日气温数据缺测的插补方法试验与误差分析  

Interpolating Method for Missing Data of Daily Air Temperature and Its Error Analysis

作　　者：王海军[1] 涂诗玉[1] 陈正洪[2] 

机构地区：[1]湖北省气象信息与技术保障中心,武汉430074 [2]武汉区域气候中心

出　　处：《气象》2008年第7期83-91,共9页Meteorological Monthly

基　　金：中国气象局气象新技术推广重点项目(CMATG2006Z03);武汉区域气象中心重点项目(QY-Z-200701、QY-Z-200708);国家科技基础条件平台工作项目(2005DKA31700)

摘　　要：对缺测气象观测记录进行插补是建立连续气象数据集的基础。将孤立1日或数日缺测资料进行插补的线性回归模型法应用于连续缺测数月的逐日最高、最低和平均气温的插补,并进行了一系列改进,包括:(1)用滑动选优法确定邻近参考气象站站数和数据样本时间窗的最佳值;(2)在记录缺测站与邻近参考站之间建立逐日气温的线性回归模型,并选取以最小绝对误差(Least Absolute Deviation,LAD)为目标函数求取模型参数的方法,取代以最小均方根误差为目标函数的最小二乘法(Least Squares Esti mate,LAD)求解模型参数的方法,可提高计算效率和参数的稳定性;(3)进一步提出将LAD法与DeGaetano标准化序列法插补结果平均的综合插补方法,以减少极端误差。通过对湖北蔡甸气象站1961-2006年插补试验表明:(1)以4个邻近站和年数为8年、日数为15天时间窗的样本资料建模进行插补误差达到最小;(2)逐日最高、最低和平均气温的平均绝对误差分别为0.32℃、0.45℃、0.28℃,误差在±0.8℃以内的频次分别占总数的94.1%、84.8%、96.1%,观测值与插补值月相关系数在0.886以上。插补与观测资料平均值和相关系数分别通过了显著水平为0.05和0.01的检验。It is the foundation of building up a continuous meteorological datasets for interpolating the missing meteorological record. A method of estimating missing daily air temperature （ daily maximum, minimum and average air temperature.） for several continuous months is proposed, which is improved from the linear regression model that are usually used to interpolate the missing data for a single day or several days. A series of improvements are taken on, such as： （1）The number of neighboring referenced meteorological stations and size of time window about sample data are determined by optimized method; （2）The linear regression model is set up between the station of missing data and neighboring referenced stations, and the least absolute deviation（LAD） instead of the least square-root estimate （LSE）is treated as the object function when the model parameter is solved for higher computing efficiency and parameter stability; （3）In order to reduce extreme error, the averaged value between LAD method and DeGaetano＇s standardized method is used as the final result. A lot of interpolating experiments are done with the data of Caidian Meteorology Station, Hubei Province from 1961 to 2006 and the results indicate （1）The least interpolating error can be obtained when 4 neighboring referenced stations, time window with 8 years and 15 days is used; （2） The frequency of mean absolute error of daily maximum, minimum and average air temperature within ± 0.8℃ is 94.1% ,84.8 % ,96.1% respectively. The monthly average and correlation coefficient between actual value and interpolated value are all over 0. 886 with the significant level of 0.05 and 0.01 by t-test respectively.

关 键 词：记录缺测 逐日气温 插补 线性回归模型 最小绝对误差 

分 类 号：P467[天文地球—大气科学及气象学] TG659[金属学及工艺—金属切削加工及机床]