【摘要】 以湖北省宜都市红花套镇典型柑橘种植区采集到的329个土壤样本为研究对象,设置土壤有机质(SOM)进行普通克里格(OK)插值的结果为参照,借助地理探测器选取与SOM相关性最大的前5种主要影响因子,分别建立全局模型多元线性回归、偏最小二乘回归和局部模型地理加权回归(GWR),再深入分析模型残差的结构性,构造GWR扩展模型GWRMLR、GWRPLSR,讨论几种SOM预测模型的差异。结果表明:使用GWRPLSR模型预测研究区SOM含量的均方误差和均方根误差可分别降低到9.834和3.136,相对分析误差提高到1.468,实测值与预测值间的相关系数(r)达0.743,具有最高的预测精度,GWRMLR其次,说明除SOM与主要影响因子间存在空间相关性,分析模型残差可进一步消除预测的不平稳性。因此,将模型残差项纳入考虑的局部扩展模型更适宜进行区域化SOM空间分布预测与数字土壤制图。更多还原

【Abstract】 329 soil samples were collected from the citrus growing areas in Honghuatao Town,Yidu City,Hubei Province.Based on the principle of spatial stratified heterogeneity,the top five major impact factors having the greatest correlation with soil organic matter(SOM)were selected with the GeoDetector software.Using the interpolation results of ordinary Kriging as control,the global model multiple linear regression(MLR),partial least squares regression(PLSR)and local model geographical weighted regression(GWR)were established by the soil organic matter and its main environmental factors.After analyzing the structure of the model residuals,GWRMLRand GWRPLSR were constructed as the extensions of GWR model.The results showed that the mean square error(MSE),root mean square error(RMSE),relative analysis error(RPD)and the correlation coefficient(r)between measured and predicted values of GWRPLSR were 9.834,3.136,1.468,0.743,respectively.The GWRPLSRmodel had the highest prediction accuracy,followed by GWRMLR.In summary,except for the spatial correlation between SOM and its major impact factors,analyzing model residuals can further eliminate the predicted instability.Therefore,taking the model residual terms into consideration is more suitable to predict the regional SOM spatial distribution and digital soil mapping.更多还原