基于不同植被指数和分类回归树的关中地区土地利用分类

Classification of Land Use in Guanzhong Area Based on Different Vegetation Indexes Time Series and CART

及时准确地获取土地利用信息可以为土地资源规划管理和生态环境保护提供重要依据。为获取关中地区土地利用信息,比较基于不同植被指数的分类效果,基于2014年10月1日至2016年3月15日33个时相的MOD13Q1数据,通过Savitzky-Golay滤波法重构了NDVI及EVI时间序列,并构建了NDVI+EVI和EVI+NDVI组合植被指数时间序列,在迭代自组织分析算法聚类的基础上采用分类回归树,分4种情形对关中地区土地利用进行了分类,采用混淆矩阵进行空间精度分析,并提取关中5个地市和整个地区的农业用地面积,进行定量精度评价。结果显示:(1)4种情形总体精度均大于96%,Kappa系数均大于0.94;(2)基于不同植被指数的分类精度和农业用地提取精度存在EVI+NDVI>NDVI+EVI>EVI>NDVI的规律。研究表明:(1)该分类方法适用于关中地区土地利用分类;(2)EVI对植被的识别能力较NDVI高;(3)基于组合植被指数进行分类时,判别依据更多,决策树准确性更高。

To obtain the land use information accurately in time could provide important basis for management of land resources and the protection of the ecological environment.In order to obtain the land use information in Guanzhong area,and compare classification precision based on different vegetation indexes,based on 33-phase MOD13Q1 data from October 1,2014 to March 15,2016,firstly,normalized difference vegetation index(NDVI)and enhanced vegetation index(EVI)time series were reconstructed with the Savitzky-Golay filtering method and combined into NDVI+EVI and EVI+NDVI time series;secondly,Classification and Regression Tree was built upon the clustering results of Iterative Selforganizing Data Analysis Techniques Algorithm to classify different land use types in four vegetation indexes cases,finally,the classification results were assessed by the confusion matrix and statistical data in terms of spatial and quantitative accuracy.The results show that:(1)the total classification accuracy and Kappa coefficient were over 96%and 0.94,respectively,in four cases;(2)the precision of classification and agricultural land extraction based on different vegetation indexes appeared to meet the order:EVI+NDVI >NDVI+EVI >EVI >NDVI.This study indicates that:(1)the method in this paper could be used in classification of land use in Guanzhong area;(2)the ability of EVI to recognize agricultural land is better than that of NDVI;(3)decision tree is more accurate due to more classification information based on the combined vegetation indexes.