Predicting Near Real-Time Inundation Occurrence from Complimentary Satellite Microwave Brightness Temperature Observations
Abstract
Throughout the world, there is an increasing need for new methods and data that can aid decision makers, emergency responders and scientists in the monitoring of flood events as they happen. In many regions, it is possible to examine the extent of historical and real-time inundation occurrence from visible and infrared imagery provided by sensors such as MODIS or the Landsat TM; however, this is not possible in regions that are densely vegetated or are under persistent cloud cover. In addition, there is often a temporal mismatch between the sampling of a particular sensor and a given flood event, leading to limited observations in near real-time. As a result, there is a need for alternative methods that take full advantage of complimentary remotely sensed data sources, such as available microwave brightness temperature observations (e.g., SMAP, SMOS, AMSR2, AMSR-E, and GMI), to aid in the estimation of global flooding. The objective of this work was to develop a high-resolution mapping of inundated areas derived from multiple satellite microwave sensor observations with a daily temporal resolution. This system consists of first retrieving water fractions from complimentary microwave sensors (AMSR-2 and SMAP) which may spatially and temporally overlap in the region of interest. Using additional information in a Random Forest classifier, including high resolution topography and multiple datasets of inundated area (both historical and empirical), the resulting retrievals are spatially downscaled to derive estimates of the extent of inundation at a scale relevant to management and flood response activities ( 90m or better) instead of the relatively coarse resolution water fractions, which are limited by the microwave sensor footprints ( 5-50km). Here we present the training and validation of this method for the 2015 floods that occurred in Houston, Texas. Comparing the predicted inundation against historical occurrence maps derived from the Landsat TM record and MODIS imagery, we find good agreement for most areas and are able to provide a daily mapping given the increased temporal coverage. These results illustrate the feasibility of a near real-time inundation prediction system driven by multi-sensor satellite microwave observations, which can be extended to provide a daily estimate of global flooding.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2017
- Bibcode:
- 2017AGUFM.H21D1487F
- Keywords:
-
- 1820 Floodplain dynamics;
- HYDROLOGY;
- 1821 Floods;
- HYDROLOGY;
- 1847 Modeling;
- HYDROLOGY;
- 4331 Disaster relief;
- NATURAL HAZARDS