Building a base Ecosim model[footnote]See Bentley et al. 2024 Supplementary Data for details about the model construction[\/footnote]<\/h2>\r\n
Ecosim simulations for Anchovy Bay were created by adding temporal trends to fishing effort and adjusting vulnerability multipliers. Simulated fishing effort trends \u00a0reflected trends often seen in reality:<\/p>\r\n\r\n
- \r\n \t
- sealers fishing effort followed an exponential decline as may be expected in response to conservation efforts\/policy,<\/li>\r\n \t
- trawlers fishing effort followed an exponential decline under the assumption that whitefish (cod and whiting) stocks have been overexploited, leading to reductions in effort to encourage stock recovery,<\/li>\r\n \t
- seiners and bait boat effort followed a slight linear increase in response to growing demand, and<\/li>\r\n \t
- shrimpers effort increased assuming fishers shifted their target species to shrimp following reduced opportunities to catch white fish.<\/li>\r\n<\/ul>\r\nVulnerability multipliers (kij<\/sub><\/em>) \u00a0were adjusted following ecological assumptions and assumptions linked to the fishing effort trajectories. To distinguish between scenarios more easily, predator vulnerability multipliers will hereafter be denoted as kj<\/sub><\/em>, while predator-prey vulnerability will remain as kij<\/sub><\/em>. For predator \u00a0estimates, a mix of high, low and default \u00a0estimates were applied. For groups which were assumed to be overexploited, \u00a0<\/em>values were estimated using the \"Estimate Vulnerabilities\" interface. For the predator-prey estimates, the Ecosim sensitivity search was used to identify the 10 most sensitive predator\/prey kij<\/sub><\/em>\u00a0parameters, which were then adjusted to ensure a range of high and low kij<\/sub><\/em>\u00a0estimates were included.\r\n
For the purpose of this exercise, these two simulations (one with predator kj<\/sub><\/em>\u00a0<\/em>and one with predator-prey kij<\/sub><\/em>) were viewed as perfect representations of their ecosystems, i.e., the biomass and catch simulations were \"real observations\" \u00a0driven by the \"true\" vulnerability multipliers. The aim of the following exercise was to test whether, when using these \"real observations\" as calibration time series, the \"true\" vulnerability multipliers would reemerge, and whether the addition of noise to the \"real observations\" had any impact on the emerging vulnerability multipliers. Biomass and catch simulations were extracted from Ecosim and four scenarios for observation data quality were prepared: noise (random noise, normally distributed around the mean (true) biomass trend to represent observation error) was added to the calibration time series with coefficients of variation (CV<\/em>) of 0 (no noise) 0.1, 0.3, and 0.5.<\/p>\r\n\r\n
Predator vulnerability multipliers<\/h2>\r\n
Vulnerability multipliers were reset to the default value of 2; fishing dynamics were not changed from those used to produce the \"real observations.\" The exported biomass and catch time series were used as calibration time series to estimate predator vulnerability multipliers for the functional groups seals, cod, whiting, shrimp, benthos, and zooplankton using the manual stepwise fitting interface. kj<\/sub><\/em>\u00a0values for groups which had values of 2 in the initial model were not altered.<\/p>\r\nFigure 1 shows how kj<\/sub><\/em>\u00a0parameters emerged after model calibration, and how this altered functional group carrying capacities in the absence of fishing and FMSY<\/sub><\/em> estimates. kj<\/sub><\/em>\u00a0<\/em>values which emerged when estimated using the calibration time series with no noise were similar to the \"true\" \u00a0<\/em>parameters (<\/strong>Figure 1a). Adding noise to the calibration time series led to divergence between the estimated kj<\/sub><\/em> values and the \"true\" parameters, highlighting the impact data quality can have on the fitting procedure and thus stressing the importance of evaluating the suitability of time series before using them to drive model calibration. The variability in kj<\/sub><\/em>\u00a0<\/em>re-emergence under the four data quality scenarios was also unique to specific functional groups, for example: kj<\/sub><\/em>\u00a0estimates for cod showed greater re-emergence accuracy (or consistency) when compared to other functional groups. Cod is highly connected within the food web (i.e., cod is an opportunistic predator which is also preyed upon by higher trophic levels), therefore vulnerability multipliers which improve the model fit tend to be more constrained due to their potential to have large cascading impacts on the wider food web.