Developing High-Resolution Inundation Estimates through a Downscaling of Brightness Temperature Measurements

Abstract:

There is currently a large demand for high-resolution estimates of inundation extent and flooding for applications in water management, risk assessment and hydrologic modeling. In many regions of the world it is possible to examine the extent of past inundation from visible and infrared imagery provided by sensors such as the Moderate Resolution Imaging Spectroradiometer (MODIS); however, this is not possible in regions that are densely vegetated or are under persistent cloud cover. As a result of this, there is a need for alternative methodologies that make use of other remotely sensed data sources to inform high-resolution estimates of inundation. One such data source is the AMSR-E/Aqua 37 GHz vertically and horizontally polarized brightness temperature measurements, which have been used in previous studies to estimate the extent of inundated areas and which can make observations in vegetated or cloudy regions. The objective of this work was to develop a decision tree classifier based downscaling methodology by which inundation extent can be obtained at a high resolution, based on microwave brightness temperature measurements and high resolution topographic information.

Using a random forest classifier that combined the AMSR-E 37GHz brightness temperatures (~12km mean spatial resolution) and a number of high-resolution topographic indices derived from the National Elevation Dataset for the United States (30m spatial resolution), a high-resolution estimate of inundation was created. A case study of this work is presented for the severe flooding that occurred in Iowa during the summer of 2008. Training and validation data for the random forest classifier were derived from an ensemble of previously existing estimates of inundation from sources such as MODIS imagery, as well as simulated inundation extents generated from a hydrologic routing model. Results of this work suggest that the decision tree based downscaling has skill in producing high-resolution estimates of inundation when informed by the brightness temperature measurements along with high quality training data and can be used to estimate the likelihood of inundation for the region of interest.