Comparison of generalized additive model and boosted regression tree in predicting fish community diversity in the Yangtze River Estuary, China

Abstract

Yangtze River Estuary is the biggest estuarine ecosystem in the western Pacific Ocean. Evaluating fish community in this ecosystem can provide scientific basis for its restoration and mana-gement. Generalized additive model (GAM) and boosted regression tree (BRT) were built to examine the relationship between fish community diversity and environmental and spatio-temporal variables based on data collected during 2012-2014. Combined with linear regression analysis, a cross validation was used to evaluate the fitness and predictive performance of both models. We plotted the spatial distribution of fish community diversity and richness in each station of the Yangtze River Estuary in 2014. The results showed that salinity, pH and chlorophyll-a had the most contribution on diversity, while pH, dissolved oxygen and chlorophyll-a were the most contributive variables on richness. BRT models showed better fitness and lower prediction error than GAM models. In contrast to GAM models, BRT models could distinguish the fish community index in each station area with respect to the spatial prediction. The diversity index in external water was obviously greater than that in internal water. Meanwhile, the station at higher latitude had a higher diversity index in both external and internal water.