摘要
为了构建区域优化的大气加权平均温度模型(ROTm)并提高可降水量(PWV)的反演精度和短时降雨预报的准确性,提出一种全球卫星导航系统(GNSS)加权平均温度确定及短时降雨预报方法:给出基于奇异谱分析(SSA)分解的大气加权平均温度非线性建模方法,并将其应用于研究区域的PWV反演;然后基于GNSS反演的PWV数据,分别采用阈值预报方法和随机森林方法进行降雨预报;最后,以临界成功指数(CSI)、命中率(POD)和误报率(FAR)等指标对比传统模型与随机森林方法的预报效果。结果表明,提出的ROTm模型在大气加权平均温度和PWV反演精度上显著优于全球气压和温度模型3(GPT3)、贝维斯(Bevis)模型和新不伦瑞克大学模型3(UNB3),其中在大气加权平均温度反演中的均方根误差(RMSE)降低约25.0%~54.6%,在PWV反演中的RMSE降低约27.3%~52.9%。随机森林方法在12 h时效降雨预报中表现出明显优势,CSI最高提升21.71%,POD提升达5.38%,FAR降低4.71%,可显著提高预报的可靠性;预报时效从12 h缩短到1 h,随着CSI从31.4%提高到37.22%,POD从78.61%提升至86.55%,FAR从64.27%降至52.11%,显示短时预报效果能够得到显著提升;该方法在提高降雨预报准确性和反演大气参数等方面具有优势。
In order to construct a regional optimization atmospheric weighted mean temperature model(ROTm)and improve the accuracy of precipitable water vapor(PWV)retrieval and short-term rainfall forecasting,the paper proposed a determination method of weighted average temperature for global navigation satellite system(GNSS)and short-term rainfall forecasting:a nonlinear modeling method for atmospheric weighted mean temperature based on singular spectrum analysis(SSA)decomposition was given and applied to the retrieval of PWV in the study area;then,based on the PWV data retrieved from GNSS,rainfall forecasting was conducted by using both threshold forecasting and random forest methods;finally,the forecasting performance was evaluated by using metrics such as critical success index(CSI),probability of detection(POD)and false alarm ratio(FAR),comparing the results of the traditional models and the random forest method.Results showed that the proposed ROTm model would significantly outperform the global pressure and temperature model 3(GPT3),the Bevis model and the University of New Brunswick model 3(UNB3)in terms of atmospheric weighted mean temperature and PWV retrieval accuracy;specifically,the root mean square error(RMSE)in atmospheric weighted mean temperature retrieval would be reduced by approximately 25.0%to 54.6%,while the RMSE in PWV retrieval reduced by 27.3%to 52.9%;moreover,the 1.71%,an increase in POD by 5.38%,and a reduction in FAR by 4.71%,significantly improving forecasting reliability;random forest method would exhibit notable advantages in 12 h rainfall forecasting,achieving a maximum increase in CSI by furthermore,as the forecasting time interval decreased from 12 hours to 1 hour,it could be known that the forecasting time interval of CSI would be improved from 31.4%to 37.22%,POD increased from 78.61%to 86.55%,and FAR decreased from 64.27%to 52.11%,highlighting the substantial enhancement in short-term forecasting performance;in conclusion,the proposed method could demonstrate significant advantages in improving the accuracy of rainfall forecasting and the retrieval of atmospheric parameters.
作者
杨静
常巧梅
YANG Jing;CHANG Qiaomei(Shanxi Engineering Technology Vocational University,Jinzhong,Shanxi 030619,China;Shanxi Vocational College of Engineering,Taiyuan 030000,China)
出处
《导航定位学报》
北大核心
2025年第6期187-199,共13页
Journal of Navigation and Positioning
关键词
全球卫星导航系统(GNSS)
大气可降水汽含量
大气加权平均温度
降雨预报
随机森林
global navigation satellite system(GNSS)
atmospheric precipitable water vapor content
atmospheric weighted average temperature
rainfall forecasting
random forest
