An altimeter height extraction technique for dynamically changing rivers of South and South-East Asia

摘要

A new approach is proposed in this study to maximize the skill of satellite altimeter-derived heights by utilizing river extent information from additional sensors. We apply this approach for JASON-3 nadir altimeter. First, locations were identified in South and South East Asia region as potential candidates for this approach based on river extent. Information on river extent composed of river width and course (i.e., path) from visible (Landsat) and Synthetic Aperture Radar (SAR) platform (Sentinel-1). The extent-based approach was applied to filter out non-water radar returns in two methods: 1) River Mask (RM) based K-means (KM) clustering (hence, RM + KM) and 2) K-means clustering embedded with River Mask (RM) (hence, KMRM). Both methods were validated over 12 locations with long records of in-situ height observations and with distinct river characteristics. The maximum benefit observed in utilizing dynamic river mask is during low flow season when the probability of non water radar signal contamination is highest. For highly varying seasonal widths, satellite imagery on river conditions, as an ensemble of techniques, yields heights with higher accuracy. An average RMSE improvement of 40% in 8 out of 12 stations is observed. Overall, at least one of the two dynamic mask-based techniques studied consistently outperformed the conventional technique of altimeter height extraction based on a fixed river mask. Information on river mask or extent can play an important role in highly meandering rivers as the deviation between actual satellite altimeter passes and nominal passes can be considered during the extraction of heights. Compared to visible (Landsat) imagery, SAR was found to be much more effective at improving altimeter height extraction because of the all-weather capability during frequently cloudy conditions. The practicality of the river mask-based approach for altimeter height extraction needs to be carefully considered by the operational end user based on a