基于GWRK的土壤有效磷空间预测及其超标风险评估

以江苏省金坛区土壤有效磷的空间预测为例，构建地理加权回归克里格（GWRK）模型，即采用地理加权回归（GWR）来量化土壤有效磷与主要土壤因子（即：土壤全磷、土壤pH值和土壤有机质）之间的局部空间关系，并结合局部回归残差的插值结果来预测土壤有效磷的空间分布状况.GWR结果显示主要土壤因子对土壤有效磷含量的影响程度随空间位置的变化而变化.同时，采用独立验证样本对比GWRK模型和普通克里格（OK）模型的空间预测精度.结果显示，GWRK预测结果具有更低的平均绝对误差（MAE）、均方根误差（RMSE）和更高的Pearson相关系数（r），且较OK预测结果的相对提高指数（RI）为19.61%.此外，根据GWRK预测结果，对金坛区土壤有效磷含量的超标风险进行了评估.结果表明土壤有效磷含量超过其环境安全阈值（40mg/kg）的区域集中分布在金坛区北部，其面积为175.58km²，约占金坛区总面积的18%.因此，GWRK模型能有效评估区域土壤元素有效量空间分布状况，且GWR局部空间回归系数能为区域土壤元素有效量的调控提供更精确空间决策支持.

Spatial prediction and standard-exceeding risk assessment of soil available phosphorus based on geographically weighted regression kriging

In this study, geographically weighted regression kriging (GWRK) model was established to predict the spatial distribution pattern of soil available phosphorus in Jintan County, Jiangsu Province. Geographically weighted regression (GWR) was first used to quantify the local spatial regression relationships between soil available phosphorus and its three main influencing factors (i.e., soil total phosphorus, soil pH, and soil organic matter). And then the final prediction value of GWRK is the sum of the GWR prediction value and the regression residuals value interpolated by ordinary kriging (OK). In this study 52 independent verification samples were used to compare the prediction accuracy of the GWRK model and the traditional OK model. Finally, the standard-exceeding risk of the soil available phosphorus was assessed based on the results generated by the GWRK model. The GWR analysis showed that the relationships between soil available phosphorus and its three main influencing factors (i.e., soil total phosphorus, soil pH, and soil organic matter) were spatial non-stationary, with local regression coefficient changing with spatial position. Model comparison showed that the GWRK prediction result had lower mean absolute error (MAE), root mean square error (RMSE) and higher Pearson correlation coefficient (r). In addition, the relative improvement index (RI) of GWRK over OK was 19.61%. The risk assessment results showed that the 175.58km² areas was divided into the risk area of the soil available phosphorus content exceeded the Environmental safety threshold (40mg/kg), which accounted for about 18% of the whole area. Therefore, the GWRK model could effectively assess the spatial distribution pattern of available content of the soil elements. And the local regression coefficient of GWR could provide more accurate spatial decision support for the regulation of soil elements available content at a regional scale.