中文摘要：

为高效利用土壤和开展精准农业生产,采用地理权重回归方法对区域土壤pH的空间分布进行预测,以获得比普通线性回归方法更高的预测精度和论证其对其他土壤属性进行预测的可行性。实测样点土壤的pH,然后选用与土壤pH相关程度较高且易于获取的环境因子（海拔高程、坡度、归一化植被指数、铁氧化物指数、样点距河流的最近距离、土壤侵蚀强度）来建立与土壤pH的回归方程。因子经过逐步回归及多重共线性检验后,用地理权重回归模型对pH进行预测。通过ME、RMSE及回归决定系数等指标,用验证样点数据对预测效果进行评价。结果表明,地理权重回归模型的预测精度高于普通线性回归模型,能大幅度降低AIC值,显著提高模型的回归决定系数,有效地降低回归的残差值;生成的预测图在空间上过渡自然,制图效果较好;利用地理权重回归法能将环境影响因子纳入模型,并能充分考虑样点的空间位置,适合于具有高度空间变异的土壤属性预测及制图。

英文摘要：

Soil pH is of great importance for nutrient form and validity. It varies greatly from space to space for the impact of environmental factors on. Hence, precisely knowing the distribution of soil pH is significant to effectively utilizing soil, fertilizing reasonably, and precision agriculture. By comparison with ordinary least square（OLS）, geographically weighted regression（GWR） was proposed by the paper for analyzing the distribution of soil pH at regional scale. The purpose was to know the prediction accuracy of GWR approach and its feasibility in predicting other soil properties. Soil pH was firstly tested using the collected soil samples at field, then the environmental factors, which were easily acquired and had close relation to soil pH, were taken into account for establishing regression function with soil pH These factors were elevation, slope, NDVI, iron oxide index, the nearest distance from samples to river, and soil erosion intensity. After tested by multi-collinearity and stepwise regression among factors, the sieved factors were employed for GWR analysis. The performances of GWR and OLS were estimated and discussed by using mean error, root mean square error, correlation coefficient （R^2） between the predicted soil pH and observed pH at validation sites. The results show that the estimation accuracy by GWR has been greatly improved than that of by OLS. It can significantly lower AIC, largely improve coefficient of determination（Adj R^2） and decrease residuals sum of squares（RSS）. The map interpolated by GWR was satisfactory in the appearance of continuous surface and gradual transition. It is concluded that GWR, by incorporating environmental factors and sample sites, is a promising method in predicting and mapping the spatial distribution of soil properties with great variations in space.