基于伪逆学习算法的地基微波辐射计反演算法研究

利用兰州大学半干旱气候与环境观测站(SACOL站)2009-2010年的地基微波辐射计亮温资料和榆中站探空资料,基于伪逆学习算法建立了应用于地基微波辐射计温度、相对湿度和水汽密度反演的神经网络(PIFN),并将反演结果与地基微波辐射计自带反演产品进行了对比,研究了伪逆学习算法在地基微波辐射计气象要素反演算法本地化的应用效果.结果表明:PIFN反演的温度、相对湿度和水汽密度的均方根误差的最大值分别为6.41K,31.21%和1.5g/m3,地基微波辐射计温度、相对湿度和水汽密度产品的均方根误差最大值分别为11.93K,53.18%和3.06g/m3,与微波辐射计自带神经网络反演结果在不同高度层进行比较可以看出PIFN对2～10km、1～7km和0～3km的大气温度、相对湿度和水汽密度廓线的反演均有明显改善,伪逆学习算法能够应用于地基微波辐射计气象要素的反演算法的本地化.

Research of an Inversion Method Based on Pseudoinverse Learning Algorithm for Ground-Based Microwave Radiometer Measurement

In order to estimate the application of pseudoinverse learning algorithm in the localization of ground-based microwave radiometer meteorological elements inversion algorithms, the observed data of the ground-based microwave radiometer from the Semi-Arid Climate and Environment Observatory (SACOL) of Lanzhou University and the radiosonde data from the Yuzhong Station during 2009 to 2010 are used to establish a neural network (PIFN) for temperature, humidity and water vapor density inversion based on the pseudoinverse learning algorithm (PLA), and the inversion results are compared with the products of the ground-based microwave radiometers. The result show that the maximum mean square root error of temperature, relative humidity and water vapor density inversed by PIFN are 6.41 K, 31.21% and 1.5 g/m³, respectively, and the maximum root mean squared root error of temperature, relative humidity, and water vapor density products recorded by the ground-based microwave radiometers are 11.93 K, 53.18% and 3.06 g/m³, respectively. PIFN significantly improves the inversion performance of temperature, relative humidity and water vapor density profiles between 2~10 km, 0~3 km, 1~7 km, respectively. It is concluded that the inversion result of PLFN has a better performance than the microwave radiometer''s own products and is more close to the radiosonde data and that PLA can be introduced to ground-based microwave radiometer inversion algorithm localization field.