Two decades of river discharge from multi-mission multispectral data

Abstract

Long-term river discharge time series are essential for assessing water availability, seasonal variability, and the impacts of climate change. However, in-situ data do not ensure continuity and large-scale availability, as they are constrained by the limitations of monitoring networks, which are affected by high maintenance costs, geopolitical factors, and the remoteness of many river basins. Satellite remote sensing offers a valuable alternative, with multispectral data providing information on river discharge dynamics. Nevertheless, also satellite data suffer discontinuity, being periodically decommissioned and substituted by sensors with different characteristics.

This study applies and refines the Calibration-Measurement (CM) approach across 54 river sites worldwide, using 10 different multispectral satellite products from 8 satellite sensors to examine long time series of data. The methodology optimizes the selection of reflectance indices based on local hydrological conditions, evaluating the best procedure to obtain river discharge proxies according to the specific flow regimes and climatic condition. Multi-mission data are then combined to improve temporal coverage and accuracy, obtaining long-term timeseries of river discharge proxies. A new uncalibrated procedure is also introduced to extract river discharge information in sites with decommissioned stations or to obtain the CM proxies in ungauged basins.

Results demonstrate that integrating multiple satellite sources substantially improves river discharge estimation, due to factor such as the superior performance of high-resolution sensors (e.g., Sentinel-2) over coarse-resolution datasets (e.g. MODIS) in narrow rivers, and the complementary value of MODIS finer temporal resolution. The proposed merging approach enhances data consistency and reduces gaps while maintaining good performance values, while the uncalibrated method proves effective in many cases but remains challenging for frozen rivers, cloud-prone regions and areas with a low ratio between river width and satellite sensor spatial resolution. Specifically, after the application of a Cumulative Distribution Function Merging (CDF) matching to obtain river discharge estimates from CM proxies, the average and maximum value of Spearman's correlation for the calibrated approaches are respectively 0.78 and 0.92 whereas for uncalibrated approach are 0.55 and 0.89; in terms of Kling-Gupta Efficiency (KGE) the average and maximum value are 0.41 and 0.85 for calibrated approach and 0.29 and 0.8 for the uncalibrated one.

These findings highlight the potential of multi-sensor approaches for global river discharge monitoring and lay the groundwork for future operational applications.