Modelling the abundance of key aquatic species in a tropical Indian estuary using biotic and abiotic predictors

Abstract

Species distribution modelling (SDM) is a method that assists in linking the biotic and abiotic features of the ecosystems. In this study, we developed SDMs for total fish abundance, abundance of Escualosa thoracata (ET), Mugil cephalus (MC), Sardinella longiceps (SL) and Fenneropenaeus merguiensis (FM) based on biotic and abiotic variables using multivariate statistical techniques such as Regression tree (RT), Support Vector Machine (SVM), K-nearest neighbours (kNN), Gaussian process regression (GPR), Random Forest (RF), eXtreme Gradient Boosting (XGB), Cubist, and Quantile random forest (QRF). A combination of abiotic and biotic variables such as salinity, temperature, mud, turbidity, zooplankton, shell debris, sand, and benthos were identified as the significant predictors for the distribution of total fish abundance and species wise fish abundance. The model performance was evaluated based on coefficient of determination (R²), d-index, mean bias error (MBE), Nash–Sutcliffe efficiency (NSE) and root mean squared error (RMSE). A unique average ranking procedure was followed using all the model evaluation indicators to select a best model for a given season or species. The top performing models did not vary much between the seasons and species. We identified the following models performed best for different species respectively during postmonsoon and premonsoon seasons- Cubist and QRF for E. thoracata, RF and GLMNET for M. cephalus, SVM and QRF for S. longiceps, GPR and QRF for F. merguiensis. The major variables influenced the species distribution were benthic substratum (sand, mud, shell debris), chlorophyll-a and depth. Thus, it is evident that though the other variables also influence the distribution and abundance of the species in an estuarine bay, the benthic features, depth and morphometry can have significant effect on the species distribution. Moreover, the advanced multivariate statistical tools can aid in the prediction of the species distribution in the tropical estuaries.