Abstract. The major goal of large-scale Earth System Models (ESMs) is to understand and predict global change. However, computational constraints require ESMs to operate on relatively large spatial grids (typically ~1 degree or ~100 km in size) with the result that the heterogeneity in land surface properties and processes at smaller spatial scales cannot be explicitly represented. Averaging over this spatial heterogeneity may lead to biased estimates of energy and water fluxes in ESMs. For example, evapotranspiration rates and the properties that regulate them are spatially heterogeneous at scales orders of magnitude smaller than typical ESM grid cells. Here we quantify the effects of spatial heterogeneity on grid-cell-averaged evapotranspiration (ET) rates, as seen from the atmosphere over heterogeneous landscapes across the globe. In an earlier study, we used a Budyko framework to functionally relate ET to precipitation (P) and potential evapotranspiration (PET), and used a sub-grid closure relation to quantify the effects of sub-grid heterogeneity on average ET at 1° by 1° grid cells- the scale of typical ESM. We showed that because the relationships driving ET are nonlinear, averaging over sub-grid heterogeneity in P and PET leads to overestimation of average ET. In this study, we extend that work to the globe and examine the global distribution of this bias, its scale dependence, and the underlying mechanisms. Our analysis shows that this heterogeneity bias is more pronounced in mountainous terrain, in landscapes where P is inversely correlated with PET, and in regions with temperate climates and dry summers. We also show that the magnitude of this heterogeneity bias grows on average, and expands over larger areas, as the size of the grid cell increases. Correcting for this overestimation of ET in ESMs is important for modeling the water cycle, as well as for future temperature predictions, since current overestimations of ET rates imply smaller sensible heat fluxes, and potential underestimation of dry and warm conditions in the context of climate change. Our work provides a basis for translating the heterogeneity bias into correction factors in large-scale ESMs, and highlights the regions where more detailed mechanistic modeling is needed.