基于双时相卫星遥感光谱指数估算土壤有机质含量

以黄淮海平原典型县——封丘县为研究区，探讨了在一年两熟、裸土时间窗口较短的区域中，基于两景影像波段组合构建的双时相光谱指数在有机质含量预测中的表现。研究共计采集117个代表性土样，以分析筛选出的裸土期(10月)内双时相(获取时间：2014年10月6日和2017年10月30日)高质量Landsat 8卫星影像作为数据源，构建了4种类型的光谱指数：比值光谱指数、差值光谱指数、归一化光谱指数以及优化光谱指数，并结合最小绝对收缩和选择算子变量筛选方法和支持向量机算法建立了有机质预测模型。留一交叉验证结果表明，与直接使用影像波段反射率或者基于单景影像构建的光谱指数(单时相光谱指数)相比，利用双时相光谱指数可以更好地利用时相信息优势，其有机质预测精度更高(R2=0.53,RMSE=2.01 g/kg)。而且，基于双时相光谱指数所构建的预测模型得到的有机质空间分布格局与真实值较为吻合。可见，本文提出的在黄淮海平原典型县域利用双时相光谱指数预测土壤有机质的方法，可以促进具有短裸土期特点区域的高分辨率土壤属性遥感预测与制图研究。

Estimation of Soil Organic Matter Content Based on Dual-temporal Satellite Remote-sensing Spectral Index

Soil organic matter (SOM) is the core of soil quality and its spatial distribution information is of great significance to the efficient utilization and management of soil resources. The study investigated the performance of the dual-temporal spectral index (the band combination of two images) in predicting SOM over a double-cropping agricultural region (Fengqiu County) in the Huang-Huai-Hai Plain, where the bare soil period is often short for remote sensing of soils. In the study, a total of 117 soil samples were collected and dual-temporal Landsat8 satellite images during the bare soil period (Acquisition Date: October 6, 2014 and October 30, 2017) were selected for building four types of spectral indices: ratio spectral index, difference spectral index, normalized spectral index and optimized spectral index. Then, these indices were used as the input in SVM (Support Vector Machine) models of SOM after being selected by the variable selection method of LASSO (Least Absolute Shrinkage and Selection Operator). The results of leave-one-out cross-validation showed that, compared with image bands or spectral indices built by single images (single-temporal spectral index), the dual-temporal spectral index could make better use of temporal information of images and its prediction accuracy was higher for SOM (R2=0.53, RMSE=2.01g/kg). Moreover, the spatial distribution pattern of SOM predicted by the dual-temporal spectral index was consistent with the real condition. The proposed method of using the dual-temporal spectral index for SOM prediction in the study could promote prediction and mapping of soil properties in areas with short bare soil periods.