热带地理

热带地理 ›› 2017, Vol. 37 ›› Issue (2): 269-276.doi: 10.13284/j.cnki.rddl.002933

• 论文 • 上一篇    下一篇

基于地理加权和最小二乘线性回归模型的 气温与NDVI聚集密度的相关分析

许剑辉1，赵 怡2,3，钟凯文1，阮惠华4，孙彩歌1,3   

  1. （1．广州地理研究所//广东省地理空间信息技术与应用公共实验室//广东省遥感与地理信息应用重点实验室，广州 510070；2．中国科学院 广州地球化学研究所，广州 510640；3．中国科学院大学，北京 100049；4．广东省气象探测数据中心，广州 510080）
  • 出版日期:2017-03-05 发布日期:2017-03-05
  • 通讯作者: 孙彩歌（1985―），女，河南人，助理研究员，博士研究生，主要从事城市生态与环境遥感，（E-mail）gecaisun@163.com。
  • 作者简介:许剑辉（1984―），男，广东人，助理研究员，博士，主要从事城市遥感与数据同化，（E-mail）xujianhui306@163.com
  • 基金资助:
    广东省科学院实施创新驱动发展能力建设专项资金项目（2017GDASCX-0804）；广州地理研究所优秀青年创新人才基金资助项目；广东省科学院平台环境与能力建设专项资金项目（2016GDASPT-0103）；广东省科技计划（2015A030401069）

Analysis of the Correlation between Air Temperature and NDVI Cohesion Density Using Geographically Weighted and Ordinary Least Square Regression Models

XU Jianhui1，ZHAO Yi2,3，ZHONG Kaiwen1，RUAN Huihua4，SUN Caige1,3   

  1. （1．Guangzhou Institute of Geography//Key Laboratory of Guangdong for Utilization of Remote Sensing and Geographical Information System//Guangdong Open Laboratory of Geospatial Information Technology and Application，Guangzhou 510070，China；2．Guangzhou Institute of Geochemistry，China Academy of Sciences，Guangzhou 510640，China；3．University of Chinese Academy of Sciences，Beijing 100049，China；4．Guangdong meteorological observation data center，Guangzhou 510080，China）
  • Online:2017-03-05 Published:2017-03-05

PDF (PC)

赞

摘要：

基于2015年广州地区1 km空间分辨率的MOD13A 3月合成NDVI数据以及春夏秋冬4个季节的气象站点近地表气温，首先利用聚集密度计算方法计算NDVI的聚集密度，构建不同季节近地表气温与NDVI聚集密度的最小二乘线性回归模型（OLS）和地理加权回归模型（GWR），分析广州市近地表气温与NDVI聚集密度的相关关系，探讨不同季节NDVI聚集密度回归系数的空间分布，并利用AICc信息准则、拟合优度和Sigma指标对GWR与OLS的结果进行比较分析。结果表明：NDVI聚集密度较好地反映了研究区建设用地、植被和水体等下垫面的综合信息；与OLS模型相比，GWR模型的拟合效果更显著，最小的拟合度从0.02提高到0.464，GWR模型的拟合度最大值达到了0.724；GWR模型回归残差的Moran’s I显著减少，如1月份Moran’s I指数从0.383减少到0.022；NDVI聚集密度对气温的影响具有空间异质性，整体上，从广州北到南，GWR模型中NDVI聚集密度与气温的回归系数由负值逐渐增加到正值，表明NDVI聚集密度对气温有着从负到正的影响；下垫面以不透水面为主的区域，GWR模型拟合度较低，以植被为主要下垫面的区域，GWR模型拟合度较高。

关键词: 广州, 气温, 归一化植被指数, 聚集密度, 地理加权回归模型, 最小二乘线性回归模型

Abstract: To study the influence of the spatio-temporal pattern of normalized difference vegetation index (NDVI) on air temperature in Guangzhou city, four seasons data of MOD13A3 monthly NDVI and monthly data of average air temperature in 2015 are used in this study. The distance-weighted NDVI cohesion density is calculated with MOD13A3 monthly NDVI, which indicates the synthetic information of construction land, vegetation, and waters. The central regions, Panyu and Nansha in Guangzhou city, have low NDVI cohesion density, while other regions with high-density vegetation have high NDVI cohesion density. Based on four seasons’ data of NDVI cohesion density and air temperature, this study quantitatively evaluates the correlation between air temperature and NDVI cohesion density using the geographically weighted regression (GWR) and ordinary least square regression (OLS) models. The GWR and OLS models are then established with NDVI cohesion density and air temperature. The results of GWR and OLS models are evaluated with AICC information criterion, goodness of fitting, and Sigma indexes. As compared with OLS model, GWR shows a much better fitting result, the minimum goodness of fitting increases from 0.02 to 0.464. The maximum goodness of fitting in GWR model reaches 0.724. The Moran’s I value of residual from GWR is much less than that from OLS. The Moran’s I values from OLS are 0.383, 0.342, 0.370, and 0.204 in four seasons, respectively. However, the Moran’s I values from GWR decrease to 0.022, -0.002, -0.022 and -0.025, respectively. Four seasonal Moran’s I values of residual from GWR approach 0, which indicates that the residual from GWR has no spatial autocorrelation. The relationship between NDVI cohesion density and monthly average air temperature has a significant spatial heterogeneity. In general, regression coefficients of NDVI cohesion density increase gradually from negative to positive values from north to south in Guangzhou city. The negative regression coefficients mainly exist in Huadu, Zengcheng, Conghua and Huangpu regions. The negative regression coefficients show a negative influence of NDVI cohesion density on air temperature, while the positive regression coefficients show a positive influence. In those regions dominated by the impervious surfaces, the GWR model has lower goodness of fitting, with the goodness of fitting less than 0.20. Conversely, the GWR model results are in high fit degree in other regions dominated by vegetation, with the goodness of fitting larger than 0.30.

Key words: Guangzhou, air temperature, Normalized Difference Vegetation Index, cohesion density, Geographically Weighted Regression model, Ordinary Least Square regression model