Operational Atmospheric Correction of Landsat TM Data

The recent algorithms developed for biophysical variables assessment require accurate surface reflectance measurements. This article describes algorithms used for atmospheric correction of Landsat Thematic Mapper (TM) data. Atmospheric corrections include Rayleigh scattering, gaseous absorption, and aerosol scattering in three visible channels (480 nm, 560 nm, and 660 nm), and the near-infrared channel (830 nm). Atmospheric constituents such as water vapor and ozone are extracted from climatology data sets, while aerosol optical depths (AODs) are derived from the TM scene itself by adopting the dark target approach. The dark target pixels are identified, and their reflectances in the visible channels are estimated using TM Channel 7 (2.1 μm). Atmospheric transmittance and aerosol optical depth are derived for 16 grid points equally distributed over the scene, then interpolated to match the TM spatial resolution. This technique considerably reduces the computing time without decreasing the accuracy. These algorithms were tested using 11 TM scenes over a wide variety of sites, including forest, crop, and semiarid areas. The AOD in the blue, green, and red channels retrieved using the dark target technique was validated using sunphotometer measurements. The absolute error associated with AOD assessment was less than 0.15. A statistical analysis was also conducted to evaluate the atmospheric correction method. Based on data from the FIFE (First ISLSCP Field Experiment) experiment, the absolute error between ground measurements and TM reflectance was less than 0.015 in the visible channels, and less than 0.08 in the near-infrared channel.