class: right, middle, my-title, title-slide .title[ # Zooplankton indices as covariates in WHAM ] .subtitle[ ## Herring Research Track Review <br /> 11 March 2025 ] .author[ ### Sarah Gaichas, Jon Deroba, Adelle Molina ] --- class: top, left # Does food drive recruitment of Atlantic herring? ## Atlantic herring, *Clupea harengus* .pull-left[  ] .pull-right[ .large[ "The herring is a plankton feeder.... Examination of 1,500 stomachs showed that adult herring near Eastport were living solely on copepods and on pelagic euphausiid shrimps (*Meganyctiphanes norwegica*), fish less than 4 inches long depending on the former alone, while the larger herring were eating both." <a name=cite-collette_bigelow_2002></a>([Collette et al., 2002](#bib-collette_bigelow_2002)) ] ] ??? When first hatched, and before the disappearance of the yolk sac, the larvae (European) feed on larval snails and crustaceans, on diatoms, and on peridinians, but they soon begin taking copepods, and depend exclusively on these for a time after they get to be 12 mm. long, especially on the little Pseudocalanus elongatus. As they grow older they feed more and more on the larger copepods and amphipods, pelagic shrimps, and decapod crustacean larvae. --- background-image: url("https://github.com/NOAA-EDAB/presentations/raw/master/docs/EDAB_images/AherringConceptualMod.png") background-size: 1070px background-position: bottom ## Which indicators are potential covariates for recruitment? --- background-image: url("https://github.com/NOAA-EDAB/presentations/raw/master/docs/EDAB_images/AherringBRT_rec.png") background-size: 800px background-position: right ## Which indicators are potential covariates for recruitment? .pull-left-30[ Boosted regression tree (Molina 2024) investigated relationships between environmental indicators and Atlantic herring recruitment estimated in the assessment. Larval and juvenile food (zooplankton), egg predation, and temperature always highest influence ] .pull-right-70[ ] --- ## Spatial partitioning: zooplankton trends at multiple scales <div class="figure"> <img src="20250311_Zoopcovariates_Gaichas_files/figure-html/maps-1.png" alt="Maps of key areas for Herring assessment indices. The full VAST model grid is shown in brown." width="33%" /><img src="20250311_Zoopcovariates_Gaichas_files/figure-html/maps-2.png" alt="Maps of key areas for Herring assessment indices. The full VAST model grid is shown in brown." width="33%" /><img src="20250311_Zoopcovariates_Gaichas_files/figure-html/maps-3.png" alt="Maps of key areas for Herring assessment indices. The full VAST model grid is shown in brown." width="33%" /> <p class="caption">Maps of key areas for Herring assessment indices. The full VAST model grid is shown in brown.</p> </div> ??? NEFSC survey strata definitions are built into the VAST `northwest-atlantic` extrapolation grid already. We defined additional new strata to address the recreational inshore-offshore 3 mile boundary. The area within and outside 3 miles of shore was defined using the `sf` R package as a 3 nautical mile (approximated as 5.556 km) buffer from a high resolution coastline from the`rnaturalearth` R package. This buffer was then intersected with the current `FishStatsUtils::northwest_atlantic_grid` built into VAST and saved using code [here](https://github.com/sgaichas/bluefishdiet/blob/main/VASTcovariates_updatedPreds_sst_3mi.Rmd#L49-L94). Then, the new State and Federal waters strata were used to split NEFSC survey strata where applicable, and the new full set of strata were used along with a modified function from `FishStatsUtils::Prepare_NWA_Extrapolation_Data_Fn` to build a custom