\u00a0 In addition, cod also experienced a period of collapse followed by recovery, which provides much needed contrast for the model to reliably estimate the vulnerability multipliers.\r\n\r\n
![\"\"](\"https:\/\/pressbooks.bccampus.ca\/ewemodel\/wp-content\/uploads\/sites\/1902\/2023\/11\/Bentley-et-al.-2024-Figure-5.png\")
\r\n\r\nFigure 1. Estimation and impact of predator vulnerability multipliers (kj<\/sub><\/em>). The Anchovy Bay ecosystem model was calibrated against generated time series with incremental coefficients of variation (CV<\/em>) to identify the impact of time series quality on (a) kj<\/sub><\/em> re-emergence and how kj<\/sub><\/em> estimates impacted (b) functional group carrying capacities in the absence of fishing and (c) estimates of relative fishing mortality consistent with achieving Maximum Sustainable Yield (FMSY<\/sub><\/em>).<\/strong>\r\n\r\nFunctional group carrying capacities and estimates of F<\/em>MSY<\/sub> were impacted by emerging \u00a0values (Figure 1b and 1c). Carrying capacities from scenarios with \u00a0\u00a0parameters calibrated against data with no noise were most similar to those achieved with the \"true\" \u00a0parameters (Figure 1b), with dissimilarity generally increasing with the addition of noise to the calibration data. The importance of acknowledging the impact of \u00a0estimates beyond model fit is demonstrated with the resulting F<\/em>MSY<\/sub> estimates: relative changes to F<\/em>MSY<\/sub> estimates mirrored the deviations of estimated \u00a0values relative to the \"true\" \u00a0values (Figure 1c). Increases in \u00a0values led to decreases in F<\/em>MSY<\/sub>, while decreases in \u00a0values led to increases in F<\/em>MSY<\/sub> This is because higher \u00a0values enable groups to recover faster with the cessation of fishing and reach a higher carrying capacity, but they also decrease stock resilience to increases in F (functional groups decline faster and more severely if you increase their ).\u00a0 It is worth noting that where differences between true and estimated F<\/em>MSY<\/sub>\u00a0occurred, they were not proportional to the difference in true and estimated vulnerability multipliers (i.e., large changes in k<\/em>ij<\/sub> do not result in equally large changes to F<\/em>MSY<\/sub>).\r\n\r\nPredator-prey vulnerability multipliers<\/span>\r\n\r\nSimilar to the predator scenario, vulnerability multipliers were reset to the default of 2, and the exported biomass and catch time series (generated with \"true\" predator-prey vulnerability multipliers) were used as calibration time series to estimate predator-prey vulnerability multipliers. Predator-prey values for the ten most sensitive predator\/prey parameters were estimated using the manual stepwise fitting interface. Figure 2\u00a0<\/strong>shows how \u00a0parameters emerged and how this altered functional group carrying capacities and FMSY estimates.<\/p>\r\n
In comparison to the emergence of predator vulnerabilities, the emergence of predator-prey vulnerabilities was less constrained with examples of poor \u00a0re-emergence accuracy across all calibration data scenarios (<\/strong>Figure 2a).\u00a0 Functional group carrying capacities showed higher dissimilarity from their baseline when compared to predator simulations and their baseline (Figure 2b). Carrying capacity dissimilarity increased with the addition of noise to the calibration data, however simulations with no\/low noise were notably more dissimilar when estimating predator-prey vulnerabilities as opposed to predator vulnerabilities (Figure 2b) which is due to the greater differences in kij<\/sub>\u00a0<\/em>estimates.