Radiance-based validation of the V5 MODIS land-surface temperature product

This paper presents the procedure and results of the radiance-based validation approach for the Moderate Resolution Imaging Spectroradiometer (MODIS) Land-Surface Temperature (LST) product. Surface emissivity spectra were retrieved by a sun-shadow method from surface-leaving radiance spectra measured with a thermal infrared (TIR) spectroradiometer in the 3.5-14 mu m spectral region under sunshine and sun-shadow conditions. By using the measured surface emissivity spectrum and atmospheric profiles obtained by radiosonde balloons, and the LST values at validation sites in the V5 MODIS level-2 LST products, radiative transfer simulations were made with the MODTRAN4 code to calculate the top-of-atmosphere (TOA) radiance values in MODIS band 31 (L31). By adjusting the LST input values in the simulations to match the calculated L31 values to the MODIS measured radiance (MOD L31) values, MOD L31 inverted LSTs can be obtained. The MODIS LST product was validated by comparison to the values of the MOD L31 inverted LSTs. This approach compares well with the conventional temperature-based approach. The results of the radiance-based validation indicate that the accuracy of the MODIS LST product is better than 1 K in most cases, including lake, vegetation and soil sites in clear-sky conditions. The errors in the split-window retrieved LSTs may be larger in bare soil sites and highly heterogeneous sites due to large uncertainties in surface emissivities. The results of the radiance-based validation also reveal the weakness of the split-window algorithm used for the generation of the MODIS LST product in two situations: one in cases where LSTs are larger than the air temperature at the surface level (T(s-air)) by more than 16 K and the columnar water vapour (cwv) is larger than 1.5 cm, and another in cases under the influence of thin cirrus clouds or heavy aerosol loadings. These two situations were not considered in the development of the current MODIS LST algorithm.