Assessment of the predictive performance of 1D/2D, fully 2D and coarse-resolution 2D HEC-RAS models by integrating different evaluation metrics and SAR-derived flood boundary information

Three HEC-RAS models (a coupled 1D/2D, a fully 2D and a coarse-resolution 2D) were tested for replicating a real flood in the Spercheios river, Central Greece. Their comparative response was investigated under the variation of the roughness coefficient across the floodplains and the main channel and the inflow discharge, aiming to provide practical insights for modelers and support decision-making. Model performance was quantified with respect to the inundation extent via six distinct measures with the purpose to examine the influence of performance metrics in the evaluation outcomes. These were the: Critical Success Index (<i>CSI</i>), Hit rate (<i>H</i>), False alarm (<i>F</i>), Success Index (<i>SI</i>), Accuracy (<i>Acc</i>) and Error bias (<i>E</i>). A satellite-derived inundation map was considered as benchmark. Overall, the fully 2D model showed greater sensitivity to the floodplain roughness whereas the 1D/2D approach tended to be primarily affected by the upstream discharge. According to the results, the applied metrics are grouped as follows: (a) <i>CSI</i>, <i>H</i> and <i>SI</i>, which rank the fully 2D model as best; and (b) <i>F</i>, <i>E</i> and <i>Acc</i>, demonstrating the 1D/2D model as best. The coarse 2D model consistently underperformed, highlighting the limitations of oversimplified approaches in capturing flow dynamics, especially in narrow rivers with complex topography.