1

Proposal to NOAA Fisheries and the Environment (FATE) program

Title: Short-term forecasting of Pacific hake distribution in the California Current Ecosystem

Lead Principal Investigator: Mary Hunsicker, Northwest Fisheries Science Center, 2032 SE OSU Drive, Newport, Oregon 97365, mary.hunsicker@noaa.gov, 541-867-0306

PIs: Richard Brodeur (NWFSC), Melissa Haltuch (NWFSC), Allan Hicks (NWFSC), Isaac Kaplan (NWFSC), Sandy Parker-Stetter (NWFSC), Nicholas Bond (PMEL/JISAO), Albert Hermann (PMEL/JISAO), Jan Newton (Univ. Washington) and Samantha Siedlecki (Univ. Washington)

Project Duration: 2 years

2

Background Pacific hake (Merluccius productus) is the most abundant groundfish species in the California Current Ecosystem (CCE) and comprises the largest groundfish landings by volume in the commercial groundfish fishery. The hake stock is exploited by both U.S. and Canadian fleets. In spring/summer hake migrate northward from spawning grounds in Mexico and the U.S., with a variable proportion of the stock reaching Canadian waters. The hake population is jointly monitored by a U.S.-Canada summer acoustic survey that provides the index of abundance that is the basis for the stock assessment and management advice. Management strategy evaluations demonstrate that a higher precision acoustic survey of hake biomass would improve the hake stock assessment (unpublished results, A. Hicks). More efficient allocation of survey effort and more precise estimates of stock size require knowledge of the expected hake habitat distribution prior to the survey. In addition, in the event that few hake are observed in the survey, knowledge of hake distribution at fine temporal and spatial scales could be used to improve subsequent targeting of the hake stock. Forecasts of the expected summer hake distribution would allow the survey to focus effort on areas with hake, reducing effort in areas with no or few fish. Further, advanced warning of hake spatial distribution would allow fishery managers and the fishing industry to anticipate the availability of hake to U.S. and Canadian fisheries. We propose to provide short-term (weekly to monthly) forecasts of hake habitat distribution based on oceanographic conditions to improve acoustic survey design and management planning. Variations in hake migrations have been attributed to topography, variable oceanographic conditions, prey distributions (primarily euphausiids), and large-scale phenomena such as El Niño and ocean regime shifts (Smith et al. 1990, Dorn 1995, Benson et al. 2002, see review by Ressler et al. 2007). In addition, previous research on the summer distribution of hake indicates that in warmer years hake tend to migrate further north (Ware and McFarlane 1995). Migration routes also vary for fish of different ages; older and larger hake migrate further north (Beamish and McFarlane 1985). Thus, forecasting the northerly extent of hake distribution is particularly important for ensuring that the allocation of sea days spent in the southern versus northern portions of the survey area are appropriate to capture the full extent of the population, resulting in a survey biomass estimate with less uncertainty. Despite numerous studies of hake (see Ressler et al. 2007), drivers of hake distribution are not fully understood, especially across hake life history stages. For instance, past authors have used acoustic survey data and in situ physical data (e.g. alongshore flow, temperature, depth) in the CCE to evaluate the relationship between the distribution of hake with ocean current strengths and oceanographic conditions (Agostini et al. 2006, 2008). However, all hake age classes were modeled together. In addition, the studies are based on only two years of data (1995 and 1998). Our study makes use of 10 additional years of age-specific survey biomass estimates and previously unavailable in situ physical data to model the relationships between hake habitat distributions by age-class and/or life stage and oceanographic conditions. Investigating the links between hake spatial distribution and dynamics with ocean conditions and ecosystem variables is listed as one of the top research needs for improving stock assessments and decision-making for Pacific hake (Taylor et al. 2015).

