Skip to main content
ARS Home » Plains Area » Bushland, Texas » Conservation and Production Research Laboratory » Soil and Water Management Research » Research » Publications at this Location » Publication #258521

Title: Downscaling of Aircraft, Landsat, and MODIS-bases Land Surface Temperature Images with Support Vector Machines

item Ha, Wonsook
item Gowda, Prasanna
item OOMMEN, THOMAS - Michigan Technological University
item Howell, Terry
item Hernandez, Jairo

Submitted to: American Geophysical Union
Publication Type: Abstract Only
Publication Acceptance Date: 10/15/2010
Publication Date: 12/15/2010
Citation: Ha, W., Gowda, P., Oommen, T., Howell, T.A., Hernandez, J.E. 2010. Downscaling of Aircraft, Landsat, and MODIS-bases Land Surface Temperature Images with Support Vector Machines [abstract]. American Geophysical Union Meeting, December 13-17, 2010, San Francisco, California. Paper No. H31B-0994.

Interpretive Summary:

Technical Abstract: High spatial resolution Land Surface Temperature (LST) images are required to estimate evapotranspiration (ET) at a field scale for irrigation scheduling purposes. Satellite sensors such as Landsat 5 Thematic Mapper (TM) and Moderate Resolution Imaging Spectroradiometer (MODIS) can offer images at several spectral bandwidths including visible, near-infrared (NIR), shortwave-infrared, and thermal-infrared (TIR). The TIR images usually have coarser spatial resolutions than those from non-thermal infrared bands. Due to this technical constraint of the satellite sensors on these platforms, image downscaling has been proposed in the field of ET remote sensing. This paper explores the potential of the Support Vector Machines (SVM) to perform downscaling of LST images derived from aircraft (4 m spatial resolution), TM (120 m), and MODIS (1000 m) using normalized difference vegetation index images derived from simultaneously acquired high resolution visible and NIR data (1 m for aircraft, 30 m for TM, and 250 m for MODIS). The SVM is a new generation machine learning algorithm that has found a wide application in the field of pattern recognition and time series analysis. The SVM would be ideally suited for downscaling problems due to its generalization ability in capturing non-linear regression relationship between the predictand and the multiple predictors. Remote sensing data acquired over the Texas High Plains during the 2008 summer growing season will be used in this study. Accuracy assessment of the downscaled 1, 30, and 250 m LST images will be made by comparing them with LST data measured with infrared thermometers at a small spatial scale, upscaled 30 m aircraft-based LST images, and upscaled 250 m TM-based LST images, respectively.