An improved approach to estimating crop lodging percentage with Sentinel-2 imagery using machine learning

It is imperative to rapidly and precisely acquire crop lodging area and severity for disaster prevention and yield prediction. However, estimation of crop lodging area at a large scale remains challenging due to the relatively low sensitivity of remote sensing signal to the lodging variation, limited availability of remote sensing images, and lodging statistical data. This study proposes a new method for lodging area estimation based on the optimal grid cell of Sentinel-2 and crop lodging percentage, overcoming the limitation of traditional pixel-based mapping approaches that fail to obtain quantitative lodging information. Basing the spatial aggregation method, we analyzed the optimal grid size of Sentinel-2 data for lodging percentage estimation. Then we investigated the spectral response for different lodging percentage levels and analyzed the potential of lodging percentage estimation for Sentinel-2 metrics (including selected spectral bands and their derived vegetation indexes (VIs)). A quantitative model was established between the training set and the Sentinel-2 metrics using the random forest (RF) algorithm. Finally, around 1462.62 ha fields from six counties or districts in Heilongjiang province in China were estimated for lodging percentage. Results indicate that the proposed method can estimate the crop lodging percentage on the testing set with an R2 and RMSE of 0.64 and 25.24, respectively, which can explain around 95 % spatial variation of lodging crop. Moreover, the overall magnitude of reflectance increased with the increase in lodging percentage. Among all Sentinel-2 optimal metrics, the Green, SWIR1, and Red edge 1 bands are the most crucial indicators for lodging percentage estimation. Our results on lodging percentage estimation in the study area indicate that there is more lodging maize in the Meilisidawoerzu district than in other areas. Although typhoons passed over Fuyu and Lindian counties, the lodging percentage in these areas is relatively low. The lodging percentage map has great value in agriculture management and insurance claim.