摘要
利用拉曼激光雷达进行高分辨率、连续的大气观测,可以为清晰理解和定量认识大气动力-物理-化学过程在垂直方向上的耦合关系提供有效支撑。为进一步提升拉曼激光雷达的探测能力,研究提出一种基于广义回归神经网络(GRNN)的拉曼激光雷达温度反演方法,通过将振动拉曼后向散射信号、高低量子数(J)转动拉曼后向散射信号、海拔高度作为自变量,真实大气温度作为因变量投入GRNN模型,训练出高精度的大气温度反演模型。该方法可以通过振动拉曼散射信号的修正有效改善低量子数转动拉曼通道高空信噪比低和弹性后向散射泄漏所造成的反演误差,优化系统定标过程,增加系统有效探测距离。团队于南京国家基准气候站(31.93°N,118.90°W)建立一套拉曼激光雷达系统,得到振转拉曼信号原始数据,结合站点的探空数据训练出大气温度反演模型并使用模型进行了反演。效果测试所得大气温度与探空数据的均方根误差(RMSE)为0.621 K,平均绝对误差(MAE)为0.128 K,相关系数为0.999,预报准确率高达99.948%,相对纯转动拉曼激光雷达温度反演方法有较大提升。模型12月的应用结果也进一步验证其应用于温度廓线反演的有效性。
Raman LiDAR systems enable high-resolution,continuous atmospheric profiling critical for elucidating vertical coupling mechanisms among dynamical,physical,and chemical processes.This study presents an optimized temperature retrieval methodology leveraging Generalized Regression Neural Networks(GRNN)to enhance traditional Raman lidar performance.By inputting vibrational Raman backscatter signals,rotational Raman backscatter signals for high and low quantum numbers(J),altitude as independent variables,and actual atmospheric temperature as the dependent variable into the GRNN model,a high-precision atmospheric temperature inversion model is trained.This method can effectively improve the inversion error caused by the low signal-to-noise ratio in high altitudes of low quantum number rotational Raman channels and leakage of elastic backscatter by correcting vibrational Raman scattering signals,optimize the system calibration process,and increase the effective detection distance of the system.A Raman lidar system was established at the Nanjing National Benchmark Climate Station(31.93°N,118.90°E),where raw vibrational-rotational Raman signal data were collected.The model was trained using radiosonde data and validated through temperature inversion experiments.Results demonstrated a root mean square error(RMSE)of 0.621 K,a mean absolute error(MAE)of 0.128 K,a correlation coefficient of 0.999,and a forecasting accuracy of 99.948%between inverted temperatures and radiosonde data,representing a significant improvement over conventional rotational Raman lidar methods.Application in December further confirmed the method's effectiveness for temperature profile inversion.
作者
张珉铨
曹华
彭智臻
蒋琦
黄驿森
杨彬
卜令兵
Zhang Minquan;Cao Hua;Peng Zhizhen;Jiang Qi;Huang Yisen;Yang Bin;Bu Lingbing(School of Atmospheric Physics,Nanjing University of Information Science and Technology,Nanjing 210044,Jiangsu,China;Key Laboratory for Aerosol-Cloud-Precipitation of China Meteorological Administration,Nanjing University of Information Science and Technology,Nanjing 210044,Jiangsu,China;Jiuquan Meteorological Administration,Jiuquan 735000,Gansu,China)
出处
《应用激光》
北大核心
2026年第1期135-148,共14页
Applied Laser
基金
国家自然科学基金(42175145)
国家级大学生创新创业训练计划支持项目(202310300031Z)。
关键词
拉曼激光雷达
神经网络
机器学习
温度反演
raman lidar
neural network
machine learning
temperature inversion
