中国农业气象

• 论文 • 上一篇    下一篇

基于邻域特征的温度缺失值的填补方法

唐云辉;高阳华;   

  1. 重庆市气象科学研究所;
  • 出版日期:2008-08-10 发布日期:2008-08-10
  • 基金资助:
    中国气象局2007年多轨道业务建设项目“西南地区滑坡和泥石流灾害气象监测预警业务服务系统”

Imputation Method of Missing Temperature Data Based on Neighborhood Features

TANG Yun-hui,GAO Yang-hua(Chongqing Institute of Meteorological Sciences,Chongqing 400039,China)   

  • Online:2008-08-10 Published:2008-08-10

摘要: 以1971-2000年重庆市35个气象站点日气温要素为基础数据,提出了一种基于气候均值、距离和高程差异等因子的新的气温插值方法,可应用于日气温要素序列缺失资料的插值填补。交叉验证结果表明:全市各季节日平均气温的MAE(平均绝对误差)在0.2~0.8℃,平均为0.4℃左右;对日最高、最低气温都具有可靠的拟合精度。通过对沙坪坝站各季节(用1月上旬、4月上旬、7月上旬、11月上旬分别代表)逐时平均温度进行拟合计算,MAE在0.66℃以下,平均相对误差(MRE)除冬季稍高外(为7.9%),其余各季节在3.2%以下,实测值与计算值的相关系数在0.95以上,说明此方法同样适用于逐时观测气温资料的填补。

关键词: 缺失资料, 邻域特征, 高差因子, 插值填补

Abstract: A new temperature interpolating method was put forward based on some factors,such as climate average value,distance and altitude variation.This method was on basis of daily temperature data collected in 35 meteorological stations of Chongqing from 1971 to 2000.It could be used to interpolate missing data in daily average temperature series.The cross verification experiment showed that the Mean Absolute Error(MAE) of the daily average temperature was between 0.2℃ to 0.8℃ with the average of 0.4℃ in the whole Chongqing municipality.It could also simulate daily maximum temperature and daily minimum temperature with highly reliable accuracy.This method was used to simulate hourly average temperature data of all seasons in Shapingba meteorological station.The MAE was less than 0.66℃,while the Mean Relative Error(MRE) was less than 3.2% except that 7.9% in winter.The correlation coefficient between observations and calculations was above 0.95.In this experiment,spring was represented by the first ten days of April,summer the first ten days of July,autumn the first ten days of November,and winter the first ten days of January.So this method could also be used to interpolate missing hourly observed temperature data.

Key words: Missing data, Missing data, Neighborhood features, Altitude variation factor, Interpolating