Land change for all municipalities in Latin America and the Caribbean assessed from 250-m MODIS imagery (2001-2010)

详细信息    查看全文

• 作者：Matthew L. Clark ; T. Mitchell Aide ; George Riner
• 关键词：Land-cover and land-use change ; MODIS ; Random Forests ; Google Earth interpretation ; Latin America and the Caribbean ; Continental scale
• 刊名：Remote Sensing of Environment
• 出版年：2012
• 出版时间：November, 2012
• 年：2012
• 卷：126
• 期：Complete
• 页码：84-103
• 全文大小：2399 K

文摘

Although over 40 years of satellite imagery has greatly helped in documenting the location and extent of human impact, especially deforestation, our ability to confidently detect current patterns of land change at broad spatial scales needs improvement. To address this challenge, we have developed a cost-effective mapping procedure based on 250-m MODIS imagery that produces annual land-use/land-cover (LULC) maps for Latin America and the Caribbean (LAC). This procedure uses annual spectral statistics, collection of references samples with a Web-based tool, and tree-based Random Forests classifiers stratified by biome map regions to produce wall-to-wall, annual LULC maps for 2001 to 2010 that cover all of LAC. Across 26 map regions, overall pixel-level accuracy averaged 80.2 ¡À 8.1 % for eight basic LULC classes, and 84.6 ¡À 6.5 % for a five-class scheme. Municipality-scale area change between 2001 and 2010 in the three dominant classes (woody vegetation, mixed woody/plantation, and agriculture/herbaceous vegetation) was then estimated using regression models fit to 10 years of data, thus minimizing the impact of inter-annual class variation on change statistics. Closed-canopy forest area and change between 2001 and 2009 were well correlated with high-resolution maps of the Brazilian Legal Amazon (PRODES project). Our LAC-wide analysis of significant change revealed the recent extent and magnitude of deforestation hotspots, such as in the Amazon moist forests and the dry forests of Argentina, Paraguay and Bolivia. Our data also revealed biome-specific clusters of municipalities with increasing woody vegetation due to forest recovery, reforestation, or woody encroachment. Taken as a whole, our MODIS-based mapping and trend modeling methodology can provide reliable land change data, not just for the tropics or for forest cover, but for all biomes and municipalities in LAC and including multiple LULC classes. Because this information can be produced quickly on an annual time scale, with internally-consistent data sources, it is a very useful tool for resource managers, policy makers, scientists and conservationists interested in tracking recent land change across broad-scale, political and environmental gradients.