Results
Demersal fish biomass was highest in the northern regions of the Northeast Pacific (Gulf of Alaska, Eastern Bering Sea and Aleutian Islands) and Northeast Atlantic (Barents Sea and Norwegian Sea) (Figure 1a). Conversely, demersal fish biomass was lowest in the Gulf of Mexico and temperate regions of the North Atlantic (Baltic Sea, southern North Sea, Gulf of Saint Lawrence).
At the ecoregion scale, Pearson correlations between demersal fish biomass and temperature (Figure 1b, r = -0.54), fishing exploitation (Figure 1c, r = -0.34), net primary production (Figure 1d, r = -0.40) and detrital bottom flux (r = -0.33) were negative, while depth correlated positively with biomass (r = 0.23). Demersal fish biomass had no correlation with zooplankton production (r = 0.04) and mean trophic level (r = -0.04). Whether the correlations with biomass were direct effects of the predictor variable or indirect effects governed by other predictor variables were examined with the SEM.
Including all predictors resulted in a SEM that was too complex for the available number of observations at the ecoregion scale to assess goodness-of-fit (Figure S2.2). Hence, we simplified the full model by removing the detrital bottom flux, which had an insignificant relation with biomass in 6 out of 6 runs (2 spatial scales × 3 time periods). We also removed depth, which became irrelevant for the SEM network after removing the detrital bottom flux. The final model, including the remaining five predictors, had a meanΧ 2 of 6.02 (standard deviation from the 6 runs is 2.4) with 6 degrees of freedom, and p-values ranging between 0.21 and 0.88, indicating that our hypothesized causal structure is supported by the data (an insignificant result indicates good model fit).
Among the individual pathways, demersal fish biomass at the ecoregion and subdivision scale was negatively related to temperature, fishing exploitation and mean trophic level and positively related to zooplankton production (Figure 2). The pronounced spatial variation in demersal fish biomass was reasonably well explained (mean R2 = 0.59) with no clear spatial pattern in the residuals (Figure S2.3). The effects of temperature and fishing were almost equally strong (Figure S2.4). For most other pathways, the directionality conformed with the initial expectations (Figure 2 vs Figure S2.2). A partial effect size plot showed that demersal fish biomass is approximately twice as high with a decline in temperature from 15 to 5°C and a decline in exploitation rate from 0.3 to 0.03, whereas the effect of mean trophic level and zooplankton production on biomass were more variable (Figure 3).
Similar to the SEM analyses, the grid-cell analysis using wavelet-revised model regression showed a negative relationship between demersal fish biomass and fishing exploitation and temperature, while zooplankton had a positive relationship with biomass for all three time periods (Table 2). In contrast to the previous analysis, the detrital bottom flux, which was excluded in the SEM, had a mixed effect on biomass (positive in one period and negative in the two others). Mean trophic level was not part of the best candidate model in any of the time periods.
The best fit between observed and predicted demersal fish biomass with the trophodynamic model (i.e., eq. 1) was obtained with a trophic transfer efficiency of 0.075 and a Q 10temperature scaling of trophic transfer efficiency between 0.4 and 0.7 (Figure 4, S2.5), implying that trophic transfer declines with increasing temperature. Replacing the temperature dependent trophic transfer efficiency with a single mean value sharply reduced the R2 of the trophodynamic model from 0.66 to 0.42. Furthermore, replacing the exploitation rate in eq. 1 with a single mean value and refitting led to an R2 of 0.55. The results of the trophodynamic model are thus consistent with the SEM in suggesting temperature-linked trophodynamic effects, i.e., a trophic transfer efficiency decrease with increasing temperature, and exploitation rates as the primary drivers of demersal fish biomass across the range of systems considered.
Finally, we found no evidence that temporal changes in temperature have impacted demersal community biomass within ecoregions during the period 1980-2015 (Table 1). All models with and without a temperature term differed less than 2 AIC units. The models without the temperature term were therefore selected as best candidate as they have the fewest parameters.