extrapolation grid for VAST as described in detail [here](https://sgaichas.github.io/bluefishdiet/VASTcovariates_finalmodbiascorrect_3misurvstrat.html). --- ## Exploratory zooplankton indices in the stock assessment .pull-left[ Use as a basis the new herring stock assessment in Woods Hole Assessment Model (WHAM) <a name=cite-stock_woods_2021></a>([Stock et al., 2021](https://www.sciencedirect.com/science/article/pii/S0165783621000953)). We are using the `devel` version of WHAM: https://github.com/timjmiller/wham/tree/devel Model [mm192](https://drive.google.com/drive/folders/1sQdDsfdnVbiiY4X7Rgr-fvegwT7Fa1Az?usp=drive_link) is our starting point. Zooplankton indices were explored as covariates on herring recruitment. Recruitment is modeled as deviations from the "recruitment scaling parameter", leaving one option for modeling effects of covariates on recruitment: "controlling". A "controlling" recruitment covariate results in a time-varying recruitment scaling parameter. ] .pull-right[ Covariates explored: * Jan-Jun (Spring) large copepods in spring herring BTS strata with lag-0 to represent food for pre-recruit juveniles * Jul-Dec (Fall) small copepods in fall herring BTS strata with lag-1 to represent food for larvae in general * Sep-Feb small copepods in herring larval area with lag-1 to represent food for larvae more specifically * Combinations of large and small copepod covariates above * Optimal temperature duration (days) during the fall larval season, September-December * Sensitivity: Turn off NAA RE and then try fitting with the most robust recruitment covariates. ] ??? --- ## Implementing each index We evaluated * Options for covariate input (millions of cells vs. log(cells), VAST estimated SE vs. WHAM estimated SE) * Options for covariate observation model ("rw" vs. "ar1") * Options for recruitment link ("none" vs. "controlling-linear" with lag-0 for large copepods, lag-1 for small copepods, and lag-1 for larval temperature duration) * No attempts were made to fit polynomial effects although this is possible in WHAM Short story: * Models with covariates input on the log scale generally converged * Models with WHAM estimated covariate SE ("est_1") generally converged * Under the above conditions, most models with and without the recruitment link converged for all covariates * Models with the Jan-Jun (Spring) large copepods covariate also converged with input as millions of cells and VAST estimated SE --- ## Overview of results shown for each model .pull-left[ Summary table comparing across models with and without recruitment covariates turned on Diagnostics shown: * WHAM's fit to the covariate time series * One step ahead residuals for WHAM's fit * Estimated time varying recruitment scaling parameter with N age 1 Finally, recruitment sigma compared with base, covariate beta with CI ] .pull-right[ The main diagnostics we used to determine if the model was improved by covariates were: * model converged and Hessian matrix invertable * dAIC lowest * recruitment sigma reduced (how much?) * estimated covariate effect CI does not include 0 * direction of covariate effect is sensible ] --- ## Results: Spring large copepods covariate: model summary *Models with no covariates had slightly better AIC than comparable models with recruitment links* ``` Model ecov_process ecov_how ecovdat conv pdHess NLL dAIC AIC rho_R rho_SSB rho_Fbar m10 ar1 none logmean-est_1 TRUE TRUE -1793.611 0 -3325.2 0.8684 0.5901 -0.2314 m14 ar1 controlling-lag-0-linear logmean-est_1 TRUE TRUE -1794.325 0.6 -3324.6 0.8999 0.5844 -0.2296 m2 rw none logmean-est_1 TRUE TRUE -1790.837 3.5 -3321.7 0.8683 0.5901 -0.2314 m6 rw