Mapping land use of tropical regions from space.

A n alarming annual average deforestation rate of 22,500 km2 in the Brazilian Amazon (1) from 2000 to 2005 has focused international attention on Brazil because of the destruction of this high-biodiversity biome (2) and its globally significant contribution to carbon emissions (3). Deforestation has already converted 680,000 km2 of pristine forests to other land covers, accounting for 75% of Brazil’s carbon emissions (1, 3). Satellite remote sensing is an indispensable tool for detecting and mapping deforestation over large areas in the region. For example, the Amazon Deforestation Monitoring Program (PRODES), conducted by the Brazilian National Institute for Space Research (INPE), has used Landsat images to map deforestation since 1988. Even though the extent and rate of deforestation are monitored, landuse changes after deforestation have not been systematically tracked. The few studies that have characterized and mapped land use after deforestation have done so only at local scales (4–6). Therefore, the amount and rates of land abandonment at regional scales are unknown. In this issue of PNAS, Morton et al. (7) present the results of a pioneering study that used remote sensing and field observations to characterize and map land use after deforestation in the Brazilian Amazon. Their work fills an important technological gap for understanding deforestation dynamics and land-use patterns in the Brazilian Amazon. Morton et al. (7) mapped and followed individual deforestation clearings ( 25 hectares) over time to track land-use history. Satellite-derived deforestation data, vegetation phenology information from MODIS (Moderate Resolution Imaging Spectroradiometer), and field observations were combined to classify all clearings as being pasture, cropland, or not in production. Their study was conducted in the state of Mato Grosso, the Amazon state that has most contributed to deforestation, from 2001 to 2004. The approach developed by Morton et al. (7) represents a breakthrough in tropical remote-sensing studies in light of the difficulty in acquiring cloud-free optical imagery (8) for mapping tropical land cover. Additionally, no study that I am aware of has used the temporal domain of remote-sensing data to characterize land use in the Amazon region. To overcome the cloud limitation, these scientists took advantage of the high-frequency MODIS images that are acquired every 1–2 days (9). The advent of fast, inexpensive computer technology and advanced imageprocessing algorithms allowed Morton et al. to rapidly process a very large volume of remotely sensed data and assign land-use categories to each individual clearing after deforestation. The changing phenological temporal signature of deforestation clearings was used to determine whether each forest area had been converted to pasture or cropland or not put into production (7). Morton et al. (7) point out that cropland is expanding at an unprecedented rate in Mato Grosso. The annual forest area directly converted to crop production ranged from 785 to 2,150 km2 during 2001–2004, totaling 5,400 km2 of the 33,200 km2 in total area deforested in this period. The establishment of cropland