Satellite Fusion Based Historical Inundation Estimates for Accurate Return Period Estimates in Bangladesh
Abstract
The economic impacts of floods push people into poverty and cause setbacks to development as government budgets are stretched and people without financial protection are forced to sell assets. Investments in flood mitigation and adaptation, such as expanding insurance coverage through index based insurance, i.e. direct payouts based on predetermined indexes of e.g. inundated area, could reduce anticipated losses from floods and increase resilience. Insurance penetration remains low ( < 1%) for climate-vulnerable populations in countries like Bangladesh, which urgently need financial protection from extreme floods to protect development. Expanding insurance coverage requires the ability to quantify flood risk and monitor it in near real-time in remote locations, which is challenging due to limited data in most areas. Satellite observations have the potential to fill this data gap and expand insurance coverage by providing globally available observations available at regular intervals.
Algorithms to map flood extent are improving by leveraging machine learning and multiple sensors. One way to estimate the frequency of extreme floods is to measure the spatial extent of inundation directly from space. Given the growing length of the satellite record, time series of inundated areas could be used for exceedance probability estimation to develop insurance products, but relies on the availability of the data over extended periods of time. High spatial resolution satellites (such as Sentinel-1) have not been available for enough time to reliably estimate exceedance probability to design index insurance triggers. Using machine learning, we fuse the daily MODIS time series with Sentinel-1 data to create 20 year historical inundated area estimates over Bangladesh. We show how these novel fusion models generate more consistent return period estimates compared to datasets with limited historical periods. We explore novel approaches to return period estimates using bayesian hierarchical modeling, and compare it to frequency distribution methods. These steps, necessary to creating accurate return period estimates, are paramount to develop robust insurance products, inform policy makers and disaster responses and enable both insurers and governments to develop trusted methods in assessing payout thresholds.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFMNH55A..07G