April it takes quality. Similar methods were used to evaluate the effect of the place in all three zones. climate regime on the rates of roe recovery, and differences in

Page 39 of 127 Decreasing seasonal sea-ice cover and the winter Bering Sea pollock fishery 1153 roe recovery between hauls in icy areas and those in ice-free areas Table 2. Mean CG and SD (i.e. 1 s.d.) of the distribution of fishing at similar times and in similar locations. Haul-level observer data effort, by year and sector. needed to be matched to weekly production data to obtain the CPs and motherships CVs rates of roe recovery. This introduced considerable coarseness into the data. The locations of hauls had to be aggregated to a Mean CG Mean CG weekly level, resulting in variables describing the weekly propor- tion of fishing in each fishing zone. In addition, there was consid- Year xySD (km) xySD (km) erable mismatch between the weekly production data and the 1999 56.11 2166.50 163.2 observer data, particularly with respect to roe. Some vessels seem 2000 56.02 2166.42 152.4 55.83 2165.65 109.2 to wait until the roe is sold to record it in formal production 2001 56.37 2167.69 199.8 56.07 2165.67 179.1 2 2 data, resulting in impossibly high rates of roe recovery at the 2002 55.92 165.62 157.1 55.62 165.11 115.8 2003 56.45 2167.59 249.6 55.78 2165.46 178.9 end of the season. We restricted the data to exclude recovery 2004 56.17 2166.06 165.7 55.73 2165.48 134.4 rates from April on, when most of the problematic observations 2005 56.10 2167.12 200.0 55.84 2165.32 184.8 were made. Finally, weekly production data that can be matched 2006 55.83 2166.62 161.2 55.63 2165.83 140.5 to vessel fishing activity are only available for CPs and mother- 2007 55.67 2166.84 181.3 55.62 2166.44 160.3 Downloaded from ships, so the analysis excludes CVs. Roe recovery is modelled 2008 56.03 2166.16 157.9 55.79 2166.12 156.3 using a linear regression of weekly roe recovery by vessel on 2009 56.29 2167.50 188.8 55.92 2166.12 192.5 vessel effects, interactions between month of the season and the The mean CG is calculated using all observed haul locations and is climate regime, the proportion of fishing in each zone during a measured in decimal degrees. week, and the ratio of roe to surimi annual average prices. Because of the lack of within-season price data, it was not possible http://icesjms.oxfordjournals.org/ to analyse differences in roe quality. However, the direct effect of Because of the progression of roe maturity and the progression ice is tested for using a linear regression of roe recovery on inter- of effort across the fishing area, some areas do not receive any actions between month and the proportion of hauls in areas with fishing effort during a part of the season. Figure 5b shows the .20% ice, the proportion of fishing from each zone, and vessel effect of ice on the number of areas with ,20% ice cover, by effects. A comparison of roe recovery from icy and non-ice areas day, that normally receive effort at that time of year. “Normally is only appropriate for 2008, though, when operators fished in receive effort” is the description applied to a statistical area both conditions simultaneously. having at least two fishing hauls during the concurrent week at any time during the period 1999–2009. Figure 5 shows that al- Results though there may be as many as 68 statistical areas available for at NW Fisheries Science Center on August 29, 2012 There is relatively little fishing where there is ice cover (Table 1). fishing, at any one time many fewer are actually fished. Figure 5 Whereas .20% of CP and mothership hauls and .30% of CV also shows that the decrease in the number of areas available for hauls are made in areas with some ice cover, the amount of fishing as a result of increases in ice cover is not nearly as dramatic fishing decreases quickly as ice cover increases. Only 2% of all when unused area–days are excluded. A linear regression of the hauls are made in areas with .20% ice cover. Despite this, and data in Figure 5a yields b ¼ 20.75 and r2 ¼ 0.81, whereas using the fact that the ice extent varies greatly between cold and warm the data in Figure 5b yields b ¼ 20.32 and r2 ¼ 0.25. When years, the variation in the mean centre of distribution of fishing only the areas affected by ice that harvesters actually visit are con- is well within the SD of the distribution in each year, suggesting sidered, a 1% increase in total ice cover is associated with a de- that there has been no significant change in the mean centre of dis- crease in the number of areas available for fishing of 0.32, less tribution of fishing (Table 2). than half the rate estimated using all areas. In other words, al- However, there is local variation in the distribution of effort. though ice restricts access to many of the statistical areas in the Figure 4 shows the percentage of total effort in each ice quintile, fishing region, many of those areas are places that harvesters by year. It also shows the average ice cover in each ice quintile. rarely visit in other years when ice does not restrict access. It During the warm years of 2001–2005, effort extended into the also shows that when the total ice extent is .40%, as few as 12 5th ice quintile (the statistical areas most affected by ice), fishing areas may have ,20% ice cover. whereas in other years, it did not. The effect of ice can be seen The areas affected by ice may be more or less valuable than the in the 4th ice quintile too. Effort was highest in 2001 and 2003– areas unaffected by ice. We analysed the characteristics of the 2005 when there was very little ice in the 4th ice quintile. In area–day combinations that were ice-restricted (by .20% ice 2002, although the year is categorized as a warm one, early ice des- cover on a particular day) by inspecting their characteristics cended into ice quintiles 4 and 5 (Figure 1), reducing effort in during area–days in which ice cover was ,20%. The characteris- these areas. Effort was greatest in the three lowest ice quintiles in tics (expected cpue, expected price, distance, and proportion of the years with the greatest ice extent, 1999 and 2006–2009. The effort) of each area–day with ,20% ice were divided into quin- data in Figure 4 are pooled over sectors; if split between the CV tiles by year. An area–day with .20% ice has no value, because and CP/mothership sectors, the pattern is similar, although CVs there is virtually no fishing. For each area–day, the quintiles of rarely fish in the 5th ice quintile, even in warm years. Rather, each characteristic are averaged over the years for which the CV effort in the 4th ice quintile increases in warm years. area–day had ,20% ice, to obtain a “non-ice average quintile”. Figure 5a shows the relationship between the number of statis- Table 3 tabulates the frequency that an ice-restricted area–day tical areas with ,20% ice cover by day and the total percentage of has a non-ice average quintile of expected cpue of zero (there ice on the fishing grounds. As total ice increases, fewer areas are was no fishing on that particular day during any of the non-ice available for fishing. However, many of the areas affected by ice years), from 0 to 1 (the average quintile of expected cpue is are areas where there is no fishing at that time of year anyway. between 0 and 1 on that particular day during non-ice years), or

Page 40 of 127 1154 L. Pfeiffer and A. C. Haynie Downloaded from http://icesjms.oxfordjournals.org/

Figure 4. Proportion of annual effort and average ice cover by ice quintile. The average ice cover over all days in the winter pollock fishing season are used to define the ice quintiles; lower ice quintiles are less often affected by ice. Effort in the areas most often affected by ice increases in warm (low-ice) years. at NW Fisheries Science Center on August 29, 2012

Figure 5. (a) The number of winter pollock fishing statistical areas (Kong et al., 2004) with ,20% ice cover decreases as the total ice cover on the fishing grounds increases. (b) The number of statistical areas with ,20% ice cover actually used by the fishery on a given day decreases to a much lesser extent because the total ice cover on the fishing grounds increases. Linear fits are shown on both panels. from 1 to 2, etc. The average quintiles categorize area–days Table 3 shows that 43% of the area–days that had at least 20% based on their quality (expected cpue, price, etc.) during ice cover were area–days when the expected cpue was zero non-ice times. (because the area was never visited by a harvester) in non-ice

Page 41 of 127 Decreasing seasonal sea-ice cover and the winter Bering Sea pollock fishery 1155

Table 3. Average quintiles of expected cpue, prices, proximity to capture the temporal differences in value only to the extent that port, and proportion of effort (during non-ice periods) of area– more vessels fish in an area when the value is high. days ice-restricted by .20% ice cover. Table 3 also shows that most of the areas with .20% ice cover Statistical areas ice-restricted in are in the lowest quintiles of the proximity to the ports of Dutch Harbor and Akutan. Except a few areas along the Aleutian All Cold Warm Islands, most areas consistently affected by ice are located farther Average quintile years (%) years (%) years (%) from the ports used by the fleet. Expected cpue during non-ice times Finally, Table 3 shows how ice affects areas with respect to their 0433681average quintiles of the proportion of effort in warmer times. The 0–1 18 20 12 proportion of effort reflects harvesters’ choices and the trade-offs 1–2 11 12 4 they have made between expected cpue, expected prices, ice, and 2–3 4 5 1 travel distance. In cold years, most of the ice-restricted areas had 3–4 1 1 0 no effort during warmer years (48%) or were in the lowest cat- 4–5 0 0 0 Expected price during non-ice times egory of average effort quintiles (20%). In warm years, 87% of 0–1 11 9 23 the areas affected by ice had no effort in other years when there Downloaded from 1–2 6 5 10 was no ice at the time, and another 9% were in the lowest category. 2–3 25 22 38 In evaluating whether the climate regime affects fishery cpue, 3–4 21 22 12 we found no evidence of systematically different cpue by regime 4–5 17 16 21 (Figure 6). The mean standardized cpue was not significantly dif- Proximity to port during non-ice times ferent between warm years and the transition years of 2000 and 0–1 30 25 60 2006, so they were combined. The sample was split by zone of http://icesjms.oxfordjournals.org/ 1–2 31 31 30 fishing location and by month, because in general, fishing pro- 2–3 13 14 10 gresses from zone 1 to the northwest over the season. In addition, 3–4 5 6 4 4–5 0 0 0 as the season progresses, roe is more likely to become overmature, Proportion of effort during non-ice times and hence of lower value. The decrease in value makes targeting 0544887higher cpue more likely. Vessels are also more likely to target 0–1 18 19 9 cpue in zones 2 and 3, where roe tends to be of lower value. 1–2 6 6 3 There is no fishing in zone 3 in January. Figure 6 shows the 2–3 1 1 0 mean standardized cpue by temperature regime, by zone, and 3–4 0 0 0

over the months January through April, with 95% confidence at NW Fisheries Science Center on August 29, 2012 4–5 0 0 0 intervals. As Figure 6 indicates, cpue increases slightly over the Number of area–days 3 299 2 652 647 season in zone 2, but there is no significant difference in cpue with .20% ice cover by the climate regime. A regression of cpue on zone and month interactions, controlling for vessel effects and total biomass, explains just 6% of the variation in fishery cpue (adjusted years (years when there was ,20% ice). Some 18% of the r2 ¼ 0.058). ice-restricted area–days were in the lowest category of expected Cpue of hauls in areas where ice was present were compared cpue and 11% were in the second lowest category. This means with those in ice-free areas. Because little fishing occurs where that 72% of the ice-restricted areas were in the lowest categories there is ice, information about why vessels avoid it is sparse. of expected cpue. Further insight can be gained by separating However, ice cover in 2008 was unusually extensive and early to the areas ice-restricted in cold years from those restricted in descend onto the fishing grounds (shown in Figure 1). The year warm years. In cold years, more area–days are ice-restricted, but 2008 is the only year in the current sample where a large portion they also tend to be higher quality area–days in terms of expected of the fishing (19% of total effort) was in areas with .20% ice cpue. Although the total proportion is still low (6%), a greater per- cover in multiple regions (Figure 7). In 2007, there was a period centage of ice-restricted area–days are in the highest three categor- in March when operators fished in icy areas in zone 2, but it ies of expected cpue in cold years. Only 1% of ice-restricted areas accounted for just 3% of total effort. If the cpue in icy areas is were in the highest three categories of expected cpue. less than the cpue in non-icy areas, then vessels presumably The mean quintiles of prices in the ice-restricted areas are also avoid ice because of the reduction in their expected cpue. In shown in Table 3. In both warm and cold years, a high percentage 2008, vessels may have had few options other than to fish in icy of the area–days with .20% ice cover were in the highest categor- areas because so much of the shelf was restricted by ice. ies of expected value per tonne. In other words, fish caught in However, if the cpue in icy areas is greater than or similar to the those areas were worth more per tonne than fish caught in other cpue in non-icy areas, then vessels likely avoid ice for other areas, because of their quality attributes or the amount of roe reasons, such as increased danger from fishing in such areas, or they contained. However, because the measure of prices used the possibility of having to steam great distances if the ice moves does not vary over a season, they may be an unrealistic measure in unexpected ways. of the true value of fish caught in an area on a particular day, so Table 4 shows that fishing in area–days of .20% ice was in they should be interpreted with caution. For example, because February in zones 1 and 2 and in March in zones 1–3. The the maturity of roe progresses spatially and temporally, a fish number of hauls, the mean standardized cpue, and the results of caught in an area at peak roe time would be more valuable than an F-test for equality of their means are shown by month and a fish caught there at another time of the season. The prices zone combination. Table 4 shows that in 2008, the cpue in icy used here are an area average over the whole season, and they areas was significantly higher than the cpue in non-icy areas in

Page 42 of 127 1156 L. Pfeiffer and A. C. Haynie

Figure 6. Mean standardized cpue by climate regime, over months and zones. Downloaded from February in zone 1, and the same in March in zones 1 and 2. The results from 2007 are similar, although fishing in ice only occurred in March. There was no significant difference in the cpue in zones 1 and 3 (but there were few ice hauls made there), but in zone 2, the cpue in icy areas was significantly higher. Both CPs and CVs

fished in the icy conditions. Again, however, the unexplained vari- http://icesjms.oxfordjournals.org/ ability is high. A regression of cpue on an indicator variable for ice .20%, and zone, month, and vessel effects explains just 10% of the variation in the cpue. Finally, there were no significant differences in the rates of roe recovery by the climate regime (regression results not shown, but are available from the authors). The recovery rate increases in February and March as the fishery moves north and vessels target more mature, lower-value roe in both cold and warm regimes. However, given the many complications with the data, Figure 7. Proportion of annual effort in icy and non-icy areas and at NW Fisheries Science Center on August 29, 2012 the unexplained variation is very large (adjusted r2 ¼ 0.034). the percentage total ice cover, by week of 2008. Similarly, there were no significant differences in the rates of roe recovery between vessels that spent a larger proportion of their time in areas with ice cover .20% in 2008 (again, the results by ice (Figure 4). However, spatial and temporal distribution of are available from the authors). ice is important to the fishery. Figure 5 and Table 3 show that many of the ice-covered areas are affected at times of the season Discussion when there is little effort directed to those areas, even when ice Seasonal sea-ice cover, the timing of its advance onto the Bering is not present. Such areas tend to have lower expected cpue and Sea shelf, and its retreat, varies significantly from year to year. be more distant from port. Although not evident from the This combination of factors is important in determining the results in Table 3 because of the temporal coarseness of the price climate regime of the Bering Sea ecosystem for the rest of the data, they may also yield a lower-value product (have less potential year. It is also important because it occurs simultaneously with for high-value roe). Price data that capture in-season variation the winter fishing season of the Bering Sea pollock fishery, the would enable a more rigorous estimation of the effect of ice on largest foodfish fishery in the world. Climate change is expected the value of the fishery. to decrease seasonal sea-ice cover in the Bering Sea by 40%, and Some evidence exists that climate regimes may affect cpue this may affect the winter pollock fishery by opening up new through biological mechanisms, such as greater abundance or a fishing grounds or reducing travel costs associated with the avoid- reduced area of fishing grounds in pollock’s preferred temperature ance of ice. band, causing a greater concentration of fish (Wyllie-Echeverria An important observation is that ice in the pollock fishing and Wooster, 1998; Kotwicki et al., 2005). However, the climate grounds of the Bering Sea is highly variable from year to year. regime does not affect mean fishery cpue in the winter pollock Even among years classified as unusually cold (Stabeno et al., in fishery. This could be due to harvester behaviour; harvesters press), the extent of ice in March varied from 30% in 1999 to make trade-offs between higher value, low cpue areas, and lower .70% in 2008. In 2008, the onset of the ice was much earlier value, high cpue areas. The targeting of value rather than cpue and lasted longer than in other years. A simple classification of could cause the responsiveness of observed fishery cpue to years into cold and warm, however, may obscure the subtleties changes in stock or concentration to be dampened, especially in associated with the variation in the spatial and temporal extent winter. In addition, in the years under study, abundance was of ice, especially from an economic perspective, because harvesters lower in cold years than in warm years [an average of 6.6 make daily decisions about where to fish. million tonnes vs. 9.8 million tonnes (Ianelli et al., 2011)]. Low Ice affects fishing by restricting the size of the fishing grounds abundance combined with low TAC may lead harvesters to fish available to harvesters. In cold years, there is some redistribution more slowly, i.e. to expend more search time and fish only when of total annual effort away from those areas most often affected cpue and value are high for each haul. The fact that fishery cpue

Page 43 of 127 Decreasing seasonal sea-ice cover and the winter Bering Sea pollock fishery 1157

Table 4. Differences in mean standardized cpue between hauls made in statistical areas with ,20% ice cover and areas with .20% ice cover, by month and zone. Ice cover <20% Ice cover >20% % CV in ice F-test of Month Zone Number of hauls Mean cpue Number of hauls Mean cpue cover >20% equality February 2008 1 1 270 25.67 219 59.12 ** 47.5 2 513 38.64 7 21.51 0.0 30 0 March 2008 1 103 53.38 304 93.71 ** 23.0 2 225 46.47 164 115.81 ** 42.1 3 1 18.93 36 56.08 0.0 March 2007 1 298 51.24 14 73.47 100.0 2 678 49.92 112 115.91 ** 40.2 3 247 61.64 2 23.72 0.0 **p , 0.01. Downloaded from is observed only in locations where harvesters actually choose to danger or interference with fishing operations, for example. It fish, as well as the non-temporal nature of the estimated price also suggests that ice-avoidance behaviour by vessels is heteroge- data used here, restrict us from separating these behavioural neous, either because of differences in the relative risk aversion effects from any biological changes. Biological surveys of the of skippers or because of differences in the danger that ice poses http://icesjms.oxfordjournals.org/ area do not take place until June and July, so they provide little in- to individual vessels. formation about the concentration of biomass during the winter Finally, although there is local variation in the extent to which fishing season. areas are affected by ice and the reduction in the distribution of Although there is no biological mechanism that suggests that effort at some times of the year because of it, the global statistics there should be differences in the rate of roe recovery by the describing the distribution of effort in the winter pollock fishery climate regime, ice (or other factors associated with high-ice are reasonably constant over time. The change in the spatial distri- years) may change harvesters’ fishing choices in a way that bution of effort from year to year does not result in a significant affects the rates of roe recovery, and hence the value of the change in the CG of the distribution. fishery. For example, if ice restricts an area with a particularly In future, available fishing grounds may expand with the high rate of recovery in cold years, harvesters may instead fish in decline in ice as warming occurs. Pollock harvesters produce a at NW Fisheries Science Center on August 29, 2012 high-cpue, low-roe areas. However, there was no evidence of dif- variety of products, the suite of which varies with the characteris- ferences in the rate of roe recovery associated with the climate tics of the fish being caught. Targeting decisions are based on many regime. factors, including price and contractual obligations to deliver par- There is limited evidence that harvesters realize higher cpue in ticular quantities and grades of the possible products. However, ice-affected areas. The mean standardized cpue in ice-affected because pollock roe is so valuable relative to other products, the areas was nearly double that in ice-free areas in 2008. There was pursuit of prespawning fish that contain roe is an important a similar difference in 2007, but the number of hauls in icy areas driver of harvester behaviour. Unless an unforeseen then was much smaller. The evidence for higher cpue in ice- climate-related regime shift causes extreme changes in the spatial covered waters is suggestive, but not conclusive because there is distribution of spawning, there is little evidence that the winter only a short period when fishing in both icy and non-icy condi- fishery will shift into the newly ice-free waters of the future. tions occurred simultaneously. In addition, it is not possible to Some areas in the north of the current fishing grounds will be control for the endogeneity of location choice [the possibility ice-free more often, but only marginal increases in effort deployed that vessels fishing in high-cpue (icy) areas were targeting high to those regions is anticipated because they have lower expected cpue, while vessels fishing in other areas were targeting price, cpue and are more distant from port, on average. Little change causing cpue to be lower than its potential in non-icy areas]. in the cost of fishing attributable to the direct effects of changes The high cpue in the areas covered by ice in 2008 may have in ice cover is expected. The predicted changes have the potential been consequential and random, and equally probable had there to be locally significant if an area is particularly sensitive to ecosys- been no ice. The model shows that a high variability in cpue tem interactions. exists that cannot be explained by controlling for vessel, time, However, these results would likely be different for fisheries and area effects. that target other species with different characteristics. In a Despite the cpue difference between icy and non-icy areas, only fishery without a strong market for roe, such as that for Pacific 50% of total effort during the high-ice period in 2008 covered cod, harvesters are less likely to constrain themselves to the south- icy areas. A significant negative effect on the rate of roe recovery ern areas of the fishing grounds where ice rarely affects the fishing from fishing in ice-covered waters would be some evidence that area. Ice may have a much greater effect on high-value (high vessels that fished in the ice in 2008 were targeting cpue, expected cpue and/or price) areas. In addition, changes in whereas those that did not were targeting value. However, there market conditions can change the characteristics of fisheries. For was no significant effect of ice on the rate of roe recovery. example, in the pollock fishery, changes in the price of roe are Although this does not negate the hypothesis that the vessels likely to affect targeting behaviour. A decrease in the price of the fishing in ice were targeting high cpue, it means that the vessels highest-value roe relative to the price of average- or low-value that avoided the ice likely did so because of some other factor: roe would likely cause producers to target higher rates of recovery

Page 44 of 127 1158 L. Pfeiffer and A. C. Haynie of lower-value roe, which tend to be found around the Pribilof Eales, J., and Wilen, J. 1986. An examination of fishing location choice Islands (zone 2) later in the season, rather than the highest value in the pink shrimp fishery. Marine Resource Economics, 2: (but low recovery) roe. This could increase potential ice interac- 331–351. tions. If average roe prices were to fall to levels where roe targeting Francis, R. C., and Bailey, K. 1983. Factors Affecting Recruitment of became less profitable, harvesters may expand their fishing to the Selected Gadoids in the Northeast Pacific and East Bering Sea. In northern regions of the fishery, where ice interactions are much From Year to Year, pp. 35–60. Ed. by W. S. Wooster. Washington Sea Grant, Seattle, WA. more likely, even in warm years. Management is also important, Grebmeier, J. M., Overland, J. E., Moore, S. E., Farley, E. V., Carmack, and in an unrationalized fishery, harvesters may be willing to E. C., Cooper, L. W., Frey, K. E., et al. 2006. A major ecosystem shift fish intensively in icy areas regardless of the danger, because if in the northern Bering Sea. Science, 311: 1461–1464. they do not, other harvesters might capture a larger share of the Hiatt, T., Dalton, M., Felthoven, R., Fissel, B., Garber-Yonts, B., TAC. Haynie, A., Kasperski, S., et al. 2009. Stock assessment and This paper has analysed the distributional changes in the Bering fishery evaluation report for the groundfish fisheries of the Gulf Sea pollock fishery resulting from inter- and intra-annual vari- of Alaska and Bering Sea/Aleutian Islands area: economic status ation in sea ice. However, several questions remain unanswered. of the groundfish fisheries off Alaska, pp. 1–254. Alaska Fisheries

First, even in a fishery with an exceptional amount of biological Science Center, National Marine Fisheries Service, Seattle. Downloaded from and economic data, it remains challenging to predict how ice Holt, R., Lawton, J., Gaston, K., and Blackburn, T. 1997. On the rela- cover will impact the fishery under future warmer climate condi- tionship between range size and local abundance: back to basics. tions. This is in part because the cold years in the sample have been Oikos, 78: 183–190. years of lower-than-average fishable pollock abundance (Ianelli Hunt, G. L., Coyle, K. O., Eisner, L., Farley, E. V., Heintz, R. A., et al., 2011). More insight into the effects of ice and cold Mueter, F., Napp, J. M., et al. 2011. Climate impacts on eastern regimes on fishing behaviour could be gained with more variation Bering Sea foodwebs: a synthesis of new data and an assessment http://icesjms.oxfordjournals.org/ of the Oscillating Control Hypothesis. ICES Journal of Marine in total abundance in both warm and cold years. Second, the true Science, 68: 1230–1243. progression of the value of the pollock harvested in winter is Ianelli, J. N., Barbeaux, S., Honkalehto, T., Kotwicki, S., Aydin, K., and poorly represented by the annual average prices used here. A far Williamson, N. 2011. Assessment of the walleye pollock stock in the better understanding of the trade-offs between cpue and value, eastern Bering Sea. In 2011 North Pacific Groundfish Stock as well as of the effects of climate change on the profitability of Assessment and Fishery Evaluation Reports, pp. 51–168. Alaska the fishery, could be achieved with better price data. Finally, Fisheries Science Center, National Marine Fisheries Service, Seattle. winter sea ice drives ecosystem processes in the Bering Sea for IPCC. 2007. Contribution of Working Groups I, II, and III to the the entire year. The effects of sea ice, the cold pool that it Fourth Assessment Report of the Intergovernmental Panel on Climate Change. In Climate Change 2007: Synthesis Report. Ed.

