GWR模型在土壤重金属高光谱预测中的应用

发布时间：2018-04-20 14:20

本文选题：GWR模型 + 土壤重金属　；参考：《地理学报》2017年03期

【摘要】：目前土壤重金属高光谱反演模型大多忽视了重金属与光谱变量间相关关系的空间异质性,这与实际情况不相吻合,而地理权重回归(GWR)模型能有效地揭示变量间关系的空间异质性。本文以福州市土壤重金属Cd、Cu、Pb、Cr、Zn、Ni为对象,构建土壤重金属预测的GWR高光谱模型,并将预测结果与普通最小二乘法回归(OLS)结果进行比较分析,探讨GWR模型在土壤重金属高光谱预测中的适用性及局限性。结果表明:(1)GWR模型在土壤重金属高光谱预测中适用与否取决于重金属对光谱变量影响的空间异质性程度:对于Cr、Cu、Zn、Pb等对光谱变量影响空间异质性大的元素,其GWR预测精度较OLS提高明显,表现为GWR模型的调节R2较OLS模型有了明显提高,分别为OLS模型的2.69倍、2.01倍、1.87倍和1.53倍;而AIC值以及残差平方和较OLS模型却明显降低,AIC值减少量均大于3个单位,残差平方和则仅分别为OLS模型的25.33%、30.09%、47.22%和86.84%;对于Cd和Ni等对光谱变量影响空间异质性小的元素,相较于OLS模型,GWR模型的调节R2分别提高了0.015和0.007,残差平方和分别减少了5.97%和4.18%,但AIC值却分别增加了2.737和2.762,GWR预测效果改善不明显;(2)光谱变换可以有效增强土壤重金属的光谱特征,其中以光谱的倒数变换效果最好,而且该变换及其微分形式可以很好地提高模型的预测效果;(3)GWR模型的应用前提是变量间关系的空间非平稳性,适合在与土壤光谱变量间关系具有显著空间异质性的重金属高光谱预测中推广。
[Abstract]:At present, the spatial heterogeneity of the correlation between heavy metals and spectral variables is neglected in the hyperspectral inversion models of soil heavy metals, which is not consistent with the actual situation. The geographical weight regression (GWR) model can effectively reveal the spatial heterogeneity of the relationship between variables. In this paper, the GWR hyperspectral model of soil heavy metal prediction was constructed by taking the heavy metal CdCuCuPbPbPbCZZZN Ni in Fuzhou as an example, and the prediction results were compared with those obtained by the ordinary least square regression method (LLS). To discuss the applicability and limitation of GWR model in soil heavy metal hyperspectral prediction. The results showed that the applicability of the GWR model to the hyperspectral prediction of soil heavy metals depended on the spatial heterogeneity of the influence of heavy metals on the spectral variables. The prediction accuracy of GWR was significantly higher than that of OLS, and the regulating R2 of GWR model was significantly higher than that of OLS model, which was 2.69 times of OLS model, 1.87 times and 1.53 times of OLS model, respectively. However, the AIC value and the sum of squared residuals were significantly decreased by more than 3 units, and the sum of squared residuals were only 25.3330.09% and 86.84% of those of the OLS model, respectively, for elements with low spatial heterogeneity, such as CD and Ni. Compared with OLS model, the adjusted R2 of GWR model increased by 0.015 and 0.007, the sum of squared residuals decreased by 5.97% and 4.18%, but the AIC value increased 2.737 and 2.762 GWR respectively. Among them, the reciprocal transformation of spectrum is the best, and this transformation and its differential form can improve the prediction effect of the model and the application of the GWR model is based on the spatial nonstationarity of the relationship between variables. It is suitable for the hyperspectral prediction of heavy metals with significant spatial heterogeneity in relation to soil spectral variables.
【作者单位】：福建师范大学地理科学学院;闽江学院地理科学系;
【基金】：国家自然科学基金项目(41601601) 福建省自然科学基金项目(2016J01194) 科技部国际合作重大专项(247608)~~
【分类号】：S151.9;O212.1

【相似文献】