controlling-lag-0-linear logmean-est_1 TRUE TRUE -1791.227 4.7 -3320.5 0.8844 0.5846 -0.23 m11 ar1 none meanmil-logsigmil TRUE TRUE -1509.570 566.1 -2759.1 0.8682 0.5901 -0.2314 m15 ar1 controlling-lag-0-linear meanmil-logsigmil TRUE TRUE -1510.349 566.5 -2758.7 0.9081 0.592 -0.2379 m3 rw none meanmil-logsigmil TRUE TRUE -1506.709 569.8 -2755.4 0.8594 0.5907 -0.2362 m7 rw controlling-lag-0-linear meanmil-logsigmil TRUE TRUE -1507.424 570.4 -2754.8 0.8104 0.594 -0.2448 ``` --- ## Results: Spring large copepods covariate: logscale ar1 diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Spring large copepods covariate: logscale ar1 recruitment .pull-left[  ] .pull-right[  ] Without covariate, recruitment variance is 0.823, and with is 0.797; lgCopeSpring2 beta_1 is -0.407, CI -1.063, 0.25 --- ## Results: Spring large copepods covariate: logscale rw diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Spring large copepods covariate: logscale rw recruitment .pull-left[  ] .pull-right[  ] Without covariate, recruitment variance is 0.823, and with is 0.804; lgCopeSpring2 beta_1 is -0.45, CI -1.438, 0.538 --- ## Results: Spring large copepods covariate: natural scale ar1 diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Spring large copepods covariate: natural scale ar1 recruitment .pull-left[  ] .pull-right[  ] Without covariate, recruitment variance is 0.823, and with is 0.791; lgCopeSpring2 beta_1 is -4.5\times 10^{-4}, CI -0.00128, 3.8\times 10^{-4} --- ## Results: Spring large copepods covariate: natural scale rw diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Spring large copepods covariate: natural scale rw recruitment .pull-left[  ] .pull-right[  ] Without covariate, recruitment variance is 0.823, and with is 0.793; lgCopeSpring2 beta_1 is -4.3\times 10^{-4}, CI -0.00126, 4\times 10^{-4} --- ## Results: Fall small copepods covariate: model summary *Models with no covariates had slightly better AIC than the ar1 model with recruitment links* ``` Model ecov_process ecov_how ecovdat conv pdHess NLL dAIC AIC rho_R rho_SSB rho_Fbar m10 ar1 none logmean-est_1 TRUE TRUE -1797.177 0 -3332.4 0.8682 0.5901 -0.2314 m2 rw none logmean-est_1 TRUE TRUE -1796.175 0.1 -3332.3 0.8682 0.5901 -0.2314 m14 ar1 controlling-lag-1-linear logmean-est_1 TRUE TRUE -1797.931 0.5 -3331.9 0.8486 0.6001 -0.2333 m6 rw controlling-lag-1-linear logmean-est_1 TRUE FALSE -1794.370 --- --- --- --- --- ``` --- ## Results: Fall small copepods covariate: logscale ar1 diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Fall small copepods covariate: logscale ar1 recruitment .pull-left[  ] .pull-right[  ] Without covariate, recruitment variance is 0.823, and with is 0.79; smcopeFall2 beta_1 is -1.013, CI -2.715, 0.689 --- ## Results: Sep-Feb small copepods in herring larval area covariate: model summary *Model with ar1 covariate had slightly better AIC than models without recruitment links* ``` Model ecov_process ecov_how ecovdat conv pdHess NLL dAIC AIC rho_R rho_SSB rho_Fbar m14 ar1 controlling-lag-1-linear logmean-est_1 TRUE TRUE -1791.900 0 -3319.8 0.814 0.5966 -0.2327 m2 rw none logmean-est_1 TRUE TRUE -1789.657 0.5 -3319.3 0.8683 0.5901 -0.2314 m10 ar1 none logmean-est_1 TRUE TRUE -1790.469 0.9 -3318.9 0.8683 0.5901 -0.2314 m6 rw controlling-lag-1-linear logmean-est_1 TRUE FALSE -1785.746 --- --- --- --- --- ``` --- ## Results: Sep-Feb small copepods in herring larval area covariate: logscale ar1 diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Sep-Feb small copepods in herring larval area covariate: logscale ar1 recruitment .pull-left[  ] .pull-right[  ] Without covariate, recruitment variance is 0.823, and