丹江口库区土壤镍含量高光谱反演方法

Soil Nickel Metal Content Estimation Based on Hyper-spectrum in Danjiangkou Reservoir Area

为探讨土壤高光谱法反演土壤重金属含量的可行性,以丹江口库区具有代表性的55个土壤样品为研究对象,应用高光谱技术对研究区土壤镍含量进行反演方法研究。对土壤的原始光谱数据,进行6种形式的微分变换。按最大正相关性和最小负相关性共筛选了12维光谱特征,利用随机森林回归(random forest regression,RFR)和极限梯度提升树回归(extreme gradient boosting regression,XGBR)建立了土壤镍含量的高光谱反演模型。分析结果表明:土壤镍含量反演的最优波段主要出现在1686 nm、2238 nm和2254 nm处;基于波段特征的XGBR模型稳定性和拟合精度总体优于RFR模型,稳定性系数R2高达0.93,均方根误差为样本镍含量均值的10.1%,拟合精度较高。文章最后利用土壤高光谱数据,采用XGBR模型对丹江口库区土壤镍含量进行了有效估测。

To explore the feasibility of soil heavy metal content estimation determined by soil-hyper-spectrometry,55 representative soil samples in the Danjiangkou reservoir area are taken as research objects,and the soil nickel content of samples is characterized by hyper-spectrum technology.Based on soil’s original hyper spectrum data,six differential transformations are adopted.Based on maximum positive correlation and least negative correlation,12 dimension of spectral features are extracted,and choose random forests regression(RFR)and extreme gradient boosting regression(XGBR)to establish the high spectral estimation model of soil nickel content.The results prove that the optimal band of soil nickel estimation mainly occurs at 1686 nm,2238 nm and 2254 nm.The stability and fitting accuracy of XGBR model based on band features are generally superior to RFR model.The determination coefficient R2 is 0.93,and the root mean square error value is only 10.1%of the sample mean nickel content,showing that the fitting accuracy is high.Based on the soil hyper spectrum data,extreme gradient boosting regression model is used for effective estimation of the nickel content in danjiangkou reservoir area.