Flood Index Insurance Trigger Development with Radar Satellites and News Media in Northern Bangladesh

Abstract:

A large portion of global flood losses (15%) are absorbed by the agricultural sector. In Asia, the majority of losses in agricultural production (60%) were due to floods (FAO 2015). Yet insurance access for farmers is estimated at less than 1% in places like Bangladesh, where extreme floods frequently destroy harvests. Innovative risk transfer mechanisms (developed by Oxfam and the International Water Management Institute) are currently piloting parametric flood index insurance in Bangladesh using optical satellite data (from MODIS, at 250m). Yet, due to clouds, optical satellite sensors may fail to capture maximum observed flood extent. Furthermore, poor access to flood damage time series data makes it difficult to assess the potential inaccuracy of satellite sensors. Here we estimated inundated area time from two public satellite radar sensors capable of detecting floods through clouds, using Sentinel-1 (10m resolution, 6-12 day frequency, since 2014) and SSM/I-SSMI/S (3.125km resolution, twice daily observations since 1992) in Northern Bangladesh. We generate a spatial flood damage time series using a Natural Language Processing algorithm to identify flood events and locations from over 10,000 scraped online news media articles with 95% accuracy. We compare both media and satellite time series of inundation to water level station data. We will demonstrate how to detrend online news media data, and leverage it to compare to the uncertainty of satellite-based flood indices. We conclude with recommendations for how to best use high spatial resolution (but less frequent) radar data, high temporal resolution (but coarse resolution) microwave data, and high spatial and temporal optical observations of inundation (but sometimes blocked by clouds) to improve flood index insurance.