3

Once the links between in situ ocean conditions and hake distribution are established, we can test our ability to predict seasonal hake distributions (1-6 months ahead) in the northern CCE based on forecasts of ocean conditions. One ocean forecasting tool that operates at this seasonal time scale and at the interface of physics and ecology is the FATE-funded project J-SCOPE: JISAO’s Seasonal Coastal Ocean Prediction of the Ecosystem. J-SCOPE has been developed for the northern CCE along the west coast of North America to provide projections of physical, chemical and biological ocean properties on six-month time horizons. These forecasts result from dynamically downscaled Climate Forecast System (CFS) forcing applied to the Regional Ocean Modeling System (ROMS). As an example of the potential utility for J-SCOPE, Kaplan et al. (submitted) have used it to forecast the spatial habitat of Pacific sardine, which are known to track ocean conditions such as sea surface temperature and salinity (Emmett et al. 2005; Zwolinski et al. 2011). Using the 2009 J-SCOPE reforecasts and field observations of sardines, Kaplan et al. (submitted) found that J-SCOPE still had moderate ability to predict May-August sardine habitat from three surveys in 2009, as well as in summer 2013 and 2014. For sardine (as desired for hake), one aim in the management context is for the forecasts to serve as an early warning to fishers and research surveys of shifts in the northward migrations of sardine. Seasonal forecasts of hake distribution (weeks to months in the future) are a natural way to leverage improved empirical understanding of hake and ocean conditions, and the J-SCOPE forecasting system (http://www.nanoos.org/products/j-scope/home.php/). In a collaborative effort among ecologists, fisheries biologists, oceanographers, climatologists and stock assessment scientists, we will: (1) Model the relationships between hake habitat distribution and oceanographic variables in the CCE by age class (age-1+) and/or life stage (juvenile, subadult, adult), (2) Assess the skill of J-SCOPE to predict ocean variables likely important for hake, such as SST, mid-water temperatures, salinity, poleward flowing subsurface California undercurrent and upwelling, (3) Force hake distribution models with seasonal lead-time forecasts of oceanographic variables produced by J-SCOPE to provide weekly to monthly to forecasts of hake distribution, and (4) Assess skill of hake distribution forecasts, by age class and month, by comparing data from the hake acoustic survey to J-SCOPE forecasts. Approach Data sources: The NOAA Northwest Fisheries Science Center (NWFSC) and the Department of Fisheries and Oceans Canada (DFO) have conducted a joint acoustic-trawl survey of hake biomass and distribution since 1992. The joint survey was conducted triennially from 1992 to 2001 (4 years) and biennially from 2003 to 2015, with an additional survey in 2012 (8 years). Starting in 2012, the U.S. portion of the hake survey has been joint with the Southwest Fisheries Science Center (SWFSC) coastal pelagic survey. The coast-wide hake acoustic survey starts in California (southern extent varies by year) and extends north into Canada (including Alaska in some years), ending when hake are no longer detected (Figure 1). Water column estimates of hake biomass and distribution within 0.5 nmi horizontal bins are available aggregated across ages (age-1+), and separated by age-groups (age-1, age-2, etc). Historical data (1992 to 2003) plus newly-collected 2015 data are being processed and we anticipate that they will be available for this project (Table 1).

4

In situ physical data collected during the joint NWFSC-DFO survey will be used to test a set of mechanistic hypotheses regarding the environmental drivers of hake distribution (see below). With a few exceptions, due to equipment failure or survey conflicts, data are available for all years that we have hake biomass estimates. Due to the use of different survey vessels, latitudinal extent of data varies among years. Data that are not currently processed will be prepared to meet the Presidential “Public Access to Research Results” mandate and will be available to this project. The survey Acoustic Doppler Current Profiler (ADCP) data will be used to characterize the strength and continuity of the poleward flow in the undercurrent. The Conductivity-Temperature-Depth (CTD) sensors (on a grid of stations), the underway sensors (collected between CTD stations), and the headrope sensors (collected to trawl headrope depth at trawl locations, SBE or MBT) will be used to characterize water column temperature and salinity across the survey area. The shipboard underway thermosalinograph (uTSG) data are collected continuously across the survey area and will be used to characterize surface temperature and salinity conditions. Other environmental covariates will include bottom

depth, distance to land and upwelling indices (SWFSC Env. Res. Lab). Changes in the intensity of upwelling will also be indexed using the upwelling indices from J-SCOPE, like the 8-day upwelling index (Austin and Barth 2002). While we recognize that prey, specifically krill, influence hake spatial dynamics (see review by Resler et al. 2007), acoustic estimates of krill are not available. However, it is likely that krill distribution and some of the physical covariates (e.g. temperature and depth) are strongly correlated.

