Exploring the spatially varying relationships between cadmium accumulations and the main influential factors in the rice-wheat rotation system in a large-scale area

Cadmium (Cd) accumulations in crops and the effects of the related soil factors on them are critical to developing precise soil management measures for food safety. Traditionally-used non-spatial multiple linear regression (MLR) cannot adequately model the spatially varying effects of the related soil properties on Cd accumulations in crop (or soil). Moreover, the traditionally-used methods for exploring the spatial accumulation characteristics (e.g., ordinary kriging) and the effects of other factors on Cd accumulations (e.g., MLR) are sensitive to outliers. In this study, robust geostatistics, enrichment index, and bioavailability index were first used to explore the spatial accumulation characteristics of Cd in wheat grain (wheat-Cd), Cd in rice grain (rice-Cd), and soil DTPA-extractable Cd (DTPA-Cd) in Jintan County, a typical rice-wheat rotation area in China. Then, robust geographically weighted regression (RGWR), established in geographic space rather than variable space, was used to explore the spatially varying relationships between Cd accumulations and the corresponding main influential factors determined by stepwise regression. Last, the modelling accuracy of RGWR was compared with those of basic GWR and MLR. Results showed that (i) outliers affected the spatial predictions of soil total Cd, soil DTPA-Cd, wheat-Cd, and rice-Cd and robust variograms should be used; (ii) the enrichment index of wheat grain was significantly higher than that of rice grain in almost the whole study area; (iii) the areas with the high bioavailability index of soil Cd mainly located in the southeast, southwest, and centre of the study area; (iv) RGWR acquired higher modelling accuracy than GWR and MLR; (v) the spatially varying relationships between Cd accumulations and the corresponding influential factors were revealed by RGWR, which cannot be determined by MLR. The methods suggested in this study provided more precise spatial decision support for soil management measures to guarantee main agricultural product safety in large-scale areas.