creates, and annual climate regimes on the pollock fishery and at NW Fisheries Science Center on August 29, 2012 other Bering Sea fisheries are the subjects of ongoing research. by R. K. Pachauri, and A. Reisinger. IPCC, Geneva, Switzerland. 104 pp. Kong, T. M., Gilroy, H. L., and Leickly, R. C. 2004. Definition of IPHC Acknowledgements Statistical Areas. Technical Report, 49. International Pacific Halibut Funding for this research was provided by the North Pacific Commission, Seattle, WA. Research Board (NPRB publication number 348) as part of the Kotwicki, S., Buckley, T. W., Honkalehto, T., and Walters, G. 2005. Bering Sea Integrated Ecosystem Research Program (BEST- Variation in the distribution of walleye pollock (Theragra chalco- BSIERP publication number 61). Angie Grieg assisted with the gramma) with temperature and implications for seasonal migra- processing of ice data and Terry Hiatt with the organization of eco- tion. Fishery Bulletin US, 103: 574–587. nomic data. Mueter, F. J., Bond, N. A., Ianelli, J. N., and Hollowed, A. B. 2011. Expected declines in recruitment of walleye pollock (Theragra chal- cogramma) in the eastern Bering Sea under future climate change. References ICES Journal of Marine Science, 68: 1284–1296. Aydin, K., Bond, N., Curchitser, E. N., Gibson, M. G. A., Hedstro¨m, Mysak, L. A., and Manak, D. K. 1989. Arctic sea ice extent and anom- K., Hermann, A. J., Moffitt, E., et al. 2010. Integrating data, field- alies, 1953–1984. Atmosphere–Ocean, 27: 376–405. work, and models into an ecosystem-level forecasting synthesis: the Overland, J. E., and Pease, C. H. 1982. Cyclone climatology of the Forage–Euphausiid Abundance in Space and Time (FEAST) Bering Sea and its relation to sea ice extent. Monthly Weather model of the Bering Sea Integrated Research Program. ICES Review, 110: 5–13. Document CM 2010/L: 21, Nantes, . Overland, J. E., and Wang, M. 2007. Future regional Arctic sea ice Bacheler, N. M., Ciannelli, L., Bailey, K. M., and Duffy-Anderson, J. T. declines. Geophysical Research Letters, 34, L17705. 2010. Spatial and temporal patterns of walleye pollock (Theragra chalcogramma) spawning in the eastern Bering Sea inferred from Schumacher, J., Bond, N., Brodeur, R., Livingston, P., Napp, J., and egg and larval distributions. Fisheries Oceanography, 19: 107–120. Stabeno, P. 2003. Climate change in the southeastern Bering Sea and some consequences for biota. In Large Marine Ecosystems of Bockstael, N. E., and Opaluch, J. J. 1983. Discrete modelling of supply response under uncertainty: the case of the fishery. Journal of the World: Trends in Exploitation, Protection, and Research, pp. Environmental Economics and Management, 10: 125–137. 17–40. Ed. by G. Hempel, and K. Sherman. Elsevier Science, . Chapman, W. L., and Walsh, J. E. 1993. Recent variations of sea ice and air temperature in high latitudes. Bulletin of the American Smart, T. I., Duffy-Anderson, J. T., and Horne, J. K. in press. Meteorological Society, 74: 33–48. Alternating temperature states influence walleye pollock Ciannelli, L., Bailey, K. M., Chan, K-S., and Stenseth, N. Ch. 2007. (Theragra chalcogramma) early life stages in the southeastern Phenological and geographical patterns of walleye pollock Bering Sea. Marine Ecology Progress Series. (Theragra chalcogramma) spawning in the western Gulf of Smith, M., and Wilen, J. 2003. Economic impacts of marine reserves: Alaska. Canadian Journal of Fisheries and Aquatic Sciences, 64: the importance of spatial behavior. Journal of Environmental 713–722. Economics and Management, 46: 183–206.

Page 45 of 127 315

Page 46 of 127 Please do not cite this chapter without permission from the authors.

316 Chapter 4: Climate fluctuations and fishing behavior in the Pacific cod (Gadus macrocephalus) 317 longline fishery 318

319 Alan C. Hayniea and Lisa Pfeifferb 320 321 aAlaska Fisheries Science Center, NOAA Fisheries, 7600 Sand Point Way NE, Seattle, WA 98115 322 [email protected] 323 b Corresponding author 324 Alaska Fisheries Science Center, NOAA Fisheries, 7600 Sand Point Way NE, Seattle, WA 98115, 325 [email protected], Phone: 206-526-4696; Present affiliation and address: Northwest Fisheries 326 Science Center, NOAA Fisheries, 2725 Montlake Boulevard East Seattle, WA 98112. 327 328 329 Abstract 330 This paper assesses the impacts of past climate variation on the Bering Sea and Aleutian Island Pacific 331 cod (Gadus macrocephalus) longline fishery. Climate-driven spatial and temporal differences in expected 332 revenue and catch per unit effort lead to differences in harvester behavior and catch patterns, with 333 implications for the cost of fishing. Retrospective data from the fishery were used to examine the 334 relationship between climate variation and catch per unit of fishing effort. Warmer conditions decrease 335 the size of the Bering Sea cold pool, resulting in less concentration of biomass and lower average CPUE. 336 Local depletion happens more rapidly when fish are not as concentrated by the cold pool, leading to more 337 movement about the fishing grounds and longer trips. Vessels traveled 14% more on warm-year trips 338 compared to cold-year trips, leading to fuel costs. Distance traveled per ton of catch was higher in warm 339 years as well, by a similar percentage. 340 341 Keywords: Climate changes, cod fisheries, fishing costs, fishery institutions, temperature effects 342 343

344 1. Introduction 345 Assessing the impacts of climate change on fisheries is an important element of fisheries 346 management, especially in Arctic and sub-Arctic regions where climate change is expected to have the 347 earliest and greatest impact (IPCC, 2007). Most research to date has focused on how the ranges and

Page 47 of 127 Please do not cite this chapter without permission from the authors.

348 dynamics of marine populations will shift, but has ignored the complexity inherent in fish harvesters’ 349 decision-making processes, which has been well studied by economists (Bockstael and Opaluch, 1983, 350 Eales and Wilen, 1986, Haynie and Layton, 2010, Smith and Wilen, 2003, Smith, 2002). This paper 351 examines the effects of past climate variation on one important fishery in the Bering Sea and Aleutian 352 Islands (BSAI), the Pacific cod (Gadus macrocephalus) longline fishery, which harvests approximately 353 50% of the annual Pacific cod catch and produces an average total revenue of $138 million per year 354 (2002-2010). The fishing choices of Pacific cod longliners are affected by a number of factors, 355 including distribution of Pacific cod over the eastern Bering Sea fishing grounds, the price that they 356 expect to receive for their catch, travel costs, the costs of fishing in a given location, vessel 357 characteristics, and the regulatory environment of the fishery. Climate-driven spatial and temporal 358 differences in these factors may lead directly to changes in catch patterns and the cost of fishing. By 359 examining the relationship among environmental conditions, the characteristics of the fishery, and the 360 factors that drive its exploitation, the effects of climate on the fishery can be identified. 361 The climatic conditions of the Bering Sea vary significantly from year to year; this is primarily 362 caused by variation in the extent of winter sea ice cover that forms in late fall (November-January), 363 advances into the Bering Sea in February and March, and can persist through May. The greatest ice 364 coverage occurs in February and March (Mysak and Manak, 1989, Chapman and Walsh, 1993). The 365 extent to which the sea ice extends into lower latitudes (as low as 58°N) and the degree to which it 366 persists into the spring is determined by atmospheric temperatures and wind generated by the Aleutian 367 Low Pressure System (Overland and Pease, 1982). Ice cover is a major component in the creation of a 368 cold pool of water (water with bottom temperature < 2 degrees Celsius) in the Eastern Bering Sea 369 (Stabeno et al., 2001). During the spring, the upper portion of the water column is warmed as air 370 temperatures and solar insolation increase, but cold water remains at depth. This “cold pool” of water 371 often persists into the summer, dividing the Bering Sea ecosystem into arctic and subarctic waters and 372 driving many ecosystem processes for the entire year (Grebmeier et al., 2006). The spatial extent of the 373 cold pool is an important factor affecting the spatial distribution of many species of fish and crabs on the 374 eastern Bering Sea shelf (Wyllie-Echeverria and Wooster, 1998, Stabeno et al., 2012, Kotwicki and 375 Lauth, 2013, Spencer, 2008).

376 Kotwicki and Lauth (2013) found that the distribution of Pacific cod in the annual bottom-trawl 377 survey in the eastern Bering Sea is affected by the extent of the cold pool (as measured by the area within 378 a 1 degree C isotherm1). Pacific cod avoid the waters of the cold pool, and are present in the highest

1 The cold pool does not have a universally defined temperature boundary. Stabeno et al., 2001 use 2 degrees Celsius. Kotwicki and Lauth (2013) emirically test the avoidance behavior of several species. They find that for Pacific cod, avoidance of the 1-degree isotherm is strongest, although they also avoid 1-2 degree C water. In this

Page 48 of 127 Please do not cite this chapter without permission from the authors.

379 densities outside of the 1 degree C isotherm. This means that during cold years, Pacific cod are more 380 concentrated in the deeper waters in the north and west of the Bering Sea shelf than in warm years. 381 However, they did not find a statistically significant cold pool-related shift in the global center of gravity 382 of the population over the time period studied (1982-2011). Kotwicki and Lauth (2013) also found that 383 the global center of gravity of the population shifts to the northwest in periods of higher Pacific cod 384 abundance.

385 The longline (hook and line) catcher-processor sector of the fishery for Pacific cod consists of a 386 total of 39 vessels (from 2000-2010). Vessels range from 98 to 196 feet in length and catch from these 387 vessels is processed, frozen, and packed in a factory on the vessel. Two seasons make up the fishery: the 388 A- or winter season begins on January 1, and the B- or summer season starts on August 15.2 The fishery 389 operated under a limited-entry system until the B-season of 2010, when Congress passed legislation 390 allowing the fleet to form a cooperative.3 Before the cooperative was created, the fishing season ended 391 when the TAC was reached. There is evidence of an escalating race to fish over this time period; the 392 length of the seasons decreased dramatically throughout the 1990s and 2000s. Under a race to fish, 393 vessels have an incentive to compete to catch the largest portion of the TAC possible before the total TAC 394 is reached and the season is closed.4 This makes the time cost of travel very high, as time spent in travel is 395 time spent not fishing that cannot be recovered later. Under a common pool TAC, vessels are less willing 396 to travel farther from port for marginal increases in catch per unit effort (CPUE) than they are when TAC 397 share is individually assigned. Eighteen vessels have endorsements to fish for Pacific cod in the Gulf of 398 Alaska, but the number of vessels that participate in a given year has ranged from 4 to all 18. Vessels also 399 fish in the Aleutian Islands, but effort there has been partially restricted by spatial regulations designed to 400 protect prey for the endangered Western stock of Steller sea lions (NMFS, 2013). 401 Longline fishing for Pacific cod involves the deployment of approximately 10-20 kilometers of 402 line with an average of over 1000 hooks per kilometer. Line is set, left to soak for an average of 13 hours, 403 and then crewmen remove fish from the individual hooks as the line is reeled in. An informal agreement

paper, we show the cold pool at 1.5 degrees because the data from the Eastern Bering Sea bottom trawl survey that is available to us defines the “2 degree isotherm” as the area in which the water was between 1.5 and 2.5 degrees Celsius. Thus the borders would be at 0.5 degrees, 1.5 degrees, etc. 2 The B-season has opened on August 15 since 2001; prior to 2001 the opening of the B season varied from Aug. 31 to Sept. 15. 3 The members of the current cooperative had been lobbying for the act since at least 2004. The cooperative includes all vessels that were participating in the fishery except 5 that participated in a federal buy-back program. 4 In addition, vessels may have been competing for quota history in the years before the cooperative was created. Given their active lobbying of decision makers, participants expected that the creation of cooperatives was imminent and that quota allocation rules, which are often based on catch history, would need to be created. This created an additional incentive to race for the annual quota, because an extra ton of fish caught during the years chosen to define the quota allocation rules could potentially mean an extra ton of quota in perpetuity after the cooperative was created.

Page 49 of 127 Please do not cite this chapter without permission from the authors.

404 within the sector regarding vessel spacing exists; vessels respect a buffer on the order of 30 km around 405 other vessels that are actively fishing. Vessels work within their 30 by 30 km grid, but may move with a 406 school of fish. If CPUE decreases in an area such that the expected value of fishing in some other 407 location, minus the cost of travel, is greater than the value of the current location, the vessel will move to 408 a new location to fish. Declines in CPUE could be due either to local depletion or movement of fish out of 409 the area. When the hold is full, the vessel will return to port to offload. 410 The purpose of this paper is to explore the links between climate and fishing behavior in the 411 Bering Sea Pacific cod longline fleet. Climate affects the local distribution of Pacific cod through the 412 spatial extent of the cold pool, and abundance affects both the local and global distribution (Kotwicki and 413 Lauth, 2013). However, research on other species has shown that changes that the fishery do not 414 necessarily mirror changes in the distribution of biomass (Haynie and Pfeiffer, 2013). Harvesters’ fishing 415 decisions may be affected by many other factors. Spatial or temporal differences in prices, for example, 416 may drive harvesters to fish in different areas at different times, emphasizing value over catch rates when 417 net revenues are increased by the choice. The cost of travel is affected by distance, vessel and fleet 418 characteristics, and the time cost of travel, and may affect a vessel’s willingness to travel. Institutions, for 419 example whether the fishery is regulated as open-access, limited-entry, cooperatives, or with individual 420 fishing quotas, affect profit-maximizing behavior and the constraints under which harvesters make 421 decisions. Consideration of these factors leads to more nuanced, complex, and accurate retrospective 422 analyses and predictions (Haynie and Pfeiffer, 2012). In the case of the Pacific cod fishery, annual 423 variation in the size of the cold pool causes variation in the concentration of biomass outside of the 1 424 degree C isotherm. In addition, the extent and timing of winter sea ice advance may impact fishing, as 425 vessels cannot physically travel where sea ice is present. These factors may affect many aspects of 426 fishing, including catch rates, the amount of search needed to encounter acceptable catch rates, the rate of 427 local depletion, the length of trips, the spatial distribution of fishing, or the temporal distribution of 428 fishing. These differences in fishing behavior may have implications for the cost of fishing or the location 429 and timing of fish removals (which may have population biology implications). 430

431 2. Methods 432 2.1. Data 433 Set-level data on the catch location, quantity, size, and composition of Pacific cod longline vessel 434 harvests from the Alaska Fisheries Science Center’s North Pacific Observer Program Database were used 435 for the analysis. One hundred percent of days at sea by vessels in the fleet over 125 feet are observed, 436 while 30 percent of days at sea by vessels less than 125 feet are observed, resulting in a total Observer

Page 50 of 127 Please do not cite this chapter without permission from the authors.

437 coverage of approximately 80% of total catch.5 We assumed that the vessels’ trips, when unobserved, can 438 be represented by the average of the observed trips. Consistent observer data is available from 1992- 439 present, but we use only data through the A-season of 2010. The formation of the cooperative in 2010 440 changed the incentives facing the fishery, from a competition for the seasonal fleet-wide TAC or “race to 441 fish”, to a fishery where vessels may fish their individual allocation of TAC in a way that is individually 442 profit maximizing. This makes the data on the spatial distribution, timing, and other characteristics of the 443 behavior of the fleet under the cooperative not directly comparable to the characteristics of the fleet 444 before the cooperative. 445 These data are used to investigate the relationship between climate and fishery-dependent CPUE. 446 Fishery-dependent CPUE is standardized by vessel to control for vessel-level differences (Maunder and 447 Punt, 2004, Hilborn and Walters, 1992). Four fishing “zones” are defined to analyze the spatial 448 distribution of effort: the north, the south, the Aleutian Islands, and the Gulf of Alaska (Figure 2). Fishing 449 in the north is defined as fishing west of 171°W, or northwest of the Pribilof Islands, and fishing in the 450 south is defined as fishing east of 171°W. The majority of catch is landed at the port of Dutch Harbor. 451 Data from Vessel Monitoring Systems (VMS) aboard all Pacific cod vessels were used to 452 determine the timing of the beginning and end of trips prior to 2008, when trips began to be recorded in 453 the Observer records. VMS collection began in the B-season of 2002. Thus, for parts of the analysis that 454 involve trips, a subset of the data from the B-season of 2002 to the A season of 2010 was used.

455 Variables describing the environmental conditions of the BSAI were collected from the bottom 456 trawl survey of the Bering Sea which is conducted annually.6 Data that are used to calculate the average 457 bottom temperature of the Eastern Bering Sea are collected during the trawls and are lower when the cold 458 pool is large. Figure 2 shows the size of the cold pool in 2009, a cold-regime year, and 2004, a warm- 459 regime year. Seasonal ice cover, the timing of its retreat, and the size of the cold pool were used to 460 categorize years into temperature regimes. Figure 3 shows the average bottom temperature, the ice cover 461 index7, and the resulting categorization of years into warm and cold regimes (categorization from Stabeno

5 Catch and general location information about unobserved catcher-vessel fishing is collected at landing; however, it is not possible to disaggregate the landings data to the level of individual hauls. 6 Information on the survey conducted in June-July by the Resource Assessment and Conservation Engineering (RACE) Division of the Alaska Fisheries Science Center is available at http://www.afsc.noaa.gov/RACE/groundfish/default.php.

7 Ice cover index obtained from www.beringclimate.noaa.gov on 12/16/2011. The ice cover index is the average ice concentration for Jan 1-May 31. Ice concentration data are from the National Snow and Ice Data Center (NSIDC) using a bootstrap algorithm for historical data (through ~2006) and the NASA Team algorithm for more current data. The data start from late 1978 and are daily time series calculated by Sigrid Salo (NOAA/PMEL) to give average ice concentration in a 2-deg x 2-deg box (56-58°N, 163-165°W). The final index is given as normalized anomalies for each year, based on the mean (7.15) and standard deviation (4.01) for the period 1981-2000.

Page 51 of 127 Please do not cite this chapter without permission from the authors.

462 et al., 2012).

463 464 Figure 2. The extent of the cold pool in 2004 (a warm year) and 2009 (a cold year), as measured by the 465 annual bottom trawl survey. 466

Page 52 of 127 Please do not cite this chapter without permission from the authors. 6 4 4 3 2 2 Ice cover index 0 1 Temperature (degrees Celsius) 0 -2

1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

Cold regime Avg. bottom temperature Transition year Ice cover index Warm regime 467 468 Figure 3. Climate regimes, average bottom temperature, and the ice cover index of the Eastern Bering 469 Sea, 1992-2010. 470 The total allowable catch for Pacific cod (TAC) is set based on projections from a biological 471 stock assessment model (Thompson et al., 2010) with data for the model primarily collected from an 472 annual bottom trawl survey. TAC has ranged from 164,500 t to 270,000 t since 1992, and total biomass 473 estimates from the model range from 0.92 million t to 1.4 million t (Figure 1).

Page 53 of 127 Please do not cite this chapter without permission from the authors.

1500

1300

1100

900

700

500 mt) (thousands TAC 300

100 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

TAC Model Estimated Biomass 474 475 Figure 1. Bering Sea and Aleutian Islands total allowable catch (TAC) and model estimated age 0+ 476 biomass (Thompson et al., 2010).

477

478 2.2. Empirical methods 479 Figure 4 illustrates the expected effects of climate on the fishery. Harvesters are assumed to make profit- 480 maximizing decisions about where to fish; expected revenue is a function of expected prices and expected 481 CPUE in the chosen location, and costs are a function of travel costs (which may include factors such as 482 fuel, the time cost of travel, and danger). 483 Prices received for the catch can be important drivers of fishing location choice (Haynie and 484 Pfeiffer, 2013). However, harvesters in this fishery, prior to rationalization, have not demonstrated a 485 significant response to the observed variation in product prices in terms of their location choice or other 486 behaviors (Pfeiffer and Haynie, 2014) . Thus, in this paper we have assumed that spatial differences in 487 prices due to production choices have not played an important role in the behavior of the longline fleet 488 during the time period considered, and can reasonably be excluded. 489 In this paper, we concentrate on the effects of climate on expected fishery CPUE, and the effects 490 that climate-related changes in expected CPUE have on harvester behaviors that potentially affect fishing 491 costs. We test the hypotheses depicted in the conceptual model in Figure 4. First, we test if the 492 distribution of water temperatures (measured by the average bottom temperature of the Eastern Bering 493 Sea) affected CPUE on the fishing grounds. Second, we test if biomass affects fishery CPUE. While 494 Pacific cod biomass may be affected by climate-related factors, the relationship is indirect and

Page 54 of 127 Please do not cite this chapter without permission from the authors.

495 complicated, and beyond the scope of this paper. Rather, we take biomass as given and control for 496 variation in biomass to identify the effect of water temperatures on CPUE. Then, we investigate how 497 variation in CPUE has affected several characteristics of harvester behavior, including the number of sets 498 in a trip, the number of moves, and the spatial distribution of fishing. The spatial distribution and timing 499 of fishing may be affected by the seasonal sea ice extent, as harvesters avoid ice-covered areas. Finally, 500 we estimate each factor’s contribution to the total distance traveled in a trip and the distance-per-ton of 501 fish caught in a trip. We can then compare costs between cold and warm climate regimes.

502 503 Figure 4. Mechanisms relating climate to Pacific cod harvester behavior and the costs of fishing. 504

505 2.2.1. Effects of climate and biomass on CPUE 506 The size of the cold pool, which varies enormously from year to year, is directly related to the 507 annual climate regime (Stabeno et al., 2012). Pacific cod avoid the cold pool, and are found in greater 508 concentrations outside of the 1 degree C isobath (Mueter and Litzow, 2008, Kotwicki and Lauth, 2013). 509 In very cold years the cold pool extends far to the east in the southeast Bering Sea into Bristol Bay, while 510 in very warm years it may only extend to approximately 58ºN latitude (Figure 2). This may result in an 511 increased concentration of biomass, and therefore higher CPUE, in cold years when the cold pool is large. 512 In addition, higher abundance may increase CPUE overall, or may increase the spatial range of a species 513 (Holt et al., 1997, Kotwicki and Lauth, 2013).

514 A linear regression relating average standardized CPUE in each fishing season to average annual 515 bottom temperatures and annual estimated biomass is used to investigate the relationship between 516 abundance and CPUE, primarily to be sure that any link relating climate to CPUE is not primarily 517 attributable to changes in abundance. Climate may affect abundance directly, but the effect is complicated 518 to identify because it is likely through recruitment, predation, and survival, and thus involves a variety of 519 temporal lags. This is a subject for future research. Seasonal effects are included because fishery CPUE 520 may vary over the course of a year. It may change due to migration patterns or other ecosystem drivers

Page 55 of 127 Please do not cite this chapter without permission from the authors.

521 that cause the population to become more or less concentrated at different times of the year. For example, 522 the disappearance of the cold pool in the fall may cause fish concentration to decrease in the summer and 523 fall of cold years. However, if the population is relatively immobile on a large spatial scale, the size and 524 location of the cold pool may determine the location of fish for the entire year. CPUE may also decrease 525 due to local depletion from fishing, or it may increase as fish feed and gain weight throughout the year.

526 2.2.2. Vessel responses to changes in CPUE 527 2.2.2.1. Number of sets 528 In general, vessels fish until they are almost full, or need to return to port for some other reason 529 because it is costly to travel to and from the fishing grounds. Thus, the number of sets in a trip is expected 530 to be directly related to the average fishing conditions during that trip (better fishing will lead to fewer 531 sets, because the vessel’s hold fills faster) and the size of the vessel’s hold; a linear regression is used to 532 estimate the effect of each factor. The predicted number of sets per trip is compared for cold and warm 533 years with a t-test.