Table 1: Type, source and available years of in situ data collected via hake acoustic survey.

Hypotheses: At the core of this proposal are mechanistic hypotheses regarding the environmental drivers of hake distribution that can be tested using the in situ oceanographic data collected during survey operations. Two hypotheses address possible climate mechanisms forcing hake summer distribution.

Variable Source YearsHake all ages (age-1+) Acoustics 1992-2013 available, 2015 in progressHake by age (age-1,2, etc) Acoustics 2009-2015 available, 1992-2007 in progressHake by age (age-1) Acoustics No 1992, 1995-2015 in progressTemp, depth, salinity CTD 1992-2007 available, 2009-2015 in progressTemp, depth, salinity uCTD, XBT All years except 2003 availableTemp, depth, salinity SBE, MBT All years except 1992 availableUndercurrent ADCP No 1992, 1995-2009 available, 2011-2015 in progressSurface salinity uTSG All years except 2003 available

Fig. 1. West coast of U.S. and Canada. Typical hake survey extent (grey) in 1995-2005.

5

H1: The intensity and location of the poleward undercurrent impacts the period of active migration, with stronger poleward flow leading to the population moving farther north (Smith et al. 1990; Dorn 1995; Saunders and McFarlane 1997; Agostini et al. 2006). The inter-annual variability in the depth and cross-shelf location of the undercurrent (Agostini et al. 2006) is likely reflected by the distribution of hake and will be quantified using weekly to monthly depth stratified temperature, salinity, and hake aggregation distance from shore.

H2: The timing of the spring transition and in turn the intensification of upwelling (Benson et al. 2002) impacts the timing and distribution of euphausiid availability (Peterson and Keister 2003) and therefore hake distribution. Changes in upwelling will be quantified using the weekly to monthly cumulative upwelling index generated by the SWFSC’c Env. Res. Lab and J-SCOPE. Modeling hake distributions: General additive models (GAMs) have been applied to numerous investigations of fisheries spatial data (see Guisan et al. 2002), including studies that modeled the relationship between species distributions and environmental covariates (e.g. Ciannelli et al. 2012, Hunsicker et al. 2013). GAM models (and their mixed effects extensions, general additive mixed effects models (GAMMs)), provide a way to account for non-linear relationships and threshold dynamics between response and explanatory covariates. Specifically, we will use variable coefficient GAMMs to model hake biomass as a function of environmental and spatial covariates. Variable coefficient GAMMs can account for spatial autocorrelation (Ciannelli et al. 2012) and assume there is a local, linear relationship among the covariates with the response variable, but the linear coefficients vary smoothly across the study area, modeling a spatially heterogeneous effect (Bacheler et al. 2009, Bartolino et al. 2011). The different age-classes and/or life stages (juvenile, subadult and adult) of hake will be modeled separately. In addition, we will assess the effect of spatial scale (i.e. coastwide vs. regional (northern and southern CCE) and fine (~1 km), inner/outer transect (~50 km), full transect (~100 km)), temporal scale (week, month), and time lags on the environment-distribution relationships based on model fits.

Analysis of species distribution data is often complicated by the presence of zero observations (Maunder and Punt 2004). This project will examine the utility of methods based on zero-inflated distributions as well as hurdle or delta models to address the issue of a high proportion of zero observations in the biological data. Hindcast skill will be evaluated using statistical model-fit criteria (such as fraction of explained deviation) and cross validation. Forecast skill will be evaluated by randomly splitting our dataset into training and validation data (70% and 30% of the dataset, respectively), fitting models to the training data, generating predictions for the validation data, and calculating prediction errors. The uncertainty for both hindcasts and forecasts for each set of predictions will be quantified. The statistical analysis will be completed using the R programming language. The selected habitat distribution models for each age class will be used as the basis for conducting short-term projections of hake summer distribution based on J-SCOPE predictions of ocean conditions. Testing J-SCOPE model skill and short-term forecasting: Once the hypotheses regarding the environmental drivers of hake distribution have been evaluated and models of hake distribution by age class are selected, we will develop tools for projecting short-term forecasts of hake distribution. J-SCOPE system will be used to produce forecasts initialized in January and April of each year, to focus on predicting the spring upwelling season and the summer northward migrations of hake. We will continue the current practice of using ‘mini-ensembles’ of three

