Using extreme gradient boosting (XGBoost) to evaluate the importance of a suite of environmental variables and to predict recruitment of young-of-the-year spotted seatrout in Florida

Abstract

AbstractEnvironmental factors strongly influence the success of juvenile fish recruitment and productivity, but species-specific environment-recruitment relationships have eluded researchers for decades. Most likely, this is because the environment-recruitment relationship is nonlinear, there are multi-level interactions between factors, and environmental variability may differentially affect recruitment among populations due to spatial heterogeneity. Identifying the most influential environmental variables may result in more accurate predictions of future recruitment and productivity of managed species. Here, gradient tree boosting was implemented using XGBoost to identify the most important predictors of recruitment for six estuary populations of spotted seatrout (Cynoscion nebulosus), an economically valuable marine resource in Florida. XGBoost, a machine learning method for regression and classification, was employed because it inherently models variable interactions and seamlessly deals with multi-collinearity, both of which are common features of ecological datasets. Additionally, XGBoost operates at a speed faster than many other gradient boosting algorithms due to a regularization factor and parallel computing functionality. In this application of XGBoost, the results indicate that the abundance of pre-recruit, juvenile spotted seatrout in spatially distinct estuaries is influenced by nearly the same set of environmental predictors. But perhaps of greater importance is that the results of this study show that this algorithm is highly effective at predicting species abundance and identifying important environmental factors (i.e. predictors of recruitment). It is strongly encouraged that future research explore the applicability of the XGBoost algorithm to other topics in marine and fisheries science and compare its performance to that of other statistical methods.