An Analysis of Decadal Land Change in Latin America and the Caribbean Mapped from 250-m MODIS Data

We present a method to map annual landuse/land-cover (LULC) change using low-cost and consistent data from the MODIS satellite and Google Earth reference data. Our LULC maps cover Latin America and the Caribbean from years 2001 to 2009, have a 250-m pixel size, and include eight general classes. Maps were produced within 27 separate zonal biomes using the Random Forests classifier. We found that accuracy of the maps is sensitive to broad climatic and vegetation differences encapsulated by biomes, with an average 84.9% (s.d.±6.3%) overall accuracy for a five-class reclassification. Trends in woody vegetation area were analyzed using all years and pixels aggregated at ecoregion and municipality scales. The largest hotspots of deforestation were in South America, concentrated mostly in moist forests around the south to west Amazon basin and western edge of Atlantic forests, and dry forests in the Chaco ecoregions of northern Argentina. Driving factors included mostly pastures for cattle and industrial-scale agriculture for global export. We also detected increasing woody vegetation in northern Mexican deserts and in dry forests of southern Bolivia. Although not thoroughly evaluated, possible factors driving increased woody vegetation could be local increase in precipitation and the mitigating influence of protected areas.