534 2.2.2.2. Number of moves and number of sampling sets 535 Harvesters in the Pacific cod longline fleet typically fish in one place until the CPUE decreases to 536 where the benefit of the expected increase in CPUE exceeds the cost of moving to a new location. This 537 basic result of a fishing location choice model for any type of fishery, that a harvester will fish in the 538 location where they expect the greatest profit, was validated in industry interviews. One of the 539 characteristics of this fishery, however, is that the timing of moves can be ascertained from the data. If 540 skippers make the “right” choices (they make economically efficient moves) on average, then the average 541 distance at which the change in CPUE from one set to the next becomes greater than zero is the distance 542 of a “move”. This distance was empirically estimated to be approximately 40 km .8 Forty km is consistent 543 with the informal agreements between captains to work within a grid of approximately 30 km by 30 km. 544 A sequence of sets is defined as consecutive sets fished without moving a distance greater than 40 km. 545 The number of moves a vessel makes within the span of a fishing trip is representative of the 546 quality of a fishing location if a) CPUE decreases at some rate due to local depletion, and b) the 547 probability that a vessel changes locations increases as CPUE in the current location decreases. We test if 548 each assumption is supported by the data. 549 To test if CPUE decreases over a sequence of sets, CPUE at the beginning of a sequence of sets in

550 one location was compared to CPUE at the end of a sequence, or ΔCPUEseq:

8 We conducted the analyses included in the paper using alternative definitions of a “move” (30km and 50km) to determine robustness. The main results and conclusions of the analyses did not change with either alternative definition.

Page 56 of 127 Please do not cite this chapter without permission from the authors.

= (1) 푵 푵 풏=ퟎ ퟏ+풏 풏=ퟎ 푻−풏 풔풆풒 ∑ 푪푷푼푬 ∑ 푪푷푼푬 551 where CPUE1 is the first set∆ 푪푷푼푬of a sequence, CPUET is the last− set of a sequence, and ΔCPUEseq is 푵 푵 552 calculated for N=1, 2, 3 (a 1-, 2- and 3-set average at the beginning and end of a sequence). The final 553 sequence at the end of each trip was not included, because it was more likely to have ended because the 554 vessel hold was full or the trip was ended for other reasons. 555 To test if CPUE increases, on average, when a move is made, the change in CPUE for moving

556 averages of consecutive sets (ΔCPUEset) was calculated:

= (2) 푵 푵 ∑풏=ퟎ 푪푷푼푬풕−풏 ∑풏=ퟎ 푪푷푼푬풕+ퟏ+풏 557 again for N=1, 2, 3. Consecutive∆푪푷푼푬 sets풔풆풕 separated by moves of− greater than 40 kilometers were compared to 푵 푵 558 consecutive sets between which the vessel did not move (“non-moves” of less than 40 km). 559 Next, moves as a function of CPUE were modeled. The probability of changing locations while 560 on a fishing trip is expected to be a function of CPUE, but if CPUE decreases due to local depletion, 561 assuming the time (number of sets) until a move is normally distributed is unreasonable. Thus, the 562 probability of moving is modeled with a survivor function (Collett, 2009). If T is a nonnegative random 563 variable denoting the number of sets until a move, the probability that a vessel continues fishing beyond 564 set t can be characterized by the survival function S(t):

( ) = ( ) = ( > ) (3)

565 where F(t) is the cumulative distribution푺 풕 function.ퟏ − 푭 The풕 density퐏퐫 푻function풕 f(t) is ( ) ( ) = = { ( )} = ( ) (4) 풅푭 풕 풅 ′ 566 The hazard function, h(t) is the풇 limiting풕 probability thatퟏ −a “failure”,푺 풕 which−푺 풕 in this case is the probability of 풅풕 풅풕 567 changing fishing locations, occurs on a given interval. Thus it measures the conditional “move” rate, 568 conditional on having fished t sets in the current location: ( + > > | > ) ( ) ( ) = = (5) ( ) 퐏퐫 풕 휟풕 푻 풕 푻 풕 풇 풕 풉 풕 휟풕퐥퐢퐦→ퟎ 569 We can model the hazard function as 휟풕 푺 풕 ( ) = ( , + ) (6)

풋 ퟎ 풋 풙 570 Where xj is a vector of predictors of the풉 probability풕 품 of풕 휷moving,풙 휷including realized CPUE and abundance, β 571 are the parameters to be estimated, and g(.) is the Weibull function, which was compared to several 572 alternative distributions and determined to have the best fit. Standard errors are clustered by vessel to 573 account for correlation among data generated by the same vessel.

Page 57 of 127 Please do not cite this chapter without permission from the authors.

574 If the probability of a move is a function of CPUE, and CPUE varies with climate, then the 575 number of moves is impacted by climate variation. While the hazard model estimation confirms that 576 vessel moves are a predictable function of CPUE, it cannot be used to estimate the probability of a move 577 because CPUE is a time-varying variable (Cleves et al., 2008). Instead, we estimate the number of moves 578 per trip as a linear function of the factors that affect CPUE (including TAC, biomass, and the location of 579 fishing), and other factors that may affect the number of moves (such as vessel effects). CPUE is not 580 included directly because realized CPUE is endogenous; it is dependent upon the choices that vessels 581 have made, including the number of moves.

582 2.2.2.3. Spatial distribution of effort 583 Changes in the spatial distribution of effort may result from a differential effect of climate on 584 CPUE over space. For example, if CPUE in the north increases more than CPUE in the south, then 585 harvesters have a greater incentive to travel to the north to fish. Traveling to the north may increase the 586 total cost of a fishing trip (by increasing the total distance traveled). 587 At the level of a fleet, the mean center of the distribution of fishing in a season is used to describe the 588 spatial distribution of fishing locations (Woillez et al., 2007). The mean center of fishing locations was 589 calculated using the longitude and latitude of sets by the 100% observed vessels:

= , = (7) 푵 푵 . ∑풊=ퟏ 풙풊 ∑풊=ퟏ 풚풊 590 North/south variation in the latitude풙� of the풏 distribution풚� 풏 was modeled as a linear function of 591 temperature regime categories. 9 592 At the level of an individual vessel, we modeled the proportion of a fishing trip spent in the 593 northern reaches of the fishing grounds (west of 171°W) as a function of the same variables, as well as 594 the number of sets in a trip and vessel effects. The dependent variable is a share in the interval [0,1], so a 595 fractional logit model was used, where E(y|x) was modeled as a logistic function (Papke and Wooldridge 596 1993; Wooldridge 2002), ( | ) = exp( ) /[1 + exp( )]. The model was estimated using maximum

597 likelihood, and the estimated퐸 푦coefficients푥 were푥훽 exponentiated푥훽 to obtain odds ratios. Odds ratios can be 598 interpreted by subtracting 1; the odds ratio minus 1 has the interpretation of the proportionate change in 599 the probability resulting from a unit change in an independent variable.

600 2.2.2.4. Timing of fishing 601 Pacific cod longline vessels avoid areas of significant ice cover. Extensive ice occurs only in the 602 winter season of cold years. In order to avoid the ice but take advantage of high CPUE, harvesters may 603 shift the timing of fishing to avoid the time periods when ice extent is at its greatest. However, the cost of

9 Average bottom temperatures were used instead of temperature regime categories, with similar results.

Page 58 of 127 Please do not cite this chapter without permission from the authors.

604 ice avoidance (in terms of lost catch) may outweigh the benefits when there is a common-pool quota. 605 The percentage of 100% observed active vessels fishing on the second day of the season is used 606 as an indicator of the timing of fishing. While this measure does not capture all of the complexities 607 relating to the choice of when to fish a season, the regulations in place during the period under study 608 tended to incite early season effort. While delaying the start of a season could be advantageous for safety 609 or travel cost reasons, it would result in the loss of catch because the limited TAC would be caught by 610 other vessels10. The percentage of active vessels fishing on the second day of the season is modeled a 611 linear function of the ice cover index, CPUE, the number of vessels actively fishing in each season, and 612 TAC.11

613 2.2.3. Cost of fishing 614 Direct data on the cost of fishing is not collected for any of the BSAI longline fisheries. However, 615 a large percentage of the cost of fishing is the cost of fuel, and the cost of fuel at any given time is directly 616 proportional to the distance traveled. Therefore, we used the distance traveled in a trip as an indicator of 617 the total cost of a fishing trip, and investigated the impact of each of the factors we have just modeled on 618 distance. The distance traveled in a trip was modeled as a linear function of the number of sets, the 619 number of >40 km moves made, biomass, TAC, and the proportion of the trip spent fishing in the 620 Aleutian Islands, the Gulf of Alaska, and the northern regions of the fishing grounds. 621 Harvesters make tradeoffs between increased travel and increased harvest rates, so total distance 622 may increase as more fish are caught. Thus, the distance traveled per kilogram of harvest in a trip may be 623 a more relevant measure of the cost of travel, and was modeled as a function of the same variables. 624 The estimated models were used to predict the average distance traveled and distance traveled per 625 ton of catch in warm and cold climate regimes, and the difference in the means tested with a t-test.

626 3. Results 627 3.1. Effects of climate and biomass on CPUE 628 Fishery CPUE is contingent upon the fishing decisions (regarding location, timing, speed, etc.) 629 made in each year. In both fishing seasons, there is a significant negative relationship between mean 630 standardized fishery CPUE and the size of the cold pool, measured here by the average bottom 631 temperature of the annually surveyed area of the eastern Bering Sea, holding biomass constant (table 1). 632 In the B season, average CPUE is positively correlated with biomass ; in the A season there is no 633 significant relationship. In fact, biomass was at its lowest level of the study period in the cold years of

10 To be precise, any harvest that a vessel would forgo would be shared by the n vessels fishing at the end of the season, so a vessel would potentially harvest 1/n of their forgone harvest at season’s end. 11 The percentage of active vessels fishing on the 3rd and 5th day of the season was used as an alternative dependent variable, with similar results.

Page 59 of 127 Please do not cite this chapter without permission from the authors.

634 2006-2009 (Figure 1), which corresponds to a period of very high average fishery CPUE in the A season 635 (Figure 5). Kotwicki and Lauth (2013) found that in years when the cold pool is larger, Pacific cod avoid 636 a larger area of the eastern Bering Sea shelf. Table 1 is consistent with this finding, and is evidence that 637 this cold pool avoidance increases CPUE for the fishery, perhaps by concentrating fish into a smaller area 638 in the north and west of the fishing area. In addition, we find that CPUE is higher in cold years than in 639 warm years in both fishing seasons, even after the cold pool has disappeared through mixing of the water 640 column. This indicates that while Pacific cod are generally thought of as a mobile species (Conners and 641 Munro, 2008), the effect of the cold pool on the location of biomass can persist through the year. 642 643 1 .8 .6 .4 Linear predictionof fishery CPUE (kg/hook)

.6 1 1.4 1.8 2.2 2.6 3 3.4 3.8 Average bottom temperature of surveyed area (Deg. C)

season=1 season=2

644 645 Figure 5. Mean standardized fishery CPUE in the winter (A) and summer (B) seasons and the average 646 bottom temperature of the Eastern Bering Sea, 1992-2010. 647 648 Table 1. Estimated effect of average bottom temperature and biomass on standardized CPUE. CPUE (kg/hook) Avg. bottom temperature (°C)*A season 0.161 (0.162) Avg. bottom temperature (°C)*B season -0.226*

Page 60 of 127 Please do not cite this chapter without permission from the authors.

(0.127) Avg. bottom temperature2 (°C)*A season -0.023 (0.080) Avg. bottom temperature2 (°C)*B season 0.108* (0.060) Avg. bottom temperature3 (°C)*A season -0.007 (0.012) Avg. bottom temperature3 (°C)*B season -0.020** (0.009) Model estimated biomass (millions mt)*A season -0.103 (0.103) Model estimated biomass (millions mt)*B season 0.467** (0.088) Constant 0.909** (0.142) B season -0.662** (0.168) N 104141

Adjusted R2 0.25

649 Note: * p<0.1, ** p<0.05. Standard errors in parentheses and clustered by vessel. Vessel fixed effects 650 included. 651

652 3.2. Vessel response to changes in CPUE and abundance 653 3.2.1. Number of sets 654 The number of sets in a trip is expected to be affected by the rate of harvest (CPUE) and vessel 655 size. When CPUE is higher, a vessel fills up more quickly with fewer sets. Vessels with larger holds can 656 fish more sets in a trip. Table 2 supports this hypothesis. Vessel fixed effects are included rather than 657 vessel length to control for additional non-time varying vessel characteristics such as horsepower and 658 hold size (R2=0.24 with vessel fixed effects vs. R2=0.10 with vessel length). Given CPUE conditions and 659 assuming that vessels make optimal choices about when and where to move, the predicted number of sets

Page 61 of 127 Please do not cite this chapter without permission from the authors.

660 is about 30% higher in warm years than in cold years. 661 662 Table 2. The estimated effect of CPUE on the number of sets per fishing trip. Number of sets TAC (thousands t) 0.143** (0.030) CPUE (kg/hook) -25.040** (2.009) Vessel effects included Constant 38.526** (7.311) N 1570 Adjusted R2 0.24 Predicted number of sets in cold years 31.1 Predicted number of sets in warm years 40.5 P-value of t-test of difference 0.000 663 Note: * p<0.1, ** p<0.05. Data is restricted to 2002B-2010A, the subset of data for which trip 664 information is available. 665

666 3.2.2. Number of moves 667 Cod longline vessels are assumed to fish in an area until the CPUE declines such that the CPUE 668 benefit from moving to a new location outweighs the cost of moving. We find that CPUE declines over a 669 sequence of sets that are followed by a move to a new fishing location (Table 3). For simplicity, results of 670 N=2 (a 2-set average) are presented, but the results from N=1, 3 were similar. CPUE declines by an 671 average of 0.068 kg/hook over a sequence of sets in one location when that sequence ends with the 672 harvester leaving to fish in another location. The decrease is significantly different from zero.

673 Table 3. Comparison of the mean change in CPUE (ΔCPUEseq) over a sequence of sets in one location Sequences not at end of trip

Mean ΔCPUEseq -0.068 (95% confidence interval) (-0.079, -0.057) Number of sequences not at end of trip 1787 Mean length of sequence (number of sets) 14.2

Page 62 of 127 Please do not cite this chapter without permission from the authors.

674 Note: Sequences tested in the 2-set average are at least 4 sets long. Data is restricted to 2002B-2010A, the 675 subset of data for which trip information is available. 676 677 678 When harvesters move to a new location to fish (greater than 40 kilometers from their previous 679 location), CPUE increases by an average of 0.065 kg/hook (Table 4). For consecutive sets between which 680 the vessel did not move, the average change in CPUE is very slightly negative. Again, we present the 681 results for N=2, but for N=1 and N=3, the results were similar. 682 683 Table 4. Mean change in CPUE between sets when a vessel moved to a new location and when it stayed 684 in the same location

ΔCPUEset moves ΔCPUEset non-moves

Mean ΔCPUEset 0.065 -0.002 95% confidence interval (0.056, 0.073) (-0.005, 0.000) N 2968 43207 685 Note: Data is restricted to 2002B-2010A, the subset of data for which trip information is available. 686 687 The number of moves is a predictable function of CPUE. Table 5 shows the results of a hazard 688 model estimation of the probability of moving, which considers that the probability of moving varies with 689 the number of sets that have already occurred in a given location. The first column shows the regression 690 coefficients. For example, if CPUE were to increase by 1 unit and if no nonlinear terms or interactions 691 were included in the model, the hazard (probability of moving) would decrease by 86.3% because exp(- 692 1.984)=0.137 and 1-0.137=0.863. Note that a 1-unit change in CPUE is very large relative to the mean; 693 CPUE has a mean of 0.64 and a standard deviation of 0.34, and that interactions and nonlinear terms are 694 included. From the first column, it is evident that the probability of moving decreases at an increasing rate 695 as CPUE increases. Lagged CPUE decreases the probability of moving at an increasing rate as well,

696 although the coefficient on “CPUE*CPUEt-1” is not significant, indicating that there is no additive effect 697 of a higher realization of CPUE for two sets in a row. The second column contains the hazard ratio, 698 accounting for the interactions. If we calculate the effects at a more reasonable increase, for example, a 699 10% increase in CPUE would decrease the probability of moving by about 4.1% (0.064*(1-0.349)). A 1- 700 unit increase in lagged CPUE decreases the probability of moving by about 3.3% (0.064*(1-0.489)). P is 701 the Weibull shape parameter. It is significantly less than 1, indicating that as the consecutive number of 702 sets in a single location increases, the probability of moving decreases (at a decreasing rate), holding all

Page 63 of 127 Please do not cite this chapter without permission from the authors.

703 other variables in the model constant. 704 705 Table 5. Results of hazard model estimation of the probability of moving a distance greater than 40 km Probability of moving Hazard ratios considering Hazard model coefficients the interactions CPUE (kg/hook) -1.984** 0.349 (0.147) CPUE2 0.318** (0.042) Average size of fish (kg) -0.098** 0.976 (0.019) CPUE*Average size 0.116** (0.021)

CPUEt-1 -1.101** 0.489 (0.108) 2 CPUEt-1 0.222** (0.092)

CPUE*CPUEt-1 0.154 (0.149) Constant -0.272** 0.761 (0.091) N 80681 P 0.817** (0.183) 706 Note: * p<0.1, ** p<0.05. Data is restricted to 2002B-2010A, the subset of data for which trip 707 information is available. 708 709 The mean number of moves per trip was nearly twice as large warm years compared to cold 710 years, controlling for the total number of sets, month of the trip, TAC, fishable biomass, the proportion of 711 effort in each area, and vessel effects (Table 6). 712 Table 6. Determinants of the number of moves >40 km per trip. Number of moves Cold years Base category

Page 64 of 127 Please do not cite this chapter without permission from the authors.

Transition years 0.701* (0.364) Warm years 1.322** (0.565) Number of sets 0.076** (0.003) Vessel effects included TAC (thousands t) -0.035** (0.014) Abundance (model estimated age 0+ biomass (thousands t)) 0.000 (0.001) Constant 3.434 (2.982) N 1570 Adjusted R2 0.26 713 Note: * p<0.1, ** p<0.05. Data is restricted to 2002B-2010A, the subset of data for which trip 714 information is available. 715

716 3.2.3. Spatial distribution of effort 717 The mean center of the distribution of fishing results from hundreds of decisions made by 718 harvesters about where to fish each individual set. It is an aggregate measure of the effects of external 719 factors, as reflected in harvesters’ endogenous decision-making process. We find no evidence that the 720 mean center of the distribution of fishing shifts due to climate regime, controlling for abundance 721 (measured with estimated age 0+ biomass), in either the A or the B seasons (Table 7). The mean center of 722 the distribution is affected by abundance, however. A 100 000 t increase in abundance is associated with a 723 southward shift in the mean center of the distribution of fishing by 0.2 degrees latitude in the A season, 724 and 0.3 degrees latitude in the B season (Table 7). Haynie and Pfeiffer (2013) found similar results for the 725 Bering Sea Alaska pollock fishery. 726 727 Table 7. Determinants of the mean center of the distribution of fishing, 1992-2010. A season B season mean center mean center

Page 65 of 127 Please do not cite this chapter without permission from the authors.

(latitude) (latitude) Abundance (model estimated 0+ biomass, -0.002** -0.003** thousands t) (0.00) (0.00) Cold years Base Base category category Transition years -0.318 -0.065 (0.29) (0.33) Warm years -0.156 0.185 (0.21) (0.27) Constant 59.656** 60.147** (0.84) (1.07) N 18 18 Adjusted R2 0.25 0.26 728 Note: * p<0.1, ** p<0.05. Data is restricted to 100% observed vessels to calculate the mean center of the 729 distribution, because the timing of observation of 30% vessels may vary by year and season. 730 731 At the level of trips made by individual vessels, the proportion of a trip fished in the northern 732 reaches of the fishing grounds, west of 171°W (to the northwest of the Pribilof Islands) was modeled 733 (using the trip-level subset of data, 2002-2010). The results of the maximum likelihood estimation of the 734 share of fishing in the north, as a function of climate regime, month of the trip, trip length, TAC, 735 abundance, vessel effects, and interaction terms between climate regime and month are summarized in 736 Figure 7 (regression results available from the authors). In January of cold years, vessels spend about 50% 737 of a trip in the north. In contrast, they spend about 25% of a trip in the north in January of warm years. 738 This difference between warm and cold years disappears after January, suggesting that vessels may be 739 fishing in the north prior to the descent of seasonal sea ice from the Arctic, which usually occurs in 740 February and March. The A-season usually closes in March and the B-season opens in August. The 741 proportion of fishing effort in the north in warm or cold years is not statistically different in the B- 742 season.12

12 Similar models were estimated for the proportion of effort in the Aleutian Islands and the Gulf of Alaska as the dependent variables, but most of the explanatory power was from vessel effects indicating that variation climate, TAC, and abundance do not affect the choice to fish in the Aleutian Islands or the Gulf of Alaska.

Page 66 of 127 Please do not cite this chapter without permission from the authors. 1 .8 .6 .4 .2 Predicted proportion of effort west of 171 deg. W 0

1 2 3 8 9 10 11 12 Month of trip

Cold years Warm years 743 744 Figure 7. Predicted proportion of effort expended west of 171°W longitude, northwest of the Pribilof 745 Islands, as a function of climate regime and month of the year. 746

747 3.2.4. Timing of fishing 748 The timing of fishing may be influenced by the expected extent of sea ice in the Bering Sea. 749 General expectations about the extent of ice are possible given winter temperatures and expected spring 750 temperatures and wind patterns. At times, ice may completely cut off access to the northwest regions of 751 the fishery and the northern portion of the southern Bering Sea. Vessels may fish intensively at the 752 beginning of the season in order to fish in the northern region before the advance of ice onto the fishing 753 grounds (as suggested by Figure 7). Alternatively, they may wait until after the ice has retreated from the 754 fishing grounds. However, we find no statistically significant evidence of a change in the timing of 755 fishing due to climate regime (Table 8). Rather, there is a significant linear time trend toward earlier 756 fishing in both seasons. The percentage of active, 100% observed vessels fishing on the second day of the 757 season increases by 2.6 % on average each year in the A season, and by 1.7% each year in the B season. 758 This is consistent with an escalating race for fish in the fishery (Brinson and Thunberg, 2013). In 759 addition, vessels began fishing earlier in years with higher TAC. A 1000 t increase in annual TAC results 760 in a 0.2% increase in the percentage of vessels fishing on the second day of a season. 761

Page 67 of 127 Please do not cite this chapter without permission from the authors.

762 Table 8. Determinants of the timing of Pacific cod fishing. Proportion of active vessels Proportion of active fishing on 2nd day of A vessels fishing on 2nd day season of B season TAC (thousands t) 0.208* 0.244** (0.10) (0.10) Cold years base Transition years -0.571 6.313 (7.97) (5.90) Warm years -4.096 3.585 (5.96) (4.73) Time trend 2.575** 1.703** (0.59) (0.73) Constant -7.981 -10.530 (24.48) (28.33) N 18 15 Adjusted R2 0.50 0.23 763 Note: * p<0.1, ** p<0.05. Data is restricted to 100% observed vessels to calculate the proportion of active 764 vessels fishing on the second day of each season, because the timing of observation of 30% vessels may 765 vary by year and season. Data does not include CDQ fishing, which may be fished when the directed 766 fishery is closed. 767

768 3.3. Cost of fishing 769 At any given time, vessels may travel a longer distance because they expect to be rewarded with 770 higher catch rates. Table 9 shows that total distance traveled in a trip increased by an average of 138 km 771 for each move >40 km made by a vessel in the A season, and by 127 km per move in the B season. 772 Distance per trip increased as the number of sets increased, by approximately 26 km in both the A and B 773 seasons. Higher abundance, holding TAC constant, decreased the total distance traveled in a trip (finding 774 sufficient levels of CPUE may be easier), while an increase in TAC, given abundance, was associated 775 with an increase in distance traveled. As the percentage of the trip that occurred outside of the most 776 southern zone (zone 1) increased, trips lengthened, with the greatest increases coming from effort in north 777 (zone 4), which is farthest from the ports. Vessel fixed effects were included in the model to account for 778 vessel characteristics and systematic differences in vessel behavior that do not vary over time.

