Seasonal inundation monitoring and vegetation pattern mapping of the Erguna floodplain by means of a RADARSAT-2 fully polarimetric time series

The Erguna floodplain, located on the boundary of China, Russia, and Mongolia, is a globally important breeding and stop-over site for many migratory bird species. However, it has a fragile ecosystem and is facing severe impacts from climate change and human activities. Motivated by the high demand for spatial and temporal information about flooding dynamics and vegetation patterns, six fully polarimetric RADARSAT-2 images acquired with the same configuration were explored to reveal the characteristics of the floodplain. The evaluation of a set of polarimetric observables derived from the series of polarimetric synthetic aperture radar (PolSAR) data was used to investigate the characteristics of the typical land-cover types in the floodplain. The behaviors of the polarimetric features highlight the benefits of interpreting the scattering mechanism evolutions at different phenological and inundation stages. The vegetation and inundation patterns were mapped by a random forest (RF) classifier with the input of these polarimetric observations. The results show that a multi-temporal classification can better capture the spatial distribution of the vegetation pattern than single-temporal classifications, which often show a large variability due to the effect of phenology and flooding. A hydroperiod map, which was generated by the time-series classification results of open water, illustrates the flooding temporal dynamics and the hydrological regime. The two main environmental factors that induced the changes in the backscattering – phenology and flooding – were identified based on the aggregated classification results of different phenological intervals. The spatial and temporal characteristics of the phenology- and flooding-induced changes were documented by the changes in the trajectories of three patch metrics at the class level. The polarimetric features' importance scores derived from the RF classifier indicate that the intensity-related features are important for land-cover mapping. The HV intensity, in particular, achieves the highest importance score. In addition, the volume scattering component shows a better performance among the composite features, revealing the importance of the quad-polarimetric information for floodplain interpretation and monitoring.