Mapping Inundation and Changes in Wetland Extent with L-band SAR: A Combined Data and Medeling Approach

Author: Arindam Samanta and John F. Galantowicz
Date: 
December 6, 2011
Type: 
Poster presentation
Venue: 
AGU Fall Meeting 2011
Citation: 

John F. Galantowicz, Arindam Samanta (2011) Mapping Inundation and Changes in Wetland Extent with L-band SAR: A Combined Data and Medeling Approach. AGU Fall Meeting 2011.

Accurate mapping of seasonal and inter-annual changes in inundation and wetland extent is a key requisite for the estimation of greenhouse gas (GHG, e.g., methane) emissions from land surfaces to the atmosphere. This task would benefit from the 1- to 3-km spatial resolution L-band synthetic aperture radar (SAR) and 3-day revisit time of NASA’s Soil Moisture Active Passive (SMAP) mission, planned for launch in 2014. With a view to utilizing this unique capability, we propose a method for mapping the fraction of area inundated using a combination of semi-empirical models of radar backscatter and L-band SAR data. Inundation exhibits a characteristic radar backscatter that is affected by a set of factors, including roughness of soil and water surfaces, and presence of vegetation cover. Further, the impact of vegetation cover on radar backscatter from underlying soil and/or water surface will depend on biome type. The effects of these factors on both the like-polarized (HH, VV) and cross-polarized (HV) radar backscatter was investigated using semi-empirical models. A key step in devising an inundation fraction retrieval algorithm is to benchmark and calibrate the backscatter simulated with semi-empirical models against SAR data. This task was undertaken using data from the Phased Array L-Band Synthetic Aperture Radar (PALSAR) instrument onboard Japan’s Earth Resources Satellite’s (JERS, e.g., Fig. 1). This calibration was performed in the following way. First, using a Monte-Carlo type of approach, a large set of random backscatter samples was extracted from different landcover classes, including dry forests and clear-cut areas, inundated forests (wetlands), and open water. Second, mean backscatter was calculated at varying spatial resolutions: 100 m, 500 m, 1 km, 2 km, 3 km and 10 km. Third, the mean model backscatter was set to the mean PALSAR backscatter for each landcover class, but the model dispersion was retained. Finally, using these calibrated values, backscatter for varying open water and inundated forest (wetland) fractions were calculated and are stored in look-up-tables (LUTs), which will be modified to include information about SMAP-specific noise, footprint geolocation errors, etc. This approach is being developed and tested in four regions: Amazon forests, Hudson Bay Lowlands (Canada), Yukon-Kushkokwim Delta (Alaska) and Mississippi Delta.