亚热带山地丘陵区气象要素空间化方法分析

A Comparative Study of Spatial Interpolation Methods for Meteorological Elements in Subtropical Mountainous and Hilly Regions of China

空间连续的气象要素是区域环境、生态系统模拟的重要参数,空间插值是实现气象观测站点数据空间化的主要方法。由于不同区域气候要素时空分布的复杂性,研究适用于特定区域的空间化方法具有重要意义。本文以湖南省为研究区,分析长江中游亚热带丘陵区气象要素的空间化方法。基于湖南省及其毗邻地区的46个气象站点14年(2000—2013年)的气象观测数据(月降水量、月均温)和DEM数据,对普通克里金法(OK)、反距离加权法(IDW)、样条法(SPLINE)和基于DEM的多元线性回归插值法(MLR)进行比较分析研究。研究结果表明:1)月降水量和月均温的MRE和RMISE值排序分别为:SPLINE>IDW>MLR>OK和SPLIN>OK>IDW>MLR;年平均降水量和年均温插值的MRE和RMISE值排序分别为:SPLINE>IDW>OK>MLR和SPLIN>OK>IDW>MLR;2)MLR法不仅插值误差小,而且能从细节上反映出气象变化的空间变异性;3)对多变量进行数据探索分析表明,MLR插值法中,经度、坡度与年降水量空间分布存在空间的一致性;纬度、高程和坡度对年均温的空间分布具有很强的解释力。

Spatial interpolation of climate data is frequently required to provide input for regional ecological models. Interpolation is a main method of spatialization of observation station data. Due to complexity of distribution of climatic elements on different regions, the research of spatialization method applied on specific region is significant. This paper is spatial analysis of meteorological elements in subtropical hilly regions of the Middle of Yangtze River, based on climatic observation data （monthly total precipitation, monthly mean temperature） and DEM data of Hunan Province, compared and analyzed Ordinary Kriging （OK） , Inverse Distance Weight tension （IDW）, SPLINE and MLR, and results indicated that： 1 ） Mean Relative Error （MRE） and Root Mean Squared Interpolation Error （RMSIE） of monthly total precipitation and monthly mean temperature calculated by methods above can be ranked respectively as： SPLINE 〉 IDW 〉 MLR 〉 OK and SPLIN 〉 OK 〉 IDW 〉 MLR; MRE and RMSIE for annual mean precipitation and annual mean tem- perature can be ranked as ： SPLINE 〉 IDW 〉 OK 〉 MLR and SPLIN 〉 OK 〉 IDW 〉 MLR. 2 ） The interpolation error of MLR is rela- tively less, and spatial variability can be indicated in detail by MLR. 3 ） Based on data mining analysis of multivariate, there was spa- tial consistency of longitude, gradient and distribution of precipitation in MLR interpolation method, in which the spatial distribution of annual mean temperature could be well explained by latitude, elevation and gradient.