Non‐parametric modeling reveals environmental effects on bluefin tuna recruitment in Atlantic, Pacific, and Southern Oceans

Abstract

Environment–recruitment relationships can be difficult to delineate with parametric statistical models and can be prone to misidentification. We use non‐parametric time‐series modeling which makes no assumptions about functional relationships between variables, to reveal environmental influences on early life stages of bluefin tuna and demonstrate improvement in prediction of subsequent recruitment. The influence of sea surface temperature, which has been previously associated with larval growth and survival, was consistently detected in recruitment time series of bluefin tuna stocks that spawn in the Mediterranean Sea, the North Pacific, and the Southern Ocean. Short time series for the Gulf of Mexico stock may have precluded a clear determination of environmental influences on recruitment fluctuations. Because the non‐parametric approach does not require specification of equations to represent system dynamics, predictive models can likely be developed that appropriately reflect the complexity of the ecological system under investigation. This flexibility can potentially overcome methodological challenges of specifying structural relationships between environmental conditions and fish recruitment. Consequently, there is potential for non‐parametric time series modeling to supplement traditional stock recruitment models for fisheries management.

Ecosystem‐based forecasts of recruitment in two menhaden species

“…These results join several other recent studies that forge inroads into adapting the EDM approach to practical fisheries management (Harford, Karnauskas, Walter, & Liu, ; Liu, Karnauskas, Zhang, Linton, & Porch, ; Ye, Beamish, et al., ). However, real forecast skill is invaluable to proactive management; without it, management can only be reactive.…”

“…The skill and error of the constant predictor (i.e., predict that JAI(t + 1) = JAI(t)) are included as dashed lines. MAE direction is reversed to correspond with the direction of ρ These results join several other recent studies that forge inroads into adapting the EDM approach to practical fisheries management (Harford, Karnauskas, Walter, & Liu, 2017;Liu, Karnauskas, Zhang, Linton, & Porch, 2017;Ye, Beamish, et al, 2015). However, real forecast skill is invaluable to proactive management; without it, management can only be reactive.…”

Environmental variability controls recruitment but with different drivers among spawning components in Gulf of St. Lawrence herring stocks

“…The recent accumulation of long time series of data on fish stock abundance and environmental conditions, coupled with the increased incentive to incorporate environmental variability into fisheries management, provides an opportunity to test how bottom-up and top-down processes influence recruitment variability. Moreover, while stock-recruitment relationships remain widely used in fisheries management even though they remain controversial (Cury, Fromentin, Figuet, & Bonhommeau, 2014), there is increasing evidence that environmental variability needs to be taken into account to predict recruitment and improve the management of commercially important pelagic fish stocks (Britten, Dowd, & Worm, 2016;Harford, Karnauskas, Walter, & Liu, 2017;White et al, 2016).…”

Quantitative argument for long-term ecological monitoring

“…These differences in predictability could partially reflect differences in the natural response times (e.g. generation times), leading to differences in the density of points on the attractor (Table 1) however, they could also reflect exogenous effects, such as environmental drivers, that are not captured in the abundance time series and may therefore need to be included explicitly in forecast models (see Dixon et al 1999, Deyle et al 2013, and Harford et al 2017 for examples). Finally, we note that in addition to leave-one-out cross validation, for the 23 taxa with the highest data availability, nearly identical out-of-sample prediction results are obtained with 2-fold cross validation.…”

“…Data-driven approaches where causal variables and functional relationships are determined empirically may offer a viable alternative to inductive equation-based approaches. For example, Sugihara et al (2012) introduce an EDM method for using time series to identify the causal drivers of ecosystem dynamics, and several others , Deyle et al 2013, Harford et al 2017 provide examples of incorporating these environmental effects into EDMs to forecast future ecosystem states -including apparently random events such as red tides (McGowan et al 2017). These approaches do not rely a priori on hypothesized equations but instead infer relationships deductively as they appear in the data.…”