西南典型岩溶区土壤硒空间分布预测

土壤硒精准预测和制图是富硒土壤资源开发利用和环境规划管理的基础。该文以西南典型岩溶区桂林永福百寿河流域为例,在分析影响土壤硒化学行为因子的基础上,通过野外样品采集和室内化学分析以及Arc GIS空间分析,获取了研究区相关地理环境因子和土壤属性因子数据。利用逐步回归方法选择土壤硒空间分布预测的辅助变量,使用协同克里格模型对非连续分布的辅助变量进行插值。在此基础上利用地理加权回归模型对土壤硒空间分布进行预测,同时以普通克里格插值结果作为参照。研究结果表明:使用地理环境因子和影响土壤硒化学行为的土壤属性因子可以提高土壤硒预测精度;协同克里格插值解决了辅助变量数据连续分布的问题;土壤硒的空间分布与地形和影响土壤硒化学行为的因子有关。

Prediction on spatial distribution of soil selenium in typical karst area of southwest China

Abstract: Selenium is an essential micronutrient for animals and humans. Extensive studies have been conducted on distribution of selenium, but seldom studies have been done in Se-enriched, longevity, and karst region. Accurately mapping the spatial distribution of soil Se is the basis for Se-enriched soil utilization, planning, and environmental management. To better understand the Se distribution in the soils in a Se-enriched, longevity, and karst area of China, this study focuses on the total Se in soil in the Guilin Yongfu Baishou river catchment, and 226 soils (0-20 cm) samples were collected by random sampling in 180 km2 research area in March, 2015. Coordinates and elevations of the sample sites were recorded using a Garmin GPS. Factors affecting the chemical behavior of soil Se were studied to obtain geographical environment factors and related soil properties in the study area through lab analysis and ArcGIS spatial analysis. Auxiliary variables, including soil properties (pH value, soil organic matter, and amorphous iron oxides) and geographical environment factors (elevation, slope, aspect, curvature, compound topographic index, and stream power index) were preliminary selected for predicting total soil Se through stepwise regression. The adjusted determination coefficient (adjusted R2) was used to select the regression model, and then elevation, soil organic matter, amorphous iron oxides and compound topographic index were selected as variables. Co-kriging (COK) was used for interpolation of discontinuously distributed auxiliary variables (soil organic matter, amorphous iron oxides), and the correlation coefficient between prediction value and chemical analysis value was used to measure the prediction accuracy of COK. The amorphous iron oxides prediction value showed significant correlation with analysis value (R=0.62, P<0.01). However, the soil organic matter prediction value was no significantly correlated with analysis value in the researched area. In the end, elevation, amorphous iron oxides, and compound topographic index were selected as variables for spatial prediction of soil Se with geographically weighted regression model (GWR). The predicted results of GWR on soil Se were compared with the results obtained from ordinary Kriging (OK), and mean absolute error (MAE) and root mean square error (RMSE) were adopted to validate the prediction of soil Se by these methods. In order to quantify the improvement on prediction precision, a relative improvement (RI) in correlation coefficient was used to measure the improvement on the prediction accuracy of GWR and OK. Compared to OK, the MAE and RMSE of GWR were 0.19 and 0.24, which reduced by 13.64% and 7.69%, respectively. The correlation coefficient between prediction value and chemical analysis value increased from 0.32 to 0.51. The application of GWR resulted in relative improvement (RI) of 59.38%. The study results showed that the precision of total soil Se prediction could be improved if geographical environment factors and related soil properties were properly selected. The problem of discontinuously distributed auxiliary variables could be resolved by COK method. The spatial distribution of soil selenium was related to topography (elevation) and soil property factors (amorphous iron oxides), which had important influence on selenium chemical behavior in soil. Because of the complexity of terrain in the karst area, the correlation coefficient is still low, and it is worth further research.