重庆市区雾的天气特征分析及预报方法研究  

Study on Fog Synoptic Characteristics and Fog Forecast Method in Chongqing

作　　者：马学款 蔡芗宁 杨贵名 刘还珠 

机构地区：[1]国家气象中心,北京100081

出　　处：《气候与环境研究》2007年第6期795-803,共9页Climatic and Environmental Research

基　　金：科技部科研院所社会公益研究专项课题2004DIB3J100;国家气象中心自筹资金课题ZK2007-0501

摘　　要：分析了重庆市区雾的特点、天气特征及温、湿等气象要素垂直分布特征,利用重庆站的观测资料选取适当的诊断因子,采用动态学习率BP算法的人工神经网络对重庆市区能见度进行了拟合和预报检验。研究表明：55年以来,重庆年雾日数总体呈逐年下降趋势,同时轻雾日数急剧上升,这种变化可能主要与城市热岛效应增强和空气污染状况加重有关;发展成熟的辐射雾大多具有逆温的稳定结构,雾顶上下温度、湿度存在明显跃变特征;神经网络模型具有较强的自适应学习和非线性映射能力,对能见度为0-1 km雾的报出率为83%,Ts评分达到69%,平均预报误差为0.384 km。除常规气象要素外,通过M指数、Ri数、凝结核、辐射状况和其他物理量的引入,以及对因子网络输入值的技术处理,明显提高了神经网络对雾尤其浓雾的预报能力,其对能见度在0.4 km以下浓雾预报的Ts评分可达89.5%。模型结果对重庆市区雾的预报具有良好的参考价值。Synoptic characteristics of fog and the vertical distribution characteristics of some meteorological elements such as temperature, humidity and so on, are analyzed in terms of 1951- 2005 observed fog data in Chongqing. By means of selecting reasonable diagnostic factors, an artificial neural network model is established with dynamic learning rate BP algorithm to simulate the visibility of Chongqing. Results show that annual mean foggy days in Chongqing have an obvious descent tendency and the light foggy days are increasing sharply. This variation is likely to be mainly associated with the enhancement of urban heat island effect and air pollution. Generally in the mature phase of radiation fog, it is featured by stable inversion structure and extremely pronounced vertical variation of the temperature and relative humidity in the vicinity of the fog top. The BP neural network model is possessed of preferable adaptive learning and non-linear mapping abilities with 99% verification forecasting accuracy, wherein the forecasting accuracy of thick fog （visibility from 0 to 1 kin） is 83%, the Ts grade is 69%, and the average forecast error is 0. 384 kin. The forecast ability of neural network to the fog （especially thick fog） is enhanced obviously due to the introduction of M-index, Richardson number, condensation nucleus, radiation condition and various physical parameters, as well as the technical processing to the network input values of some diagnostic factors in addition to conventional meteorological elements. Ts grade of thick fog with visibility lower than 0. 4 km can reach 89. 5% and model results can provide favorable reference to the fog forecast of Chongqing.

关 键 词：雾 能见度 天气特征 神经网络 诊断因子 

分 类 号：P426.4[天文地球—大气科学及气象学]