C of plant leaf and soil endmembers. Our approach relies upon the near-universal restriction of C 4 photosynthesis to the herbaceous growth form and the differing performance of C 3 and C 4 grasses in various climates, along with land-cover and crop-type distributions. Specifically, we predict the percentage cover of C 3 and C 4 vegetation in each 5-minute grid cell (;10 km) based on input gridded layers of vegetation growth form fractional cover, crop-area/crop-type distributions, and a high spatial resolution climate data. We develop a consistent set of rules to harmonize the different data layers. The d 13 C of vegetation in South America is then estimated based on the C 3 /C 4 composition in each land grid cell, assuming constant mean values for closed C 3 tropical forest (À32.3%), open C 3 forest ecosystems (À29.0%), C 3 herbaceous cover (À26.7%) and C 4 herbaceous cover (À12.5%). In addition to using the mean isotope values, we also incorporate the measured standard deviation for each category. Soil d
13C is estimated for the C 4 -favored climate regions of South America using two, largely independent approaches: one that is derived from our vegetation d 13 C prediction and one that is based on a previously published relationship between fractional woody cover and the d 13 C of soil organic carbon. Finally, we present preliminary maps of relative uncertainty in the estimates of vegetation growth form, generated by integrating global measures of accuracy with local measures of neighborhood variability. These maps demonstrate that the highest uncertainty is found in savanna ecosystems, which contain the most heterogeneous vegetation cover and structure along with a high percentage of C 4 grass cover.