| 基于Parsivel天气现象仪资料的郑州“7·20”罕见特大暴雨微物理特征分析 |
| |
| 引用本文： | 周丹,周淑玲,田金华,等.基于Parsivel天气现象仪资料的郑州“7·20”罕见特大暴雨微物理特征分析[J].气象与环境科学,2022,45(1):93-101. |
| |
| 作者姓名： | 周丹 周淑玲 田金华 等 |
| |
| 作者单位： | 深圳市气象局;深圳市南方强天气研究重点实验室;深圳市国家气候观象台;浙江省气象台 |
| |
| 基金项目： | 国家重点研发计划政府间/港澳台重点专项项目(2019YFE0110100);国家自然科学基金(41975124、41405047);中国气象局预报员专项项目(CMAYBY2019045、CMAYBY2019081);广东省自然科学基金(2019A1515010814、2019B020208016)。 |
| |
| 摘 要： | 基于2016-2018年ECMWF模式温度预报和浙江省72个国家基本站观测资料,根据温度日变化特征,采用K-近邻(KNN)回归算法进行误差订正,改进浙江省172 h精细化温度预报。在KNN回归算法中,将模式起报时刻的温度视作“背景”,由模式预报减去起报时刻温度消除“背景”影响,得到温度日变化曲线,通过温度日变化曲线构建差异指标,选取历史相似个例。根据历史相似个例的误差特征,对温度预报进行订正,得到改进的温度预报。检验结果表明,KNN方案的温度预报平均绝对误差较ECMWF和30 d滑动平均误差订正方案(OCF)的分别减小26.2%和5.2%;日最高和最低温度预报误差绝对值小于2℃,准确率较ECMWF的分别提高14.8%和4.3%,较OCF的分别提高3.0%和1.3%。KNN方案对地形复杂地区的温度预报改进效果更为明显,对冷空气活动和夏季高温等天气过程预报改善效果也较稳定。
|
| 关 键 词： | 精细化预报 K-近邻回归 温度日变化 相似个例 |
Application of KNN Approach in Improvement of Temperature Forecast in Zhejiang |
| |
| Li Chao,Li Minghua,Zhou Kai,Hao Shifeng,Chen Xunlai,Zhao Chunyang.Application of KNN Approach in Improvement of Temperature Forecast in Zhejiang[J].Meteorological and Environmental Sciences,2022,45(1):93-101. |
| |
| Authors: | Li Chao Li Minghua Zhou Kai Hao Shifeng Chen Xunlai Zhao Chunyang |
| |
| Institution: | (Shenzhen Meteorological Bureau, Shenzhen 518040, China;Shenzhen Key Laboratory of Severe Weather in South China, Shenzhen 518040, China;Shenzhen National Climate Observatory, Shenzhen 518040, China;Zhejiang Meteorological Observatory, Hangzhou 310056, China) |
| |
| Abstract: | Based on the ECMWF model output and observational data from 72 basic meteorological stations in Zhejiang Province in 20162018,combined with the diurnal cycle features of temperature,the short-range(172h)temperature forecast is improved by using the K-nearest neighbor(KNN)regression algorithm.In the KNN algorithm,the initial temperature is removed as background from total temperature to obtain diurnal cycle curve of temperature.The Euclidean Distance is conducted as the metric of difference of temperature diurnal curves and used for more similar cases selection.The temperature is then corrected by using the mean error of selected similar cases.The mean absolute error of KNN is reduced by 26.2%compared with that of ECMWF model,and 5.2%compared with the scheme of 30 d running error correction(OCF).The accuracy of daily high and low temperature errors less than 2℃in the KNN scheme is improved by 14.8%and 4.3%compared with that of ECMWF forecast,and 3.0%and 1.3%compared with the OCF scheme.The KNN scheme performs better in improving the temperature forecast for complex terrain area and is also stable in correcting the forecasts of cold air activities and long-duration high temperature cases. |
| |
| Keywords: | refined forecast K-nearest neighbor(KNN)regression diurnal cycle of temperature similar historical cases |
| 本文献已被 维普 等数据库收录！ |
| 点击此处可从《气象与环境科学》浏览原始摘要信息 |
| 点击此处可从《气象与环境科学》下载免费的PDF全文 |
|