Page 68 of 127 Please do not cite this chapter without permission from the authors.

779 While distance traveled per trip represents a measure of the total cost of a trip, distance-per-ton 780 more accurately represents the average variable cost of catching a ton of fish. In both seasons, moves 781 >40km increased the distance-per-ton by approximately 0.4 km per move. This means that while vessels 782 trade off the costs involved with moving to a new location for increased CPUE, their cost-per-ton 783 increased with the number moves they needed to do. Distance-per-ton decreased as the number of sets 784 increased, which was expected. Greater TAC, given abundance, increased distance-per-ton slightly, while 785 abundance did not have any effect. Distance-per-ton was higher in the northern eastern Bering Sea (zone 786 4) and the Aleutian Islands. 787 At the bottom of the table, mean predicted distance-per-trip and distance-per-ton per trip are 788 compared by climate regime, calculated at the average of relevant covariates. In warm years, lower CPUE 789 contributed to more moves, more sets, and a change in the location of fishing, increasing total distance 790 traveled by about 14% compared to cold years. In addition, while vessels can increase CPUE by changing 791 their fishing behavior, distance-per-ton in warm years was about 16% higher compared to cold years. 792 793 Table 9. Distance traveled per trip and distance-per-ton of catch in a trip as measures of the cost of fishing Distance/trip (km) Distance/ton/trip (km) A season Base category Base category B season -31.607 3.844** (63.782) (0.522) A season*Number of moves 138.142** 0.437** (14.218) (0.116) B season*Number of moves 126.599** 0.480** (7.415) (0.061) A season*Number of sets 24.822** -0.138** (1.695) (0.014) A season*Number of sets 25.557** -0.131** (1.034) (0.008) Abundance (model estimated age 0+ biomass (thousands t)) -0.590** 0.001 (0.186) (0.002) TAC (thousands t) 5.669** 0.043** (1.612) (0.013) % of trip east of 171°W (south) Base category Base category

Page 69 of 127 Please do not cite this chapter without permission from the authors.

% of trip west of 171°W (north) 7.901** 0.016** (0.434) (0.004) % of trip in AI 1.866* 0.062** (0.976) (0.008) % of trip in GOA -2.514 -0.028 (2.489) (0.020) Vessel effects included included Constant -22.270 3.461** (213.422) (1.747) N 1570 1570 Adjusted R2 0.73 0.37 Predicted distance in cold years 2106.13 7.95 Predicted distance in warm years 2404.784 9.21 P-value of t-test of difference 0.000 0.000 794 Note: * p<0.1, ** p<0.05. Data is restricted to 2002B-2010A, the subset of data for which trip 795 information is available. 796

797 4. Discussion 798 The results suggest that climate significantly influences fishing conditions for Pacific cod in the Bering 799 Sea and that vessels respond to the changing environment by adjusting many aspects of their behavior. 800 The characteristics of the fishery affect how harvesters respond to these changing conditions. Our results 801 corroborate Kotwicki and Lauth’s (2013) findings that Pacific cod populations are displaced by the larger 802 extent of the cold pool in cold climate regime years. Additionally, we find that a larger cold pool may 803 benefit the fishery by increasing average CPUE and decreasing the rate of local depletion while a vessel is 804 fishing in an area, which results in fewer moves around the fishing grounds. Higher average CPUE also 805 decreases the length of trips (in terms of the number of sets necessary to fill a vessel’s hold). Fewer 806 moves and fewer sets-per-trip result in less distance traveled on a trip. Vessels spend a larger percentage 807 of early A-season trips in the northern reaches of the fishing grounds, which increases total trip distance, 808 but the increase in CPUE attainable there likely compensates for the increase in distance. Trip-level 809 differences in the spatial distribution of effort did not contribute to an overall difference in the mean 810 center of fishing effort. While direct measures of the cost of fishing are not available with current data, 811 the cost of travel is a significant part of fishing costs. Total distance traveled in a trip and distance-per-ton 812 of catch in a trip were significantly lower in cold climate regime years.

Page 70 of 127 Please do not cite this chapter without permission from the authors.

813 The characteristics of the fleet, the fishery, and the management of the Pacific cod longline 814 fishery have shaped how harvesters respond to climate variation. A fishery with different characteristics 815 would be expected to respond quite differently. For example, longline fishing involves the placement of 816 gear in the water and retrieval some hours later. Harvesters often have several sets of gear in the water at 817 the same time, and work back and forth in a relatively compact area. This is in contrast to trawlers, who 818 tend to constantly move about the fishing grounds. Thus, it may be more costly for a longline vessel to 819 move to a new location to fish. Climate-driven differences in the frequency at which these moves occur 820 were found to be the largest source of differences in distance traveled. In the Eastern Bering Sea 821 catcher/processor pollock trawl fishery, on the other hand, the propensity of vessels to travel farther north 822 in cold years to take advantage of the increased concentration of pollock was found to be a very important 823 source of differences in travel costs (Haynie and Pfeiffer, 2013). Thus, cold climate regime years resulted 824 in less costly fishing for the Pacific cod longline fleet, but more costly fishing for the pollock 825 catcher/processor fleet in the same region. In addition, during the time period under study (1992-2010), 826 the Pacific cod longling fishery was managed as a limited entry, TAC-regulated fishery. This means that 827 vessels competed for a limited quantity of TAC, making the attainment of high rates of CPUE with 828 minimal travel time important. Beginning in the B season of 2010, the fishery underwent a profound 829 institutional shift. All vessels in the fishery are now members of the newly-formed cooperative, and 830 TAC is individually allocated among members by the cooperative. This change eliminated the race to fish 831 (the number of days spent fishing the A-season TAC increased from 38 to 158 days from 2010 to 2011), 832 and is expected to greatly alter the behavior of harvesters in the fishery (Grafton, 1996, Homans and 833 Wilen, 2005). The time cost of travel will decrease, potentially resulting in more moves and more travel 834 to the north. The value of harvested fish is expected to become much more important in the harvesters’ 835 decision-making process; small differences in prices due to quality, size, and product form can be 836 exploited by targeting the fish most suitable for the highest value product. Fishing speed is expected to 837 slow, and more robust markets for byproducts may develop. In years when TAC is low, some vessels may 838 not fish at all. These changes, and their interaction with climate-related changes, will be the subject of 839 future research. 840 In addition to cold pool-related changes in distribution, the total abundance of Pacific cod was 841 shown to affect harvester behavior. While climate factors have thus far not been shown to directly affect 842 the Pacific cod population, the climate links are complex and are the subject of ongoing ecological 843 research. However, the population of Pacific cod in the eastern Bering Sea experiences strong abundance 844 cycles, and it is important to control for changes in abundance even if the mechanistic link to climate 845 factors is incomplete at this time. Higher abundance is associated with higher average CPUE, but in 846 contrast to Kotwicki and Lauth’s (2013) finding that greater abundance is associated with a shift in the

Page 71 of 127 Please do not cite this chapter without permission from the authors.

847 center of gravity of the population to the northwest (a range expansion), we found that greater abundance 848 was associated with a shift in the center of gravity of commercial fishing effort to the south. This result 849 underscores the importance of understanding how economics drives the behavior of fish harvesters in 850 research on the effects of climate change on fisheries. Harvesters makes trade-offs between expected 851 returns and travel costs, and if an increase in abundance increases CPUE uniformly over the fishing 852 grounds, a vessel would not need to travel as far to find profit-maximizing rates of CPUE. Thus, the 853 center of gravity of fishing effort may shift to the south even if the range of the species has shifted to the 854 north. Notably, this is inconsistent with the “northward march” prediction of climate envelope-type 855 models (Cheung et al., 2010, Lehodey et al., 2006), which predict that fisheries will follow fish 856 populations northward as climate change affects their ranges. This study shows that the institutions and 857 economic drivers of a fishery must be thoroughly considered to determine the potential effects of climate 858 variation and climate change. 859 860 Acknowledgements 861 We would like to acknowledge helpful discussions with Bob Lauth, Stan Kotwicki, and Franz Mueter, 862 assistance with bottom trawl survey data from Angie Grieg and Bob Lauth, and additional climate data 863 from Franz Mueter. 864 865 Chapter References 866 867 Bockstael, N.E., Opaluch, J.J. (1983) Discrete modelling of supply response under uncertainty: The case 868 of the fishery. Journal of Environmental Economics and Management 10, 125-137.

869 Brinson, A.A., Thunberg, E.M. (2013) The economic performance of u.S. Catch share programs 160.

870 Chapman, W.L., Walsh, J.E. (1993) Recent variations of sea ice and air temperature in high latitudes. 871 Bulletin of the American Meteorological Society 74, 33-48.

872 Cheung, W.W.L., Lam, V.W.Y., Sarmiento, J.L., et al. (2010) Large-scale redistribution of maximum 873 fisheries catch potential in the global ocean under climate change. Global Change Biology 16, 24- 874 35.

875 Cleves, M.A., Gould, W., Gutierrez, R., Marchenko, Y. (2008) An introduction to survival analysis using 876 stata, Vol., Stata Press, College Station.

877 Collett, D. (2009) Modelling survival data in medical research, Vol., CRC Press, Boca Raton.

Page 72 of 127 Please do not cite this chapter without permission from the authors.

878 Conners, M.E., Munro, P. (2008) Effects of commercial fishing on local abundance of pacific cod (gadus 879 macrocephalus) in the bering sea. Fisheries Bulletin 106, 281-292.

880 Eales, J., Wilen, J. (1986) An examination of fishing location choice in the pink shrimp fishery. Marine 881 Resource Economics 2, 331-351.

882 Grafton, R.Q. (1996) Individual transferable quotas: Theory and practice. Reviews in Fish Biology and 883 Fisheries 6, 5-20.

884 Grebmeier, J.M., Overland, J.E., Moore, S.E., et al. (2006) A major ecosystem shift in the northern bering 885 sea. Science 311, 1461-1464.

886 Haynie, A., Layton, D. (2010) An expected profit model for monetizing fishing location choices. Journal 887 of Environmental Economics and Management 59, 165-176.

888 Haynie, A.C., Pfeiffer, L. (2012) Why economics matters for understanding the effects of climate change 889 on fisheries. ICES Journal of Marine Science 69, 1160-1167.

890 Haynie, A.C., Pfeiffer, L. (2013) Climatic and economic drivers of the bering sea walleye pollock 891 (theragra chalcogramma) fishery: Implications for the future. Canadian Journal of Fisheries and 892 Aquatic Sciences 70, 841-853.

893 Hilborn, R., Walters, C. (1992) Quantitative fisheries stock assessment: Choice, dynamics and 894 uncertainty. New York, Chapman and Hall Publishers.

895 Holt, R., Lawton, J., Gaston, K., Blackburn, T. (1997) On the relationship between range size and local 896 abundance: Back to basics. Oikos 78, 183-190.

897 Homans, F., Wilen, J. (2005) Markets and rent dissipation in regulated open access fisheries. Journal of 898 Environmental Economics and Management 49, 381-404.

899 IPCC (2007) Contribution of working groups i, ii, and iii to the fourth assesment report of the 900 intergovernmental panel on climate change. In: Climate change 2007: Synthesis report. (Eds. 901 R.K. Pachauri, A. Reisinger), IPCC, Geneva, Switzerland, p. 104.

902 Kotwicki, S., Lauth, R.R. (2013) Detecting temporal trends and environmentally-driven changes in the 903 spatial distribution of bottom fishes and crabs on the eastern bering sea shelf. Deep-Sea Research 904 Part II: Topical Studies in Oceanography 94, 231-243.

905 Lehodey, P., Alheit, J., Barange, M., et al. (2006) Climate variability, fish, and fisheries. Journal of

Page 73 of 127 Please do not cite this chapter without permission from the authors.

906 Climate 19, 5009-5030.

907 Maunder, M.N., Punt, A.E. (2004) Standardizing catch and effort data: A review of recent approaches. 908 Fisheries Research 70, 141-159.

909 Mueter, F.J., Litzow, M.A. (2008) Sea ice retreat alters the biogeography of the bering sea continental 910 shelf. Ecological Applications 18, 309-320.

911 Mysak, L.A., Manak, D.K. (1989) Arctic sea ice extent and anomalies, 1953-1984. Atmosphere-Ocean 912 27, 376-405.

913 NMFS (2013) Status review of the eastern distinct population segment of steller sea lion (eumetopias 914 jubatus). (Ed. A.R. Protected Resources Division, National Marine Fisheries Service), Juneau, 915 AK, USA.

916 Overland, J.E., Pease, C.H. (1982) Cyclone climatology of the bering sea and its relation to sea ice extent. 917 Monthly Weather Review 110, 5-13.

918 Pfeiffer, L., Haynie, A.C. (2014) The effects of individual fishing quotas on rent generation through 919 targeting and production choices National Marine Fisheries Service.

920 Smith, M., Wilen, J. (2003) Economic impacts of marine reserves: The importance of spatial behavior. 921 Journal of Environmental Economics and Management 46, 183-206.

922 Smith, M.D. (2002) Two econometric approaches for predicting the spatial behavior of renewable 923 resource harvesters. Land Economics 78, 522-538.

924 Spencer, P.D. (2008) Density-independent and density-dependent factors affecting temporal changes in 925 spatial distributions of eastern bering sea flatfish. Fisheries Oceanography 17, 396-410.

926 Stabeno, P.J., Bond, N.A., Kachel, N.B., Salo, S.A., Schumacher, J.D. (2001) On the temporal variability 927 of the physical environment over the south-eastern bering sea. Fisheries Oceanography 10, 81- 928 98.

929 Stabeno, P.J., Kachel, N.B., Moore, S.E., et al. (2012) Comparison of warm and cold years on the 930 southeastern bering sea shelf and some implications for the ecosystem. Deep-Sea Research Part 931 II 65-70, 31-45.

932 Thompson, G.G., Ianelli, J.N., Lauth, R.R. (2010) Assessment of the pacific cod stock in the eastern 933 bering sea and aleutian islands area.

Page 74 of 127 Please do not cite this chapter without permission from the authors.

934 Wyllie-Echeverria, T., Wooster, W.S. (1998) Year-to-year variations in bering sea ice cover and some 935 consequences for fish distributions. Fisheries Oceanography 7, 159-170.

936

Page 75 of 127 Please do not cite this chapter without permission from the authors.

937 Chapter 5: The effects of catch share management on rent generation through targeting and 938 production choices in the Pacific cod longline catcher processor fishery 939 940 Lisa Pfeiffer and Alan C. Haynie 941 This Draft: 6/13/2014 942 943 Abstract

944 Catch share management has been designed and used as a means for improving efficiency in fisheries by 945 reducing the dissipation of fishery resource rents. Important impacts often include a reduction in the 946 number of vessels fishing and the elimination of the incentive to overinvest along other dimensions, such 947 as crew, gear, and effort. In addition, catch shares may create new rents by creating the incentive for 948 harvesters to increase the value of each fish caught. We investigate changes in the fishing and processing 949 strategies of Bering Sea and Aleutian Islands (BSAI) Pacific Cod (Gadus macrocephalus) longline 950 catcher/processor harvesters due to the formation of a fishing cooperative in 2010 that involves catch 951 allocation to individual vessels. We focus on the creation of value in the fishery through two mechanisms: 952 the propensity of harvesters to target sizes of fish that increase product value, and the development of 953 markets for byproducts. We find the volume and the value of byproducts increased after the formation of 954 the cooperative, reversing a period of decreasing recovery rates and byproduct production that 955 corresponded with an escalating race-for-fish prior to cooperative formation. An entirely new byproduct, 956 roe for bait, began to be marketed under the cooperative. Finally, we note that the potential for broad- 957 scale changes in product form and profitability appear relatively limited compared to fisheries with high- 958 value product potential, such as roe in the Alaska pollock fishery and fresh (vs. frozen) fish in the Alaska 959 and British Columbia halibut fisheries.

960

961 1. Introduction

962 Different forms of catch share management (e.g., fishing cooperatives, individual fishing quotas (IFQs), 963 sectors, or territorial use right fisheries) have been designed and used as a means for improving efficiency 964 in fisheries by eliminating the “race to fish” that occurs because harvesters in fisheries that limit mortality 965 through seasonal catch limits have the incentive to compete with one another for the total allowable catch 966 (TAC). This “rule of capture” externality occurs because harvest by one vessel reduces the remaining 967 TAC for all other vessels in the fishery (e.g., Gordon 1954; Grafton 1996; Scott 1955) and results in the 968 incentive for vessels to overcapitalize and maximize fishing speed and the amount of fish caught each day

Page 76 of 127 Please do not cite this chapter without permission from the authors.

969 (e.g., Dupont 1990; Wilen 1979). If the number of vessels is not fixed by limited entry regulations, the 970 fishery will become overcapitalized via entry (Gordon 1954). If regulation attempts to limit fishing 971 mortality along other dimensions such as trip limits, area closures, effort controls, or other input controls, 972 harvesters typically respond by expanding effort along the unregulated factors of production (Wilen 973 1979). 974 The adoption or implementation of catch shares reduces the dissipation of fishery resource rents 975 through a variety of means. Important impacts include the reduction in overcapitalization through a 976 reduction in the number of vessels fishing (which is often mandated through vessel buybacks or license 977 limitations) and the elimination of the incentive to continue to overinvest along other dimensions, such as 978 crew, gear, energy, and materials (e.g., Dupont et al. 2002; Grafton 1996; Grafton et al. 2006). In 979 addition, new rents can be created by increasing the value of each fish caught, which can include 980 increasing product recovery rates, changing the production process and product mix to optimize 981 differences in product value, targeting fish spatially and temporally to obtain the highest value, 982 developing markets for byproducts or new products, decreasing the speed of fishing and processing to 983 increase quality, or developing trade cooperatives and contracting (Casey et al. 1995; Dupont et al. 2005; 984 Homans and Wilen 2005). Pre- and post-catch shares management comparisons often demonstrate 985 increases in revenue and average prices post-catch shares (Arnason 2005; Brinson and Thunberg 2013; 986 Grafton et al. 2000; Hannesson 2013; Scheld et al. 2012; Wilen and Richardson 2008), and a few 987 investigate the specific production-related sources of the increase (Casey et al. 1995; Herrmann 2000; 988 Morrison-Paul et al. 2009). 989 Characterizing the change in fishing and production behavior is an important component of the 990 analysis of changes in fisheries management. In 2010, National Oceanic and Atmospheric Administration 991 (NOAA) Fisheries, the body regulating fisheries management in the federal waters of the United States, 992 began advocating the implementation of catch shares in fisheries “wherever appropriate”, and mandated 993 the collection of data and development of indicators to evaluate the ability of catch shares to achieve the 994 economic and sustainability goals of the Magnusson-Stevens Fishery Conservation and Management Act 995 in different fisheries (NOAA 2010). These indicators, while useful for comparisons and summaries, 996 generally do not provide information about how and why changes have occurred. The mechanisms 997 through which average revenue changes, for example, are arguably more important to understanding 998 catch share management than the fact that average revenue changed. Disentangling the effects that are 999 unrelated to catch shares (e.g., changes in TAC and demand) from those that were mandated (changes in 1000 the number of vessels participating in the fishery) and those caused by harvester or processor behavior 1001 (changes in quality, targeting, production, or recovery rates) is essential to understanding how catch 1002 shares can change how value is created in the fishery. More broadly, understanding these mechanisms is

Page 77 of 127 Please do not cite this chapter without permission from the authors.

1003 essential to predicting if and how catch shares may have similar effects in other sectors, fisheries, or 1004 regions, and why management via catch shares may be more successful or create more wealth in some 1005 fisheries than others depending on the types of production changes that might occur in the fishery. 1006 A number of fisheries that have transitioned to catch-share management have demonstrated 1007 changes in their production processes. For example, in the three years prior to a trial period of catch 1008 shares in the British Columbia Pacific halibut fishery (which was implemented in 1991), 42 % of halibut 1009 was marketed as fresh. In the three years after catch shares, 89 % of the catch was marketed as fresh, 1010 resulting in an increase in ex-vessel price of about 55%. The elimination of the race for fish allowed 1011 harvesters to lengthen the period in which they fished for halibut (from 10 days in 1990 to 245 days in 1012 1991) to meet the demand for fresh product, rather than glutting the market which required that excess 1013 harvest be frozen (Casey et al. 1995). Alaska started its own catch share program for Pacific halibut in 1014 1995 that resulted in similar changes in season length, the marketing of fresh rather than frozen product, 1015 and increases in prices and revenue (Herrmann 2000). Market characteristics and the potential for drastic 1016 changes in product form vary widely by fishery. Catch shares in the Alaska pollock catcher-processor 1017 fishery were institutionalized through the formation of fishing cooperatives in 1999. The production 1018 process has since focused on factory efficiency: on-board factory processing lines are optimally tuned to 1019 the size of fish that is being caught, fishing occurs at the most efficient rate for factory through-put, and 1020 the production process of primary and by-products maximize product recovery rates (Wilen and 1021 Richardson 2008) The characteristics of the fishery and the market determine the means in which 1022 harvesters and processers might increase resource rents to maximize the expected value of the TAC after 1023 catch shares. In the halibut fishery, a complete transformation of product quality and form occurred. In 1024 the pollock at-sea fishery, profits were increased by fine-tuning harvesting and production efficiency. 1025 In this paper, we investigate changes in the production and size-targeting strategies of Bering Sea 1026 and Aleutian Islands (BSAI) Pacific Cod (Gadus macrocephalus) longline catcher/processor harvesters 1027 due to the formation of a fishing cooperative in 2010, which allowed the longline sector’s share of the 1028 TAC to be allocated to vessels based on historical catch, and thus operates as a de-facto catch shares 1029 fishery albeit with voluntary (currently 100 %) participation.13 In particular, we focus on the creation of 1030 value in the fishery through three mechanisms: the propensity of harvesters to target sizes of fish that 1031 increase product value, changes in recovery rates, and the development of markets for byproducts, which

13 Catch shares/individual fishing quotas can be institutionalized in a variety of ways and the regulations involved in different programs vary widely. In the BSAI Pacific Cod longline fishery, an act of congress allowed the formation of a fishing cooperative between participants in the fishery. A coalition of fishery participants had lobbied for the act with express intention of allocating quota to individual vessels based on historical catch. The sector is not designated as a “catch share program” by the National Marine Fisheries Service (NMFS), nor does it receive any exclusive privilege from NMFS. The cooperative is entirely voluntary, based on private contractual arrangements, and self-regulated.

Page 78 of 127 Please do not cite this chapter without permission from the authors.

1032 increased the average value of fish harvested. Pacific Cod is the second most valuable species in the 1033 commercial groundfish catch off Alaska, behind Alaska pollock (Fissel et al. 2012). 1034 While only a short period of post-cooperative data is available (2.5 years), we find some evidence 1035 of a change in the profit maximization strategy of Pacific Cod harvesters as a result of the formation of 1036 the cooperative. Prior to the cooperative, production choices did not show a reaction to changing product 1037 price ratios even though harvesters had the ability to target products because of consistent spatial patterns 1038 in fish size throughout the fishing grounds. Rather, production was better explained by the size profile of 1039 the Pacific Cod population. The size of annual age classes varies considerably, so while larger fish can 1040 consistently be found in certain areas of the fishery, the overall average size of available fish varies by 1041 year as large year classes have aged. A shift toward the creation of resource rents through value- 1042 maximization is expected post-cooperative formation. We find that byproduct production and revenue 1043 from byproducts have increased post-cooperative. These results support the hypothesis that harvesters are 1044 more likely to focus on value-per-ton of catch when they have an individual allocation, thus increasing 1045 resource rents through targeting quality, changing the product mix to optimize differences in product 1046 value, and developing markets for byproducts.

