Can regional to continental river hydrodynamic models be locally relevant? A cross-scale comparison

Regional to global river hydrodynamic models are now a reality, and have largely improved in the recent years due to increasing computation power, remote sensing datasets and new modeling methods. It is fundamental then to understand to which extent these models can provide locally relevant estimates of variables as discharge and water levels. It is addressed here by setting up three different versions of the large scale MGB hydrologic-hydrodynamic model with increasing spatial scale (global, regional and local), based on different topography datasets, cross section parameterization and reach lengths. These model versions are then compared to a detailed HEC-RAS model, set up with >620 locally surveyed cross sections in the Itajai-Acu basin in Brazil. All models are run with calibrated rainfall-runoff parameters, so that effects of non-calibrated runoff fields or uncertain forcing data are not taken into consideration. To assist interpretation, three main requirements to define estimates from a river hydrodynamic model as locally relevant are defined: the model errors should be equal or smaller (i) than the accuracy requirement for a particular application and location, (ii) than typical local, reach scale models' errors, and (iii) than observation uncertainties. Results then indicate that the regional and local MGB versions (and to a lesser extent, the global one) could estimate locally relevant discharges and water level anomalies for many parts of the basin. For absolute water levels, all models failed to provide RMSE smaller than 0.4 m basin-wide, although the local MGB model yields an average 2.1 m RMSE, in comparison to 4.4 m (regional) and 26.1 m (global). For flood extent, all model versions generally yielded unsatisfactory results with average Fit metric values smaller than 65%. Lastly, different setups of the regional MGB model were run to investigate the role of different factors in improving model estimates. Results showed that a coarse reach length (15 km) is unable to capture flood dynamics and water level variations in the basin, while a smaller 1 km long reach improves discharges, water level and flood extent. A locally derived DEM did not lead to significant improvements in relation to a global DEM. In turn, the smallest water level errors would be attained by using information from the in-situ cross sections in the regional models. This study's outcomes are encouraging for the future development of regional to global hydrodynamic models, with new remote sensing techniques and higher computational power improving cross section estimation and model discretization. (C) 2019 The Authors. Published by Elsevier B.V.