Surface water detection from passive microwave and optical data using deep learning

Keywords: surface water, MODIS, AMSR2, convolutional network, data fusion, floods, passive mircrowave
Session Type: Virtual Paper Abstract
Day: Tuesday
Session Start / End Time: 3/1/2022 09:40 AM (Eastern Time (US & Canada)) - 3/1/2022 11:00 AM (Eastern Time (US & Canada))
Room: Virtual 36

Abstract

An increasing number of people are exposed to floods which cause significant damage to livelihood and properties annually. During such flood events, timely detection of surface water extents can be crucial. However, land surface observations via optical imagery are often significantly affected by cloud cover. Passive microwave sensors capture land surface information that penetrates clouds, such as brightness temperature. Surface brightness temperature information from underneath cloud cover can potentially improve surface water detection. Therefore, surface water extent information from optical imagery might be improved by incorporating data from passive microwave sensors during overcast conditions. However, combining passive microwave data with optical imagery is a challenge due to significant spatial resolution differences. For example, AMSR2 captures imagery between 6km to 36km, depending on the temporal resolution, and MODIS captures optical imagery in finer spatial resolutions - 250m, 500m, and 1km. Deep learning algorithms have been successful in extracting information from multi-modal satellite datasets to detect land surface features. Additionally, deep learning algorithms can downscale imagery from coarse to fine resolutions. In this study, we will incorporate coarse resolution AMSR2 and fine resolution MODIS data to map surface water globally across various magnitudes of cloud cover at a 1km resolution. Primarily, we will investigate the value of incorporating cloud penetrating observations from passive microwave sensors with cloud-affected optical data for detecting surface water. This research is a part of a larger objective of combining multi-source multi-modal data to improve urban flood detection.