Assimilating in situ and radar altimetry data into a large-scale hydrologic-hydrodynamic model for streamflow forecast in the Amazon
- 1Instituto de Pesquisas Hidráulicas IPH, Universidade Federal do Rio Grande do Sul UFRGS, Brazil
- 2Géosciences Environnement Toulouse GET, UMR5563, CNRS IRD UPS – OMP, Université Toulouse III Paul Sabatier, Toulouse, France
- 3CPTEC/INPE – Centro de Previsão de Tempo e Estudos Climáticos, Instituto Nacional de Pesquisas Espaciais, Cachoeira Paulista, Brazil
- 4Laboratoire d'Etudes en Géophysique et Océanographie Spatiales LEGOS (UMR 5566 CNES CNRS IRD UPS), OMP, Université Toulouse III Paul Sabatier, Toulouse, France
- 5NASA Goddard Space Flight Center, Hydrological Sciences Lab,Greenbelt, USA
- 6Universidade do Estado do Amazonas UEA, Brazil
Abstract. In this work, we introduce and evaluate a data assimilation framework for gauged and radar altimetry-based discharge and water levels applied to a large scale hydrologic-hydrodynamic model for stream flow forecasts over the Amazon River basin. We used the process-based hydrological model called MGB-IPH coupled with a river hydrodynamic module using a storage model for floodplains. The Ensemble Kalman Filter technique was used to assimilate information from hundreds of gauging and altimetry stations based on ENVISAT satellite data. Model state variables errors were generated by corrupting precipitation forcing, considering log-normally distributed, time and spatially correlated errors. The EnKF performed well when assimilating in situ discharge, by improving model estimates at the assimilation sites (change in root-mean-squared error Δrms = −49%) and also transferring information to ungauged rivers reaches (Δrms = −16%). Altimetry data assimilation improves results, in terms of water levels (Δrms = −44%) and discharges (Δrms = −15%) to a minor degree, mostly close to altimetry sites and at a daily basis, even though radar altimetry data has a low temporal resolution. Sensitivity tests highlighted the importance of the magnitude of the precipitation errors and that of their spatial correlation, while temporal correlation showed to be dispensable. The deterioration of model performance at some unmonitored reaches indicates the need for proper characterisation of model errors and spatial localisation techniques for hydrological applications. Finally, we evaluated stream flow forecasts for the Amazon basin based on initial conditions produced by the data assimilation scheme and using the ensemble stream flow prediction approach where the model is forced by past meteorological forcings. The resulting forecasts agreed well with the observations and maintained meaningful skill at large rivers even for long lead times, e.g. >90 days at the Solimões/Amazon main stem. Results encourage the potential of hydrological forecasts at large rivers and/or poorly monitored regions by combining models and remote-sensing information.