Cloud-based typhoon-derived paddy rice flooding and lodging detection using multi-temporal Sentinel-1&2

Rice production in China's coastal areas is frequently affected by typhoons, since the associated severe storms, with heavy rain and the strong winds, lead directly to the rice plants becoming flooded or lodged. Long-term flooding and lodging can cause a substantial reduction in rice yield or even destroy the harvest completely. It is therefore urgent to obtain accurate information about paddy rice flooding and lodging as soon as possible after the passing of the storm. This paper proposes a workflow in Google Earth Engine (GEE) for mapping the flooding and lodging area of paddy rice in Wenzhou City, Zhejiang, following super typhoon Maria (Typhoon No.8 in 2018). First, paddy rice in the study area was detected by multi-temporal Sentinel-1 backscatter data combined with Sentinel-2-derived Normalized Difference Vegetation Index (NDVI) using the Random Forests (RFs) algorithm. High classification accuracies were achieved, whereby rice detection accuracy was calculated at 95% (VH + NDVI-based) and 87% (VV + NDVI-based). Secondly, Change Detection (CD) based Rice Normalized Difference Flooded Index (RNDFI) and Rice Normalized Difference Lodged Index (RNDLI) were proposed to detect flooding and lodged paddy rice. Both RNDFI and RNDLI were tested based on four different remote sensing data sets, including the Sentinel-1-derived VV and VH back-scattering coefficient, Sentinel-2-derived NDVI and Enhanced Vegetation Index (EVI). Overall agreement regarding detected area between the each two different data sets was obtained, with values of 79% to 93% in flood detection and 64% to 88% in lodging detection. The resulting flooded and lodged paddy rice maps have potential to reinforce disaster emergency assessment systems and provide an important resource for disaster reduction and emergency departments.