Water Level Reconstruction and Prediction Based on Space-Borne Sensors: A Case Study in the Mekong and Yangtze River Basins

Water level (WL) measurements denote surface conditions that are useful for monitoring hydrological extremes, such as droughts and floods, which both affect agricultural productivity and regional development. Due to spatially sparse in situ hydrological stations, remote sensing measurements that capture localized instantaneous responses have recently been demonstrated to be a viable alternative to WL monitoring. Despite a relatively good correlation with WL, a traditional passive remote sensing derived WL is reconstructed from nearby remotely sensed surface conditions that do not consider the remotely sensed hydrological variables of a whole river basin. This method’s accuracy is also limited. Therefore, a method based on basin-averaged, remotely sensed precipitation from the Tropical Rainfall Measuring Mission (TRMM) and gravimetrically derived terrestrial water storage (TWS) from the Gravity Recovery and Climate Experiment (GRACE) is proposed for WL reconstruction in the Yangtze and Mekong River basins in this study. This study examines the WL reconstruction performance from these two remotely sensed hydrological variables and their corresponding drought indices (i.e., TRMM Standardized Precipitation Index (TRMM-SPI) and GRACE Drought Severity Index (GRACE-DSI)) on a monthly temporal scale. A weighting procedure is also developed to explore a further potential improvement in the WL reconstruction. We found that the reconstructed WL derived from the hydrological variables compares well to the observed WL. The derived drought indices perform even better than those of their corresponding hydrological variables. The indices’ performance rate is owed to their ability to bypass the influence of El Niño Southern Oscillation (ENSO) events in a standardized form and their basin-wide integrated information. In general, all performance indicators (i.e., the Pearson Correlation Coefficient (PCC), Root-mean-squares error (RMSE), and Nash–Sutcliffe model efficiency coefficient (NSE)) reveal that the remotely sensed hydrological variables (and their corresponding drought indices) are better alternatives compared with traditional remote sensing indices (e.g., Normalized Difference Vegetation Index (NDVI)), despite different geographical regions. In addition, almost all results are substantially improved by the weighted averaging procedure. The most accurate WL reconstruction is derived from a weighted TRMM-SPI for the Mekong (and Yangtze River basins) and displays a PCC of 0.98 (and 0.95), a RMSE of 0.19 m (and 0.85 m), and a NSE of 0.95 (and 0.89); by comparison, the remote sensing variables showed less accurate results (PCC of 0.88 (and 0.82), RMSE of 0.41 m (and 1.48 m), and NSE of 0.78 (and 0.67)) for its inferred WL. Additionally, regardless of weighting, GRACE-DSI displays a comparable performance. An external assessment also shows similar results. This finding indicates that the combined usage of remotely sensed hydrological variables in a standardized form and the weighted averaging procedure could lead to an improvement in WL reconstructions for river basins affected by ENSO events and hydrological extremes.