基于多要素协助的气象降水数据质量控制研究

Study on Quality Control ofMeteorological Precipitation Data Based on Multi factor Assistance

针对气象降水数据质量控制难度大,准确性低等问题,提出了基于多要素协助的气象降水数据质量控制方法,使用福州市区域站逐小时数据,分析降水数据的单站要素相关性和邻近站点的降水空间相关性,使用集成学习算法XGBoost(极度梯度提升算法)训练模型,综合考虑查全率和查准率选取异常判断的阈值,最终形成降水异常检测模型,并与多种质控方法进行结果比较。结果表明:(1)单站要素之间有弱相关性,邻近站点的降水数据相关性与空间分布有关,具有强相关性。(2)与传统的变化率判断法,单站单要素方法,单站多要素方法进行结果比较,该方法可以明显区分出异常降水值,其准确性高效果好。(3)该方法泛化能力更好,总体性能优于传统的变化率判断法。

Aiming at the problems of dificult quality control and low accuracy of meteorological precipitation data,this paper puts forward a quality control method of meteorological precipitation data based on multi-factor assistance.By using the hourly data of Fuzhou regional stations,it analyzes the correlation of single station elements and the spatial correlation of precipitation in neighboring stations,uses XGBoost(Extreme Gradient Lfting Algorithm)training model,comprehensively considers the recall rate and precision rate,selects the threshold of abnormal judgment,and fnally forms a precipitation anomaly detection model,and The results show that:(1)There is a weak correlation between the elements of a single station,and the correlation between pre-cipitation data of neighboring stations is related to spatial distribution,which has a strong correlation.(2)Compared with the tra-ditional rate of change judgment method,single-station single-factor method and single-station multi-factor method,this method can clearly distinguish the abnormal precipitation value,with high accuracy and good effect.(3)The generalization ability of this method is better,and its overall performance is better than that of the traditional change rate judgment method.