摘 要 大面积范围内准确获取土地利用/覆盖变化信息对生产管理和生态环境评价都具有重要的意义。本研究联合2010年25 m分辨率的ALOS PALSAR L波段雷达和2009～2010年多时相Landsat TM/ETM+影像，利用决策树分类方法对海南岛土地利用类型进行分类。结果表明，PALSAR雷达对森林和水体均有较高的识别精度，生产者精度和用户精度均超过88%，但耕地与建筑分类精度最高仅为77%。通过结合森林、建筑与耕地的光谱信息及其年际变化特征，采用多时相TM/ETM+影像合成的NDVI最大值和最小值对PALSAR分类结果进行修正。修正后结果的精度均有显著提升，森林、水体和建筑的生产者和用户精度均超过了92%，相应耕地的精度也分别达到了91%和74%，总体分类精度达到94%，Kappa系数为0.92。本研究表明，联合PALSAR雷达和多时相Landsat系列光学遥感影像，在解决热带地区影像数据源匮乏的同时，能够显著提高土地利用分类精度，具有良好的应用前景。
关键词 PALSAR；TM/ETM+；海南岛；NDVI
中图分类号 S127 文献标识码 A
Land Utilization Mapping in Hainan Island by Using ALOS
PALSAR and Multi-temporal Landsat TM/ETM+ Imagery
CHEN Bangqian1，2， LI Xiangping2， XIAO Xiangming2， SUN Rui1， WU Zhixiang1，
QI Dongling1， YANG Chuan1， TAO Zhongliang1
1 Rubber Research Institute（RRI）， Chinese Academy of Tropical Agricultural Sciences（CATAS）/ Investigation &
Experiment Station of Tropical Cops， Ministry of Agriculture， Danzhou， Hainan 571737， China
2 School of Life Science， Fudan University， Shanghai 200438， China
Abstract Accurately mapping land use / cover change over large area is of vital importance for production management practices and eco-environmental assessment. A decision tree-based land utilization mapping in Hainan Island by using the 25 m ALOS Phased Arrayed L-band Synthetic Aperture Radar（PALSAR）in 2010 and multi-temporal 30-m Landsat TM/ETM+ imagery in 2009-2010 was made. The results indicated that forest and water could be easily identified by PALSAR imagery. Both the producer"s accuracy（PA）and user"s accuracy（UA）were more than 88%. However， the PA and UA were low for the cropland and built-up land， with a max value of about 77%. On the basis of analyzing the spectral reflectance and their seasonal dynamics of the forest， built-up land and cropland， the maximum/minimum value composited Normalized Difference Vegetation Index（NDVI）were used to revise the PALSAR-based mapping results. Significant improvement in the accuracies was obtained for all the classes. For the forest， water and built-up land， the PA and UA were moe than 92%. The accuracy of the cropland also improved， with PA and UA of 91% and 74%， respectively. The overall accuracy was 94% and kappa was 0.92. This study demonstrated the great application prospective of integrating PALSAR and multi-temporal TM/ETM+ imagery for land use classification， which can not only overcome the shortage of the optical images in tropical area， but also greatly improve the mapping accuracy.