大规模降雨监测数据异常识别方法  

Anomaly recognition method of massive rainfall data

作　　者：田济扬[1,2] 刘含影 刘荣华[1,2] 丁留谦[1] 刘宇 TIAN Jiyang;LIU Hanying;LIU Ronghua;DING Liuqian;LIU Yu(China Institute of Water Resources and Hydropower Research,Beijing 100038,China;Research Center on Flood and Drought Disaster Reduction of Ministry of Water Resources,Beijing 100038,China)

机构地区：[1]中国水利水电科学研究院,北京100038 [2]水利部防洪抗旱减灾工程技术研究中心,北京100038

出　　处：《中国水利水电科学研究院学报（中英文）》2022年第5期438-448,463,共12页Journal of China Institute of Water Resources and Hydropower Research

基　　金：国家自然科学基金项目(51909274);国家重点研发计划项目(2019YFC1510605);中国水科院五大人才计划项目(JZ0199A022021)。

摘　　要：为实现大规模降雨监测数据的异常识别和快速处理,基于Hampel法、格拉布斯准则、周边测站分析法和雷达辅助校验等方法,建立了递进式异常站点筛查体系,通过K-dtree(K-dimensiontree)高级数据结构和并行计算方法提高计算效率,并以福建省5234个具有雨量监测功能的地面站2015—2021年雨量数据进行了验证,结果表明福建省地面站雨量监测数据质量逐年提升;各类测站中,雨量站异常站点占全部异常站点的比例最高,各类异常站点在全省相应类型站点中,雨量站异常站点的占比也最高;雷达辅助校验能够有效解决在雨区与非雨区边界、雨强差异较大的雨区边界的正常站点易被误判为异常值的问题,校验前异常识别准确率为90%左右,校验后准确率提高为95%左右。通过K-dtree和并行计算,全省测站完成一次异常识别需约5~8min,为大规模降雨监测数据异常识别、充分利用雨量监测站有效信息提供可靠的方法。To achieve anomaly recognition and quick processing of the massive rainfall data, a Progressive Screening System(PSS) is established in this study based on the Hampel method, Grubbs criterion, Radar Auxiliary Verification, and Peripheral Station Analysis Method.Meanwhile, the K-dimension tree advanced data structure and parallel computing method are also applied to improve the calculation efficiency.Based on the established PSS,the anomaly recognition on the rainfall data from 5234 stations in Fujian Province from 2015 to 2021 are conducted, and the results show that the rainfall data qualities of these stations have been improved by years.Among all kinds of stations, the rain gauging stations account for the highest proportion of 5234 stations with abnormal data.It is proved that, the misjudgment on the abnormal value can be reduced greatly by radar auxiliary verification, which occurs generally at the boundary of rainfall areas or the rainfall areas with extreme variation of rainfall intensity.The accuracy rate of anomaly recognition can be improved from 90% to 95% by radar auxiliary verification.Additionally, based on K-dimension tree advanced data structure and parallel computing method, the anomaly recognition for 5234 stations only takes 5 to 8 minutes.All of these indicate that the PSS is not only effective on anomaly recognition and quick processing of massive rainfall data, but also reliable on extracting valuable information from mass data.

关 键 词：Hampel法 格拉布斯准则 周边测站分析法 雷达校验 异常识别 

分 类 号：P426.62[天文地球—大气科学及气象学]