with is 0.77; smcopeSepFeb2 beta_1 is -0.85, CI -1.883, 0.182 --- ## Alternate: Sep-Feb small copepods in fall herring survey strata covariate *Both rw and ar1 models worked with covariates, rw better fit* ``` Model ecov_process ecov_how ecovdat conv pdHess NLL dAIC AIC rho_R rho_SSB rho_Fbar m6 rw controlling-lag-1-linear logmean-est_1 TRUE TRUE -1791.981 0.0 -3322.0 0.7592 0.6023 -0.2327 m14 ar1 controlling-lag-1-linear logmean-est_1 TRUE TRUE -1792.464 1.1 -3320.9 0.7943 0.5971 -0.2329 m2 rw none logmean-est_1 TRUE TRUE -1790.196 1.6 -3320.4 0.8682 0.5901 -0.2314 m10 ar1 none logmean-est_1 TRUE TRUE -1790.734 2.5 -3319.5 0.8683 0.5901 -0.2314 ``` --- ## Results: Sep-Feb small copepods in herring larval area covariate: logscale rw diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Sep-Feb small copepods in herring larval area covariate: logscale rw recruitment .pull-left[  ] .pull-right[  ] Without covariate, recruitment variance is 0.823, and with is 0.762; smcopeSepFeb2 beta_1 is -1.01, CI -2.11, 0.089 --- ## Results: Sep-Feb small copepods in herring larval area covariate: logscale ar1 diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Sep-Feb small copepods in herring larval area covariate: logscale ar1 recruitment .pull-left[  ] .pull-right[  ] Without covariate, recruitment variance is 0.823, and with is 0.757; smcopeSepFeb2 beta_1 is -0.976, CI -2.044, 0.093 --- ## Results: Both Spring large copepods and Sep-Feb small copepods in herring larval area: model summary *Models with and without ar1 covariates linked to recruitment had the same AIC* ``` Model ecov_process ecov_how ecovdat conv pdHess NLL dAIC AIC rho_R rho_SSB rho_Fbar m2 ar1 none logmean-est_1 TRUE TRUE -1771.328 0.0 -3272.7 0.8682 0.5901 -0.2314 m4 ar1 controlling-lag-0/1-linear logmean-est_1 TRUE TRUE -1773.347 0.0 -3272.7 0.8529 0.5922 -0.2313 m1 rw none logmean-est_1 TRUE TRUE -1767.742 3.2 -3269.5 0.8683 0.5901 -0.2314 m3 rw controlling-lag-0/1-linear logmean-est_1 TRUE TRUE -1769.627 3.4 -3269.3 0.8001 0.5976 -0.2321 ``` --- ## Results: Both Spring large copepods and Sep-Feb small copepods: logscale rw diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Both Spring large copepods and Sep-Feb small copepods: logscale rw diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Both Spring large copepods and Sep-Feb small copepods: logscale rw recruitment .pull-left[  ] .pull-right[  ] Without covariates, recruitment variance is 0.823, and with is 0.755; smcopeSepFeb2 beta_1 is -0.88, CI -1.929, 0.169 ??? - --- ## Results: Duration of optimal larval temperature, Sept-Dec: model summary ``` Model ecov_process ecov_how ecovdat conv pdHess NLL dAIC AIC rho_R rho_SSB rho_Fbar m4 rw controlling-lag-1-linear logmean-est_1 TRUE TRUE -1827.543 0 -3393.1 0.6543 0.5984 -0.2323 m3 rw none logmean-est_1 TRUE TRUE -1826.370 0.4 -3392.7 0.8682 0.5901 -0.2314 m8 ar1 controlling-lag-1-linear logmean-est_1 TRUE TRUE -1826.627 3.8 -3389.3 0.6714 0.598 -0.2324 m7 ar1 none logmean-est_1 TRUE TRUE -1825.511 4.1 -3389 0.8682 0.5901 -0.2314 m2 rw controlling-lag-1-linear mean-est_1 TRUE TRUE -1650.465 354.2 -3038.9 0.6473 0.5976 -0.2321 m1 rw none mean-est_1 TRUE TRUE -1649.265 354.6 -3038.5 0.8594 0.5907 -0.2362 m6 ar1 controlling-lag-1-linear mean-est_1 TRUE TRUE -1649.607 357.9 -3035.2 0.6643 0.5971 -0.2322 m5 ar1 none mean-est_1 TRUE FALSE -1644.890 --- --- --- --- --- ``` --- ## Results: Duration of optimal larval temperature, Sept-Dec: logscale rw diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Duration of optimal larval temperature, Sept-Dec: logscale rw recruitment .pull-left[  ] .pull-right[  ] Without covariate, recruitment