Soil Moisture Estimation Using Hyperspectral SWIR Imagery

The U.S. Geological Survey (USGS) is engaged with the U.S. Department of Agriculture's (USDA) Agricultural Research Service (ARS) and the University of Georgia's National Environmentally Sound Production Agriculture Laboratory (NESPAL) both in Tifton, Georgia, USA, to develop transformations for medium and high resolution remotely sensed images to generate moisture indicators for soil. The Institute for Technology Development (ITD) is located at the Stennis Space Center in southern Mississippi and has developed hyperspectral sensor systems that, when mounted in aircraft, collect electromagnetic reflectance data of the terrain. The sensor suite consists of sensors for three different sections of the electromagnetic spectrum; the Ultra-Violet (UV), Visible/Near InfraRed (VNIR) and Short Wave InfraRed (SWIR). The USDA/ ARS' Southeast Watershed Research Laboratory has probes that measure and record soil moisture. Data taken from the ITD SWIR sensor and the USDA/ARS soil moisture meters were analyzed to study the informatics relationships between SWIR data and measured soil moisture. The geographic locations of 29 soil moisture meters provided by the USDA/ARS are in the vicinity of Tifton, Georgia. Using USGS Digital Ortho Quads (DOQ), flightlines were drawn over the 29 soil moisture meters. The SWIR sensor was installed into an aircraft. The coordinates for the flightlines were also loaded into the navigational system of the aircraft. This airborne platform was used to collect the data over these flightlines. In order to prepare the data set for analysis, standard preprocessing was performed. These standard processes included sensor calibration, spectral subsetting, and atmospheric calibration. All 60 bands of the SWIR data were collected for each line in the image data, 15 bands of which were stripped from the data set leaving 45 bands of information in the wavelength range of 906 to 1705 nanometers. All the image files were calibrated using the regression equations generated by using radiometer data collected over calibration tarps. Regions of Interest (ROI) were drawn over the image data set corresponding with the location of the soil moisture meters. Scripts written in ENVI's Interactive Data Language (IDL) were developed to extract the spectra from each of the processed hyperspectral image data over each soil moisture meter from its corresponding ROI. The informatics relationship between soil moisture and SWIR spectra was identified by using the resulting data set.