| 动态时变参数方法在云量精细化预报中的应用研究 |
| |
| 引用本文： | 杜晖,殷启元,贾晓红,姬雪帅,尚可政,尚子溦,曾瑛.动态时变参数方法在云量精细化预报中的应用研究[J].热带气象学报,2020,36(6):834-845. |
| |
| 作者姓名： | 杜晖 殷启元 贾晓红 姬雪帅 尚可政 尚子溦 曾瑛 |
| |
| 作者单位： | 1.广东省气象公共安全技术支持中心，广东 广州 510641 |
| |
| 基金项目： | 灾害天气国家重点实验室开放课题2019LASW-B04 |
| |
| 摘 要： | 利用2008年1月—2013年12月以及2017年1—11月全球天气预报系统（GFS）预报场资料，采用自适应线性最小二乘回归（LS）和自适应递推卡尔曼（Kalman）滤波两种动态时变参数方法，建立了河套周边地区0~168 h预报时效的总云量精细化预报，并与GFS模式直接输出的总云量、线性预报模型逐步回归预报方法得到的总云量以及非线性预报模型BP神经网络和最小二乘支持向量机回归方法（LSSVM）得到的总云量进行了对比，结果如下:（1）相比GFS模式直接输出的总云量，LS、BP神经网络、LSSVM得到的总云量与实况值的平均绝对误差均明显减小。LS方法误差最小，LS方法的年MAE均在20%~25%，且随着预报时效的延长，改进效果越大。LS方法、多元逐步回归方法、BP神经网络、LSSVM四种方法在6—8月的改进效果最大。（2）LS方法预报的总云量与实况云量的相关性最好，即使168 h预报时效的相关系数依然在0.64以上，远高于其他几种模型的预报结果。（3）LS方法能够明显地提高少云和多云天空状况下预报的击中率，且最优（少云击中率平均提高24 %，多云击中率平均提高34 %）。（4）自适应递推Kalman滤波方法存在预报滞后现象，改进效果不明显。
|
| 关 键 词： | 总云量 精细化预报 动态时变 自适应最小二乘回归 自适应递推卡尔曼滤波 |
| 收稿时间： | 2020-03-19 |
RESEARCH ON APPLICATION OF DYNAMIC TIME-VARYING PARAMETER METHODS IN REFINED FORECAST OF CLOUD COVER |
| |
| DU Hui,YIN Qi-yuan,JIA Xiao-hong,JI Xue-shuai,SHANG Ke-zheng,SHANG Zi-wei,ZENG Ying.RESEARCH ON APPLICATION OF DYNAMIC TIME-VARYING PARAMETER METHODS IN REFINED FORECAST OF CLOUD COVER[J].Journal of Tropical Meteorology,2020,36(6):834-845. |
| |
| Authors: | DU Hui YIN Qi-yuan JIA Xiao-hong JI Xue-shuai SHANG Ke-zheng SHANG Zi-wei ZENG Ying |
| |
| Institution: | 1. Guangdong Technical Support Center of Meteorological Public Security,Guangzhou 510641,China;
2. Key Laboratory of Arid Climate Changes and Disaster Reduction of Gansu Province,College of Atmospheric Sciences, Lanzhou University, Lanzhou 730000, China;;1. Guangdong Technical Support Center of Meteorological Public Security,Guangzhou 510641,China;3. State Key Laboratory of Severe Weather, Chinese Academy of Meteorological Sciences, Beijing 116024, China;;Inner Mongolia Service Center of Meteorology, Hohhot 010051, China;;Zhangjiakou Meteorological Observatory, Zhangjiakou 075000, China;Meteorological Bureau of Lanzhou, Lanzhou 730000, China;; Meteorological Bureau of Foshan, Foshan 528000, China |
| |
| Abstract: | Based on the Global Weather Forecast System (GFS) forecast field data from January 2008 to December 2013 and from January 2017 to November 2017, the present study used two dynamic time-varying parameter methods, namely adaptive linear least squares regression and adaptive recursive Kalman filter, to establish a 0~168 hour forecast for the surrounding area of Hetao Area (the Great Bend of the Yellow River). The time-efficient refined cloud forecast was compared with the total cloud cover provided by the GFS model, and the results of the linear forecast model stepwise regression forecast method, the nonlinear forecast model BP neural network and the least square support vector machine (LSSVM) regression method. The results are as follows: (1) Adaptive LS, BP neural network, and LSSVM all significantly reduce the mean absolute error (MAE) of the total cloud cover and observed value directly output from the GFS model. The MAE of the adaptive LS prediction method is the smallest, and the annual MAE of the LS method at different forecast times is around 20~25%. With the extension of forecast time, the improvement becomes greater. The improvement brought by the self-adaptive LS method, multiple stepwise regression method, BP neural network, and LSSVM is the greatest from June to August. (2) The correlation between the adaptive LS method and the live cloud cover is the best. The correlation coefficient of the 168h forecast time is above 0.64, which is much higher than those of other models. (3) Compared with other methods, the adaptive LS method can significantly improve the forecasting hit rate when it is cloudy or partly cloudy. The average hit rate for partly cloudy sky is increased by 24%, and the hit rate for cloudy sky is increased by 34%. (4) There is a lag in forecasting using the adaptive recursive Kalman filtering method, and the improvement is not obvious. |
| |
| Keywords: | total cloud cover refined forecasting dynamic time-varying adaptive least squares regression adaptive recursive Kalman filter |
|
| 点击此处可从《热带气象学报》浏览原始摘要信息 |
| 点击此处可从《热带气象学报》下载免费的PDF全文 |
|
