Test of the MODIS Land Surface Temperature and Emissivity Separation Algorithm With Ground Measurements Over a Rice Paddy

The Moderate Resolution Imaging Spectroradiometer (MODIS) land surface temperature (LST) and emissivity separation (MODTES) algorithm is the basis of the MOD21 product, which provides 1-km LSTs and emissivities for bands 29 (8.55 mu m), 31 (11 mu m), and 32 (12 mu m). The MODTES algorithm uses the TES method with the water vapor scaling (WVS) method for refined atmospheric correction. The performance of the MODTES algorithm was tested with a set of MODIS data concurrent with ground LST and emissivity measurements. The test site is a large area of homogeneous full-cover rice crops (graybody), with high atmospheric water vapor. The data included LSTs measured along transects with multiple calibrated radiometers and emissivity measurements in different bands within 8-13 mu m. We applied the full MODTES algorithm and that without the WVS method for comparison. For the data used here, MODTES minus in situ LSTs yielded a median difference of 0.5 K and a robust standard deviation (RSD) of 0.4 K. When the WVS method was not applied, we obtained a median bias of 1.3 K and an RSD of 0.9 K. MODTES retrieved emissivities were (median +/- RSD) 0.966 +/- 0.011 for band 29, 0.976 +/- 0.004 for band 31, and 0.981 +/- 0.003 for band 32 (0.961 +/- 0.029, 0.971 +/- 0.013, and 0.978 +/- 0.007 without WVS), underestimating the ground values by 0.019, 0.009, and 0.001, respectively. Emissivities from the full MODTES algorithm presented less dispersion and were closer to the ground data, particularly for lower water vapor cases. These results show that MODTES performs within the predicted algorithm uncertainty and the importance of the WVS method for reducing residual errors in atmospheric correction.