Atmospheric boundary layer is a key parameter for boundary layer studies,including meteorology,air quality and climate.The atmospheric boundary layer height estimates are inferred from local radiosonde measurements or remote sensing observations from instruments like laser radar,wind profiling radar or so-dar.Methods used to estimate atmospheric boundary layer height from radiosonde profiles are also used with atmospheric temperature and humidity profiles retrieved by microwave radiometers.An alternative approach to estimate atmospheric boundary layer height from microwave radiometer data is proposed based on microwave brightness temperatures,instead of retrieved profiles.Using the ground-based microwave radiometer and laser radar atmospheric boundary layer height obtained in 2013 at Xianghe Station,algo-rithms for retrieving atmospheric boundary layer height from 14-channel microwave brightness tempera-tures are developed based on the nonlinear neural network and multiple linear regression methods.The at-mospheric boundary layer height is derived from laser radar backscattering data using the algorithm that retrieves the most significant gradients in profiles using gradient method.Root mean square errors (RM-SEs)and correlation coefficient with two kinds of method are obtained to analyze which method is better through comparison.Retrieval results with the neural network method are compared in different periods of time and weather conditions.It shows that neural network algorithm is better than the multiple linear re-gression algorithm because results are more consistent with the observation.The correlation coefficient be-tween the lidar-detected and neural network algorithm retrieved boundary layer height is 0.83,which is a-bout 26% higher than the multiple linear regression algorithm retrieved result.Also,RMSEs of the neural network algorithm retrieved values (268.8 m)are less than the multiple linear regression algorithm re-trieved values (365.1 m).For different time periods and weather conditions,retrievals in spring are best of four seasons,retrievals in the clear sky are better than those in the cloudy sky.But RMSEs in the cloud sky are less than those in the clear sky.Overall,correlation coefficients in four seasons are close to 0.80. It suggests that in order to improve the retrieval precision,specific retrievals under different conditions (such as different seasons and different skies)should be carried out.%采用2013年中国科学院大气物理研究所香河大气综合观测试验站的地基微波辐射计和激光雷达观测数据，以激光雷达探测的大气边界层高度为参考，分别利用非线性神经网络和多元线性回归方法建立微波亮温直接反演大气边界层高度的算法，并对比两种方法的反演能力，同时分析非线性神经网络算法在不同时段及不同天气状况下反演结果的差异。结果表明：非线性神经网络算法的反演能力优于多元线性回归算法，其反演结果与激光雷达探测的大气边界层高度有较好一致性，冬、春季的相关系数达到0．83，反演精度比线性回归算法约高26％；对于不同时段和不同天气条件，春季的反演结果最好，晴空的反演结果好于云天；四季和不同天气状况的划分也有利于提高反演精度。
展开▼
