Incorporation of optimal environmental signals in the prediction of fish recruitment using random forest algorithms

Abstract

The drivers of recruitment of selected Baltic sprat (Sprattus sprattus) and herring (Clupea harengus) stocks were investigated. Data on the interaction dynamics among fish species, the biological characteristics of the stocks, the biomass of the main predators, and the hydroclimatic environmental factors (Baltic Sea Index and sea surface temperature) were used in the analysis. The combination of random forest (Boruta algorithm) and “sliding window” approaches was tested on the simulated data and then used for the selection of relevant predictors and the optimal time window for real environmental variables. Sea surface temperature had a significant positive effect on the recruitment processes. Moreover, contrasting effects were observed in the mean Baltic Sea Index from two different time windows. The same environmental variable generated contrasting short-term and long-term effects on fish recruitment. This paper highlights the potential benefits of random forest and data mining applications for the incorporation of environmental factors in the assessment of stocks. The proposed analytical approach may be valuable for the investigations of complex environmental impacts in a broad range of ecological studies.