基于Sentinel-2A的孙吴地区土壤有机质反演研究

Sentinel-2A based inversion of the organic matter content of soil in the Sunwu area

利用Sentinel-2A多光谱遥感影像,结合实测土壤信息,对黑龙江省孙吴县黑土区土壤有机质含量进行反演研究。对影像进行预处理后,通过相关分析和随机森林(RF)选取特征波段,采用偏最小二乘法和BP神经网络构建土壤有机质含量多光谱模型反演红旗林场土壤有机质含量。研究表明:相关性选取的倒数对数一阶微分反射率波段和RF选择的组合波段能够有效提高土壤反演精度,组合波段的RF-BP神经网络模型反演效果最佳,R^(2)=0.7245,RMSE=1.3127%。本次研究可为实现土壤有机质动态监测提供技术支持和参考。

This study conducted the inversion of the organic matter content in the soil of the black soil area in Sunwu County,Heilongjiang Province using the Sentinel-2A multispectral remote sensing images and the surveyed soil data.After preprocessing the images,the characteristic bands were selected through correlation analysis and using the random forest(RF)method.Subsequently,a multispectral inversion model for the organic matter content of the soil was built using the partial least square method and the BP neural network,and the inversion of the organic matter content of the soil in the Hongqi Forest Farm was conducted.According to the obtained results,the bands selected based on the reciprocal of the logarithm of the first-order differential of reflectance through the correlation analysis and the combined bands selected using the RF method can effectively improve the inversion precision of the organic matter content in the soil,and the RF-BP neural network model for the combined bands yielded the optimal inversion performance(R^(2)=0.7245 and RMSE=1.3127%).The results of this study will provide technical support and reference for the dynamic monitoring of the organic matter content in soils.