Main content area

Modeling and mapping of cadmium in soils based on qualitative and quantitative auxiliary variables in a cadmium contaminated area

Cao, Shanshan, Lu, Anxiang, Wang, Jihua, Huo, Lili
The Science of the total environment 2017 v.580 pp. 430-439
cadmium, geostatistics, kriging, land use, models, prediction, residential areas, soil types, spatial data, topography, China
The aim of this study was to measure the improvement in mapping accuracy of spatial distribution of Cd in soils by using geostatistical methods combined with auxiliary factors, especially qualitative variables. Significant correlations between Cd content and correlation environment variables that are easy to obtain (such as topographic factors, distance to residential area, land use types and soil types) were analyzed systematically and quantitatively. Based on 398 samples collected from a Cd contaminated area (Hunan Province, China), we estimated the spatial distribution of Cd in soils by using spatial interpolation models, including ordinary kriging (OK), and regression kriging (RK) with each auxiliary variable, all quantitative variables (RKWQ) and all auxiliary variables (RKWA). Results showed that mapping with RK was more consistent with the sampling data of the spatial distribution of Cd in the study area than mapping with OK. The performance indicators (smaller mean error, mean absolute error, root mean squared error values and higher relative improvement of RK than OK) indicated that the introduction of auxiliary variables can improve the prediction accuracy of Cd in soils for which the spatial structure could not be well captured by point-based observation (nugget to sill ratio=0.76) and strong relationships existed between variables to be predicted and auxiliary variables. The comparison of RKWA with RKWQ further indicated that the introduction of qualitative variables improved the prediction accuracy, and even weakened the effects of quantitative factors. Furthermore, the significantly different relative improvement with similar R2 and varying spatial dependence showed that a reasonable choice of auxiliary variables and analysis of spatial structure of regression residuals are equally important to ensure accurate predictions.