H197-0022
Using remote sensing to collect data on the impact of flooding on the built environment in Kerala, India

Wednesday, 16 December 2020
Poster
Eleanor A Ainscoe1, Barbara Hofmann1, Steven Reece2 and Quillon K Harpham1, (1)HR Wallingford, Wallingford, United Kingdom, (2)University of Oxford, Oxford, United Kingdom
Abstract:
Flooding displaces millions of people worldwide every year, causes billions of dollars’ worth of damage, and is only expected to worsen in coming decades. Measurements of floods’ socioeconomic impact are needed both to enable efficient targeting of relief resources for that event itself, and to act as verification data for the development and testing of impact-based forecast models. However, data about a flood event’s impacts can take months to collect in the aftermath of an event and are often not collected in a consistent fashion. Remote sensing data, while not equivalent to ground truth data, offers the capability to systematically collect data over large regions in a time-efficient way.

Here we present research into methodologies for using remote sensing to collect data on the impact of flooding on the built environment, using the case study of the 2018 Kerala, India floods. According to figures published by the Kerala State Disaster Management Authority, the 2018 floods affected over 800,000 houses and damaged over 9,000 km of roads. Methods based on interferometric coherence decrease have proved successful in detecting structural damage in urban areas caused by natural disasters such as typhoons, tsunamis and landslides. In this situation, however, those methods perform less well because they struggle to detect flood impact in non-urban areas which did not have high pre-event coherence and at buildings that were damaged by inundation but did not sustain the type of structural or roof damage that causes coherence decrease. We therefore assess the alternative approach of combining flood and built environment datasets and we compare the results of doing so using existing or routinely-produced datasets versus using datasets processed especially for this application.