The monitoring of crop disasters is extremely important for food security, farmers' income increase and crop insurance. However, due to the ability to obtain remote sensing data and the difficulty in identifying crop disasters, remote sensing monitoring of crop rainstorm and typhoon disasters still faces great challenges. This study firstly explored the crop classification of south China by using time series of Sentinel-1 data and crop phenological information., in combination with field data. In order to achieve high-precision crop type mapping, this paper proposed a field-scale classification approach based on XGBoost machine learning. Aiming at identifying paddy rice lodging in Guangdong, China, caused by heavy rainfall and strong wind, a decision-tree model was constructed using multiple-parameter information from Sentinel-1 SAR images and the in situ lodging samples using five backscattering coefficients with five polarization decomposition parameters and quantifying the importance of each parameter feature. Finally, a typhoon and rainfall induced damage assessment method of sugarcane crops is proposed using remote sensing images.

 This study had the following findings. 1) Comparing with the classification result based on time series features of pixel, the classification method based on time series features of fields could effectively suppress the generation of speckle noises in SAR images, as well as the overall accuracy and Kappa coefficient in Nansha district of Guangzhou were improved by 12.5% and 0.19 respectively. 2) Compared with the classification method based only on the time series features of Sentinel-1 (VV+VH) after spatio-temporal filtering, the method of adding phenological feature variables presented the better accuracy, Kappa coefficient was 0.91 and the sown area accuracy of sugarcane and banana reached 82.04% and 71.01% respectively. 3)The decision-tree method coupled with polarization decomposition can be used to obtain an accurate distribution of paddy rice-lodging areas, such as Jiangmen and Zhanjiang of Guangdong, China. Radar parameters can best capture the changes of lodged paddy rice by VV, VV+VH, VH/VV, and Span. 4) Span is the parameter with the strongest feature importance. 5) The dual-polarized Sentinel-1 database classification model can effectively extract the area of lodging paddy rice with an overall accuracy of 84.38%, and a total area precision of 93.18%. 6) The NDVI percentage difference between the amount of vegetation index change and the baseline change based on the crop in the year of typhoon and rainfall damage were used to describe the extent of typhoon and rainfall damage. In Dagang Town in Nansha district of Guangzhou most severely affected, and the damage distribution area of sugarcane by remote sensing was verified using the same period disaster statistics at the town level with an accuracy of 85%. These observations can guide the future use of SAR-based information for agricultural crop-lodging assessment and post-disaster management.

Keywords: crop, classification, lodging area, extent of loss, typhoon and rainfall, remote sensing