Communicating Flood Risk in Near Real-Time - Turning Pixel Flood Maps Into Three Word Warnings

Abstract:

In communicating flood events, scientists may overlook the interpretability of the graphics or maps they provide disaster management and emergency response stakeholders. Clear categorical rankings of risk (e.g. high, medium, or low) that benchmark an observed flood against the historical satellite record can help responders develop a preliminary understanding of a flood’s severity as well as the urgency for dispatching aid. This project presents a methodology developed by Cloud to Street for categorizing risk of a near real time flood event compared to historic observed flood frequency. By leveraging the computing of Google Earth Engine, we map water across Landsat, Sentinel-1, and Sentinel-2 satellites following heavy rainfall flooding along the Volta river in Northern Ghana in 2018. We use the historical satellite observation record in Ghana to identify if current flood observations are lower, higher, or congruent with historical flood observations. Our approach adjusts for sensor bias in flood area predictions. We find Sentinel-1 (a radar satellite) consistently detects areas that more frequently experience flooding, while Landsat 8 and Sentinel-2 are more likely to capture areas that are less commonly flooded. We examine the relationship between flood return period estimates and cloud cover by simulating clouds (based on historic cloud patterns) to determine the minimum cloud free area required to make a risk classification. These results highlight the need to account for imagery sources and visibility when assigning flood risk categories. Leveraging historic satellite records, we can communicate risk from near real-time satellite observations to non-experts in an intuitive and simple way.