| 高原地区基于微波辐射计反演大气廓线的神经网络算法研究 |
| |
| 引用本文： | 田建兵,张玉欣,颜海前,唐文婷,康晓燕.高原地区基于微波辐射计反演大气廓线的神经网络算法研究[J].高原山地气象研究,2021,41(2):125-134. |
| |
| 作者姓名： | 田建兵 张玉欣 颜海前 唐文婷 康晓燕 |
| |
| 作者单位： | 青海省人工影响天气办公室，西宁 810001 |
| |
| 基金项目： | 第二次青藏高原综合科学考察研究课题（2019QZKK010406）；青海省科技计划项目（2021-ZJ-T04）；青海省气象局重点项目（QXZ2021-09，QXZ2020-07） |
| |
| 摘 要： | 以2007~2018年西宁二十里铺气象站探空资料为模拟样本，利用MonoRTM模式模拟中心频率21.985~58.759GHz的35通道亮温，应用BP神经网络对模拟数据进行反复训练，构建最优反演模型，并以2019年探空资料为测试样本，对比分析了不同季节和不同天气条件下BP神经网络与微波辐射计的反演效果。结果表明:晴空条件下，BP神经网络与微波辐射计在温度反演上效果最佳，水汽密度次之，相对湿度最差，其中冬春季BP神经网络反演效果优于微波辐射计，夏秋季反之；有云条件下，BP神经网络温度反演效果在冬、春和夏季均优于微波辐射计，其水汽密度反演效果在四季均较微波辐射计有明显提升，其相对湿度反演效果在冬、春和夏季均较微波辐射计更佳。晴空和有云条件下，BP神经网络在不同季节反演温度、水汽密度和相对湿度的平均绝对误差和标准偏差均小于微波辐射计，尤其是相对湿度的反演精度提升最为明显。晴空条件下，BP神经网络反演温度廓线在春、夏和秋季效果最佳，反演水汽密度廓线在中低层精度较高，反演相对湿度廓线的精度较差，但基本和探空资料趋势一致；有云条件下，BP神经网络反演温度廓线与晴空时基本一致，较微波辐射计精度更高，反演水汽密度和相对湿度廓线在8km以上效果较好。
|
| 关 键 词： | 微波辐射计 BP神经网络 大气廓线 反演精度 高原 |
| 收稿时间： | 2021-04-06 |
Study on Neural Network Algorithm for Retrieving Atmospheric Profile Based on Microwave Radiometer in Plateau Region |
| |
| Institution: | Weather Modification Office of Qinghai Province,Xining 810000,China |
| |
| Abstract: | Using the sounding data of Xining Ershilipu meteorological station from 2007 to 2018, the MonoRTM simulated the brightness temperature of 35 channels within 21.985~58.759GHz of the center frequency, and obtained the inversion model through repeated training of BP neural network. The 2019 data set was used as the test sample to respectively invert the temperature, relative humidity and water vapor density, and then compared with the sounding data to discuss the accuracy of the BP inversion algorithm. The results show that under clear sky conditions, BP neural network and microwave radiometer have the best inversion effcet on temperature, followed by water vapor density and relative humidity. The inversion results of BP neural network in winter and spring are better than that of microwave radiometer, and vice versa in summer and autum. Under cloud conditions, the temperature inversion effect of BP neural network is better than that of microwave radiometer in winter, spring and summer. The inversion effect of water vapor density is significantly improved than that of microwave radiometer in four seasons. The inversion effect of relative humidity is better than that of microwave radiometer in winter, spring and summer. Under sunny and cloudy conditions, the average absolute error and standard deviation of BP neural network inversion temperature, water vapor density and relative humidity in different seasons are less than those of microwave radiometer, especially the relative humidity. Under clear sky conditions, the BP neural network inversion temperature profile is the best in spring, summer and autumn. The inversion of water vapor density profile has high accuracy in the middle and low layers, and the inversion of relative humidity profile has poor accuracy, but it is basically consisitent with the trend of sounding data. Under cloud conditions, the BP neural network inversion temperature profile is basically consisitent with the clear sky time, and the accuracy is higher than that of microwave radiometer.The inversion of water vapor density and relative humidity profile is better above 8 km. |
| |
| Keywords: | |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《高原山地气象研究》浏览原始摘要信息 |
| 点击此处可从《高原山地气象研究》下载免费的PDF全文 |
|
