针对Terra/MODIS数据提出改进的分裂窗地表温度反演算法。充分考虑了传感器观测角度(VZA)的影响,并对地表和有效大气辐射按照不同的亮度温度区间分别进行Planck函数简化。利用TIGR3大气廓线库中的875条晴空大气廓线,ASTER波谱库中的106条地物发射率波谱,结合MODTRAN4大气辐射传输模型模拟得到分裂窗算法系数。利用MODTRAN4模拟数据对算法精度进行验证,结果表明本文的改进算法和原算法的均方根误差RMSE分别为0.34 K 和0.65 K。敏感性分析表明,在中等湿润的大气条件下,算法对大气水汽含量并不敏感。该算法降低了传感器观测角度带来的地表温度反演误差。利用2009年6月美国SURFRAD辐射观测网6个站点的实测数据对改进算法、原算法以及MOD11_L2地表温度产品进行了对比验证,RMSE分别是0.93 K、1.49 K和1.0 K,表明本文算法可以提高反演精度。

This paper presents an algorithm for the retrieval of daytime land surface temperature (LST) from the Terra/MODIS data, which considers the atmospheric radiation effects due to the viewing zenith angle (VZA) variation. The MODTRAN4 model, 875 profiles of TIGR3 database and 106 surface emissivity spectra of the ASTER spectral library were used to obtain the Split-Window Algorithm (SWA) coefficients. The Root Mean Square Errors (RMSEs) of LST retrieval using the MODTRAN4 simulation are 0.34 K. Sensitivity analysis confirmed that the algorithm is not sensitive to total column water vapor content (TCWVC) for the moderately moist atmospheric conditions. In addition, LST retrieval error due to the VZA effect was reduced. Retrieved LSTs have compared with Mao, et al.'s LST and MOD11_L2 LST. Surface Radiation (SURFRAD) budget network measurements have been used for LST validation over six sites during the entire month of June 2009. The RMSE values of LST were 0.93 K, 1.49 K and 1.0 K for this new algorithm, Mao, et al.'s algorithm and MOD11_L2 LST, while the average biases were -0.66 K, 1.34 K and -0.38 K, respectively.