6

model runs per initialization month, to capture uncertainty in forecasts. We will use in situ physical data from the coast-wide hake survey to test the skill of J-SCOPE to capture variability (anomalies relative to 2009-2014 averages) in environmental variables. Ideally, these variables would include SST, mid-water temperatures, salinity, poleward flowing subsurface California Undercurrent and upwelling. This comparison is focused specifically on 100-400 m, summer months (June-September) and on the shelf/slope from latitude 43°N to 50°N. Model skill will be assessed using standard metrics (Stow et al. 2009) and Taylor and target diagrams (Jolliff et al. 2009). The focus will be on identifying precision and bias of the forecasts, for these variables relevant for hake, and how this varies by month and lead time. Finally, we will use the J-SCOPE forecast of ocean conditions to predict distribution of each hake age class or life stage, using the same GAMM methodology described above. Hake distribution forecasts may extend up to 6 months in advance; the precise lead time will be dictated by the skill assessment described above for ocean conditions. In addition to skill metrics such as fraction-of-explained-deviation, cross validation, and prediction error (subsetting data into training and validation sets), we can also test the model skill of true forecasts, fitting the GAMM to previous years (2009-2014) and testing model performance for additional true forecast years. Predictions will also be assessed via AUC (area-under-the-curve), which tests the ability of a model to discriminate between presence and absence (Fielding and Bell 1997). As suggested by other successful forecasting work in Australia (Hobday et al. 2015), this final stage of skill assessment will be tailored to ask whether the forecast system can predict specific phenomena that trigger decisions by fishers or the acoustic survey team -- such as presence of particular age classes in northern waters. Benefits This project leverages a successful FATE-funded project, J-SCOPE, to develop forecast tools for making short-term seasonal predictions of hake distributions in the CCE (FATE Research Priority 4). In doing so, we will create spatial models that can be used to examine the influence of environmental and climate variability on a commercially valuable transboundary fish stock managed by the U.S. and Canada (FATE Research Priority 2). This will include testing hypotheses concerning the specific mechanisms driving the distribution of different hake life stages (FATE Research Priority 1). Overall, the modeling tools developed in this study will directly improve hake stock assessments via improved acoustics survey planning (i.e. allocating survey days at sea) and increased precision of hake biomass estimates. The forecasting tools will directly enhance management of the hake population by providing fishery managers and fishers with advanced warning of the availability of hake to U.S. and Canadian fisheries. Deliverables (1) Habitat distribution models for Pacific hake by age-class and/or life stage. (2) Short-term forecasting tool of Pacific hake distribution in the northern CCE. (3) Integration of project results into Pacific hake stock assessment and management process. (4) Further testing of J-SCOPE model skill in predicting ocean conditions in the northern CCE. (5) Preparation of two manuscripts concerning the modeling and forecasting of hake distributions by age-class and/or life stage in relation to environmental variability. (6) Presentation of project findings at the annual FATE meeting (2017, 2018).