variance is 0.823, and with is 0.79; LarvalTempDuration beta_1 is 2.066, CI -0.657, 4.789 --- ## Results: Duration of optimal larval temperature, Sept-Dec: logscale ar1 diagnostics .pull-left[  ] .pull-right[  ] --- ## Results: Duration of optimal larval temperature, Sept-Dec: logscale ar1 recruitment .pull-left[  ] .pull-right[  ] Without covariate, recruitment variance is 0.823, and with is 0.791; LarvalTempDuration beta_1 is 2.032, CI -0.711, 4.775 --- ## Sensitivity: Remove NAA RE, add September - December duration of larval optimal temperature: model summary *Both rw and ar1 models worked with covariates, rw better fit* ``` Model ecov_process ecov_how ecovdat conv pdHess NLL dAIC AIC rho_R rho_SSB rho_Fbar m4 rw controlling-lag-1-linear logmean-est_1 TRUE TRUE -1762.281 0 -3266.6 1.6108 0.3339 -0.121 m8 ar1 controlling-lag-1-linear logmean-est_1 TRUE TRUE -1761.214 4.2 -3262.4 1.6872 0.3341 -0.1213 m3 rw none logmean-est_1 TRUE TRUE -1748.086 26.4 -3240.2 3.5003 0.3664 -0.1424 m7 ar1 none logmean-est_1 TRUE TRUE -1747.227 30.1 -3236.5 3.5003 0.3664 -0.1424 m2 rw controlling-lag-1-linear mean-est_1 TRUE TRUE -1585.263 354.1 -2912.5 1.5766 0.3325 -0.1206 m1 rw none mean-est_1 TRUE TRUE -1570.980 380.6 -2886 3.5014 0.3662 -0.1421 m5 ar1 none mean-est_1 TRUE TRUE -1570.163 384.3 -2882.3 3.5003 0.3673 -0.1437 m6 ar1 controlling-lag-1-linear mean-est_1 TRUE FALSE -1584.188 --- --- --- --- --- ``` --- ## Duration of optimal larval temperature, Sept-Dec: logscale rw diagnostics .pull-left[  ] .pull-right[  ] --- ## Duration of optimal larval temperature, Sept-Dec: logscale rw recruitment .pull-left[  ] .pull-right[  ] Without covariate, recruitment variance is 0.833, and with is 0.503; LarvalTempDuration beta_1 is 5.122, CI 2.709, 7.535 --- # Discussion The duration of optimal larval temperature covariate resulted in slightly better model fits and reduced recruitment variability relative to the model without covariates. The effect of optimal larval temperature on recruitment was postive, as hypothesized, with fewer days of optimal temperature resulting in a lower recruitment scaling parameter. However, the confidence interval of the effect included 0. While some of the zooplankton time series also resulted in slightly better model fits and reduced recruitment variability relative to the model without covariates, the effects of zooplankton were negative on recruitment. This relationship is opposite the hypothesized relationship between herring and food, which was expected to be positive. The sensitivity run removing NAA RE resulted in a much stronger impact of the optimal larval temperature on recruitment. # Thoughts? --- ## Thank you! References .contrib[ <a name=bib-collette_bigelow_2002></a>[Collette, B. B. et al.](#cite-collette_bigelow_2002) (2002). _Bigelow and Schroeder's Fishes of the Gulf of Maine, Third Edition_. 3rd ed. edition. Washington, DC: Smithsonian Books. ISBN: 978-1-56098-951-6. <a name=bib-stock_woods_2021></a>[Stock, B. C. et al.](#cite-stock_woods_2021) (2021). "The Woods Hole Assessment Model (WHAM): A general state-space assessment framework that incorporates time- and age-varying processes via random effects and links to environmental covariates". En. In: _Fisheries Research_ 240, p. 105967. ISSN: 0165-7836. DOI: [10.1016/j.fishres.2021.105967](https://doi.org/10.1016%2Fj.fishres.2021.105967). URL: [https://www.sciencedirect.com/science/article/pii/S0165783621000953](https://www.sciencedirect.com/science/article/pii/S0165783621000953) (visited on May. 26, 2021). ] .footnote[ Slides available at https://noaa-edab.github.io/presentations Contact: <Sarah.Gaichas@noaa.gov> ]