AbstractThe interpretation of towerābased eddyācovariance (EC) turbulent flux measurements above forests hinges on three key assumptions: (1) steadiness in the flow statistics, (2) planar homogeneity of scalar sources or sinks, and (3) planar homogeneity in the flow statistics. Large eddy simulations (LESs) were used to control the first two so as to explore the breakādown of the third for idealized and real gentle topography such as those encountered in Amazonia. The LES runs were conducted using uniformly distributed sources inside homogeneous forests covering complex terrain to link the spatial patterns of scalar turbulent fluxes to topographic features. Results showed strong modulation of the fluxes by flow features induced by topography, including large area with negative fluxes compensating āchimneyā regions with fluxes almost an order of magnitude larger than the landscape flux. Significant spatial heterogeneity persisted up to at least two canopy heights, where most eddyācovariance measurements are performed above tall forests. A heterogeneity index was introduced to characterize and contrast different scenarios, and a topography categorization was shown to have predictive capabilities in identifying regions of negative and enhanced fluxes.