<\/p>\r\nRelative FMSY<\/sub><\/em> estimates, influenced by predator-prey k<\/i>ij<\/sub>\u00a0<\/i>values, showed higher dissimilarity from their baseline (Figure 2C) when compared to FMSY<\/sub><\/em> estimates influenced by predator kj<\/sub> <\/em>values (Figure 1C). The links between predator-prey kij<\/sub> <\/em>values and FMSY<\/sub><\/em> are less obvious than the links between predator kj<\/sub> <\/em>values and F<\/em>MSY<\/sub> due to the more complex interaction-specific consumption limits. This is particularly true for groups with mixed diets (e.g., cod, whiting, seals, and mackerel) while links between predator-prey \u00a0values and the FMSY<\/sub><\/em> estimates for groups, which are heavily dependent on a single prey group were observed for anchovy (FMSY<\/sub><\/em> mirrors the anchovy\/zooplankton kij<\/sub><\/em>\u00a0<\/em>estimates) and shrimp (FMSY<\/sub><\/em> mirrors the shrimp\/benthos kij<\/sub><\/em>\u00a0<\/em>estimates).\r\n\r\n<\/div>\r\n
![\"\"](\"https:\/\/pressbooks.bccampus.ca\/ewemodel\/wp-content\/uploads\/sites\/1902\/2023\/11\/Bentley-et-al.-2024.-Figure-6.png\")
<\/div>\r\nFigure 2. Estimation and impact of predator-prey vulnerability multipliers (kij<\/sub><\/em>). The Anchovy Bay ecosystem model was calibrated against generated time series with incremental coefficients of variation (CV<\/em>) to identify the impact of time series quality on (a) kij<\/sub><\/em> re-emergence and how kij<\/sub><\/em> estimates impacted (b) functional group carrying capacities in the absence of fishing and (c) estimates of relative fishing mortality consistent with achieving Maximum Sustainable Yield (FMSY<\/sub><\/em>).<\/strong><\/div>\r\nAttribution <\/strong>This chapter is based on Bentley JW, Chagaris D, Coll M, Heymans JJ, Serpetti N, Walters CJ and Christensen V. 2024. Calibrating ecosystem models to support marine Ecosystem-based Management. ICES Journal of Marine Science, https:\/\/doi.org\/10.1093\/icesjms\/fsad213<\/a>. Adapted based on CC BY License. Rather than citing this chapter, please cite the source.<\/div>","rendered":"The Anchovy Bay ecosystem model that is used to describe and test EwE scenarios throughout this text book was used to investigate how vulnerability multipliers emerge (and whether they re-emerge through fitting) and how this process is influenced by:<\/p>\n
- \n
- noise in the calibration data, and<\/li>\n
- the chosen approach for estimating vulnerability multipliers: “predator” or “predator-prey” vulnerability multipliers.<\/li>\n<\/ol>\n
We investigated the impact of emerging vulnerability multipliers on biomass and catch simulations and estimates of fishing mortality consistent with maximum sustainable yield (FMSY<\/sub><\/em>).<\/p>\n
Building a base Ecosim model[1]<\/sup><\/a><\/h2>\n
Ecosim simulations for Anchovy Bay were created by adding temporal trends to fishing effort and adjusting vulnerability multipliers. Simulated fishing effort trends \u00a0reflected trends often seen in reality:<\/p>\n
- \n
- sealers fishing effort followed an exponential decline as may be expected in response to conservation efforts\/policy,<\/li>\n
- trawlers fishing effort followed an exponential decline under the assumption that whitefish (cod and whiting) stocks have been overexploited, leading to reductions in effort to encourage stock recovery,<\/li>\n
- seiners and bait boat effort followed a slight linear increase in response to growing demand, and<\/li>\n
- shrimpers effort increased assuming fishers shifted their target species to shrimp following reduced opportunities to catch white fish.<\/li>\n<\/ul>\n
Vulnerability multipliers (kij<\/sub><\/em>) \u00a0were adjusted following ecological assumptions and assumptions linked to the fishing effort trajectories. To distinguish between scenarios more easily, predator vulnerability multipliers will hereafter be denoted as kj<\/sub><\/em>, while predator-prey vulnerability will remain as kij<\/sub><\/em>. For predator \u00a0estimates, a mix of high, low and default \u00a0estimates were applied. For groups which were assumed to be overexploited, \u00a0<\/em>values were estimated using the “Estimate Vulnerabilities” interface. For the predator-prey estimates, the Ecosim sensitivity search was used to identify the 10 most sensitive predator\/prey kij<\/sub><\/em>\u00a0parameters, which were then adjusted to ensure a range of high and low kij<\/sub><\/em>\u00a0estimates were included.<\/p>\n
