We have investigated mesoscale circulations and atmospheric CO2 variations over a heterogeneous landscape of forests, pastures, and large rivers during the Santar¿m Mesoscale Campaign (SMC) of August 2001. The atmospheric CO2 concentration variations were simulated using the Colorado State University Regional Atmospheric Modeling System with four nested grids that included a 1-km finest grid centered on the Tapaj¿s National Forest. Surface CO2 fluxes were prescribed using idealized diurnal cycles over forest and pasture that derived from flux tower observations; while surface water CO2 efflux was prescribed using a value suggested by in situ measurements in the Amazon region. Our simulation ran from 1 August through 15 August 2001, which was concurrent with the SMC. Evaluation against flux tower observations and Belterra meteorological tower measurements showed that the model captured the observed 2-m temperatures and 10-m winds reasonably well. At 57 m the model reproduced the daytime CO2 concentration better than the nighttime concentration but missed the observed early morning CO2 maxima, in part because of the difficulties of simulating stable nocturnal boundary conditions and subgrid-scale intracanopy processes. The results also suggested that the topography, the differences in roughness length between water and land, the "T" shape juxtaposition of Amazon and Tapaj¿s Rivers, and the resulting horizontal and vertical wind shears all facilitated the generation of local mesoscale circulations. Possible mechanisms producing a low-level convergence (LLC) line near the east bank of the Tapaj¿s River were explored. Under strong trade wind conditions, mechanical forcing is more important than thermal forcing in LLC formation. Persistent clouds near the east side of the Tapaj¿s River may have a significant impact on observed ecosystem carbon flux and should be taken into account if tower fluxes are to be generalized to a larger region.