Enhancing flood inundation mapping technique using Landsat-based remote sensing and 2D hydrodynamic modeling: a case study of the Bihar flood caused by the Koshi embankment breach

Floods have become more unpredictable and erratic due to the influence of extreme hydroclimatic events. Therefore, obtaining near-real-time, accurate flood inundation maps for such events is essential for effective flood emergency response, which can be achieved by leveraging remotely sensed data. This study integrated high-resolution remote sensing data to enhance flood inundation mapping in a data-scarce South Asian basin, the Koshi River basin. The study considered the 2008 Bihar flood event, caused by an embankment breach on the Koshi River at Kusaha (12 km upstream of the Kosi barrage) in Nepal, as a case study. An important feature of this event was that it occurred at a relatively low peak discharge of ∼4078 m3/s, far lower than the design capacity of ∼26,901 m3/s for the barrage downstream. The 1500-m wide breach in the embankment resulted in a 15–20 km wide and 150 km long sheet of water, creating an inundation zone of 2722 km2. This study used Landsat satellite surface reflectance data to map flood inundation using the commonly used water index known as the Modified Normalized Difference Water Index (MNDWI) to detect open surface water features. Further, the Normalized Difference Vegetation Index (NDVI), permanent water bodies, and Height Above the Nearest Drainage (HAND) datasets are used to mask the MNDWI-based flood inundation map, thereby improving its accuracy. In addition, different thresholding values are applied to the final masked MNDWI map to obtain more accurate maps. Furthermore, two-dimensional (2D) hydrodynamic modeling, using observed flow data near the breaching location, was set up to simulate the flooding event, provide a near-best estimate of the flood-inundated areas, and validate the accuracy of the remotely sensed inundation map using different binary classification evaluation metrics. The total flooded area within the 2D flow area defined in the model using the MNDWI threshold value of 0.2 is about 1,036 km2, whereas, for the threshold value of 0.3, the total flooded area is about 846 km2. The total flooded area obtained from the 2D modeling was about 1,162 km2. The overall accuracy of the flood inundation map using the 0.2 MNDWI threshold was 79%, whereas the accuracy with the 0.3 MNDWI threshold increased slightly to 81%. However, the precision and recall values for both thresholds were below 50%, as the number of correctly predicted flooded areas was much lower than that of correctly predicted non-flooded areas.