An objective methodology for infrared land surface emissivity evaluation

Land surface emissivity (LSE) in the infrared (IR) window region (8 -12 μm) governs the thermal emissions from the Earth's surface. Many LSE databases, retrieved from various satellite instruments, are available for studying climate, earth-atmosphere interaction, weather and the environment. The precision (standard deviation) and accuracy (bias) of these databases remain unclear. In this study, we introduce an efficient method for quantitatively evaluating the LSE precision using satellite radiance observations. The LSE Tb (brightness temperature) deviations, defined as the standard deviations of Tb differences between satellite observations and radiative transfer calculations, can be estimated by minimizing the impacts from land surface temperature (LST) and atmospheric profiles. This is followed by the estimation of LSE precision. This method does not need the true LSE measurements. It only needs ancillary information such as atmospheric profiles and LST, both of which do not require high accuracy and thus can be obtained from a numerical weather prediction (NWP) forecast or analysis. The method is applied to six different monthly LSE databases of August: the University of Wisconsin – Madison (UW) High Spectral Baseline Fit (UWiremis) database, the IASI research products, the operational AIRS ascending orbit (AIRS-A) products (version 5), the operational AIRS descending orbit (AIRS-D) products (version 5), the operational MODIS/Aqua (MODIS-A) and MODIS/Terra (MODIS-T) products (version 4.1). The satellite radiance observations are from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) at 6 UTC on 01 August 2006. For 8.7 μm the UWiremis is the most precise (0.027) since laboratory hyperspectral emissivity measurements are incorporated. The operational MODIS products (0.028 and 0.029 for MODIS-T and MODIS-A) have better precisions than the emissivity databases derived from AIRS and IASI. The operational AIRS databases (≤ 0.036) have better precisions than IASI (0.038), indicating the physical algorithm is able to improve the regression retrieval. The AIRS-D (0.035) has a slightly better precision than AIRS-A (0.036) because there is less moisture in the air at night than during the day. For 10.8 and 12 μm, the AIRS physical algorithm fails to improve the first guess; the precisions are worse than 0.020. The precisions (≤ 0.017) of the MODIS products (UWiremis, MODIS-T, and MODIS-A) are surprisingly better than both the regression of IASI and the physical retrievals of AIRS, indicating the classification-based modeling might be a better way to derive LSE for these two channels, although this may be in part due to processing differences. The disadvantage though is its inability to capture the LSE temporal variations.