基于高分5号影像的东北典型黑土区土壤分类

Soil classification in typical black soil areas of Northeast China based on Gaofen-5 images

高精度的土壤分类及制图结果有助于更好地制定土地环境保护和土地资源利用策略。为探究星载高光谱影像实现区域尺度高精度土壤分类及制图的可能性,该研究获取东北黑土区拜泉县、明水县共计4幅高分5号(GF-5)星载高光谱遥感影像。首先,将原始反射率数据(Original Reflectance,OR)进行包络线去除处理获得去包络线数据(Continuum Removal,CR);其次,对OR和CR进行主成分分析(Principal Component Analysis,PCA)处理,分别得到反射率主成分信息(OR-PCA)和去包络线主成分信息(CR-PCA),并在OR-PCA和CR-PCA的基础上结合地形因子(Terrain,TA)。最后,OR、CR、OR-PCA、CR-PCA、OR-PCA-TA、CR-PCA-TA分别作为输入量结合随机森林分类模型,进行土壤分类并实现数字土壤制图。结果表明:1)包络线去除法可有效地提高星载高光谱土壤分类精度,与OR相比,CR的总精度提高了5.48%,Kappa系数提高了0.12。2)PCA可有效地降低高光谱数据的冗余性,提高模型的运算效率以及分类精度;与CR作为输入量相比,CR-PCA的土壤分类总精度提高了3.67%,Kappa系数提高了0.02。3)TA的引入显著提升了土壤分类精度,以CR-PCA-TA作为输入量的土壤分类精度最高,总精度为81.61%,Kappa系数为0.72,实现了高精度的土壤分类模型及土壤制图。研究结果可为大范围、高精度的土壤分类及制图提供新的思路。

High accuracy soil classification and model-based mapping can provide a better evaluation for sustainable management in intelligent agriculture.Nevertheless,most previous studies on the type classification and mapping were derived mainly from hyperspectral images that captured indoor measurement,or airborne hyperspectral and spaceborne multispectral images.Only a few studies were associated with the satellite hyperspectral images in soil classification and mapping,due partly to the lack of satellite hyperspectral data.Thanks to the Gaofen-5(GF-5)satellite,the high-resolution hyperspectral data can currently be returned,and thereby can make it possible to realize soil classification and mapping using satellite hyperspectral images.However,it has become a great challenge to achieve a high accuracy mapping only by using soil spectral characteristics.Alternatively,the introduction of soil formation factors can effectively improve the accuracy of soil classification.In this study,taking the Baiquan and Mingshui county as the study areas,the GF-5 satellite hyperspectral data was selected to verify whether the combination of satellite hyperspectral data and terrain data(one of the five factors of soil formation)can enhance the performance accuracy of soil classification and mapping.The continuum removed(CR)analysis was performed on the hyperspectral reflectance data(OR),and then the obtained CR and OR data were processed by a principal component analysis,in order to gain the inputs of OR-PCA(Principal Component Analysis)and CR-PCA.Subsequently,some terrain(TA)factors were added to the program of OR-PCA and CR-PCA,particularly including elevation,slope,aspect,curvature,and relief degree of land surface.As such,the soil hyperspectral remote sensing classification models were established in combination with a random forest classifier.The results are as follows:1)The spectral phase differences among various soil types increased significantly after the continuum removal treatment,indicating a great improvement on the accuracy of the hyperspectral soil classification.Compared with that of OR as input data,the correct classification number of Phaeozems,Chernozems and Cambisols increased by 20,17 and 27,respectively.The total accuracy and Kappa coefficient were also improved by 5.48%and 0.12,respectively.The findings demonstrated that the CR treatment can be used to greatly increase the accuracy of soil classification and mapping.2)The PCA method can be used to dramatically reduce the redundancy data of hyperspectral images,while improve the classification efficiency and the accuracy of model.In the mapping data,the misclassification of pixels was significantly reduced after the PCA treatment.Compared with that of OR as the input,the total accuracy of soil classification in the OR-PCA method was improved by 1.71%,and the Kappa coefficient was enhanced by 0.02.Compared with that of CR as the input,the total accuracy of soil classification in the CR-PCA method was improved by 3.67%,and the Kappa coefficient was improved by 0.02.3)The obtained results indicated that the addition of TA factors can significantly improve the accuracy of soil classification.Among the different input combinations,the CR-PCA-TA group showed the highest accuracy of soil classification when using as the input data,with the classification total accuracy of 81.61%and the Kappa coefficient of 0.72.Compared with that of CR-PCA as the input,the total accuracy of soil classification based on the combination of CR-PCA-TA was improved by 13.01%,and the Kappa coefficient was improved by 0.20,indicating that the high precision soil classification model and soil mapping were realized during this time.The findings can provide new insightful ideas for soil classification and mapping with a high accuracy in a wide range for intelligent agriculture.