3D Water Science/Engineering

For Climate Extremes and Water Resources

Extreme flooding, whether due to natural or anthropogenic causes, poses a significant threat to human society, the economy, and the environment, often resulting in the loss of human lives. The availability of data from Earth-observation satellites enhances communities' capacity for a timely response to flood events, as they offer an efficient means of mapping flood extent over a large area. Synthetic Aperture Radar (SAR) imaging provides an all-weather sensing technique that is well-suited for near-real-time disaster mapping, especially for events like floods. As part of my doctoral degree, I worked on a probabilistic flood-mapping approach to investigate hazard and exposure.
Most flood mapping algorithms provide an estimate of flood extent in the form of a binary map. Despite their usefulness, such binary maps do not provide any information on the uncertainty associated with the pixel class. Due to the ability to characterize the uncertainty associated with each pixel class, compared with the traditional deterministic approach, I applied the Bayesian framework developed by Giustarini et al. (2016) to a set of SAR images acquired by Sentinel-1 C-band satellites over the Kerala state of India before, during, and after the flood event in August 2018. 
My method generates a priori distribution functions of backscattering values for flooded and non-flooded pixels using ensembles of observed SAR amplitude histograms. Next, I apply a Bayesian framework to update a priori distribution functions using observed SAR backscattering values at individual pixels and obtain the posterior flood probability distribution of a pixel. I compare the probabilistic flood map obtained from the SAR-based Bayesian approach against the flood extent obtained from visual inspection of available optical data acquired by Sentinel-2, Landsat satellites, and images from Moderate Resolution Imaging Spectroradiometer (MODIS). Following figure shows an example of this work from Sherpa et al 2020 (IEEE).

Left: Flooded and non-flooded pixels histograms. The yellow and magenta curves are, respectively, the best fitting Gaussian distributions to flooded (F), non-flooded (F¯).

Right:  Close-up map of flood probability at selected sites in Kerala, whose locations are shown in panel (a). (b) Flood map for the zone 1 in north Kerala covered by path 63 on August 2, August 14, August 26, and September 7. (c) Flood maps for zones 2 and 3 in south Kerala covered by path 165 on August 9, August 21, August 27, and September 2. The bottom row on both panels (b) and (c) shows flood probabilities for preflood images, where pixels with high flood probability mark the existing water bodies (Source: Sherpa et al. (2020)).

At Brown, I am expanding my knowledge to 3D surface water mapping using the NASA/CNES/CSA SWOT (Surface Water and Ocean Topography) satellite mission. This work advances the scientific use of water surface elevations, discharge retrievals, and storage changes using SWOT satellite missions and in situ measurements for hydrological, climate, and hazard implications.

Related Publications




Conference Abstract (#Talk)