Remote Sensing Data Assimilation With a Chained Hydrologic-Hydraulic Model for Flood Forecasting

Abstract

Effective flood risk management requires reliable forecasts with extended lead times to enable the implementation of cost-effective and timely measures. In this study, we present a chained hydrologic-hydraulic modeling framework designed for effective near-real-time flood forecasting. The system integrates runoff predictions from a large-scale hydrologic model (ISBA-CTRIP) as inputs into high-resolution, local hydrodynamic models (TELEMAC-2D) to forecast water levels and flood extents. To improve the forecast accuracy, an Ensemble Kalman Filter is employed to reduce uncertainties in hydrological forcing and friction parameters by jointly assimilating in situ water level measurements and satellite-derived flood maps. Such a data assimilation framework operates in a real-time forecasting configuration, consisting of a reanalysis phase having an assimilation window up to the current time, followed by a forecast phase from the current time to the expected lead time. Three forecasting strategies were evaluated: (a) using CTRIP-predicted runoff for both reanalysis and forecast phases, (b) using observed discharge for reanalysis and CTRIP runoff for forecast, and (c) using observed discharge for reanalysis and keeping a constant discharge during the forecast. Results show that using observed discharge during reanalysis combined with CTRIP-predicted runoff for forecasting yields the most consistent performance. However, for all three strategies, forecast accuracy declines with longer lead times, especially when errors in the CTRIP forcings are non-stationary. This work highlights the potential for hydrologic model, despite their inherent imperfections, to serve as effective inputs for local hydraulic models, enabling near-real-time flood forecasting through the assimilation of in situ and remote sensing data.