Reflectance factor retrieval from Landsat TM and SPOT HRV data for bright and dark targets

Abstract

In recent years, there have been many land-surface studies based on visible and near-infrared reflectance values retrieved from the Landsat Thematic Mapper (TM) and SPOT High Resolution Visible (HRV) sensors. Retrieval of reflectance from satellite sensor digital count requires knowledge of the atmospheric conditions and the sensor absolute calibration. In most cases, atmospheric conditions are simulated with a radiative transfer code and sensor calibration coefficients are obtained from preflight sensor calibrations or in-flight calibrations over bright surfaces (such as White Sands, New Mexico, USA, or La Crau, France). Though these procedures are well accepted, there have been few studies specifically designed to validate the accuracy of such reflectance factor retrievals (RFR) for both bright and dark targets. Data from two experiments conducted in an agricultural region in central Arizona were analyzed to quantify the accuracy of RFR from the Landsat TM and SPOT HRV sensors. These data included measurements made with groundbased and aircraft-based four-band radiometers and the NASA Advanced Solid-State Array Spectrometer (ASAS) aboard a C130 aircraft, and TM and HRV images acquired at nadir and off-nadir viewing angles. Results showed that the off-nadir reflectance factors measured using ground- and aircraft-based instruments, including ASAS, were comparable. The RFR from the satellite-based TM and HR V sensors generally resulted in an overestimation of dark target reflectance (up to 0.05 reflectance in the visible) and an underestimation of bright target reflectance (up to 0.1 reflectance in the near-infrared). Even greater error was possible when RFR was based on outdated sensor calibrations, particularly those conducted prelaunch. There was supporting evidence from studies at three sites (White Sands, New Mexico; Maricopa, Arizona; and Walnut Gulch, Arizona) that the Landsat-5 TM sensor sensitivity may have degraded by as much as 20% from the prelaunch calibration. Regarding the potential error in RFR related to recent changes in the processing of Landsat TM data (Level-0 and Level-1) by EOSAT Corporation, we found that the Level-0 data was slightly greater (∼2 digital counts) than the Level-1 data for all bands and all targets in our study.