For the purpose of this exercise, these two simulations (one with predator kj<\/sub><\/em>\u00a0<\/em>and one with predator-prey kij<\/sub><\/em>) were viewed as perfect representations of their ecosystems, i.e., the biomass and catch simulations were “real observations” \u00a0driven by the “true” vulnerability multipliers. The aim of the following exercise was to test whether, when using these “real observations” as calibration time series, the “true” vulnerability multipliers would reemerge, and whether the addition of noise to the “real observations” had any impact on the emerging vulnerability multipliers. Biomass and catch simulations were extracted from Ecosim and four scenarios for observation data quality were prepared: noise (random noise, normally distributed around the mean (true) biomass trend to represent observation error) was added to the calibration time series with coefficients of variation (CV<\/em>) of 0 (no noise) 0.1, 0.3, and 0.5.<\/p>\n
Predator vulnerability multipliers<\/h2>\n
Vulnerability multipliers were reset to the default value of 2; fishing dynamics were not changed from those used to produce the “real observations.” The exported biomass and catch time series were used as calibration time series to estimate predator vulnerability multipliers for the functional groups seals, cod, whiting, shrimp, benthos, and zooplankton using the manual stepwise fitting interface. kj<\/sub><\/em>\u00a0values for groups which had values of 2 in the initial model were not altered.<\/p>\n
Figure 1 shows how kj<\/sub><\/em>\u00a0parameters emerged after model calibration, and how this altered functional group carrying capacities in the absence of fishing and FMSY<\/sub><\/em> estimates. kj<\/sub><\/em>\u00a0<\/em>values which emerged when estimated using the calibration time series with no noise were similar to the “true” \u00a0<\/em>parameters (<\/strong>Figure 1a). Adding noise to the calibration time series led to divergence between the estimated kj<\/sub><\/em> values and the “true” parameters, highlighting the impact data quality can have on the fitting procedure and thus stressing the importance of evaluating the suitability of time series before using them to drive model calibration. The variability in kj<\/sub><\/em>\u00a0<\/em>re-emergence under the four data quality scenarios was also unique to specific functional groups, for example: kj<\/sub><\/em>\u00a0estimates for cod showed greater re-emergence accuracy (or consistency) when compared to other functional groups. Cod is highly connected within the food web (i.e., cod is an opportunistic predator which is also preyed upon by higher trophic levels), therefore vulnerability multipliers which improve the model fit tend to be more constrained due to their potential to have large cascading impacts on the wider food web.\u00a0 In addition, cod also experienced a period of collapse followed by recovery, which provides much needed contrast for the model to reliably estimate the vulnerability multipliers.<\/p>\n
<\/p>\nFigure 1. Estimation and impact of predator vulnerability multipliers (kj<\/sub><\/em>). The Anchovy Bay ecosystem model was calibrated against generated time series with incremental coefficients of variation (CV<\/em>) to identify the impact of time series quality on (a) kj<\/sub><\/em> re-emergence and how kj<\/sub><\/em> estimates impacted (b) functional group carrying capacities in the absence of fishing and (c) estimates of relative fishing mortality consistent with achieving Maximum Sustainable Yield (FMSY<\/sub><\/em>).<\/strong><\/p>\n
