A comparative study of spatial interpolation methods fordetermining fishery resources density in the Yellow Sea

Spatial interpolation is a common tool used in the study of fishery ecology, especially for the construction of ecosystem models. To develop an appropriate interpolation method of determining fishery resources density in the Yellow Sea, we tested four frequently used methods, including inverse distance weighted interpolation(IDW), global polynomial interpolation(GPI), local polynomial interpolation(LPI) and ordinary kriging(OK).A cross-validation diagnostic was used to analyze the efficacy of interpolation, and a visual examination was conducted to evaluate the spatial performance of the different methods. The results showed that the original data were not normally distributed. A log transformation was then used to make the data fit a normal distribution. During four survey periods, an exponential model was shown to be the best semivariogram model in August and October 2014, while data from January and May 2015 exhibited the pure nugget effect.Using a paired-samples t test, no significant differences(P>0.05) between predicted and observed data were found in all four of the interpolation methods during the four survey periods. Results of the cross-validation diagnostic demonstrated that OK performed the best in August 2014, while IDW performed better during the other three survey periods. The GPI and LPI methods had relatively poor interpolation results compared to IDW and OK. With respect to the spatial distribution, OK was balanced and was not as disconnected as IDW nor as overly smooth as GPI and LPI, although OK still produced a few 'bull's-eye' patterns in some areas.However, the degree of autocorrelation sometimes limits the application of OK. Thus, OK is highly recommended if data are spatially autocorrelated. With respect to feasibility and accuracy, we recommend IDW to be used as a routine interpolation method. IDW is more accurate than GPI and LPI and has a combination of desirable properties, such as easy accessibility and rapid processing.

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),p H,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 p H 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.