ABSTRACT

The red-edge bands of Sentinel-2 allow for a greater diversity of spectral Vegetation Indices (VIs) to be calculated and used for vegetation characterization. We evaluated the utility of a selection of 40 VIs to derive Leaf Area Index (LAI), fraction of Absorbed Photosynthetically Active Radiation (fAPAR) and fraction of vegetation Cover (fCover) of winter wheat crop using regression method. We calibrated models for specific winter wheat development stages and compared the predictions with all-season models. The most useful VIs could be grouped into several types: (1) indices which use green and NIR band, (2) indices based on red edge bands, (3) indices which use red and NIR band and (4) the MCARI/OSAVIre index. It was found that fAPAR and fCover could be predicted with good accuracy using all-season models (rRMSE of 14% and 23% respectively), while LAI showed lower accuracy (rRMSE = 45%). The LAI model calibrated over the tillering stage was recommended for usage in the early stages of crop development. Compared with the existing methods for biophysical variables retrieval from Sentinel-2 data (i.e. the Level2B processor in SNAP) the regression approach based on VIs showed to be a viable alternative.

摘要

Sentinel-2的红边带允许更大范围的光谱植被指数（VI）被计算并用于植被表征。我们使用回归方法评估了选择40种VI的效用，以得出冬小麦作物的叶面积指数（LAI），吸收的光合有效辐射分数（fAPAR）和植被覆盖率（fCover）。我们对特定冬小麦发育阶段的模型进行了校准，并将预测与全季节模型进行了比较。最有用的VI可以分为几种类型：（1）使用绿色和NIR波段的索引，（2）基于红色边缘带的索引，（3）使用红色和NIR波段的索引，以及（4）MCARI / OSAVIre指数。发现使用全季节模型可以很好地预测fAPAR和fCover（rRMSE分别为14％和23％），而LAI的准确性较低（rRMSE = 45％）。建议在分ing期进行校准的LAI模型用于作物发育的早期阶段。与从Sentinel-2数据中检索生物物理变量的现有方法（即SNAP中的Level2B处理器）相比，基于VI的回归方法被证明是可行的选择。