PARENT SESSION
Oral Session #71: Remote Sensing and GIS.
Presiding: J.D. Allan
Wednesday, August 7. 1:00 PM to 4:45 PM. Palo Verde Room, Radisson.

Agricultural land use changes in Amazonia in 2000-2001: results of an experimental technique for merging agricultural censuses and satellite reflectances.

Cardille, Jeffrey^*,1, Foley, Jonathan¹, ¹ University of Wisconsin-Madison, Madison, WI

ABSTRACT- As part of our research within the Large-scale Biosphere-Atmosphere Experiment in Amazonia (LBA), we are developing a time series of the spatial distribution and abundance of major agricultural activities within the Amazon and Tocantins basins. In previous papers, we have described a new method for integrating land cover classifications from remote sensors with land use information from agricultural censuses. These fused data products, available for the mid-1980s and mid-1990s at five minute (~ 9 km) resolution, have much of the spatial detail of the satellite information and the useful attribute detail from censuses; they show snapshots of the density of cropland, natural pasture, and planted pasture across Amazonia in these two periods. In this presentation, we extend the technique and update the time series to the present day for all of Amazonia. Using 16-day composites of MODIS NDVI and band information for 2000-2001 developed at the Global Land Cover Facility, we describe a new technique for statistically combining agricultural census data with unclassified satellite reflectance data. The adopted technique differs from typical classification algorithms that identify "pure" pixels of desired classes and seek similar characteristics in the image. Instead, the method simultaneously considers the relation between reflectance characteristics and agricultural census values across administrative units, and optimizes the relation between them to produce the classification. The result is a new basin-wide spatially explicit estimate of present-day agricultural land use activity, consistent with census and satellite information and suitable for input into ecosystem models. This technique allows the combination of polygon-based GIS data with raster-based remotely sensed data, and we believe it has promise as a generally applicable strategy for merging these complementary but distinct data sources.

KEY WORDS: amazon, remote, GIS, census