Comparison between different sources of atmospheric profiles for land surface temperature retrieval from single channel thermal infrared data

摘要

Different sources of atmospheric water vapor and temperature profiles were used with a radiative transfer model for retrieving land surface temperature (LST) from thermal infrared remote sensing data with the so-called single channel (SC) method. Retrieved LSTs were compared to concurrent ground measurements over homogeneous rice fields to assess the accuracy of the atmospheric profiles. These included radiosonde balloons launched at the test site near-concurrently to satellite overpasses, re-analysis profiles from the National Centers for Environmental Prediction (NCEP), and satellite sounder products from the Atmospheric Infrared Sounder (AIRS) and the Moderate Imaging Spectroradiometer (MODIS; MOD07 product). SC LSTs were computed for Enhanced Thematic Mapper+ (ETM+), Advanced Spaceborne Thermal Emission and Reflection radiometer (ASTER), MODIS, and Advanced Along-Track Scanning Radiometer (AATSR). Results show that radiosonde profiles provided the best agreement between ground-measured and satellite-derived LSTs, with root mean square difference (RMSD) better than 1.0K and biases within ±0.5K for most of the cases. As an alternative to radiosonde profiles, NCEP and MOD07 data yielded reasonable results with RMSDs around 1.0K, although LSTs derived from MOD07 profiles showed a slight overestimation (0.5 to 1.0K) of the ground LSTs. AIRS profiles usually underestimated the ground LSTs by 1 to 2K, probably due to the large temporal gap (2-3h) with the other satellite measurements. We propose a test to assess the suitability of atmospheric profiles applicable to sensors with bands at 11 and 12μm in the split-window. So this test plus the SC method may be called split-window SC algorithm, being significantly different from the simple SC method. It implies the calculation of the difference between the LST derived from both bands (T _(ig)-T _(jg)), which should be close to zero if the atmospheric profile is accurate and the surface emissivity is well known. A small range of T _(ig)-T _(jg) values around zero can be set for which the LST derived at 11μm is accurate. Using only the profiles passing the test, LSTs derived from MODIS band 31 agreed with the ground data with a mean bias of -0.1K (ground minus satellite) and RMSD of 0.6K, while AATSR band at 11μm yielded a -0.1K bias and RMSD=0.5K.