基于随机森林的土地利用分类与景观格局分析

Land Use Classification and Landscape Pattern Analysis Based on Random Forest Method

基于景观格局分析方法,分别选取8个反映景观格局类型和6个反映景观水平格局指数用于探索土地利用变化状况。实验结果表明:①基于选择合适的训练样本下,对3期遥感影像采用随机森林的监督分类方法,总体精度均在94%以上,kappa系数均为90%以上;②综合来看,研究期内建设用地为优势景观类型。研究区斑块数量增加,破碎化程度有所加剧,景观斑块的形状复杂程度加深,空间分布由分散趋向于聚合与均衡化,景观多样性不断增加,景观格局的变化表明土地利用/土地覆盖变化主要受人类活动影响较大。

Based on multi-spectral data including Landsat TM data in 2005,2010 and Landsat OLI in 2015,we used random forest classification method to obtain land use maps for the corresponding years.Then we used the landscape pattern analysis method composed of 8 landscape pat-tern types and 6 landscape level pattern indices to explore land use/land cover changes.The experiment results show that①with suitable training samples,the random forest supervised classification method used for the 3 phases of remote sensing images has the overall accuracy of above 94%and Kappa coefficient above 90%.②During the study period,construction land was the dominant landscape type in the study area.The num-ber of patches and the fragmentation degree increased.The shape complexity of landscape patches continued to increase.Spatial distribution changed from dispersion to aggregation and equalization,and landscape diversity continued to increase.Landscape pattern changes show that land use/land cover changes are mainly affected by human activities.