Development of a near-infrared band derived water indices algorithm for rapid flash flood inundation mapping from sentinel-2 remote sensing datasets

Abstract

Rapid satellite-based flash flood inundation mapping and the delivery of flash flood inundation maps during a flash flood event for wetland communities can provide valuable information for decision-makers to put relief measures and emergency responses in place without delay. With remote sensing techniques, flash flood mapping of large areas, basically wetlands, can be done quickly with a high level of precision through different water indices. This study developed an algorithm for rapid flash flood inundation mapping for crisis management through the demarcation of the most flash flood-inundated areas in the Haor Basin (wetlands) of Bangladesh by utilizing high-resolution Sentinel-2 remotely sensed data. The algorithm applied here involves near-infrared (NIR) spectral band-derived indices, namely, a normalized difference vegetation index (NDVI) and normalized difference water index (NDWI) to develop a rapid flash flood water detection technique integrating three year (2017–2019) datasets before and after flash floods. A simple threshold method was created to cluster the data and identify the flash flood pixels in the imagery using a density slicing technique followed by natural break analysis. Calculations were then made to estimate the flash flood (inundated), mixed pixels and non-inundated pixels for each year and three combinations. NDVI and NDWI, as well as their combinations (NDVI-NDWI), were remarkably effective for extracting inundation, non-inundation and mixed pixels. Additionally, highly consistent results were obtained for all inundation classes in the studied areas, confirming that NIR-derived indices can effectively detect water pixels. However, a higher inundation pixel value was observed in the Tahirpur Subdistrict compared with the other two study areas (Gowainghat and Kulaura). The developed NIR band-derived water indices algorithm produced more than 80.0% accuracy to detect water-related pixels when verified with ground reference points. As shown by these results, the developed NIR band-derived water indices were capable of effectively detecting flash flood water turbidity in wetland areas. Therefore, these NIR band-derived water indices can be applied for rapid flash flood inundation mapping just after a flash flood occurrence for immediate decisions to support affected farmers.