Evaluating aspects of the Community Land and Atmosphere Models (CLM3 and CAM3) using a dynamic global vegetation model

The Community Land Model (CLM3) Dynamic Global Vegetation Model (CLMDGVM) is used diagnostically to identify land and atmospheric model biases that lead to biases in the simulated vegetation. The CLM-DGVM driven with observed atmospheric data (offline simulation) underestimates global forest cover, overestimates grasslands, and underestimates global net primary production. These results are consistent with earlier findings that the soils in CLM3 are too dry. In the offline simulation an increase in simulated transpiration by changing this variable’s soil moisture dependence and by decreasing canopy-intercepted precipitation results in better global plant biogeography and global net primary production. When CLM-DGVM is coupled to the Community Atmosphere Model (CAM3), the same modifications do not improve simulated vegetation in eastern United States and Amazonia where the most serious vegetation biases appear. The dry bias in eastern United States precipitation is so severe that the simulated vegetation is insensitive to changes in the hydrologic cycle. In Amazonia, strong coupling among soil moisture, vegetation, evapotranspiration, and precipitation produces a highly complex hydrologic cycle in which small perturbations in precipitation are accentuated by vegetation. These interactions in Amazonia lead to a dramatic precipitation decrease and a collapse of the forest. These results suggest that the accurate parameterization of convection poses a complex and challenging scientific issue for climate models that include dynamic vegetation. The results also emphasize the difficulties that may arise when coupling any two highly non-linear systems that have only been tested uncoupled.