Publications

Mohanty, MP; Simonovic, SP (2021). Fidelity of reanalysis datasets in floodplain mapping: Investigating performance at inundation level over large regions. JOURNAL OF HYDROLOGY, 597, 125757.

Abstract
Surface runoff estimates from atmospheric reanalysis datasets are increasingly preferred by hydrologists for modeling floods in regions where traditional observations are not sufficiently available. This study explores the fidelity of four widely used reanalyses runoff products from CFSR (6 hourly), ERA-Interim (3 hourly), MERRA (1 hourly), and NARR (3 hourly) as hydraulic forcings to a flood inundation model in describing inundation dynamics over Canada. The reanalysis datasets (duration: 1970 to 2010) are fed to the Catchment-based Macroscale Floodplain (CaMa-Flood) global hydrodynamic model, along with other relevant river and floodplain characteristic inputs, to derive high-resolution floodplain maps for 100 and 200-yr return periods. The floodplain maps derived from each reanalysis dataset are compared with the regional developed or 'benchmark floodplain maps' over six selected flood-prone basins (test basins) in Canada through a set of performance statistics. The NARR and ERA-Interim simulated flood maps are found to capture the high and very-high inundation spots fittingly over entire Canada than the remaining reanalysis datasets. While examining inundation dynamics over the six test regions, the NARR simulated floodplain maps are found to capture the majority of inundated areas by exhibiting the highest Hit-Rates between 0.78 and 0.88, and Critical Success Indexes between 0.7 and 0.82. ERA-Interim followed NARR in most of the cases, while CFSR and MERRA did not perform well on several occasions. On the basis of superior performance of NARR, a few historic flood events over the test basins are simulated and subsequently compared with MODIS satellite-derived floodplain information. We notice that more than 75% of the inundation is precisely captured for these events, leaving behind a few spots either due to the presence of a complex terrain and river geometry or due to impacts from other flood drivers such as snow-melt and tide inflow, which are not considered in the present analysis. This study strongly supports the need for careful selection of a reanalysis dataset while performing inundation modeling for large regions. Otherwise, an arbitrary choice may result in larger uncertainties in damage costs, population exposure, and flood risk values during the process. We hope that this study will provide additional insights into floodplain management practices and assist flood modelers and practitioners in a quest for selection of appropriate reanalysis data inputs for floodplain mapping.

DOI:
10.1016/j.jhydrol.2020.125757

ISSN:
0022-1694