地理科学进展 ›› 2012, Vol. ›› Issue (1): 56-62.doi: 10.11820/dlkxjz.2012.01.008

• 水文与气候变化 • 上一篇    下一篇

薄盘光滑样条插值中三种协变量方法的降水量插值精度比较

刘正佳1,2,3, 于兴修1, 王丝丝4, 商贵铎2,3   

  1. 1. 临沂大学山东省水土保持与环境保育重点实验室/资源环境学院。山东临沂 276005;
    2. 中国科学院地理科学与资源研究所,北京 100101;
    3. 山东师范大学人口资源与环境学院,济南 250014;
    4. 香港中文大学地理与资源管理学系,香港
  • 收稿日期:2011-10-01 修回日期:2012-02-01 出版日期:2012-01-25 发布日期:2012-01-25
  • 通讯作者: 于兴修(1967-), 男, 教授, 博士生导师, 主要从事资源利用与环境效应研究。E-mail: yuxingxiu@lyu.edu.cn
  • 基金资助:
    教育部新世纪优秀人才支持计划项目(NCET-08-0877);临沂市重大科技创新项目(201011019)。

Comparative Analysis of Three Covariates Methods in Thin-Plate Smoothing Splines for Interpolating Precipitation

LIU Zhengjia1,2,3, YU Xingxiu1, WANG Sisi4, SHANG Guiduo2,3   

  1. 1. Linyi University, The Key Laboratory of Soil &Water Conservation and Environment Protection of Shandong Province/College of Resources Environment, Linyi, 276005, Shandong, China;
    2. Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing, 100101, China;
    3. College of Population, Resource and Environment, Shandong Normal University, Jinan 250014, China;
    4. Department of Geography and Resource Management, the Chinese University of Hong Kong, Hong Kong, China
  • Received:2011-10-01 Revised:2012-02-01 Online:2012-01-25 Published:2012-01-25

摘要: 在薄盘光滑样条插值中,高相关协变量的选取决定了插值结果的精确性。以2001-2009 年全国728 个气象站点日降水为数据源,提取年降水量数据,在分析多年平均降水量与两协变量高程(DEM)和距海岸线距离(DCL)的空间相关性基础上,利用ANUSPLIN软件,比较不同协变量下降水量插值结果精度在全国尺度以及区域尺度上的差异。以DEM、DCL及DEM-DCL分别为协变量对降水量数据进行空间插值发现:①在全国尺度上,DEM法的平均绝对误差(MAE)为47.79,略低于DEM-DCL法(48.90),但显著低于DCL法(55.54);且DEM法的平均相对误差和均方根误差也明显低于其它两种方法。②在区域尺度上,除西藏地区外的其他7 个区域,3 种方法的插值误差与全国尺度上相一致。西藏地区降水插值结果以DCL法的精度最高,而DEM法则较差。研究建议除在西藏地区的降水量插值研究中采用DCL法,在全国其他大部分区域采用DEM法。

关键词: ANUSPLIN, 降水量插值, 精度, 协变量

Abstract: In the thin-plate smoothing splines interpolation, the accuracy of interpolation results is mainly determined by choosing the independent covariate. Annual precipitation data were extracted by using daily precipitation data of 728 meteorological stations from 2001 to 2009 in China. We evaluated spatial correlation relationships between annual precipitation and two covariates such as DEM and distance from the coastline to each point (DCL) and compared the accuracy difference of precipitation interpolation results from different covariates in the national scale and regional scale. All interpolation work has been conducted with the aid of the software of ANUSPLIN. We used three interpolation methods, which respectively considered DEM, DCL and DEM-DCL as the covariates to obtain spatial distribution of precipitation. Our analyses show that, (1) in the national scale, the mean absolute error (MAE) of interpolation method of DEM is 47.79, which is slightly lower than that of the method of DEM-DCL (48.90), while obviously lower than that of the method of DCL (55.54), and MRE and RMSE of the method of DEM were also lower than other two methods significantly. (2) In regional scale, the errors of three methods of interpolation are the same as that in national scale except Tibet. The accuracy of precipitation interpolation results was the highest using DCL method, and the poorest using DEM method. Results suggest that precipitation interpolation method of DEM could be widely used in some relevant national scale researches, and precipitation interpolation method of DCL was strongly recommended in Tibet.

Key words: accuracy, ANUSPLIN, covariates, precipitation interpolation