Towards a large-scale locally relevant flood inundation modeling framework using SWAT and LISFLOOD-FP

Abstract

Lack of geospecificity or local relevance is a major limitation in contemporary large-scale flood modeling frameworks. There is a little practical value for configuring a large-scale model if the model produces streamflow and/or inundation maps only along the large rivers while numerous lower order streams remain overlooked. This study fills the gap through a new flood prediction framework based on the loose coupling of a hydrologic model Soil and Water Assessment Tool (SWAT) and a 1D/2D hydrodynamic model LISFLOOD-FP (hence, SWAT-LISFP). The prototype SWAT-LISFP framework was configured with ~26,000 stream reaches across the ~500,000 km2 Ohio River Basin, United States. After being calibrated against 50 gauge stations across the basin, SWAT simulated streamflow outputs were fed as upstream boundary conditions in LISFLOOD-FP. The resultant flood inundation extents consistently captured 70–80% of the remotely sensed inundation, irrespective of the flood events or locations within the basin. This was also confirmed via cross-validation with an existing flood modeling framework AutoRAPID (Follum et al., 2017). Additional modeling experiments were conducted to facilitate two critical discussions – how simulated inundation extent is affected by the uncertainty in streamflow prediction and the density of streamflow boundary conditions. Taking into account the uncertainties in SWAT streamflow, LISFLOOD-FP showed a remarkable improvement with more than 95% of remotely sensed inundation captured within the simulated extent. While this approach produces a variable-area flood map (i.e., a range of areas likely to be inundated at a particular point of time), inundation in the lower order streams can still remain undetected. A solution to this problem was demonstrated by setting up streamflow boundary conditions across further lower order streams, which subsequently justified the need for high-resolution stream network, and hence, the essence of locally relevant flood inundation modeling. The new contributions of his study, particularly through introducing SWAT as a functional hydrologic alternative to supplement a hydrodynamic model such as LISFLOOD-FP and the series of experiments to draw insights on addressing lack of accuracy and local relevance, will enhance the global flood modeling initiatives.