基于地理权重回归模型的土壤有机质空间预测

准确了解土壤有机质的空间分布是合理施肥的重要前提,也是水土流失控制及保护环境的重要基础。利用113个土壤有机质样点数据,以海拔高度、土壤侵蚀强度、土地利用、比值植被指数、样点至河流的欧氏距离、亚铁矿物指数及坡度为参考因子,来尝试利用GWR(Geographically Weighted Regression)模型探索多重因素作用下的土壤有机质空间分布,并通过与普通线性回归(ordinary least squares,OLS)相比较,来了解GWR模型的精度,进而进行了土壤有机质的空间制图,并对其制图效果进行了评价。结果表明,与OLS模型相比,GWR预测模型它能显著降低AIC(Akaike Information Criterion)值,较大程度地提高模型的决定系数,并有效地减少模型的回归残差值。从制图的总体效果看,GWR模型的预测结果与实测值的吻合程度要优于OLS模型。文章还对利用GWR模型进行回归时的样点数量、因子筛选及因子定量化等方面进行了相应的讨论。

Spatial Estimation of Soil Organic Matter by Using Geographically Weighted Regression Model

The spatial distribution information of soil organic matter(SOM) is important for soil erosion control,fertilization and environmental protection.Spatial interpolation methods have been often used to get the regional distribution map of SOM.The objective of this study is to test the effect of GWR approach on estimating the spatial distribution of SOM by using limited sampling data and other geographical data.The spatial distribution of SOM is interpolated based on 113 samples with the incorporation of multiple impact factors,such as elevation,soil erosion intensity,land use types,RVI,sample distance to river,ferrous minerals index,and slope.The results show that the estimation accuracy is acceptable using the limited available sample data.The paper also compares GWR with the traditional OLS(ordinary least squares) regression approach in estimating the spatial distribution of SOM.The comparison results indicate that the estimation accuracy by GWR has been apparently improved.It can significantly lower AIC,largely improve coefficient of determination(R2 and Adj-R2) and effectively decrease residual sum of squares(RSS).In general,values estimated by GWR model are more fitted to observed values than that of by OLS model.The paper also points out that more attention should be paid to the selection of independent impact factors for the GWR model.If impact factors were poorly selected,the R2 coefficient of determination) and adj-R2 could decrease significantly and worse results might happen.