1047 2. Background and data

1048 Since 2002, a total of 39 catcher/processor vessels have fished in the BSAI Pacific Cod longline 1049 fishery, several of which have left the fishery since 2002. The fishery targets Pacific Cod with hook-and- 1050 line gear in the Bering Sea, Aleutian Islands, and Gulf of Alaska. Gulf of Alaska fishing is excluded from 1051 this analysis because a separate permit endorsement is needed to participate, fishing there occurs after the 1052 Bering Sea season is closed, and Pacific Cod from the Gulf of Alaska are prone to protistan parasites, 1053 making them less marketable (Westrheim 1996). 1054 Two main products are produced in the factories on-board the vessels: fish are “headed and 1055 gutted” and frozen into either Eastern cuts or Western cuts. Western cuts are made from larger fish, and 1056 the pectoral collar or girdle is left intact. Western cuts are destined for the salt cod markets in Southern 1057 Europe and Brazil. Smaller fish are most commonly processed into Eastern cuts in which the pectoral 1058 collar is removed. Eastern cut Pacific Cod is exported to and Japan, mainly for reprocessing and re- 1059 export to Europe and the United States (Fissel et al. 2012). Byproducts such as roe, heads, and stomachs 1060 represent a small portion of production (3%) and revenue (5%).14 1061 The time period considered for this study is from 2002 through 2012. 2002 is the first year in 1062 which vessel-specific prices are available. In the summer (“B-season”) of 2010, following an act of 1063 Congress the previous year, a voluntary cooperative within the longline fishery was formed, although the

14 Average of 2002-2011.

Page 79 of 127 Please do not cite this chapter without permission from the authors.

1064 fleet had displayed some cooperation prior to the formation of the cooperative. The formation of the 1065 cooperative significantly changed the incentives facing the fishery from a competition for the seasonal 1066 fleet-wide TAC or “race to fish” to a fishery where vessels may fish their individual allocation of TAC at 1067 their desired pace. Five vessels were removed from the fleet (3 by a federal buy-back program, and 2 in a 1068 buy-out by the cooperative) prior to the formation of the cooperative in 2010. 1069 Figure 1 displays average revenue and the number of active vessels for two distinct sets of 1070 vessels: vessels that remained in the fishery after the formation of the cooperative and those that were 1071 bought out (excluding 2007-2009 during which fewer than three non-co-op vessels were fishing).. 1072 Average revenue per vessel was lower for the vessels that were bought out prior to cooperative formation 1073 (“non-co-op”) than for those that remained, partially contributing to a substantial increase in average 1074 revenue per vessel in 2011-2012. However, figure 2 shows that this increase was primarily due to an 1075 increase in TAC in 2011-2012. Average revenue-per-ton of catch increased by a smaller amount, and 1076 average revenue-per-ton was approximately equal for vessels that were bought out and for those that 1077 remained. This implies that the buy-out itself did not cause substantial increases in average revenue-per- 1078 ton for the fleet. Average revenue-per-ton is highly correlated with the average prices of the two principal 1079 products produced by the fleet: Eastern and Western cut headed and gutted fish (figure 3).

1080 1081 Figure 1. Average vessel revenue and the number of active vessels in the BSAI Pacific Cod longline 1082 fishery. Average revenue for non-co-op vessels (vessels that were bought out) is suppressed for

Page 80 of 127 Please do not cite this chapter without permission from the authors.

1083 confidentiality 2007-2009. 1084

2500 140

120 2000

100

1500 80

1000 60 40 TAC (thousands t) (thousands TAC 500 20

0 0 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Average revenue per t of TAC caught ($) Average revenue per ton non-co-op Average revenue per ton co-op Total allowable catch 1085 1086 Figure 2. Average vessel revenue per ton of catch, and total allowable catch for the BSAI Pacific Cod 1087 longline fishery. Average revenue for non-co-op vessels is suppressed for confidentiality 2007-2009. 1088 1089 2 1.8 1.6 1.4 1.2 Average ($/lb) prices product 1

2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 year

Western cut Eastern cut 1090 1091 Figure 3. Annual reported fleet-wide average prices for each product. 1092

Page 81 of 127 Please do not cite this chapter without permission from the authors.

1093 Price data were obtained from the Alaska Department of Fish and Game’s Commercial 1094 Operator’s Annual Production Report (COAR). Prices are annual, vessel-specific averages of prices 1095 received for each product type sold (catch-weighted average prices are shown in figure 3).15 1096 Production data were obtained from the National Marine Fisheries Service (NMFS) Alaska 1097 Region’s production database. Revenue was calculated by applying vessel-level annual average prices to 1098 the production data which have been reported on a daily basis since 2009 and on a weekly basis in earlier 1099 years. Production weights by product are aggregated to obtain annual production quantities for each 1100 vessel in the fleet. These are quantities of product sold, however, not round weight (i.e., fish catch) 1101 quantities, so the amount of catch required to product the amount of finished product is calculated using 1102 average recovery rates. The official NMFS average recovery rate of Western cuts is 57%, which is higher 1103 than that of Eastern cuts (47%, because the pectoral collar is removed) (Bearden). One kilogram of 1104 Western cut product would have come from 1*(1/0.57)=1.75 kilograms of catch. One kilogram of Eastern 1105 cut product would have come from 1*(1/0.47)=2.13 kilograms of catch. The share of each product type is 1106 calculated as the ratio of each product to total production, where both quantities have been converted to 1107 round weight using the recovery rates. 16 1108 1109 The share of catch processed into Western cuts has varied over the study period (Figure 4). 1110

15 Catcher processors have been required of to report COAR data only since 2002, limiting earlier analysis of Pacific Cod catcher processor product prices. 16 In other words, the share of catch processed into Western cuts is calculated as:

. . . = 1 . . . 푃푟표푑푢푐푡 푤푒푖푔 ℎ푡 .푊 푐푢푡∗�푅푒푐표푣푒푟푦 푟푎푡푒 푊 푐푢푡� . 1 1 푆ℎ푎푟푒푊 푐푢푡 푃푟표푑푢푐푡 푤푒푖푔ℎ푡 퐸 푐푢푡∗�푅푒푐표푣푒푟푦 푟푎푡푒 퐸 푐푢푡�+푃푟표푑푢푐푡 푤푒푖푔ℎ푡 푊 푐푢푡∗�푅푒푐표푣푒푟푦 푟푎푡푒 푊 푐푢푡�

Page 82 of 127 Please do not cite this chapter without permission from the authors. 150000 100000 50,000 Weighted production (MT) of BSAI longline fleet 0 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Western cut Eastern cut

1111 1112 Figure 4. Annual production of Western and Eastern cut products weighted by the inverse of the recovery 1113 rate of each product to obtain the round weight of catch processed into each product for the BSAI Pacific 1114 Cod longline fleet. Variation in total production is driven by annual variation in TAC (annual TAC shown 1115 in figure 2). 1116 1117 The mean length of fish in the Pacific Cod population is estimated by calculating the average 1118 length of 350 mm and larger fish from all sampled fish in the annual bottom trawl survey (Table 2.8 in 1119 Thompson et al. 2010). 350 mm is the approximate size of an age 3 fish, an estimate of the minimum age 1120 of fish generally caught by the fishery.17 The survey area approximately corresponds with the actively 1121 exploited fishing grounds. Catch per unit of effort (CPUE) and average size of fish in the population 1122 varies as large year classes age through the population structure. Figure 5 shows how mean CPUE 1123 (kg/hectare BSAI bottom trawl survey, a standard measure of CPUE for biological trawl surveys) and 1124 mean length of age 3+ Pacific Cod have varied over the study period. 1125

17 Personal communication with Bob Lauth, National Marine Fisheries Service–Alaska Fisheries Science Center, 10/20/2011

Page 83 of 127 Please do not cite this chapter without permission from the authors.

16 560

540 12

520

8 500 Length (mm) Length CPUE (kg/ha)CPUE 480 4 460

0 440 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

Mean CPUE (kg/ha) Mean length of age 3+ (mm)

1126 1127 Figure 5. Mean CPUE (kg/ha) and mean length (mm) of 350 mm and larger Pacific Cod sampled in the 1128 BSAI bottom trawl survey. 1129 1130 There is substantial spatial variation in the sizes of fish surveyed in the Eastern Bering Sea in a 1131 given year (Figure 6). Except for the Aleutian Islands, the survey covers the area that is targeted by the 1132 fishery. The spatial extent of various sizes of age 3+ Pacific Cod is shown for two example years, 2005 1133 and 2009. Larger fish are found in the north and Western portions of the fishing grounds and fish of 1134 smaller than average size can consistently be found in the southern portion of the Eastern Bering Sea and 1135 in the shallower waters of the Bering Sea shelf during the time of the survey.18 These spatial differences 1136 in size are relatively consistent over time, even as the overall mean size varies due to the influence of 1137 large year classes (Figure 5). The spatial persistence of different sizes of fish indicates that it is possible 1138 for harvesters to effectively target fish size by choosing a particular area to fish, if it is profit maximizing

1139 to do so.

18 The location of Pacific Cod is also influenced by the size of the “cold pool”, a characteristic of the Eastern Bering Sea that is caused by sea ice coverage. In warmers years less ice cover contributes to faster warming and a smaller area that persists at <2ºC, temperatures that Pacific Cod avoid (Kotwicki and Lauth, 2013).

Page 84 of 127 Please do not cite this chapter without permission from the authors.

1140 1141 Figure 6. Spatial variation in average sizes of age 3+ Pacific Cod collected by the by Bering Sea bottom 1142 trawl survey in 2005 and 2009. Figure depicts a kernel density map of surveyed points using a search 1143 radius of 60 km. Overall mean length of surveyed age 3+ Pacific Cod was 494 mm in 2005 and 547 mm 1144 in 2009. 1145

1146 3. Methods 1147 3.1 Production choices as a function of product prices and fish size

1148 Because the principal products produced from Pacific Cod are dependent on the size of fish 1149 caught (small fish are generally processed into Eastern cuts, while larger fish are generally processed into 1150 Western cuts), we investigate whether the production mix of Eastern and Western cuts is determined by 1151 product prices. This would imply that harvesters target the size of fish that can be processed into the 1152 highest-value product depending on relative product prices (for example, targeting large fish to be 1153 processed into Western cuts when the price of Western cuts is relatively high). Alternatively, harvesters 1154 may target areas of high CPUE and/or low fishing cost (i.e., shorter distances from port), with less regard 1155 for the size of fish caught, and then process their catch into the most appropriate product for the size of 1156 fish caught. In this case, observed production is more likely to be a function of the size profile of Pacific 1157 Cod in the targeted population; as large year classes become older and transition through the population, 1158 the average size of fish in the ecosystem changes. For example, in years where the average size is smaller, 1159 a larger proportion of catch would be processed into Eastern cuts.

Page 85 of 127 Please do not cite this chapter without permission from the authors.

1160 In the years prior to the formation of the cooperative, we expect that size/product targeting is less 1161 likely to occur. Before the cooperative, each harvester was subject to a catch constraint that was affected 1162 by others’ catch, which resulted in a fishing strategy of maximizing the number of fish caught per day in a 1163 so-called “derby fishery” (Dupont 1990; Gordon 1954; Levhari and Mirman 1980). If size targeting 1164 involves any trade-off with CPUE and/or costs, a harvester would only be willing to target size if the 1165 product profit differential were large enough to make up for the loss of catch to other harvesters. Size 1166 targeting more likely in the post-cooperative fishery where harvesters are subject to individual quantity 1167 constraints, resulting in a fishing strategy that involves maximizing profit per pound of fish caught. 1168 However, the actual effect in both cases will depend on the price differential between products and the 1169 trade-off between expected CPUE and the profit benefit from size targeting. For example, if Western cuts 1170 are more valuable but larger fish can only be found in low CPUE conditions, it may be profit-maximizing 1171 to target smaller fish, even in a rationalized fishery. 1172 We investigate whether size targeting as a function of product prices occurred in the fishery by 1173 estimating a revenue function for the periods before and after the formation of the cooperative. While the 1174 use of a profit function would give a more complete treatment to the trade-offs between CPUE, value, and 1175 the cost of targeting, cost data does not exist. Revenue functions have been estimated in liu of cost data in 1176 many other fisheries, justified by assuming that most inputs are fixed at the level of the trip (Asche and 1177 Hannesson 2002; Jensen 2002; Thunberg et al. 1995; Torres and Felthoven 2012). The revenue function 1178 takes the form R(P, Z), where P is a vector of M output prices and Z is a vector of L input levels, and R 1179 satisfies regularity conditions (R linearly homogenous in P and non-decreasing in P and Z). By

1180 Hotelling’s Lemma, = is the supply function for product Ym. The two products are denoted YW 휕푅 푚 푚 1181 (Western cut) and YE푌 (Eastern휕푃 cut) and Pm is the price received for each product. The response of 1182 harvesters to variation in product prices (in terms of how much of each product to produce) is therefore

1183 the own-price supply response (in elasticity form, ). We specify the inputs to the production 휕푌푚 푃푚 1184 process (Z) as total catch (in metric tons, denoted C휕푃)푚 and푌푚 the mean length of age 3+ Pacific Cod in the 1185 ecosystem (denoted S). Size can also be thought of as a quality-shifter rather than an input, but its 1186 specification in the revenue function would be equivalent. The response of harvesters to variation in fish

1187 size is the elasticity of the demand for each product with respect to average size, . If harvesters 휕푌푚 푍푚 1188 target the size of fish that can be processed into the highest-value product depending휕푍푚 on푌푚 relative product 1189 prices, the own-price supply elasticity will be significant, and the ratio of production will track the ratio of 1190 product prices. If, on the other hand, harvesters target areas of high CPUE or low cost, the size of fish 1191 caught is more likely to track the average size of fish in the population. Thus, the elasticity of Western 1192 cut production with respect to increases in average fish size will be positive, while the elasticity of

Page 86 of 127 Please do not cite this chapter without permission from the authors.

1193 Eastern cut production with respect to fish size will be negative and the own-price supply elasticity is 1194 likely to be small or insignificant. 1195 We assume a fully flexible generalized Leontief function for estimation of the revenue function 1196 and associated supply functions: 1197 ( , ) = . . + + , = ( , ) (2) 0 5 0 5 ( , ) 1198 푅푖푡 푃=푖푡 푍푖푡 ∑= ∑푛푚 훼푚푛푃푚푖푡푃. 푛푖푡/ . ∑+ ∑푙푚 훿푚푙푃푚푖푡+푍푙푖푡 ∑푚 푃푚푖푡�∑ ∑ 푞푙 훿 푙푞푍, 푙푖푡=푍푞푖푡 ( � 푚, )푛. 푊 퐸 (3) 휕푅푖푡 푃푖푡 푍푖푡 0 5 0 5 푚푖푡 푛 푚푛 푙 푚푙 푙푖푡 푞푙 푙푞 푙푖푡 푞푖푡 1199 The푌 subscript휕푃푚푖푡 i indicates∑ 훼 the푃푛푖푡 vessel,푃푚푖푡 t the∑ 훿 year,푍 and∑ l and ∑ 훿 q 푍the 푍inputs. The푚 푛 system푊 퐸 of equations is estimated 1200 using Seemingly Unrelated Regression (SUR) (Zellner 1962). Due to the coarseness of the available price 1201 data, the analysis is at the seasonal level.19 Ideally, more frequent price data (e.g., per-trip or per-week) 1202 would allow better exploitation of the weekly production data, but such data do not exist.

1203 3.2 Byproducts in production

1204 Another way to increase revenue per pound of fish caught would be to retain more byproducts 1205 from the production of the standard Eastern and Western cut products. Secure allocations allow harvesters 1206 to fish their portion of the quota at any point during the regulated season. This often results in a 1207 lengthening of the season, which allows harvesters to fish more slowly, and perhaps catch fewer fish per 1208 trip but use more space in their holds for byproducts that would have been discarded during the race for 1209 fish. Harvesters may work to develop new or more robust markets for byproducts, and thus increase 1210 revenue from byproducts given the proportion of byproducts in total production. More ad-hoc methods 1211 are used to investigate changes in the production of byproducts (as opposed to the revenue function). 1212 While byproducts could be included in the revenue function as additional outputs, the variety of 1213 byproducts produced in each year varies significantly, making their inclusion difficult because prices for 1214 some products do not exist in some years. Thus, we estimate changes in proportion of revenue from 1215 byproducts with a system of equations that allows us to differentiate increases in revenue due to volume, 1216 and increases in revenue due to prices: 1217 = + + + + .

1218 푅푒푣푒푛푢푒푆 ℎ푎푟푒퐵푦푝푟표푑푢푐푡푠 푖푡 훽 푖 훽1푆ℎ 푎푟푒퐵푦푝푟표푑 푢푐푡 (4)푠 푖푡 훽2푃표푠푡푟푎푡푖표푛푎푙푖푧푎푡푖표푛 훽3푡 휔푖푡 1219 where 1220 = + + + + +

1221 푆ℎ푎푟푒퐵푦푝푟표푑푢푐푡+ 푠 푖푡 훾 푖 훾 1푇표푡푎푙퐶푎푡푐 ℎ 푖푡 훾2 푀푒푎푛퐿푒푛푔푡 ℎ푡 훾 3 푃푟푖푐 푒 푖푡 훾 (5)4푃표 푠푡푟푎푡푖표푛푎푙푖푧푎푡푖표푛 1222 훾The7푡 two휀푖푡 equations are estimated as a system using two stage least squares. The share of byproducts 1223 (ShareByproducts) is calculated as byproduct production (including whole bait, roe, roe for bait, heads,

19 An annual model would treat the A and B seasons of 2010, the year of cooperative formation, as separate years. However, model results do not drastically change if an annual model is estimated, or if 2010 is dropped.

Page 87 of 127 Please do not cite this chapter without permission from the authors.

1224 cheeks, chins, belly flaps, fish meal, milt, stomachs, and other products) divided by total production. 1225 Seasonal TotalCatch is included in case vessels may be lower on space for ancillary products in high 1226 catch (or TAC) years. MeanLength is the mean length of Pacific Cod in the ecosystem, defined as in the 1227 previous section. Price is the vessel-specific annual price ratio of Western cuts to Eastern cuts. The 1228 product price ratio is used because the production of Eastern cuts may leave more of the fish available for 1229 processing into byproducts, and the product price ratio may affect the production ratio. A time trend is 1230 included (t) to capture trends in the dependent variables resulting from the escalating pre-cooperative race 1231 for fish. If the time period after cooperative formation were longer, an additional time trend could be 1232 included to capture evolution in markets, adjustments, and learning that is likely to occur over time. 1233 Instead, the coefficient on Post-co-op measures the change in the share of byproducts in total production 1234 due to having a secure TAC allocation. The share of revenue from byproducts (RevenueShareByproducts)

1235 is the revenue from byproducts divided by total revenue. β4 estimates the effect of catch shares on the 1236 average value of a ton of byproduct. Equations 4-5 with the dependent and independent variables in 1237 natural logs, resulting in the estimated coefficients interpretable as elasticities (dependent upon non-zero 1238 byproduct production and revenue).

1239 3.3 Types of byproducts produced

1240 Production of each byproduct (roe, roe for bait, heads, cheeks, chins, milt, stomachs, and other 1241 products) is compared over time to look for the expansion of markets and the emergence of new 1242 byproducts and byproduct markets. 1243

1244 4 Results 1245 4.1 Production as a function of product prices

1246 The parameter estimates given by equations 2 and 3 do not provide intuitively interpretable 1247 information, so are combined into elasticity estimates and shown in table 1 for the pre-cooperative and 1248 post-cooperative time periods. The elasticities are calculated at the mean values of the data, and standard 1249 errors are computed by the delta method. 1250 The first-order elasticities in Table 1 show that as expected, increases in total catch and prices 1251 increased a vessel’s revenue. Increases in the mean length of harvestable Pacific Cod in the ecosystem, all

1252 else equal, did not result in an increase in revenue ( , ). This is credible because although larger sizes of

1253 fish are more often processed into Western cuts Eastern휀푅 푆 cuts are often relatively more valuable (Figure 3). 1254 The second-order price elasticities (or the price elasticities of supply) are each significantly different from 1255 zero in the pre-coop period. In other words, the price of each product affected the quantity produced of

Page 88 of 127 Please do not cite this chapter without permission from the authors.

1256 each product, directly and through substitution (the cross-price elasticity). Additionally, the supply 1257 elasticity of Western cuts with respect to the average size of Pacific Cod in the ecosystem is positive and

1258 large in magnitude ( , ); a 1 % increase in the mean size resulted in a 8 % increase in the production of

1259 Western cut Pacific 휀Cod.푌푊 푆 Correspondingly, the supply elasticity of Eastern cuts ( , ) is negative. 1260 Together, this is evidence that in the period prior to catch shares, the size of fish 휀in푌퐸 the푆 ecosystem 1261 mattered more for production choices than did prices. Harvesters processed what they caught (as a result 1262 of targeting high CPUE or location choices for other, non-product-price reasons) into the product most 1263 appropriate for the size of fish they caught, which were larger on average when there were more large fish 1264 present in the population. This resulted in an increase in Western cut production when the average size of 1265 fish was larger. 1266 While there is much less data for the period after the formation of the cooperative, there is some 1267 evidence for a change in production strategy. Both the supply elasticity of Western cuts with respect to 1268 the average size of Pacific Cod in the ecosystem and the price elasticities of supply become insignificant, 1269 This may be due to the very short time period, but may be evidence of an increase in size-targeting that is 1270 beginning to occur in the catch shares fishery. 1271 Finally, increases in total catch increased the supply of each product by approximately equal

1272 amounts ( , and , ). We have no a priori expectation that variation in total catch, all else

1273 equal, would휀푌푊 퐶affect 휀production푌퐸 퐶 choices, so the result is incidental. 1274 1275 Table 1. Elasticities of revenue and product supply. Elasticity estimates Pre-cooperative Post-cooperative 0.23** 0.25** , (0.01) (0.00) 푅 퐶 휀 0.23 0.43 , (0.24) (0.57) 푅 푆 휀 0.29** 0.18** , (0.01) (0.02) 푅 푃푊 휀 0.75** 0.84** , (0.01) (0.02) 푅 푃퐸 휀 -0.29** 0.19 , = , = , 2 (0.05) (0.14) 휀푅 푃퐸∗푃푊 휀푌퐸 푃푊 휀푌푊 푃퐸

Page 89 of 127 Please do not cite this chapter without permission from the authors.

0.30** -0.17 , = , 2 2 (0.05) (0.12) 푅 푃퐸 푌퐸 푃퐸 휀 휀 0.74** -1.03 , = , 2 2 (0.13) (0.74) 푅 푃푊 푌푊 푃푊 휀 휀 0.47** 1.01** , = , 2 (0.10) 0.18) 푅 푃푊∗퐶 푌푊 퐶 휀 휀 8.02** -5.12 , = , 2 (0.83) (3.34) 푅 푃푊∗푆 푌푊 푆 휀 휀 0.67** 1.04** , = , 2 (0.04) (0.03) 푅 푃퐸∗퐶 푌퐸 퐶 휀 휀 -2.72** 1.67** , = , 2 (0.36) (0.63) 휀푅 푃퐸∗푆 휀푌퐸 푆 1276 Notes: Significance codes: * p<0.05, ** p<0.01.

1277 4.2 Byproducts in production

1278 Table 2 shows how the revenue from byproduct production has changed due to the formation of 1279 the cooperative in the longline fishery. More flexible (and often longer) seasons allow harvesters to 1280 potentially increase profits by retaining and marketing a greater proportion of each fish caught. Revenue 1281 from byproducts can increase due to increases in volume, and/or increases in value (prices of byproducts, 1282 although because such a variety of byproducts are produced which vary by year, here we define “price” as 1283 total revenue from byproducts divided by total weight of byproducts). 1284 The results of the first stage of the system of equations (equation 4) show the change in the share 1285 of byproducts in total production. The results shows that there was a downward trend in the production of 1286 byproducts of about 0.1 percent per year in the years prior to the cooperative, which is consistent with the 1287 escalating race to fish that had been taking place in the fishery (table 2). After the cooperative, the share 1288 of byproducts in total production increased by 0.76 percent. This coefficient should be interpreted as a 1289 comparison to the first year of the pre-coop period (because of the inclusion of the time trend). Thus, from 1290 a mean of about 5 percent in the 2002, the share of byproducts in production decreased by a total of about 1291 0.9 percent over the 9-year pre-coop period, and then increased by a total of 1.66 percent (0.9+0.76) from 1292 the end of the pre-coop period to the post-coop period. Total catch, the price ratio, and mean length are 1293 not statistically significant. 1294 The results of the second stage (equation 5), show that the share of revenue from byproducts, 1295 controlling for the change in the production of byproducts, also increased. Again, there was a downward

Page 90 of 127 Please do not cite this chapter without permission from the authors.

