Testing the skill of numerical hydraulic modeling to simulate spatiotemporal flooding patterns in the Logone floodplain, Cameroon

Abstract

Recent innovations in hydraulic modeling have enabled global simulation of rivers, includingsimulation of their coupled wetlands and floodplains. Accurate simulations of floodplains using these approaches may imply tremendous advances in global hydrologic studies and in biogeochemical cycling. One such innovation is to explicitly treat sub-grid channels within twodimensional models, given only remotely sensed data in areas with limited data availability.However, predicting inundated area in floodplains using a sub-grid model has not been rigorously validated. In this study, we apply the LISFLOOD-FP hydraulic model using a subgrid channel parameterization to simulate inundation dynamics on the Logone River floodplain, in northern Cameroon, from 2001 to 2007. Our goal was to determine whether floodplain dynamics could be simulated with sufficient accuracy to understand human and natural contributions to current and future inundation patterns. Model inputs in this data-sparse region include in situ river discharge, satellite-derived rainfall, and the shuttle radar topography mission (SRTM) floodplain elevation. We found that the model accurately simulated total floodplain inundation, with a Pearson correlation coefficient greater than 0.9, and RMSE less than 700 km 2 , compared to peak inundation greater than 6,000 km 2 . Predicted discharge downstream of the floodplain matched measurements (Nash-Sutcliffe efficiency of 0.81), and indicated that net flow from the channel to the floodplain was modeled accurately. However, the spatial pattern of inundation was not well simulated, apparently due to uncertainties in SRTM elevations. We evaluated model results at 250, 500 and 1000-m spatial resolutions, and found that results are insensitive to spatial resolution. We also compared the model output against results from run of LISFLOOD-FP in which the sub-grid channel parameterization was disabled, finding that the 6 sub-grid parameterization simulated more realistic dynamics. These results suggest that analysis of global inundation is feasible using a sub-grid model, but that spatial patterns at sub-kilometer resolutions still needs to be adequately predicted. Highlights*Sub-grid channel modeling in hydraulic simulations has not been fully validated in floodplains *We apply the LISFLOOD-FP model to simulate inundation dynamics on the Logone floodplain *Total inundation and flow into the floodplain are accurately simulated, but no spatial patterns *Mismatch in spatial patterns is likely due to uncertainties in the digital elevation model *Results suggest that analysis of global inundation is feasible using a sub-grid model

Utilizing Flood Inundation Observations to Obtain Floodplain Topography in Data-Scarce Regions

“…Globally available open-source products are easy to obtain, but are not accurate. Hence, they not suitable for flood inundation modeling (Fernández et al, 2016). It is not always physically or financially feasible to obtain lidar DEMs in data-poor regions.…”

“…However, it is a commercial product and is not freely available. Because the vertical inaccuracies of the SRTM DEM, flood inundation modeling using SRTM produces spatial inconsistencies (Sanders, 2007;Yamazaki et al, 2012;Yan et al, 2015;Fernández et al, 2016;Sampson et al, 2016). Hence, these DEMs are not suitable for flood simulations as they do not represent the topography well (Sanders, 2007).…”

Water Surface Elevation Constraints in a Data Assimilation Scheme to Infer Floodplain Topography: A Case Study in the Logone Floodplain

“…Between September and October, the floodplain is inundated by overbank flow from the Logone River and its branches and remains flooded for about three months (Laborde et al., 2018; Scholte, 2007). Flooding is dictated by the flow in the Logone River, and the role of local rainfall is small (Evans & Loth, 2004; Fernández et al., 2016, 2018). Regional ecology and economics are controlled by the floodplain dynamics.…”

“…SRTM 90% absolute error at continental scale varies between 6 and 10 m (Rodriguez et al., 2006). Using these DEMs for flood modeling produces spatial inconsistencies and hence are not suitable for flood simulations (Fernández et al., 2016; Sampson et al., 2016; Sanders, 2007; Yamazaki et al., 2012; Yan et al., 2015). DLR (German Aerospace Center)'s TanDEM‐X has better spatial resolution (10–12 m) and vertical accuracy (1–2 m), (Eineder et al., 2012; Krieger et al., 2006) but is a commercial product and not freely available.…”

Co‐producing research in the “Red Zone”: Adaptation to fieldwork constraints with a transdisciplinary approach

“…Remotely sensed climate and hydrology data would have been analysed even in normal field accessibility conditions, but not with the same emphasis. Extra research assistants were hired in a Columbus laboratory to test methods to map the extent of seasonal flooding using different sensors (e.g., Landsat MSS, TM and ETM+), allowing the team to study the performance of a numerical hydraulic model to simulate flooding dynamics (Fernandez et al., ). To obtain rainfall data without the option of direct measurements via rain gauges, team members compiled and processed short and discontinuous precipitation time series (e.g., from Lake Maga between 1994 and 2012) in conjunction with satellite‐derived rainfall from the Tropical Rainfall Measurement Mission (TRMM).…”