Comparative performance of modelling approaches for predicting fish species richness in the Yangtze River Estuary

Abstract

Listen

Translate article

Fish species richness (FSR) has been widely used as an indicator of both ecological variation and fish biodiversity. On global and local scales, FSR has been declining due to human impact. Species distribution models, such as conventional regression techniques and machine learning methods, have played a critical role in predicting the distribution of FSR in estuarine systems. The Yangtze River Estuary (YRE) is the largest estuarine ecosystem in the western Pacific Ocean. It is a muddy estuary providing many important ecosystem functioning services acting as spawning, feeding, and nursing grounds for a variety of valuable commercial species such as Coilia ectenes and Muraenesox cinereus. Despite this, the spatial distribution of FSR in the YRE has not been estimated. In this study, four modelling approaches (generalized linear model, generalized additive model, regression tree, and boosted regression tree) are applied to predict FSR in the YRE. Model performance was compared by evaluating the fit and predictive performance via deviation explained, root mean squared error, and cross-validation. Finally, the spatial distribution of FSR was estimated by the optimal model. It was shown that a regression tree approach outperformed the other methods. Additionally, regression tree modelling captured the overall distribution pattern of FSR in the YRE. The factors that appear to be contributing to the distribution pattern of FSR in the YRE included environmental conditions, fish migrations, and local physical characteristics (e.g. runoff and tide). The results from this study could provide guidance for the conservation planning and spatial management of the YRE.