1296 trend in the pre-cooperative revenue share from byproducts, meaning that the total increase from the end 1297 of the pre-coop period to the post-coop period was about 2.9 percent (0.19*9+1.2). This indicates that not 1298 only were harvesters retaining more byproducts from production, but they were also more valuable than 1299 they were prior to the formation of the cooperative. 1300 1301 Table 2. The effects of cooperative the share of byproducts in total production and the share of revenue 1302 from byproducts in total revenue. First stage Second stage Share of Share of revenue byproducts in from byproducts total production in total revenue 1.187* Post-cooperative 0.760** (0.690) (0.135) Time trend for pre- -0.194** cooperative period -0.099** (0.095) (0.019) Total catch (mt) -0.008 (0.086) Price ratio -0.355 (0.305) Mean length (mm) 0.492 (1.009) Share of byproducts in -0.383

total production (0.914)

-4.328 Constant -6.301 (2.998) (6.329) N 671 671 1303 Notes: Significance codes: * p<0.05, ** p<0.01.

Page 91 of 127 Please do not cite this chapter without permission from the authors.

1304

1305 4.3 Types of byproducts produced

1306 Figure 7 shows that the mix, as well as the quantity per metric ton of TAC, of specific byproducts 1307 has changed significantly in the short period of time after the formation of the fishing cooperative. Roe 1308 (for food) and stomachs were the most important byproducts prior to 2010, although roe had been 1309 decreasing in importance throughout the period. After the formation of the cooperative in the B season of 1310 2010, harvesters marketed the largest amount of byproduct (per ton of TAC) in recent history, and nearly 1311 70 percent of it was roe for bait, a product that had only been produced in extremely small quantities prior 1312 to the cooperative formation. Roe for bait continued to be an important product in 2011-2012, even as roe 1313 for food continued to decrease.

1314 1315 Figure 7. Production of byproducts per ton of total allowable catch, 2003-2012 1316

1317 5 Discussion

1318 The adoption of catch shares for fisheries management is often associated with vessel buy-back 1319 programs or other means by which overcapitalization in fisheries is reduced. Often overlooked, however, 1320 are the behavioral changes in fishing and processing that create resource rents by increasing the value of 1321 the catch. This analysis provides insight into some of the mechanisms that can cause revenue, primary and 1322 by-product production choices, average prices, and profits to change after catch shares are implemented in

Page 92 of 127 Please do not cite this chapter without permission from the authors.

1323 a fishery. 1324 Prior to the formation of the cooperative, the fleet was operating in a limited-entry, TAC- 1325 regulated fishery which resulted in a “race to fish” for the available TAC. Season length decreased 1326 substantially throughout the 2000s—evidence that harvesters may have been more concerned with 1327 catching the maximum quantity of fish than with optimizing their production choices. 1328 A limited amount of data is available for the period after cooperative formation, but it depicts a 1329 behavioral change toward rent creation in the fishery. We find that in the years prior to the formation of 1330 the cooperative, the size profile of Pacific Cod in the ecosystem affected production more than product 1331 prices, suggesting that the observable product price differential was insufficient to cause extensive size 1332 targeting in the derby fishery, and/or it was more profitable for harvesters to target areas of high CPUE 1333 than areas of the most valuable size of fish.20 Catch shares have allowed vessels to fish their share of 1334 quota at the profit-maximizing speed, location, and time of the year. We expect catch shares to increase 1335 the extent to which product prices affect production choices, but have not yet accumulated enough data to 1336 quantify these changes. In addition, the degree to which harvesters adjust their behavior will depend on 1337 the size of the product price differential and the CPUE and cost trade-off from targeting fish size. 1338 However, harvesters have been creating resource rents in the post-cooperative period in other 1339 ways that are measureable from the available data. The elimination of the race to fish has resulted in a 1340 lengthening of the season (Brinson and Thunberg 2013). Slower fishing allows harvesters to focus on 1341 maximizing quality and recovery rates. With the available data, we can’t measure changes in quality, so 1342 we focus on byproduct production. Slower fishing allows harvesters to devote more space in their holds 1343 for byproducts that prior to the formation of the cooperative, may have been discarded in favor of filling 1344 the hold with their primary product, headed and gutted frozen fish. The production of byproducts 1345 increased. In addition, Table 2 shows that those by-products have become more valuable. Anecdotal 1346 evidence suggests that some harvesters and their managing companies are becoming more adept at 1347 marketing their primary products and by-products directly to buyers (such as restaurants), and buyers are 1348 rewarding the quality improvements and year-round availability of product that are possible in a catch- 1349 share managed fishery. 1350 Finally, figure 7 shows that harvesters began producing an entirely new type of byproduct post- 1351 cooperative: roe for bait. Available data does not allow us to determine if this was driven by a shift in the 1352 market for roe, away from food and toward bait, or processor-driven, by the ability, storage space, and

20 Unfortunately, within-year variation in product prices is not reported by this fleet and no fleet-wide source exists. Harvesters may have varied their product choices as prices varied within the year, but the differences were not large enough to capture with annual average prices. Thus, we could be missing a substantial portion of price-sensitivity of harvesters both before and after catch shares. The availability of within-year prices would improve our ability to identify size-targeting behavior as a function of prices.

Page 93 of 127 Please do not cite this chapter without permission from the authors.

1353 time to process roe from fish caught in the B season that developed because of the change in fishing 1354 incentives brought about by the formation of the cooperative. Regardless, this contributed to both an 1355 increase in the share of byproducts produced and an increase in the value of byproducts. In the future, if 1356 the Pacific Cod fishery follows other rationalized fisheries in the BSAI, vessels are likely to re-tool their 1357 factories to support the production of new byproducts and products for niche markets. In fact, two 1358 companies have begun building entirely new vessels to replace ones currently fishing.21 These new 1359 vessels will have larger holds, more efficient engines, and are designed for full utilization of the target 1360 species. These results, as well as the anecdotal evidence of vessel re-tooling and new construction, 1361 contribute to an understanding of why markets for different products may develop. Catch share 1362 management decreases the opportunity cost of product development. Slowing the pace of fishing allows 1363 byproducts to be retained and processed, as well as decreasing the opportunity cost of keeping lower- 1364 value products in the hold. Vessels may experiment with different product forms or timing (such as the 1365 shift from a winter roe-for-food market to a summer roe-for-bait market). New products may also be 1366 developed from the marketing side. The BSAI longline sector had a marketing cooperative even before 1367 the formation of the fishing cooperative, and is active in seeking out contracts and sales. With catch 1368 shares, vessels are more likely to be willing to risk targeting certain types of fish or producing niche 1369 products at the behest of their controlling company because they do not risk losing fish to other 1370 harvesters. 1371 Pacific Cod is virtually all sold frozen, and byproducts are generally low-value. Headed and 1372 gutted product is primarily exported to be reprocessed and is a substitute, considered somewhat inferior, 1373 for Atlantic cod (Fissel et al. 2012). As such, the potential for sweeping changes in the product form, such 1374 as a switch from frozen to fresh fish as occurred with halibut or the production of more higher value roe 1375 in pollock, and value appear relatively limited. Rather, increases in value have occurred at the margin, in 1376 the form of increasing recovery rates and changes in byproduct retention and production. The processes of 1377 market development, experimentation, learning, and the development of profit-maximizing responses to 1378 market conditions under the incentive structure are likely to evolve over time, however. In the long term, 1379 there is potential for further rent creation if the market for Pacific Cod changes significantly; for example, 1380 if a demand for fresh product were to develop domestically. 1381 A final implication of this analysis is that the assumptions about fishery selectivity that biological 1382 models use, such as those used for stock assessments can have important implications for robustness of 1383 those models. If harvesters are both able to and have the incentive to effectively target fish size, then

21 http://freezerlonglinecoalition.com/news.html, http://www.fishermensnews.com/story/2013/03/01/features/two- new-vessels-coming-online-to-groundfish-fisheries-in-may

Page 94 of 127 Please do not cite this chapter without permission from the authors.

1384 differences in product prices may affect the sizes of fish captured by the fishery and removed from the 1385 ecosystem. This can have implications for management. For example, the targeted removal of specific 1386 sizes of cod (which roughly correspond to age classes) can have implications for the biology of the 1387 species in future years. Currently, stock assessment models for the North Pacific Pacific Cod fishery 1388 define model selectivity coefficients for the fishery based on the average fishery selectivity, by gear and 1389 season, from blocks of past years (usually about 5 years) (Thompson et al. 2010). The stock assessment 1390 model is then used to predict the size of the stock and set total allowable catch (TAC) for the following 1391 year. The results of this analysis show that these assumptions were unlikely to have caused substantial 1392 bias due to price-related size targeting in the past, because the fleet was involved in a race for fish that 1393 resulted in the targeting of catch rates rather than fish size (that corresponded to product value). However, 1394 the use of historic size selectivity coefficients is unlikely to optimal for the catch shares fishery. Catch 1395 shares allow small differences in product prices to be exploited by targeting the optimal size of fish for 1396 the most valuable product. Changes in demand for products derived from Pacific Cod (from Europe or 1397 China, for example) could have large impacts on the sizes of fish removed from the Bering Sea and for 1398 the species' population dynamics. When profit can be affected by targeting certain sizes of fish that are 1399 not constant over time, stock assessment models that use constant or backward moving-average 1400 selectivity coefficients could be substantially biased. The incorporation of product prices and harvesters’ 1401 response to price variation in the estimation of the fishery selectivity is an important area for future 1402 research. 1403 1404

1405 Chapter References

1406 Arnason, R. 2005. Property Rights in Fisheries: Iceland’s Experience with ITQs. Rev Fish Biol Fisheries 1407 15(3): 243-264. 1408 1409 Asche, F., and Hannesson, R. 2002. Allocation of Fish Between Markets and Product Forms. Marine 1410 Resource Economics 17(3). 1411 1412 Bearden, P. National Marine Fisheries Service Alaska Regional Office Recordkeeping and Reporting. 1413 Available from https://alaskafisheries.noaa.gov/rr/tables/tabl3.pdf [accessed 3/28 2014]. 1414 1415 Brinson, A.A., and Thunberg, E.M. 2013. The Economic Performance of U.S. Catch Share Programs U.S. 1416 Department of Commerce.

Page 95 of 127 Please do not cite this chapter without permission from the authors.

1417 1418 Casey, K., Dewees, C., Turris, B., and Wilen, J. 1995. The effects of individual vessel quotas in the 1419 British Columbia halibut fishery. Marine Resource Economics 10: 211-230. 1420 1421 Dupont, D.P. 1990. Rent dissipation in restricted access fisheries. Journal of Environmental Economics 1422 and Management 19(1): 26-44. 1423 1424 Dupont, D.P., Fox, K.J., Gordon, D.V., and Grafton, R.Q. 2005. Profit and Price Effects of Multi-species 1425 Individual Transferable Quotas. Journal of Agricultural Economics 56(1): 31-57. 1426 1427 Dupont, D.P., Grafton, R.Q., Kirkley, J., and Squires, D. 2002. Capacity utilization measures and excess 1428 capacity in multi-product privatized fisheries. Resource and Energy Economics 24(3): 193-210. 1429 1430 Fissel, B., Dalton, M., Felthoven, R., Garber-Yonts, B., Haynie, A., Himes-Cornell, A., Kasperski, S., 1431 Lee, J., Lew, D., Pfeiffer, L., Sepez, J., and Seung, C. 2012. Stock assessment and fishery evaluation 1432 report for the groundfish fisheries of the Gulf of Alaska and Bering Sea/Aleutian Islands area: Economic 1433 status of the groundfish fisheries off Alaska, 2011. Economic and Social Sciences Research Program, 1434 Alaska Fisheries Science Center, National Marine Fisheries Service, National Oceanic and Athmospheric 1435 Administration. 1436 1437 Gordon, H.S. 1954. The Economic Theory of a Common-Property Resource: The Fishery. Journal of 1438 Political Economy 62(2): 124-142. 1439 1440 Grafton, R.Q. 1996. Individual transferable quotas: theory and practice. Rev Fish Biol Fisheries 6(1): 5- 1441 20. 1442 1443 Grafton, R.Q., Arnason, R., Bjørndal, T., Campbell, D., Campbell, H.F., Clark, C.W., Connor, R., 1444 Dupont, D.P., Hannesson, R., Hilborn, R., Kirkley, J.E., Kompas, T., Lane, D.E., Munro, G.R., Pascoe, 1445 S., Squires, D., Steinshamn, S.I., Turris, B.R., and Weninger, Q. 2006. Incentive-based approaches to 1446 sustainable fisheries. Can. J. Fish. Aquat. Sci. 63(3): 699-710. 1447 1448 Grafton, R.Q., Squires, D., and Fox, K.J. 2000. Private Property and Economic Efficiency: A Study of 1449 Common-Pool Resource. J. Law Econ. 43: 679-713. 1450

Page 96 of 127 Please do not cite this chapter without permission from the authors.

1451 Hannesson, R. 2013. Norway's experience with ITQs. Marine Policy 37(0): 264-269. 1452 1453 Herrmann, M. 2000. Individual Vessel Quota Price-induced Effects for Canadian Pacific Halibut: Before 1454 and After Alaska IFQs. Canadian Journal of Agricultural Economics/Revue canadienne d'agroeconomie 1455 48(2): 195-210. 1456 1457 Homans, F., and Wilen, J. 2005. Markets and rent dissipation in regulated open access fisheries. Journal 1458 of Environmental Economics and Management 49(2): 381-404. 1459 1460 Jensen, C.L. 2002. Applications of Dual Theory in Fisheries: A Survey. Marine Resource Economics 1461 17(4). 1462 1463 Kotwicki, S., and Lauth, R.R. 2013. Detecting temporal trends and environmentally-driven changes in the 1464 spatial distribution of bottom fishes and crabs on the eastern Bering Sea shelf. Deep-Sea Research Part II: 1465 Topical Studies in Oceanography 94: 231-243. 1466 1467 Levhari, D., and Mirman, L.J. 1980. The Great Fish War: An Example Using a Dynamic Cournot-Nash 1468 Solution. The Bell Journal of Economics 11(1): 322-334. 1469 1470 Morrison-Paul, C., Torres, M., and Felthoven, R. 2009. Fishing Revenue, Productivity and Product 1471 Choice in the Alaskan Pollock Fishery. Environmental and Resource Economics 44(4): 457-474. 1472 1473 NOAA. 2010. NOAA Catch Share Policy. Edited by N.O.a.A. Administration. p. 25. 1474 1475 Scheld, A.M., Anderson, C.M., and Uchida, H. 2012. The economic effects of catch share management: 1476 the Rhode Island fluke sector pilot program. Marine Resource Economics 27(3): 203-228. 1477 1478 Scott, A. 1955. The Fishery: The Objectives of Sole Ownership. Journal of Political Economy 63(2): 116- 1479 124. 1480 1481 Thompson, G.G., Ianelli, J.N., and Lauth, R.R. 2010. Assessment of the Pacific Cod Stock in the Eastern 1482 Bering Sea and Aleutian Islands Area. U.S. Department of Commerce National Oceanic and Atmospheric 1483 Administration National Marine Fisheries Service Alaska Fisheries Science Center, Seattle. 1484

Page 97 of 127 Please do not cite this chapter without permission from the authors.

1485 Thunberg, E.M., Bresnyan, E.W., and Adams, C.M. 1995. Economic Analysis of Technical 1486 Interdependencies and the Value of Effort in a Multi-Species Fishery. Marine Resource Economics 10(1). 1487 1488 Torres, M.d.O., and Felthoven, R.G. 2012. Productivity Growth and Product Choice in Fisheries: the Case 1489 of the Alaskan pollock Fishery Revisited. In Agricultural & Applied Economics Association Annual 1490 Meeting, Seattle, WA. 1491 1492 Westrheim, S. 1996. On the Pacific Cod (Gadus macrocephalus) in British Columbia Waters, and a 1493 Comparison with Pacific Cod Elsewhere, and Atlantic Cod (G. Morhua). Can. Tech. Rep. Fish. Aquat. 1494 Sci. 2902. 1495 1496 Wilen, J.E. 1979. Fisherman Behavior and the Design of Efficient Fisheries Regulation Programs. Journal 1497 of the Fisheries Research Board of Canada 36(7): 855-858. 1498 1499 Wilen, J.E., and Richardson, E. 2008. Rent generation in the Alaskan pollock conservation cooperative. 1500 Food and Agriculture Organization. 1501 1502 Zellner, A. 1962. An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests for 1503 Aggregation Bias. Journal of the American Statistical Association 57(298): 348-368.

1504

Page 98 of 127

1505 BSIERP B72 Project Conclusions 1506

1507 A number of general conclusions result from our examination of the impacts of changing climate on the 1508 Bering Sea pollock trawl and Pacific cod longline catcher processor fisheries. Many processes are 1509 involved and we believe that it has been important to evaluate these processes extensively outside of 1510 integrated model to understand their complexity.

1511 One general conclusion is that there are both direct and indirect means for how climate impacts fisheries, 1512 as we discuss in detail in Haynie and Pfeiffer (2012). In this project, we have focused on direct means 1513 through which fisheries are impacted by ice cover, water temperature, and the size and persistence of the 1514 cold pool. In other BSIERP work in which we participated, Ianelli et al (2011) explored how recruitment 1515 impacts from warmer conditions are likely to impact future pollock total allowable catch (TAC).

1516 This conceptual view led us to an empirical examination of how climate variation impacts the pollock and 1517 Pacific cod fisheries.

1518 A second general conclusion is captured in the title of the Project “Headline” piece that we produced: 1519 “Climate and Bering Sea Fisheries: Beyond a Northward March.” While the ecological literature of range 1520 shifts has indicated in many environments that both terrestrial and marine species are moving poleward 1521 (e.g., Box 1981; Guisan and Zimmermann 2000; Parmesan and Yohe 2003; Richardson and Schoeman 1522 2004, Atkinson et al. 2004, Perry et al. 2005)., we find that fisheries are impacted by more than these 1523 shifts alone. In the Bering Sea, although there has been evidence from summer biological surveys of 1524 spatial shifts (e.g., Klyashtorin 1998, Mueter and Litzow 2008), we have found that the cold pool has 1525 increased fish concentrations northward so that the relationship between ocean temperature and fishing 1526 effort has been more complicated than a “march to the north.” The following figure, for example, 1527 illustrates how the size of the cold pool is positively correlated with the ratio of pollock CPUE in the 1528 northern region of the fishing grounds to CPUE in the southern region. We found that this resulted in 1529 more fishing effort in the north in colder years, not warmer years as predicted by the “march to the north”. 1530

Page 99 of 127

1531

1532 Figure 1. Time series of the ratio of average CPUE in fishing zones 4 (the northern region) and 1 (the 1533 southern region) (black line), the size of the cold pool (percentage of the area surveyed by bottom trawl 1534 survey with temperature <1.5 wlrage CPUE in e), and the abundance of age 3+ Walleye pollock (grey 1535 line). Abundance is normalized to the year of minimum abundance (2008). 1536 1537 Not surprisingly, economic factors are an important component of what motivates where commercial 1538 fisheries fish, so that more fish alone does not determine where the fishery is located. For pollock, we 1539 found that there is no meaningful shift of A season fishing because of the pursuit of more valuable roe- 1540 bearing fish that are concentrated in the southern Bering Sea. Thus far even in warmer years with less ice 1541 cover, there has not been a winter shift to the north. This suggests that unless there is a future change in 1542 the spawning, vessels will be likely to remain in the southern Bering Sea in the A season.

1543 As discussed in this project has focused on the catcher processor fishery, building upon previous analysis 1544 of the pollock catcher vessel fishery (Haynie 2005, Haynie and Layton 2010). Going forward, we will 1545 model the mothership and catcher vessel sectors of the pollock fishery.

1546 For Pacific cod, we find that there are longer trips with more frequent moves in warmer years. This is a 1547 matter of concern going forward, when warmer years have the potential to increase fishing costs and 1548 reduce profits. However, institutional changes in the Pacific cod fishery mean that our historical analysis 1549 may not yield completely consistent predictions for the future. The formation of a fishing cooperative 1550 means that harvesters can partially adapt to climate changes in different ways. The costs involved with 1551 delaying the timing of fishing, spending more time searching for fish, or traveling greater distances are 1552 likely to be lower, for example.

1553

Page 100 of 127

1554 Being deeply involved in the Bering Sea Project was an extremely valuable experience for us. The 1555 project created a national and international inter-disciplinary network that has few rivals. We have 1556 formally and informally related to a large number of fantastic scientists and have an intimate sense of a 1557 much broader and more diverse research community than we did before the project. We have gained a 1558 deeper understanding of the research challenges and priorities of similar national research organizations 1559 around the world (e.g., CSIRO). A whole generation of researchers, including us, will do better work for 1560 decades as a result.

1561 PI Meetings and Other scientific interactions 1562 We found the BSIERP PI meetings an extremely valuable opportunity to see the evolution of other 1563 researchers’ work over time. The first PI meeting was very interesting but felt rather general; most of us 1564 did not have a complete grasp of the issues facing other researchers, most research projects were still 1565 getting started, and methodologies were adjusting to the real challenges that PIs faced. Still, this was a 1566 useful introduction to people and to the wide range of approaches and problems that were part of 1567 BSIERP.

1568 It was interesting to consider how we as economic modelers fit into different “camps” at this point. There 1569 were divisions between field researchers and modelers, there were physical and natural scientists, and 1570 there were social scientists exploring local and traditional knowledge (LTK) and we were economists 1571 modeling fleet behavior. We felt at different times like we had more in common with different groups. 1572 Having small break-outs in these meetings was useful.

1573 The second PI Meeting was an important part of the project evolution, but felt the least directly 1574 productive. We had seen the projects before and different projects had made different degrees of progress 1575 so the conversations felt uneven at times. However, what started to occur here was that researchers began 1576 to have real insight into each other’s projects, although it felt like this process really produced benefits in 1577 the third meeting.

1578 The third PI meeting was one of the best experiences that we have seen in interdisciplinary collaboration. 1579 Researchers really understood each other’s work and a large degree of complementarity occurred. 1580 Researchers would say things like “That’s really interesting – we found something similar in our project 1581 and we did this to address it.” For example, we found that pollock harvesters began fishing earlier warm 1582 years. We hypothesized that this could be due to increases in abundance, concentration, earlier roe 1583 maturity, or avoidance of some late-season aspect of the fishery. When BSIERP colleagues Smart and 1584 Duffy-Anderson presented their new results that pollock spawn earlier in warm years but that their spatial 1585 abundance remains relatively constant, we could narrow down our hypotheses. The fourth meeting was

Page 101 of 127

1586 also productive, although it felt to some degree like we were switching to a project reporting mode more 1587 than a collaborative mode. It was exciting to see results and discuss the implications of different projects 1588 for future research.

1589 One notion that was voiced from the start of the project in some of the PI meetings was that observing one 1590 or several warm/cold years would provide insight into how climate would impact the ecosystem. One 1591 would hear the idea that the field research would “validate” the models. Our feeling was that given the 1592 complexity of the system and the fact that multi-year processes were part of the system, this approach to 1593 the field data overstated its value, because it was different to tell whether a cold year was actually 1594 representative of cold years in general or rather one draw of a more complicated, multifaceted system. In 1595 the future it would be useful to be sure that the connection of field research to modeling efforts is clear.

1596 Past, current, and future collaborations 1597 The BSIERP project has led directly to several valuable collaborations.

1598 In 2010, Alan Haynie worked with Jim Ianelli, Anne Hollowed, Nick Bond, and Franz Mueter to evaluate 1599 how predicted changes in climate are likely to translate pollock biomass and total allowable catch (TAC) 1600 in the future, resulting in a publication in the ICES Journal (Ianelli et al 2011). While three of the co- 1601 authors are in the Resource Ecology and Fisheries Management (REFM) Division at AFSC, this project 1602 would not have occurred without our spending time together and with Nick Bond and Franz Mueter in 1603 various BSIERP and related scientific meetings, especially the Sendai Symposium on the Effects of 1604 Climate Change on Fish and Fisheries.

1605 Through participation in BSIERP and international meetings, Alan Haynie also became involved in a 1606 large modeling project with BSIERP PI’s Enrique Curchitser and Kate Hedstrom that lead to an 1607 additional manuscript (Rose et al. 2014).

1608 Franz Mueter and Alan Haynie are jointly supervising a student, Jordan Watson, at UAF in Juneau, who 1609 has contributed to the data analysis and vessel monitoring system (VMS) analysis in this project. Alan 1610 Haynie has become an affiliate faculty member at UAF in Juneau, which will allow for easier joint 1611 supervision of students in the future.

