Comparison Of LST Retrieval Precision And Sensitivity Analysis Between Two Split-Window Algorithems

Land Surface Temperature(LST), as an important parameter of energy balance and water balance on land surface, is widely applied in many fields especially aerography, geology, hydrology, ecology and so on. LST retrieval is one of the most important quantitative application of thermal infrared remote sensing. Some Split window algrithms have been developed well for LST retrieval. In this paper, we selected two split-window algorithms, as QIN split-window and Wan-Dozier generalized split-window algorithm, on the purpose of comparing of LST retrieval precision based on Moderate Resolution Imaging Spectroradiometer (MODIS) data between them and studying the factors affecting the LST retrieval. For this object, firstly, MODIS 1B data and two split-window algorithms including Qin and Wan-Dozier were adopted to retrieve land surface temperature. Secondly, MODTRAN4.0 was employed in order to ascertain the atmospheric coefficients of QIN split-window algorithm and Wan-Dozier generalized split-window algorithm based on Thermodynamic Initial Guess Retrieval (TIGR) database. Algorithm fitting precision analysis and the sensitivity analysis of the two algorithms were conducted. The total errors of land surface temperature were estimated by means of evaluationg the influencecs of several parameters, such as atmospheric water vapor content, land surface emissivity and the noise of the sensor. The results show that when the real LST is 300K, on the condition of mid-latitude standard summer atmosphere, the main error sources are the algorithm fitting precision and the uncertainty of the land surface emissivity for both Qin split-window algorithm and Wan-Dozier generalized split-window algorithm, a retrieval test is performed with QIN split-window algorithm and MOD021KM data acquired on June 23th,2009, which covered the southwest of Shandong province, China. LST results are compared with MODIS LST products imaged at the same time and location. The comparison showed that the distribution of LST retrieved by the two models is similar. And average LST of the study area calculated using the Qin model is 0.22K higher than that of MODIS LST products. The difference of land surface emissivity estimation is one of the main reasons contributing to the LST distraction over 1 k.