Comparing different spatial interpolation methods to predict the distribution of fishes: A case study of Coilia nasus in the Changjiang River Estuary

Spatial-temporal distribution of marine fishes is strongly influenced by environmental factors. To obtain a more continuous distribution of these variables usually measured by stationary sampling designs, spatial interpolation methods (SIMs) is usually used. However, different SIMs may obtain varied estimation values with significant differences, thus affecting the prediction of fish spatial distribution. In this study, different SIMs were used to obtain continuous environmental variables (water depth, water temperature, salinity, dissolved oxygen (DO), pH, chlorophyll a and chemical oxygen demand (COD)) in the Changjiang River Estuary (CRE), including inverse distance weighted (IDW) interpolation, ordinary Kriging (OK) (semivariogram model: exponential (OKE), Gaussian (OKG) and spherical (OKS)) and radial basis function (RBF) (regularized spline function (RS) and tension spline function (TS)). The accuracy and effect of SIMs were cross-validated, and two-stage generalized additive model (GAM) was used to predict the distribution of Coilia nasus from 2012 to 2014 in CRE. DO and COD were removed before model prediction due to their autocorrelation coefficient based on variance inflation factors analysis. Results showed that the estimated values of environmental variables obtained by the different SIMs differed (i.e., mean values, range etc.). Cross-validation revealed that the most suitable SIMs of water depth and chlorophyll a was IDW, water temperature and salinity was RS, and pH was OKG. Further, different interpolation results affected the predicted spatial distribution of Coilia nasus in the CRE. The mean values of the predicted abundance were similar, but the differences between and among the maximum value were large. Studies showed that different SIMs can affect estimated values of the environmental variables in the CRE (especially salinity). These variations further suggest that the most applicable SIMs to each variable will also differ. Thus, it is necessary to take these potential impacts into consideration when studying the relationship between the spatial distribution of fishes and environmental changes in the CRE.