1612 We expect that a variety of other collaborations will continue to evolve from our project involvement in 1613 the future. Several specific benefits are flowing directly out of the BSIERP project.

1614 First, as NOAA expands its Integrated Ecosystem Assessment (IEA) Program, the involvement of Alan 1615 Haynie, Kerim Aydin, and other BSIERP PI’s will transfer the knowledge that we have gained in the

Page 102 of 127

1616 BSIERP Program to the large ongoing IEA effort. BSIERP has provided a unique perspective for us all 1617 on how to think about both integrated and complementary models.

1618 A large benefit of BSIERP to NOAA Fisheries was hiring Dr. Lisa Pfeiffer and having her serve as a 1619 post-doc in the project and beyond. Lisa Pfeiffer is now working at the NOAA Fisheries NWFSC. She 1620 has worked on a manuscript examining how weather impacts the risk that fishers are willing to take on the 1621 West Coast sablefish fishery. She is working on a project that will analyze how climate cycles can affect 1622 fishing practices for many target species within a multispecies complex, even if climate directly affects 1623 the abundance of only one species. The recent availability of cost data for some West Coast fisheries will 1624 allow her to evaluate how climate can affect the cost of fishing in the future, as more years of data are 1625 collected. This work draws directly on the knowledge that she gained in the BSIERP project. More 1626 broadly, Lisa’s involvement in BSIERP has prepared her well to make important contributions to research 1627 and management on the West Coast.

1628 Working with other modelers 1629 Integrating different models is very challenging. A key question when planning an integrated model is 1630 whether the priority is on integration or on each model being “correct” before the models are integrated. 1631 From our perspective, the integrated model focused on justifiable concerns of improving different model 1632 components to ensure that, for example, ROMS properly captured the heat content of the Bering Sea and 1633 the spring bloom was correctly reflected. This attention on the individual models came at the expense of 1634 applying more focus on the many challenges of integration. To give an example from our modeling 1635 work, while vessels can be observed to make fishing location choices based on the expected catch and 1636 revenue in different areas, the relationship between fish abundance and fishers’ knowledge of that 1637 abundance is imperfect. There may be times where lots of fish are concentrated in an area but this may be 1638 beyond the standard fishing grounds of the fishery. Thus the linkage between how fish populations may 1639 spatial shift due to climate change and whether or not the fishery pursues the fish there is a topic of 1640 important future research. Because there were not multiple outputs from FEAST across different years, 1641 we were not able to explore the degree to which FEAST output appears to be reasonable, given the 1642 actions of the fishery. For example, the fishery will not be concentrated in an area with no fish, nor will 1643 they repeatedly pass over consistently large aggregations of fish without fishing in an area (although they 1644 may do so at times due to fish size or value). Great attention needs to be given to how uncertainty should 1645 be considered in integrated models in the future. Additional work is needed to answer the question “How 1646 confident are we of the model output?”

Page 103 of 127

1647 During the first few years of BSIERP, we consistently participated in the modeling meetings and phone 1648 calls. Later in the project, we participated in breakout meetings at the PI Meetings but only occasionally 1649 participated in the calls as it did not feel like we were providing value beyond our interactions in meetings 1650 and workshops.

1651 Interaction/connection with the LTK (local and traditional knowledge) component of BSIERP 1652 A key element of the project is recognizing that people are a key component of the Bering Sea 1653 Ecosystem. Throughout the project, we participated in breakout groups and had many conversations with 1654 LTK researchers, especially with Henry Huntington, George Noongwook, and Phil Zavadil. We had 1655 hoped that we would identify a clear complementary project to our primary modeling efforts for the 1656 commercial Pacific cod and pollock fisheries, but the available data and knowledge is very different for 1657 subsistence and commercial harvesters. We found these conversations very interesting but we also found 1658 that this was a large, distinct modeling effort. It was very interesting to consider the different meaning of 1659 climate change for subsistence harvesters and commercial fishers. Our experience has primarily been in 1660 commercial fishing, where for Bering Sea fisheries we have a large amount of data over several decades. 1661 Commercial harvesters, especially large catcher processor vessels that take fishing trips that last several 1662 weeks, can travel large distances in pursuit of mobile fish populations. Place-based harvesters are less 1663 flexible, and therefore would be expected to make very different adjustments to climate change than 1664 commercial harvesters. Modeling and measuring these types of adjustments requires very different types 1665 of data than we had access to during the duration of the project. However, the research that we completed 1666 is relevant to Alaskan coastal communities in many aspects. Some communities derive significant income 1667 from leasing their Community Development Quota to catcher processors, particularly pollock. In 1668 addition, many Alaska residents are employed in the fishing and processing industries, and pollock and 1669 Pacific cod are two of the highest-volume fisheries.

1670 Next Steps 1671 The BSIERP project offered an important opportunity to expand the Alaska Fishery Science Center’s 1672 modeling of fleet behavior to consider how climate currently impacts key fleets. Going forward, Alan 1673 Haynie and collaborators are working to extend this work in many ways. NOAA Fisheries has a large 1674 project, the Spatial Economics Toolbox for Fisheries (FishSET), which will incorporate many of the 1675 lessons learned from the BSIERP modeling effort to apply to current and future modeling of Bering Sea 1676 pollock and cod and other fisheries. This platform will allow spatiotemporal estimates of fish 1677 abundances and climate information to be included in spatial economics models. Lisa Pfeiffer is involved 1678 with related modeling on the U.S. West Coast.

Page 104 of 127

1679 In related research, we have worked to develop algorithms to estimate fishing effort for the unobserved 1680 Bering Sea pollock catcher vessel fishery from 2003-2010 when many catcher vessels had onboard 1681 observers for 30 percent of their days at sea. As we integrate all available data, we will utilize the output 1682 from the entire fisheries in spatial analyses. For the Pacific cod longline fishery, to the degree that VMS 1683 data quality allows, we will also model the fishery with data for all fishing effort, observed and 1684 unobserved.

1685 One current FishSET application is an analysis of the impact of 2011 Steller Sea Lion (SSL) protective 1686 measures on the Pacific cod catcher vessel trawl fishery, a smaller component of the Pacific cod fishery. 1687 We will integrate climate data into this analysis to examine the manner in which the fishery is impacted 1688 by changing climate conditions.

Page 105 of 127

1689 Management and policy implications 1690 1691 Our work has a number of direct and indirect policy implications. First, the “march to the north” implied 1692 by the range shift literature will interact with the climate, economic, port, and institutional factors that 1693 impact fisheries, as we have demonstrated in the papers and presentations from BSIERP. 1694 1695 In the pollock fishery, vessels frequently target roe-bearing fish in the winter “A” season. We have not 1696 seen a significant change in spawning grounds, so we expect that vessels will continue to fish in the 1697 southern portion of the Eastern Bering Sea in the A Season. We have seen the B Season fishery shift 1698 north over time, but this shift is complicated by the timing of warm and cold years, as the fishery was 1699 actually more concentrated in the north in cold years than in the warm years in the middle of the decade. 1700 In the Pacific cod longline fishery, we have observed greater travel and fuel costs in warmer years, 1701 indicating that costs for the fishery may increase in a future, warmer climate. 1702 1703 The NOAA Catch Share Policy is leading to catch share implementation and analysis in a range of 1704 projects. Catch shares are not being developed within a static environment, but rather an environment in 1705 which both economics and the environment change rapidly. The 2010 adoption of catch shares in the 1706 Pacific cod fishery, for example, is expected to change the way that harvesters respond to climate change. 1707 Increased flexibility of fishing can lead to temporal and/or spatial differences in fishery exploitation, as 1708 we found for the pollock catcher-processer fishery in the B season. There is no reason to expect the 1709 effects to be completely parallel, however, as differences in output prices, costs, market structures, and 1710 fleet cooperation affect each species and fleet differently. We found that in the Pacific cod fishery, some 1711 of the most immediate changes after catch shares involved targeting production processes. Our research 1712 highlights how it is essential to consider the details of the physical environment and management 1713 structure that are unique to individual fisheries when modeling their response to exogenous climate (and 1714 other) changes. 1715 1716 In the Bering Sea pollock fishery, a major management issue is the incidental catch of Chinook and chum 1717 salmon. Alan Haynie, Jim Ianelli, and others have been involved in the analysis and development of new 1718 management measures. The modeling and data integration conducted as part of BSIERP is contributing 1719 to the on-going analysis of evaluating the avoidance of salmon in the pollock fishery. 1720 1721 As discussed above under ‘Next Steps,’ the spatial economics toolbox for fisheries (FishSET) provides a

Page 106 of 127

1722 new framework to assess the impacts of management actions and changing environmental conditions on 1723 fisheries. 1724 1725 Finally, members of the North Pacific Ground Fish Plan Teams and the Scientific and Statistical 1726 Committee (SSC) have collaborated broadly in BSIERP which has made us more aware of each other’s 1727 research and intellectual focus. This will continue to pay dividends in the fishery management process. 1728

1729 Publications 1730 1731 Haynie, A.C., Pfeiffer, L. (2012) Why economics matters for understanding the effects of climate change 1732 on fisheries. ICES Journal of Marine Science 69, 1160-1167. 1733 Haynie, A.C., Pfeiffer, L. (2013) Climatic and economic drivers of the bering sea walleye pollock 1734 (theragra chalcogramma) fishery: Implications for the future. Canadian Journal of Fisheries and 1735 Aquatic Sciences 70, 841-853. 1736 Ianelli, J.N., Hollowed, A.B., Haynie, A.C., Mueter, F.J., Bond, N.A. (2011) Evaluating management 1737 strategies for eastern bering sea walleye pollock (theragra chalcogramma) in a changing 1738 environment. ICES Journal of Marine Science: Journal du Conseil 68, 1297-1304. 1739 Pfeiffer, L., Haynie, A. (2012) The effect of decreasing seasonal sea-ice cover on the winter bering sea 1740 pollock fishery. ICES Journal of Marine Science 69, 1148-1159.

1741 Documents in Preparation:

1742 Pfeiffer, L. and A. Haynie. 2014. “The effects of catch share management on rent generation through 1743 targeting and production choices.” In Prep.

1744 Pfeiffer, L. and A. Haynie. 2014. “Climate fluctuations and fishing behavior in the Pacific cod (Gadus 1745 macrocephalus) longling fishery.” In Prep.

1746 Watson, J. and A. C. Haynie. 2014. “Utilizing VMS data to estimate unobserved Pacific cod fishing effort 1747 in the Bering Sea.” Working Paper. 1748

Page 107 of 127

1749 BSIERP oral and poster presentations at scientific conferences or seminars 1750 We presented our work at range of scientific meetings and seminars. Whenever Alan Haynie attended a 1751 scientific meeting where BSIERP work would be relevant, he presented work from the project along with 1752 work from other projects, as appropriate. 1753

1754 Oral Presentations 1755 As our project description implies, one core element is identifying the pathways and means through which 1756 changing climate will impact fisheries. To address this objective, we attended many interdisciplinary 1757 scientific meetings, discussed our work with a wide range of biologists, oceanographers, and climate 1758 scientists, and wrote a ICES paper that proposed a general framework for why economics is important for 1759 understanding the impacts of climate change on fisheries. 1760

1761 “Not just a march to the north: How climate variation affects the Bering Sea pollock trawl and Pacific cod 1762 longline fisheries.” International Institute of Fisheries Economics and Trade (IIFET) 2014, Brisbane, 1763 Australia, July 2014.

1764 “The effects of catch share management on rent generation through targeting and production choices.” 1765 International Institute of Fisheries Economics and Trade (IIFET) 2014, Brisbane, Australia, July 2014.

1766 “Not just a march to the north: How climate variation affects the Bering Sea pollock trawl and Pacific cod 1767 longline fisheries.” Bering Sea Open Science Meeting, Honolulu, HI, February 2014.

1768 “Using vessel monitoring system data to estimate spatial effort: an example from the Bering Sea.” 1769 (Presented by J. Waston). Kodiak Area Marine Science Symposium, Kodiak, AK, April 2014.

1770 “Using vessel monitoring system data to estimate spatial effort: an example from the Bering Sea.” 1771 (Presented by J. Waston). Alaska Marine Sciences Symposium, Anchorage, AK, January 2014.

1772 “The effects of catch share management on rent generation through targeting and production choices.” 1773 (Presented by L. Pfeiffer). Western Economics Association International, Seattle, WA, July 2013.

1774 “Climate change and fisher behavior in the Bering Sea pollock trawl and Pacific cod longline fisheries.” 1775 Challenges of Natural Resource Economics and Policy (CNREP) 2013, New Orleans, LA, March 2013.

1776 “A tale of two fisheries: Climate change and fisher behavior in the Bering Sea pollock trawl and BSAI

Page 108 of 127

1777 Pacific cod longline fisheries.” PICES Annual Science Meeting, Hiroshima, Japan, October 2012.

1778 “Modeling the Impacts of Climate Change on Fleet Behavior In the Bering Sea Pollock Fishery.” 1779 (Presented by L. Pfeiffer). American Fisheries Society (AFS) Annual Meeting, Seattle, WA, September 1780 2011.

1781 “Climate Change and Location Choice in the Pacific Cod Longline Fishery.” (Presented by L. Pfeiffer). 1782 American Fisheries Society (AFS) Annual Meeting, Seattle, WA, September 2011.

1783 “Introduction to modeling fisher behavior in commercial fisheries.” The 10th R.O.K – U.S. Fisheries 1784 Panel Conference. Seattle, WA, June 2011.

1785 “Economic Drivers Key Factor in the Adaptation of Fisheries to Climate Change: The Bering Sea Pollock 1786 Fishery.” Conference of the North American Association of Fisheries Economists (NAAFE), Honolulu, 1787 HI, May 2011.

1788 “Climate Change and Location Choice in the Pacific Cod Longline Fishery.” ESSAS 2011, Seattle, WA, 1789 May 2011.

1790 “Economic Drivers Key Factor in the Adaptation of Fisheries to Climate Change: The Bering Sea Pollock 1791 Fishery.” (Presented by L. Pfeiffer). ESSAS 2011, Seattle, WA, May 2011.

1792 “Modeling fleet behavior in the Bering Sea pollock fishery under climate change.” Bering Sea Integrated 1793 Ecosystem Research Program (BSIERP) PI Meeting, Anchorage, AK, March 2011.

1794 “The Impacts of Catch Share Institutions on Fishing and Processing: the AFA Pollock Catcher/Processor 1795 Fishery.” (Presented by L. Pfeiffer). NMFS Economics and Social Science Workshop, Orlando, FL, 1796 September 2010.

1797 “Climate Change and Fisher Behavior in the Bering Sea Pollock Fishery.” (Presented by L. Pfeiffer). 1798 Conference of the International Institute for Fisheries Economics and Trade, Montpellier, France, July 1799 2010.

1800 “Modeling fleet behavior in the Bering Sea pollock fishery under climate change.” Fisheries- 1801 Oceanography Coordinated Investigations (FOCI) Seminar, May 2010.

1802 “Modeling fleet behavior in the Bering Sea pollock fishery under climate change.” International 1803 Symposium: Climate Change Effects on Fish and Fisheries, Sendai, Japan, April 2010.

1804 “Climate Change and Fisher Behavior in the Bering Sea Pollock Fishery.” Alaska Marine Sciences

Page 109 of 127

1805 Symposium, Anchorage, AK, January 2010.

1806 “Bering Sea Integrated Ecosystem Research Program: Upper Trophic Level Models Synthesis Talk.” 1807 Bering Sea Integrated Ecosystem Research Program (BSIERP) PI Meeting, Girdwood, AK, October 1808 2009.

1809 “How Fishermen Decide When and Where to Fish.” University of Western Washington Guest Lecture, 1810 Bellingham, WA, May 2009.

1811 “Projects B71 & B72: Adding Fisheries Economics Models to BSIERP,” EMC Presentation, January 1812 2009.

1813 “Climate Change and Changing Fisher Behavior in the Bering Sea Pollock Fishery.” International 1814 Symposium on Effects of Climate Change on the World's Oceans, Gijon, Spain, May 2008.

1815 Poster Presentations 1816 “Using vessel monitoring system (VMS) data to estimate spatial effort for unobserved trips in Bering Sea 1817 fisheries.” Jordan Watson, Bering Sea Open Science Meeting, Honolulu, HI, February 2014.

1818 “Climate change and fisher behavior in the Bering Sea pollock trawl and Pacific cod longline fisheries.” 1819 AMSS 2013, Anchorage, AK.

1820 "Using Longline Vessel Behavior to Understand the Effects of Climate on Pacific Cod.” AMSS, 1821 Anchorage, AK, January 2012.

1822 “Using vessel monitoring system data to estimate spatial effort: an example from the Bering Sea.” Bering 1823 Sea Open Science Meeting, Honolulu, HI, February 2014. 1824

1825 Outreach / Workshops 1826 1827 “Climate Change and Fisher Behavior in the Bering Sea Pollock Fishery.” (Presented by L. Pfeiffer). 1828 Presentation to the NPFMC Scientific and Statistical Committee (SSC), Seattle, WA, February 2011.

1829 “Bering Sea Project: Management Strategies Evaluation Workshop.” AFSC, Seattle, WA, October 2011.

1830 “Why economics matters for understanding the effects of climate change on fisheries,” (Presented by L. 1831 Pfeiffer). NMFS/CSIRO Climate and Fisheries Workshop, Seattle, WA, August 2013.

1832 “Global assessment of the implications of climate change on the spatial distribution of fish and fisheries,”

Page 110 of 127

1833 PICES/ICES workshop, St. Petersburg, Russia, May 2013.

1834 "Climate change and fisher behavior in the Bering Sea pollock trawl and Pacific cod longline fisheries.” 1835 Poster selected to display during the NPFMC Meeting, Portland, OR, February 2013.

1836 “The Effects of Climate Regimes on the Pacific Cod Longline Fishery.” Selected poster at NOAA 1837 Fisheries Science Board Reception, Seattle, WA, August 2012.

1838

1839 “NSF Resource Conservation Network RCN) Project Meeting,” Alaska Fisheries Science Center, Seattle, 1840 WA, February 2012.

Page 111 of 127

1841 Acknowledgements 1842 In addition to the acknowledgements contained in the chapter publications, we wish to acknowledge the 1843 input of a large number of researchers in improving our efforts in this project. We owe enormous 1844 gratitude to Tom Van Pelt, Francis Wiese, Carrie Eischens, Clarence Pautzke, Denby Lloyd, Matt Baker, 1845 and other staff and board members at NPRB who have made this project possible and for helping to 1846 ensure the project evolved effectively. Anne Hollowed provided new avenues for contact with researchers 1847 from PICES, NPRB, CSIRO, and other forums. Mike Sigler managed the BSIERP program at AFSC and 1848 continued to asked good questions to focus our efforts throughout the project. Meeting the range of PIs 1849 and students involved in the program as well as new colleagues at national and international sciences 1850 meetings was fantastic and has created lifelong opportunities. We greatly appreciate the many people at 1851 the Alaska SeaLife Center who have assisted with the project over the years: Kellee Weaver, Tara 1852 Riemer, Denise Cerniglia, Kristin Thoresen, Jeanette Hanneman, Pam Jarosz, and Suzan Armstrong. 1853 1854 Thanks to Dona Cocking, Jennifer Ferdinand, Mike Hoang, and Lori Budbill at AFSC for great work in 1855 handling the administration of the project. Thanks to our colleagues Ron Felthoven, Steve Kasperski, and 1856 others at AFSC for their input into our papers and data analysis. Thanks for Franz Mueter and Pat 1857 Sullivan for great input and for chairing/ serving on Jordan Watson’s PhD committee.

Page 112 of 127

1858 Literature cited 1859 1860 Please see the references in each chapter for additional references. 1861 1862 Conrad, J.M. and M.D. Smith. 2012. “Nonspatial and Spatial Models in Bioeconomics,” Natural 1863 Resource Economics 25(1): 52-92. 1864 1865 Haynie, A., R. Hicks and K. Schnier. 2009. “Common Property, Information, and Cooperation: 1866 Commercial Fishing in the Bering Sea,” Ecological Economics 69(2): 406-413. 1867 1868 Haynie, A., P. Sullivan and J. Watson. In prep. “Using Vessel Monitoring System Data to Estimate 1869 Spatial Effort in Bering Sea Fisheries for Unobserved Trips.” Working paper. 1870 1871 Holland, D.S. and J.G. Sutinen. 2000. “Location Choice in New England Trawl Fisheries: Old Habits Die 1872 Hard,” Land Economics, 76(1), pp. 133-49. 1873 1874 Mistiaen, J.A. and I.E. Strand. 2000. “Location Choice of Commercial Fishermen with Heterogeneous 1875 Risk Preferences,” American Journal of Agricultural Economics 82: 1184-1190. 1876 1877 Pfeiffer, L. and A. Haynie. 2014. “The effects of catch share management on rent generation through 1878 targeting and production choices.” Under review. 1879 1880 Pfeiffer, L. and T. Gratz 2014. “Risk Taking Before and After Catch Shares Implementation: Evidence 1881 from the Limited Entry Sablefish Fixed Gear fishery,” Unpublished manuscript. 1882 1883 Rose, K. et al. 2014. “Demonstration of a Fully-Coupled End-to-End Model for Small Pelagic Fish Using 1884 Sardine and Anchovy in the California Current.” Under Review. 1885 1886 van Putten, I. E., Kulmala, S., Thébaud, O., Dowling, N., Hamon, K. G., Hutton, T. and Pascoe, S. 1887 (2012), Theories and behavioural drivers underlying fleet dynamics models. Fish and Fisheries, 13: 216– 1888 235.

Page 113 of 127 Please do not cite this appendix without permission from the authors.

1889 Appendix 1: Working paper: Utilizing VMS data to estimate unobserved Pacific cod fishing effort 1890 in the Bering Sea 1891

1892 By Jordan T. Watson and Alan C. Haynie

1893 A.1 Introduction

1894 While Alaska fisheries are widely recognized as having an exemplary observer program (Alaska Fisheries 1895 Science Center 2014), numerous fisheries in Alaska and elsewhere have only partial observer coverage. 1896 In Chapter 4 of this report, the Haynie and Pfeiffer manuscript examines fleet behavior for the Pacific cod 1897 “freezer longliner” sector. This manuscript reports on our work to improve the data available for a 1898 subsequent analysis of that sector.

1899 In Alaska, a number of fisheries are required to carry vessel monitoring system (VMS) units, including all 1900 vessels that target pollock and Pacific cod (cod). Integrating VMS and observer data allows us to 1901 examine observed fishing behavior and the associated vessel movement signatures to determine if the 1902 vessel is fishing. Such analyses have been applied to several foreign fishing fleets to improve estimates 1903 of effort (e.g., Deng et al. 2005; Joo et al. 2011; Skaar et al. 2011) using a basic speed filter approach 1904 (e.g., all VMS observations with a vessel speed between 2 and 4 knots are deemed “fishing”). Related 1905 work, Haynie ,Sullivan, and Watson (in prep) finds that models can predict fishing behavior with a high 1906 degree of accuracy in the Bering Sea pollock catcher vessel trawl fishery. This chapter aimed to 1907 generalize the algorithmic and modeling approaches being developed for the pollock fishery to a form 1908 that could also resolve vessel behaviors in the cod longline sector.

1909 While most of the attention for describing fishing behaviors typically focuses on the time that gear was in 1910 the water and actively fishing, this is only one piece of the fishing effort puzzle. Even if fishing is 1911 observed, effort data alone may not be sufficient to adequately link periods of fishing to their landings or 1912 production data without knowing when a trip started. Even among observed vessels in Alaska, 1913 information on the start and end of trips has only been collected since late 2007, while VMS data are 1914 available for vessels that target pollock, cod, and Atka mackerel since late 2002. In such cases recent and 1915 future data are beneficial, but they offer limited retrospective value for understanding how vessel 1916 behaviors and fleet dynamics have evolved during the past decade. Furthermore, observer coverage 1917 throughout these fleets has varied during this time so analyses have been limited to observed trips. In 1918 order to best understand how environmental and other dynamics may be affecting fishers and stocks for

Page 114 of 127 Please do not cite this appendix without permission from the authors.

1919 unobserved or partially observed fleets, we need to be able to understand how fishers have responded in 1920 the past, when their fishing and trip characteristics have been observed and unobserved. Furthermore, by 1921 only including active fishing periods in order to describe effort, the actual effort (opportunity and 1922 economic costs) expended by fishers is typically overlooked or underestimated, and our ability to observe 1923 vessel actions while not fishing can contribute to a more holistic understanding of vessel behavior.

