一个从MODIS数据同时反演地表温度和发射率的神经网络算法

MODIS的三个热红外波段29、31、32建立了三个辐射传输方程,这三个方程包含了5个未知数(大气平均作用温度、地表温度和三个波段的发射率)。用JPL提供的大约160种地物的波谱数据对MODIS三个波段(29/31/32)发射率之间的关系和用MODTRAN4对大气透过率和大气水汽含量之间关系进行模拟分析。分析结果表明地球物理参数之间存在着大量的潜在信息。由于潜在的信息难以严格地用数学表达式来描述,因此神经网络是非常适合被用来解这种病态反演问题。利用辐射传输模型(RM)和神经网络(NN)反演分析表明神经网络能够被用来精确地同时从MODIS数据中反演地表温度和发射率。地表温度的平均反演误差在0.4°C以下;波段29/31/32发射率平均反演误差都在0.008以下。

An NN Algorithm for Retrieving Land Surface Temperature and Emissivity from MODIS Data

Three radiative transfer equations are built for MODIS bands 29,31 and 32 to involve five unknown parameters(average atmospheric temperature,land surface temperature and three band emissivities).Temperature and emissivity retrieval from the equations can be thus defined as an ill-posed problem.In the paper thorough analysis has been given to the relationships between different band emissivities in MODIS bands 29,31 and 32 for above 160 types of features provided by JPL,and the relationship between transmittance and water vapor.Due to the fact that some interconnections may exist among the emissivities and the temperatures,we use neural network algorithm for resolution of the five unknown parameters from the three equations.The result indicates that the algorithm is applicable for the ill-posed problem.Therefore,it can be concluded that combination of radiative transfer model(RM)with neural network(NN)algorithm can be an applicable alternative for simultaneous retrieval of land surface temperature and emissivity from MODIS thermal band data.