利用AMSR-E被动微波数据反演地表温度的神经网络算法

摘要

This paper utilizes the characteristic of multiple-sensor/multiple-resolution of the AQUA (an earth observing satellite) and the neural network to retrieve land surface temperature from the AMSR-E data. The MODIS land surface temperature ( LST) product is made as the ground data, and the average value of part MODIS pixels in an AMSR-E pixel can be used to overcome the influence of cloud. The retrieval result and analysis indicate that the neural network can be used to accurately retrieve land surface temperature from AMSR-E data. The accuracy is the highest when five frequencies (ten channels) are used, which shows that using more channels can better eliminate the influence of soil moisture, roughness, atmosphere and other influence factors. The average land surface temperature error is under 2 K relative to the MODIS LST product.%结合对地观测卫星AQUA多传感器/多分辨率的特点,研究了利用AMSR-E被动微波数据反演地表温度的神经网络算法.MODIS地表温度(LST)产品被作为地表温度实测数据,对应的平均温度被用作对应AMSR-E像元的实际地表温度,从而克服由于AMSR-E像元尺度太大和云的影响而难以获得地表实测数据的难点.反演结果分析表明,利用神经网络能够精确地由AMSR-E数据反演地表温度.当使用5个频率10个通道反演时,反演精度最高,说明使用更多的通道能更好地消除土壤水分、粗糙度、大气和其它因素的影响.相对于MODIS温度产品,用此算法反演的平均误差约低于2K.