A Multi-Method Approach to Flood Mapping: Reconstructing Inundation Changes in the Cambodian Upper Mekong Delta

Abstract : As in many tropical deltas globally, annual floods shape the livelihoods of the largely rural population in the Cambodian Mekong delta. Agricultural cycles are keyed to the flood arrival, peak, and recession, and fish populations depend on inundated floodplains for their regeneration. However, as factors like climate change and hydropower infrastructure development are altering the Mekong's hydrology, the inundation dynamics of its deltaic floodplains are shifting as well. Several studies have assessed the general changes of river discharge and flood extent on a basin-or delta-wide scale. Yet the sustainable development of this region is relying on dynamics at more local and specific scales, which have not been addressed so far. This paper presents a methodology to track the evolution of hydrological regimes and associated inundations in tropical deltas such as the upper Mekong delta in Cambodia, where it is applied over the past 30 years. Data scarcity and heterogeneity of the environment in this region necessitated the use of combined approaches. We established a link between water levels measured in situ and flood maps derived from optical and radar satellite images (Sentinel-1 and -2).The robustness of the link was assessed using Sentinel, Landsat imagery and the TanDEM-X (12 m) elevation model. This water level-flood link (WAFL) was then used to reconstruct a daily time series of inundation extents reaching back to the beginning of hydrological measurements in 1991 (30 years). On this basis, changes in the incidence, duration, and spatial distribution of floods were analysed. The results indicated that WAFL can be used to reconstruct inundation maps with an overall robustness of 87% in comparison to historical inundation maps derived from remote sensing imagery. Comparisons of WAFLderived flood extents with Landsat images further underscored the significant role of local infrastructure, sedimentation dynamics, and land cover to explain changes in inundation dynamics. WAFL-based analyses revealed that inundation durations have decreased by an average of 19 days when comparing the periods before and after 2008, which was identified as a break point in the hydrological time series. Furthermore, a drastic decrease in inundation the annual frequency with which individual pixels are flooded can be detected during the first half of the traditional flood season, with an average of-21% in early August, negatively impacting water-based livelihoods, from agriculture to fisheries.