AltHydro: an operational system for real-time water level forecasting at virtual stations on the Odra River

Floods are among the most disastrous natural hazards. Therefore, issuing accurate river water level forecasts is one of the key tasks of the hydrologic community. Such forecasts are usually computed only for gauging stations. Many basins, however, are poorly gauged with only a few monitoring stations available. In contrast, satellite altimetry provides regular water level measurements globally at the so-called virtual stations (VS), i.e. unmonitored river sites observed only by altimetry satellites. The temporal resolution of water level time series at VS is approximately 10-35 days. Due to such a long repeat cycle, altimetry observations have not been used very often for forecasting purposes.

In this study, we present the AltHydro system which represents the first approach to issue forecasts for VS of altimetry satellites. AltHydro computes hourly updated forecasts for VS with a 72-hour lead time. First, vector autoregressive models are employed to calculate water level predictions at gauge stations. Next, linear regressions between gauge and altimetry water levels are established and updated in real time. Finally, the predictions for gauge stations are transferred to the neighbouring VS using the regression coefficients and considering the along-river time lag, driven by the downward water propagation, calculated in real time. 

Our approach has been applied to 8 VS of the Sentinel-3A satellite located on the middle Odra/Oder River in southwestern Poland. The Odra/Oder is a transboundary river originating in the Sudetes Mountains. Major floods hit the Odra/Oder river basin regularly, e.g. in 1997, 2010 and 2024. The in situ data were taken from the gauges owned and maintained by the Polish Institute for Meteorology and Water Management — State Research Institute. To properly validate water level predictions at VS, we use both Sentinel-3A (since 2017) and the Surface Water and Ocean Topography (SWOT) measurements (since 2023). The accuracy assessment revealed root mean squared error (RMSE) of 0.17 m (ranging from 0.11 to 0.22 m) and the Nash-Sutcliffe efficiency (NSE) of 0.95 (ranging from 0.92 to 0.98) for the 24-hour predictions. Satisfactory accuracies were also found for the predictions with a lead time of 72 hours, with mean RMSE and NSE of 0.30 m and 0.88, respectively. The system showed robust performance during the major flood of September 2024, especially for the 24-hour lead time. The AltHydro system can lead to increasing the number of stations with water level predictions worldwide, especially when using the unprecedented geometry of the SWOT measurements.

The research presented in this paper has been carried out in frame of the project no. 2020/38/E/ST10/00295 within the Sonata BIS programme of the National Science Centre, Poland. The research has also been supported by the Bekker Programme of the Polish National Agency for Academic Exchange, as well as by the program “Excellence Initiative — Research University”. The experimental version of the system works in an operational fashion, and its real-time predictions are available at: http://althydro.uwr.edu.pl.