1924 In order for models of fishing effort to be most efficacious for describing fishing behaviors from the 1925 perspectives of both economics and stock management, these models must be capable of including when 1926 fishers began a trip and when that trip ended. Our approach was thus multi-pronged; we sought (1) to 1927 resolve when fishing vessel trips started and ended; (2) to determine trip durations and distances traveled 1928 by vessels; and (3) to identify the periods of time during trips when vessels were actively fishing.

1929 This appendix of the report represents a report of work that is still in progress and was enabled to be 1930 extended as part BSIERP B72 project resources. We will complete this analysis and submit an extended 1931 version of this appendix to a peer-reviewed journal in late 2014.

1932 A.2 Methodology

1933 The methodological development of our approaches focused on a subset of observer and VMS data from 1934 2010 – 2012. These data include observed and unobserved fishing and non-fishing trips and thus enabled 1935 us to examine a suite of different behaviors for the development of both the trip algorithm and our models 1936 to identify active fishing. We have similar data for 2003-2009 and will extend our analysis to that time 1937 period when the methods are finalized.

1938 A.2.1 Trip algorithm

1939 In theory, a fishing trip should be easy to identify; it begins when a vessel leaves port, continues 1940 while the vessel fishes and travels around the fishing grounds, and sometime thereafter, it ends when the 1941 vessel returns to port and its catch is offloaded. In reality however, trip identification can be quite 1942 complex. Vessel monitoring systems sometimes continue to transmit while a vessel is in port, creating 1943 continuous strings of static observations. In other cases, VMS transmissions may cease prior to a vessel’s 1944 return to port, or they may fail to transmit until after a vessel has already begun a new trip. Similarly, in 1945 the middle of a trip, VMS transmissions may be interrupted, or a vessel may stop in port (e.g., St. Paul) 1946 but not offload its catch. The latter situation becomes a more subjective instance in which the definition of 1947 a “trip” may require re-examination or clarification depending on the goal of a particular analysis. Our 1948 methodology attempts to account for these data nuances by integrating vessel speeds, the gap (time 1949 difference) between VMS observations and the distance of each VMS observation from the nearest port.

Page 115 of 127 Please do not cite this appendix without permission from the authors.

1950 Vessel speed was calculated for each VMS observation based on the distance traveled and the gap 1951 between VMS observations. A weighted average of two speeds (one from the previous observation and 1952 one to the proceeding observation) was determined for each observation, where the weighting was a 1953 function of the size of the gap between observations (greater weight was placed on a smaller gap)

1954 The distance was calculated between each VMS observation and a latitude-longitude coordinate at the 1955 “center” of each port. The foundation of our in port designation was whether a vessel was within 10 miles 1956 (mi) of the nearest port. However, as vessels may fish closer to some ports than others, or they may pass 1957 within 10 mi of a port while in transit, several exceptions were made for the in port designation based on 1958 a combination of distance to port, speed and the gap size between VMS observations. For example, a 1959 single VMS observation within 8 mi of a port, traveling 6 knots and with a gap of 30 minutes (min) on 1960 each side of the observation, would likely be deemed to not be in port if neither of the adjacent 1961 observations were in port. If the gap was >120 min however, it was assumed that the vessel was in port 1962 and had thus deactivated their VMS upon entering port. The variable conditions explain our use of the 1963 weighted average for speed, which better accounts for vessel behaviors over increasingly unsampled time 1964 intervals.

1965 For those trips with VMS observations clearly originating in port (e.g., speed near zero while < 10 mi 1966 from port and regularly sampled VMS observations [e.g., 30 min gaps]) at the beginning of a trip and 1967 returning to port at the end of a trip, the trip duration was calculated based on the time that the vessel left 1968 port and the time that the vessel returned to port. In a large port like Dutch Harbor, depending on where a 1969 vessel begins its trip, a different amount of time is required to go from the dock to the in port threshold of 1970 10 mi. To standardize the calculation of trip durations, we started Dutch Harbor trips (nearly all trips 1971 examined) when the vessel was 10 mi from port. In most cases, no VMS observations were transmitted at 1972 the exact point when a vessel was 10 mi from port (observations were more likely to occur at say, 8 mi 1973 and then again at 13 mi), so we calculated the speed of the vessel between the VMS observation prior to 1974 the 10 mi boundary (e.g., at 8 mi) and the observation after the 10 mi boundary (e.g., at 13 mi). Using this 1975 speed, we interpolated the vessel track to identify the time at which the vessel would have crossed the 10 1976 mi point. A similar interpolation was performed at the conclusion of the trip.

1977 To avoid under-estimating trip durations by omitting observations that occurred while inside the 10 mi 1978 port threshold, we determined the average time that vessels took to transit between the dock and the 10 mi 1979 threshold. We used vessel tracks where the VMS data started with a speed of zero to determine the 1980 average time vessels took to transit between the dock and the 10 mi boundary. For Dutch Harbor, the 1981 mean transits were 75 min at the start of the trip and 80 min at the end of the trip. Vessels traveled an

Page 116 of 127 Please do not cite this appendix without permission from the authors.

1982 average of 11 mi before reaching the 10 mi boundary. These transit times (155 min total) and distances 1983 (22 mi) were added to the trip durations and distances calculated between the trip starts (leaving the 10 mi 1984 boundary) and the trip ends (returning to the 10 mi boundary).

1985 For those trips whose starts and stops are more ambiguous, methodological development is still 1986 underway. Once this step is completed, trip statistics can be calculated for each year, different levels of 1987 observer coverage or other relevant classifications of data. Such statistics may include summaries of trip 1988 distances and durations themselves, the spatial distribution of the fleet (e.g., Pfeiffer & Haynie, 2012; 1989 Woillez et al. 2007) and comparisons of observed versus unobserved trips.

1990 A.2.2 Modeling when fishing occurs

1991 Haynie, Sullivan and Watson (in prep) have combined fishery observer and VMS data for the Bering Sea 1992 pollock catcher vessel fleet to consistently predict (accuracy > 90 %) when vessels are fishing based on 1993 their movement characteristics. We adapted their generalized additive modeling (GAM) technique for the 1994 cod longline sector. A suite of model covariates were explored including the weighted average speed 1995 (described above), the speeds of the previous and proceeding VMS observations (to better capture speed 1996 dynamics as vessels transition among transiting, gear setting and gear hauling), time of day, distance from 1997 port, vessel size and turn angles. Model performance was evaluated via the percent of deviance explained 1998 by models and the out-of-sample prediction accuracy (the percent of VMS observations that were 1999 correctly predicted to be fishing (p(fishing) ≥ 0.5) or not fishing (p(fishing)< 0.5).

2000 A.3 Results

2001 For the sake of describing trip algorithm performance, our presentation of results focuses primarily on 2002 aspects of observed fishing trips that occurred from 2010 – 2012. Identification of trip starts and ends and 2003 the calculation of trip durations were necessarily associated; durations were estimated for all trips whose 2004 ends could be identified. The accuracy of distance traveled during a trip however, decreases as gaps 2005 between VMS observations increase. Our distance methodology presently relies on interpolating straight- 2006 line distances between observations and thus as gaps increase in size, the viability of straight-line 2007 interpolation diminishes. Modeling/interpolating movement patterns between distant VMS observations 2008 (e.g., using random walks or Lévy flights) was beyond the scope of our project at this stage and our 2009 presentation of results focuses instead on identifying trip starts and ends (and thus durations) and the data 2010 issues encountered thus far.

2011 The estimation of trip durations is limited to those trips for which VMS data clearly originated within port 2012 and clearly ended within port (N=348, 53.5 % of all observed trips). The mean percent error in trip

Page 117 of 127 Please do not cite this appendix without permission from the authors.

2013 durations estimated from our algorithm versus those from observer records was -9.1% (algorithm 2014 overestimated durations; median 0.2%). This error rate includes trips that observers reported as 2015 embarking or disembarking from (1) St. Paul; (2) at-sea transfers (when the observer is transferred to a 2016 vessel while the vessel is not in port); (3) ports designated as “other.” None of these criteria are currently 2017 included in our trip algorithm as they do not support our goal of identifying trips that start when the vessel 2018 leaves port and ends when it returns to port to offload its catch. When such trips were excluded, the 2019 remaining trips (N=297, 45.7 % of all observed trips) had a mean error rate of 1.8% (algorithm 2020 underestimated durations; median 0.2%); 96% of estimated trip durations had error rates with magnitudes 2021 less than 5% and 87% of estimated trip durations had errors with magnitudes less than 1% (Fig. 1).

2022

2023 Figure 1. Percent errors in the duration of trips estimated from the trip algorithm versus those recorded by 2024 fishery observers. Trips (N=297) do not include those that embarked or disembarked from St. Paul, ports 2025 listed as “Other” or trips “At-Sea transfers.”

2026 Most of the errors in trip duration and identification of trip starts and ends were the result of gaps in the 2027 data (Table 1). Missing VMS observations were quantified based on the number of observations expected 2028 to be transmitted given 30 min intervals; a 60 min gap would be missing a single observation, a 90 min 2029 gap would be missing two observations and so on (Table 1). Gaps were pervasive across years and levels 2030 of observer coverage and somewhat surprisingly gaps were considerably more common for smaller, 2031 partially observed vessels.

Page 118 of 127 Please do not cite this appendix without permission from the authors.

2032 Table 1. Missing VMS observations based on gaps in data for vessels with full observer coverage and 2033 those with partial coverage. Hours of missing data are based on the number of missing observations and 2034 an expected 30 min VMS transmission interval.

Coverage Missing Valid % observations Year level observations observations missing Partial 2010 1104 61986 1.8 Full 2010 7548 98233 7.7 Partial 2011 400 71962 0.6 Full 2011 5146 149255 3.4 Partial 2012 465 78772 0.6 Full 2012 5539 157782 3.5 Total 20202 617990 3.3 2035

2036 If VMS observations were transmitted at 30 min intervals, the number of observations expected for a 2037 given trip would equal the number of observations observed. In some cases, VMS observations are 2038 transmitted with greater frequency than expected (< 30 min intervals) but typically gaps and more lengthy 2039 transmission intervals result in fewer observations than expected (Fig. 2).

Page 119 of 127 Please do not cite this appendix without permission from the authors.

2040

2041 Figure 2. The number of VMS observations expected (based on observed trip durations and 30 min VMS 2042 transmission intervals) versus the number actually observed.

2043

2044 Table 2. Distribution of the number and size of gaps in VMS transmissions.

Gap size Observed trips > 120 min > 240 min > 720 min 2010 185 615 335 77 2011 225 361 120 28 2012 240 486 201 29 All Years 650 1462 656 134 2045

2046 Our initial efforts to identify fishing from vessel movement characteristics using the GAM approach that 2047 was successful with the pollock fishery was not very successful. However, there are strong visible 2048 signatures in the fishery and we are exploring other methods to identify when (and where) fishing is 2049 occurring.

Page 120 of 127 Please do not cite this appendix without permission from the authors.

2050 A.4 Discussion

2051 The trip algorithm we have developed is independent of observer data and can thus be implemented for 2052 both observed and unobserved vessels and trips. This is valuable not only for characterizing fishing trips 2053 that are unobserved but also for characterizing non-fishing trips (which are typically unobserved) for all 2054 vessels that have their fishing completely observed. While these non-fishing trips represent a relatively 2055 small portion of trips in the Pacific cod freezer longline fleet, all of behaviors combined may help to more 2056 fully resolve the changing dynamics of fleets over time. Furthermore, the ability to identify and 2057 subsequently remove non-fishing behaviors/ trips from VMS data, will help to reduce errors as we move 2058 forward with the development of models for identifying fishing periods.

2059 Our initial efforts to identify vessel trips and their associated statistics (e.g., duration, distance traveled, 2060 northernmost latitude, time in port between trips, etc.) and to identify fishing from vessel movement 2061 behaviors, encountered several challenges. The first of these challenges, gaps in the VMS data, is the 2062 result of a technological or data collection issue. The second of these challenges, difficulty extending our 2063 models from the pollock trawl fishery to the cod longline fishery, is the result of substantial differences in 2064 the operating speeds of vessels across gear types during fishing episodes.

2065 Large and inconsistent gaps between VMS observations offer several challenges for characterizing trips. 2066 The first challenge lies in the calculation of vessel speeds. The greater the gap between VMS 2067 observations, the greater the probability that a vessel’s path departs from a straight line and subsequently 2068 leads to underestimation of vessel speeds. Speed is an important factor in the modeling/ prediction of 2069 fishing behaviors as well as the identification of when a vessel is in port or not. We weighted the 2070 calculation of vessel speeds by the size of the gap on either side of an observation, so as gaps increase 2071 towards 120 min (the upper threshold of gap size for which a non-zero weight is assigned), their influence 2072 diminishes. However, subtle changes in a variable may still affect apparent behaviors (e.g., if a vessel 2073 enters port during a given observation and thus triggers the start/ end of a new trip) so even relatively 2074 small gaps (between 30 min and 60 min) can be substantial.

2075 For the sake of the trip algorithm, the impacts of gaps are most relevant when a gap becomes large 2076 enough that a vessel could have made a port call during the gap without a new trip being triggered by the 2077 algorithm. If, for example, the minimum time required for a vessel to make a port call and offload its 2078 catch is 12 hours (720 min), 134 trips between 2010 and 2012 (table 2) could have made port calls that 2079 were undetected by the VMS data. Such instances would have impacts on our estimation of the number of 2080 trips that occurred as well as trip durations. Similarly, when large gaps occur, entire fishing events (e.g., a 2081 longline set) could go undetected, no matter how good of a model is developed for estimating fishing

Page 121 of 127 Please do not cite this appendix without permission from the authors.

2082 behaviors.

2083 Gaps at either the start or the end (in some cases, both the start and the end) of observed trips precluded 2084 the determination of nearly 50 % of trip durations. When the first observation of a trip does not occur at 2085 or near the dock but rather somewhere just outside of port, it may be impossible to know if that VMS 2086 observation occurred at the start of the vessel’s trip or if that vessel was fishing just outside of port (many 2087 vessels fish within a few miles of Dutch Harbor) or elsewhere for some time and VMS transmissions only 2088 just resumed. Conversely, if the last VMS observation occurs just outside of port and there is a large gap 2089 proceeding that observation, it may be impossible to know if that vessel remained outside of port for 2090 some time or if it returned to port shortly thereafter. Integration of daily production reports may help to 2091 resolve some of these uncertainties but they will likely only improve precision to within 1-2 days, leaving 2092 potentially large errors in the calculation of trip durations (and potentially, distances traveled).

2093 While gaps in the VMS data clearly provide a major source of uncertainty and pose a challenge as 2094 we continue to develop the trip algorithm further, the algorithm in the absence of large gaps works quite 2095 well; nearly 90 % of these trips had < 1 % error in their estimated durations and 42 % of them had ≤ 0.1% 2096 errors. Given the nature of VMS sampling intervals, a 30 min interval by definition has a uniform 2097 probability of capturing an event for 15 min on either side of a VMS observation. For example, if VMS 2098 observations were recorded at 8:00 and 8:30 and the vessel left the dock at 8:15, the trip would be 2099 recorded as beginning at 8:30, the nearest VMS observation within the observed activity. Given our mean 2100 observed trip durations (31471 min), 30 min represents a 0.1 % error and 42 % of estimated trip durations 2101 had errors ≤ 0.1%, while nearly 90 % had errors < 1 %.

2102 Additional challenges occur with ports like St. Paul, where a vessel may go to port and visit or it may 2103 seek shelter from weather. While the vessel is “in port,” whether or not such occasions should be 2104 considered as the starts or stops of trips are left to more subjective decisions. If the vessel does not 2105 offload, then it might be argued that this port call should not be considered as the resetting of a trip 2106 because the purpose of a “fishing trip” is to leave port, catch fish and return to port for delivery of 2107 product. St. Paul has consequently been omitted as a port from our analyses but admittedly, this creates an 2108 apparent source of error when comparing the duration of trips as identified by our trip algorithm with the 2109 duration of trips as reported by fishery observers. Further investigation of such instances is still needed.

2110 Efforts to estimate fishing by modeling vessel movements met greater challenges for the Pacific cod 2111 longline fishery than they did for similar efforts with the pollock trawl fishery. The majority of trawling 2112 activities occur while a vessel is moving slowly relative to their transit speeds. However, longline vessels 2113 may speed up and slow down through the course of a set as they deploy and retrieve their gear. They may

Page 122 of 127 Please do not cite this appendix without permission from the authors.

2114 also deploy multiple sets of gear consecutively so that multiple fast-slow cycles may occur between the 2115 time that the first hook entered the water (the observers’ cue for fishing to start) and the last hook exited 2116 the water (the cue for the end of fishing). Thus a speed profile of a longline vessel (Fig. 3) fails to reveal a 2117 clear distinction between fishing speeds and non-fishing speeds. In order to mitigate the ambiguity of the 2118 speed profiles, we examined several different lagged speed combinations in addition to non-vessel 2119 characteristics like time of day. Future modeling efforts will explore additional covariates for the GAM 2120 approach but it is likely that different modeling frameworks altogether will be more effective for 2121 identifying fishing behavior.

2122

2123 Figure 3. Profiles of speeds for VMS observations during fishing and non-fishing operations in the Pacific 2124 cod freezer longline sector.

2125 Because we began working with these data late in the scale of the project, this data integration is ongoing. 2126 We will continue to refine our trip estimation and will complete an integrated VMS, weekly/daily 2127 production, and observer database for the Pacific cod longline fleet from 2003-2013. We will utilize

Page 123 of 127 Please do not cite this appendix without permission from the authors.

2128 these data in subsequent analysis of spatial fishing behavior in the fishery.

2129

2130 Appendix 1 References

2131 Alaska Fisheries Science Center. 2014. NOAA Fisheries 2013 Science Program Reviews: Stock 2132 Assessment Data, [Online]. Available http://www.afsc.noaa.gov/program_reviews/2013/. Accessed June 2133 2014.

2134 Deng, R., C. Dichmont, D. Milton, et al. 2005. Can vessel monitoring system data also be used to study 2135 trawling intensity and population depletion? The example of Australia’s northern prawn fishery. Can. J. 2136 Fish. Aquat. Sci. 62: 611-622

2137 Joo, R., S. Bertrand, A. Chaigneau, and M. Niquen. 2011. Optimization of an artificial neural network for 2138 identifying fishing set positions from VMS data: an example from the Peruvian anchovy purse seine 2139 fishery. Ecological Modelling, 222(4): 1048–1059.

2140

2141 Skaar, K. L., Jørgensen, T., Ulvestad, B. K. H., and Enga˚s, A. 2011. Accuracy of VMS data from 2142 Norwegian demersal stern trawlers for estimating trawled areas in the Barents Sea. ICES J. Mar. Sci. 68: 2143 1615–1620.

2144 Woillez, M., J.C. Poulard, J. Rivoirard, P. Petigas, N. Bez. 2007. Indices for capturing spatial patterns and 2145 their evolution in time, with application to European hake (Merluccius merluccius) in the Bay of Biscay. 2146 ICES J. Mar. Sci. 64: 537 – 550.

Page 124 of 127

2147 Appendix 2: Headline document 2148 The following 2-page document was prepared as a summary of our findings.

Page 125 of 127 Page 126 of 127

UNDERSTANDING ECOSYSTEM PROCESSES IN THE BERING SEA 2007–2013

Climate and Bering Sea Fisheries: Beyond a Northward March SURPRISING IMPACTS ON BERING SEA POLLOCK AND PACIFIC COD FISHERIES

While some other global scale How We Did It research has suggested that a warm- A significant component of our The Big Picture ing climate will propel marine research has been focused on iden- The BEST-BSIERP Bering Sea Project species northward, our work has tifying the mechanisms by which recognized from its outset that humans demonstrated that for the biggest climate impacts fisheries. We use are an important component of the fisheries in the Bering Sea, this has data on fishing locations, fish and ecosystem, and that we cannot under- not occurred as expected. For pol- fuel prices, and how these interact stand the system without understanding lock between 1999 and 2009, the with biological survey information how they use and adapt to the changing fishery shifted northward in the sum- and environmental data. After col- environment. By examining the response mer, but this occurred in cold years lecting data and talking to fisher- of largest Bering Sea fisheries to the more than warm years. Similarly, for men, we used a variety of statistical changing environment, we have illus- Pacific cod, a larger cold pool (where methods to see how management, trated that people will not respond in a bottom water temperatures are below changing prices, and changing bio- simple manner to the changing environ- 2°C) in cold years has led to fish logical and environmental measures ment. A better understanding of how the fishery behaves in warm, low-abundance being more concentrated in northern have impacted the fisheries (Figures years will help inform how the fishery areas and consequently to more fish- 2 and 3). continued on page 2 will react in the future. Managers can use ing in those areas (Figure 1). this information to better anticipate how Fig. 1 fisheries will interact with other parts of the ecosystem, which can contribute to better-managed fisheries.

The Eastern Bering Sea and the fishing areas of the catcher–processor fleet. Points represent the catch-weighted mean center of the distribution of fishing hauls by season. Note the large distinction in the movement of the fishery over time that oc- curs in the summer fishery B season as well as the lack of movement in the winter fishery A season. [From Haynie, A. and L. Pfeiffer. 2013. “Climatic and economic drivers of the Bering Sea pollock (Theragra chalcogramma) fishery: Implications for the future.” Canadian Journal of Aquatic and Fisheries Science. 70(6): 841-853, 10.1139/cj- fas-2012-0265.]

SPATIALLY EXPLICIT INTEGRATED MODEL OF POLLOCK AND COD A component of the BEST-BSIERP Bering Sea Project, funded by the National Science Foundation and the North Pacific Research Board with in-kind support from participants. Page 127 of 127

From this research, we have has variation in climate conditions Fig. 3 seen that abundance and environ- affected the spatial extent of Bering mental conditions both directly Sea fisheries? How do we expect impact where the fisheries occur. predicted changes in future climate Other BSIERP work has indicated to impact fisheries and fishing that we are likely to see more low- communities? Informing decision- abundance years with a warming makers on how climate and fisheries climate (Mueter et al., ICES 2011), are interacting is essential to the but in recent times, warm years have effective management of marine also been high-abundance years. As resources in the future. The deci- shown in Figure 3 and discussed in sions that managers make now the Haynie and Pfeiffer ICES 2012 will impact the welfare of fishers, article referenced in Figure 2, we communities, the nation, and the have not yet experienced a likely ecosystem over the next century. future state of warm, low-abundance Summary of the effects of the size of the cold pool Alan Haynie, National Oceanic and Atmospheric Administration conditions. (NOAA) Fisheries, Alaska Fisheries Science Center (AFSC) and total pollock abundance on the intensity of Lisa Pfeiffer, NOAA Fisheries, AFSC early A-season (winter season) effort, B-season Why We Did It (summer season) catch per unit effort (CPUE), B-season effort, and B-season travel costs. Years Fishers are the apex predators The Bering Sea Project is a partnership between in the sample characterized by varying abundance of the Bering Sea ecosystem, and the North Pacific Research Board’s Bering Sea and cold pool levels are listed on the horizontal and their spatial behavior can tell us a Integrated Ecosystem Research Project and the vertical axes. [Also from Haynie and Pfeiffer CJFAS great deal about the way in which National Science Foundation’s Bering Ecosystem 2013, referenced above.] fish populations are shifting under Study. www.nprb.org/beringseaproject changing climate conditions. After controlling for other factors, how

Fig. 2

A catch of pollock on the NOAA ship Miller Free- man, a fisheries oceanography vessel that works A conceptual model of how the environment affects the distribution of fishing effort, including the total predominantly in the Bering Sea and the North allowable catch (TAC) and the cost per unit effort (CPUE). Arrows represent the direction of causality, Pacific Ocean. and dotted lines represent mechanisms that may occur on a non-contemporaneous time scale. [From Haynie, A. and L. Pfeiffer. 2012. “Why economics matters for understanding the effects of climate change on fisheries.” ICES Journal of Marine Science, 69 (7): 1160-1167, doi:10.1093/icesjms/fss021.]

SPATIALLY EXPLICIT INTEGRATED MODEL OF POLLOCK AND COD A component of the BEST-BSIERP Bering Sea Project, funded by the National Science Foundation and the North Pacific Research Board with in-kind support from participants.