Functional group carrying capacities and estimates of F<\/em>MSY<\/sub> were impacted by emerging \u00a0values (Figure 1b and 1c). Carrying capacities from scenarios with \u00a0\u00a0parameters calibrated against data with no noise were most similar to those achieved with the “true” \u00a0parameters (Figure 1b), with dissimilarity generally increasing with the addition of noise to the calibration data. The importance of acknowledging the impact of \u00a0estimates beyond model fit is demonstrated with the resulting F<\/em>MSY<\/sub> estimates: relative changes to F<\/em>MSY<\/sub> estimates mirrored the deviations of estimated \u00a0values relative to the “true” \u00a0values (Figure 1c). Increases in \u00a0values led to decreases in F<\/em>MSY<\/sub>, while decreases in \u00a0values led to increases in F<\/em>MSY<\/sub> This is because higher \u00a0values enable groups to recover faster with the cessation of fishing and reach a higher carrying capacity, but they also decrease stock resilience to increases in F (functional groups decline faster and more severely if you increase their ).\u00a0 It is worth noting that where differences between true and estimated F<\/em>MSY<\/sub>\u00a0occurred, they were not proportional to the difference in true and estimated vulnerability multipliers (i.e., large changes in k<\/em>ij<\/sub> do not result in equally large changes to F<\/em>MSY<\/sub>).<\/p>\n
Predator-prey vulnerability multipliers<\/span><\/p>\n
\nSimilar to the predator scenario, vulnerability multipliers were reset to the default of 2, and the exported biomass and catch time series (generated with “true” predator-prey vulnerability multipliers) were used as calibration time series to estimate predator-prey vulnerability multipliers. Predator-prey values for the ten most sensitive predator\/prey parameters were estimated using the manual stepwise fitting interface. Figure 2\u00a0<\/strong>shows how \u00a0parameters emerged and how this altered functional group carrying capacities and FMSY estimates.<\/p>\n
In comparison to the emergence of predator vulnerabilities, the emergence of predator-prey vulnerabilities was less constrained with examples of poor \u00a0re-emergence accuracy across all calibration data scenarios (<\/strong>Figure 2a).\u00a0 Functional group carrying capacities showed higher dissimilarity from their baseline when compared to predator simulations and their baseline (Figure 2b). Carrying capacity dissimilarity increased with the addition of noise to the calibration data, however simulations with no\/low noise were notably more dissimilar when estimating predator-prey vulnerabilities as opposed to predator vulnerabilities (Figure 2b) which is due to the greater differences in kij<\/sub>\u00a0<\/em>estimates.<\/p>\n
Relative FMSY<\/sub><\/em> estimates, influenced by predator-prey k<\/i>ij<\/sub>\u00a0<\/i>values, showed higher dissimilarity from their baseline (Figure 2C) when compared to FMSY<\/sub><\/em> estimates influenced by predator kj<\/sub> <\/em>values (Figure 1C). The links between predator-prey kij<\/sub> <\/em>values and FMSY<\/sub><\/em> are less obvious than the links between predator kj<\/sub> <\/em>values and F<\/em>MSY<\/sub> due to the more complex interaction-specific consumption limits. This is particularly true for groups with mixed diets (e.g., cod, whiting, seals, and mackerel) while links between predator-prey \u00a0values and the FMSY<\/sub><\/em> estimates for groups, which are heavily dependent on a single prey group were observed for anchovy (FMSY<\/sub><\/em> mirrors the anchovy\/zooplankton kij<\/sub><\/em>\u00a0<\/em>estimates) and shrimp (FMSY<\/sub><\/em> mirrors the shrimp\/benthos kij<\/sub><\/em>\u00a0<\/em>estimates).<\/p>\n<\/div>\n
<\/div>\nFigure 2. Estimation and impact of predator-prey vulnerability multipliers (kij<\/sub><\/em>). The Anchovy Bay ecosystem model was calibrated against generated time series with incremental coefficients of variation (CV<\/em>) to identify the impact of time series quality on (a) kij<\/sub><\/em> re-emergence and how kij<\/sub><\/em> estimates impacted (b) functional group carrying capacities in the absence of fishing and (c) estimates of relative fishing mortality consistent with achieving Maximum Sustainable Yield (FMSY<\/sub><\/em>).<\/strong><\/div>\nAttribution <\/strong>This chapter is based on Bentley JW, Chagaris D, Coll M, Heymans JJ, Serpetti N, Walters CJ and Christensen V. 2024. Calibrating ecosystem models to support marine Ecosystem-based Management. ICES Journal of Marine Science, https:\/\/doi.org\/10.1093\/icesjms\/fsad213<\/a>. Adapted based on CC BY License. Rather than citing this chapter, please cite the source.<\/div>\nMedia Attributions<\/h2>