7

References Agostini, V.N., R.C. Francis, A.B. Hollowed, S.D. Pierce, C. W. Wilson, and A. N. Hendrix. 2006. The relationship between hake (Merluccius productus) distribution and poleward subsurface flow in the California Current System. Can. J. Fish. Aquat. Sci. 63:2648–2659. Agostini, V.N., A.N. Hendrix, A.B. Hollowed, C.D. Wilson, S.D. Pierce, and R.C. Francis. 2008. Climate-ocean variability and Pacific hake: a geostatistical modeling approach. J. Mar. Syst. 71:237-248. Austin, J. A. and Barth, J.A. 2002. Variation in the position of the upwelling front on the Oregon shelf, J. Geophys. Res. 107:3180. Bacheler, N.M., Bailey, K.M., Ciannelli, L, Bartolino, V. and Chan, K-S. 2009. Density-dependent, landscape, and climate effects on spawning distribution of walleye pollock Theragra chalcogramma. Marine Ecology Progress Series 391:1-12. Bartolino, V., Ciannelli, L., Bacheler, N.M., and Chan, K-S. 2011. Ontogenetic and sex-specific differences in density-dependent habitat selection of a marine fish population. Ecology. 92(1): 189-200. Beamish, R. J., and G. A. McFarlane. 1985. Pacific whiting, Merluccius productus, stocks off the west coast of Vancouver Island, Canada. Mar. Fish. Rev. 47(2):75–81. Benson, A. J., G. A. McFarlane, S. E. Allen, and J. F. Dower. 2002. Changes in Pacific hake (Merluccius productus) migration patterns and juvenile growth related to the 1989 regime shift. Can. J. Fish. Aquat. Sci. 59:1969–1979. Ciannelli L., Bartolino V, Chan K.S.2012. Nonadditive and nonstationary properties in the spatial distribution of a large marine fish population. Proceeding Royal Society London B 279: 3635-3642. Dorn, M. W. 1995. The effects of age composition and oceanographic conditions on the annual migration of Pacific whiting, Merluccius productus. Calif. Coop. Oceanic Fish. Invest. Rep. 36:97–105. Emmett, R.L., R.D. Brodeur, T.W. Miller, S.S. Pool, G.K. Krutzikowsky, P.J. Bentley, and J. McCrae. 2005. Pacific sardines (Sardinops sagax) abundance, distribution, and ecological relationships in the Pacific Northwest. CalCOFI Rep. 46:122-143. Fielding, A.H. and Bell, J.F. 1997. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environmental Conservation 24: 38–49. Guisan, A., Edwards, T.C., Hastie, T. 2002. Generalized linear and additive models in studies of species distribution: setting the scene. Ecol. Model. 157:89-100.

8

Hobday, A.J., Spillman, C.M., Hartog, J.R., and Eveson, J.P. 2015. Seasonal forecasting for decision support in marine fisheries and aquaculture. Fisheries Oceanography 10. Hunsicker, M.E., Ciannelli, L., Bailey, K.M., Zador, S. & Stige, L.C. 2013. Climate and demography dictate the strength of predator-prey overlap in a sub-arctic marine ecosystem. PLoS ONE 8(6): e66025. doi:10.1371/journal.pone.0066025 Jolliff, J.K., Kindle, J.C., Shulman, I., Penta, B., Friedrichs, M.A., Helber, R., and Arnone, R.A. 2009. Summary diagrams for coupled hydrodynamic-ecosystem model skill assessment. Journal of Marine Systems 76(1): 64–82. Kaplan, I.C., Williams, G.D, Bond, N.A., Hermann, A.J. and Siedlecki, S.A. Cloudy with a chance of sardines: forecasting sardine distributions using regional climate models. Submitted Maunder, M.N and A.E. Punt 2004. Standardizing catch and effort data: a review of recent approaches. Fisheries Research 70 (2): 141-159. Ressler, P.H., J. A. Holmes, G.W. Fleischer, R. E. Thomas, and K.C. Cooke. 2007. Pacific hake, Merluccius productus, autecology: a timely review. Mar. Fish. Rev. 69. Smith, B. D., G. A. McFarlane, and M. W. Saunders. 1990. Variation in Pacific hake (Merluccius productus) summer length-at-age near southern Vancouver Island and its relationship to fishing and oceanography. Can. J. Fish. Aquat. Sci. 47:2195–2211. Stow, C.A., Jolliff, J., McGillicuddy Jr, D.J., Doney, S.C., Allen, J., Friedrichs, M.A., Rose, K.A., and Wallhead, P. 2009. Skill assessment for coupled biological/physical models of marine systems. Journal of Marine Systems 76(1): 4–15. Taylor, I.G., Grandin, C., Hicks, A.C., Taylor, N. and Cox, S. 2015. Status of Pacific Hake (whiting) stock in U.S. and Canadian waters in 2015. Prepared by the Joint Technical Committee of the U.S. and Canada Pacific Hake/ Whiting Agreement; National Marine Fishery Service; Canada Department of Fisheries and Oceans. 159 p. Ware, D. M. and G. A. McFarlane. 1995. Climate induced changes in Pacific hake (Merluccius productus) abundance and pelagic community interactions in the Vancouver Island upwelling system. Can. Spec. Publ. Fish. Aquat. Sci. 121:509–521. Zwolinski, J.P., Emmett, R.L. and Demer, D.A. 2011. Predicting habitat to optimize sampling of Pacific sardine (Sardinops sagax). ICES Journal of Marine Science 68: 867–879.