东北典型黑土区有机碳遥感定量反演研究

利用多元逐步回归分析法，结合Landsat8 OLI遥感数据对该地区土壤有机碳进行定量反演.试验采集了164个土壤样品，通过3倍标准差准则对样品进行奇异点去除及数据集划分，其中120个样品作为训练集，44个样品作为验证集，建立土壤有机碳的多元逐步回归预测模型.结果表明：有机碳与Landsat8各波段反射率均显著相关；黑土有机碳光谱预测最优模型以倒数为自变量模型最优，决定系数R²=0.180，均方根误差RMSE=0.558，海伦地区适于C_org含量遥感反演，预测模型稳定性好，可以用于揭示黑土典型区C_org含量的空间分布特征.同时认为在不对土壤进行地面光谱测试的情况下，直接采用化学分析数据与遥感卫星相关联的方法预测模型拟合度有限，光谱对有机碳可解释性较低.

STUDY ON QUANTITATIVE INVERSION OF REMOTE SENSING FOR ORGANIC CARBON IN THE TYPICAL BLACK SOIL AREAS OF NORTHEAST CHINA

The quantitative inversion of soil organic carbon (C_org) in the study area is conducted by using multiple stepwise regression analysis method in combination with Landsat8 OLI remote sensing data. For the test, 164 soil samples are collected. Singular points are removed and data sets are divided by tripled standard deviation. Among the total, 120 samples are chosen as the training set and the other 44 as the validation set to establish the multiple stepwise regression prediction model for C_org. The results show that the organic carbon is significantly correlated with the reflectivity of Landsat8 bands. The optimal model for the prediction of black soil organic carbon spectrum is the one that takes the reciprocal as the independent variable, with the determination coefficient R²=0.180, and root-mean-square error(RMSE)=0.558. Hailun area is suitable for remote sensing inversion of C_org content, with a stable prediction model, which can be used to reveal the spatial distribution of C_org content in typical black soil areas. Meanwhile, it is believed that without ground spectral test for the soil, the fitting degree of prediction model by simply using the method of associating chemical analysis data with remote sensing satellite is limited and the interpretation of C_org by spectrum is poor.