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

• Published in: 海洋学报：英文版 2016, Vol.35 (12), p.65-72
• Main Author: CHEN Yunlong SHAN Xiujuan JIN Xianshi YANG Tao DAI Fangqun YANG Dingtian
• Format: Article
• Language: English
• Subjects: density; fishery; interpolation; methods; resources; Sea; spatial; Yellow
• ISSN: 0253-505X; 1869-1099

Spatial interpolation is a common tool used in the study of fishery ecology, especially for the construction ofecosystem models. To develop an appropriate interpolation method of determining fishery resources densityin 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 wasconducted to evaluate the spatial performance of the different methods. The results showed that the originaldata were not normally distributed. A log transformation was then used to make the data fit a normaldistribution. During four survey periods, an exponential model was shown to be the best semivariogrammodel in August and October 2014, while data from ]anuary 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 werefound in all four of the interpolation methods during the four survey periods. Results of the cross-validationdiagnostic demonstrated that OK performed the best in August 2014, while IDW performed better during theother three survey periods. The GPI and LPI methods had relatively poor interpolation results compared toIDW and OK. With respect to the spatial distribution, OK was balanced and was not as disconnected as IDWnor 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 highlyrecommended if data are spatially autocorrelated. With respect to feasibility and accuracy, we recommendIDW to be used as a routine interpolation method. IDW is more accurate than GPI and LPI and has acombination of desirable properties, such as easy accessibility and rapid processing.