Abstract.A process-based fire parameterization of intermediate complexity has been developed for global simulations in the framework of a Dynamic Global Vegetation Model (DGVM) in an Earth System Model (ESM). Burned area in a grid cell is estimated by the product of fire counts and average burned area of a fire. The scheme comprises three parts: fire occurrence, fire spread, and fire impact. In the fire occurrence part, fire counts rather than fire occurrence probability are calculated in order to capture the observed high burned area fraction in areas of high fire frequency and realize parameter calibration based on MODIS fire counts product. In the fire spread part, post-fire region of a fire is assumed to be elliptical in shape. Mathematical properties of ellipses and some mathematical derivations are applied to improve the equation and assumptions of an existing fire spread parameterization. In the fire impact part, trace gas and aerosol emissions due to biomass burning are estimated, which offers an interface with atmospheric chemistry and aerosol models in ESMs. In addition, flexible time-step length makes the new fire parameterization easily applied to various DGVMs.Global performance of the new fire parameterization is assessed by using an improved version of the Community Land Model version 3 with the Dynamic Global Vegetation Model (CLM-DGVM). Simulations are compared against the latest satellite-based Global Fire Emission Database version 3 (GFED3) for [1997][1998][1999][2000][2001][2002][2003][2004]. Results show that simulated global totals and spatial patterns of burned area and fire carbon emissions, regional totals and spreads of burned area, global annual burned area fractions for various vegetation types, and interannual variability of burned area are reasonable, and closer to GFED3 than CLM-DGVM simulations with the commonly used Glob-FIRM fire parameterization and the old fire module of CLM-DGVM. Furthermore, average error of simulated trace gas and aerosol emissions due to biomass burning is 7 % relative to GFED3. Results suggest that the new fire parameterization may improve the global performance of ESMs and help to quantify fire-vegetationclimate interactions on a global scale and from an Earth system perspective.

[1] Arid and semiarid regions represent a large fraction of global land, but most of the existing dynamic global vegetation models (DGVMs) do not include shrubs or do not effectively distinguish shrubs from grasses, and hence cannot realistically reproduce the ecosystem formation and variability there. A shrub submodel is developed here for the Community Land Model-DGVM (CLM-DGVM), and the major revisions include (1) explicit consideration of shrubs' drought tolerance in the photosynthesis computation; (2) use of appropriate phenology type and morphology parameters for shrubs; (3) consistent treatment of fractional vegetation coverage; (4) development of tree/grass/ shrub hierarchy for light competition; and (5) improvement of the allocation scheme to avoid unrealistic behaviors. Preliminary global offline CLM-DGVM simulations for 400 years show that, with the shrub submodel, the simulated global distribution of temperate shrubs agrees with Moderate Resolution Imaging Spectroradiometer (MODIS) data. The simulated shrub coverage reaches its peak around annual precipitation (P ann ) of 300 mm, the grass coverage reaches its peak over a broad range of P ann (from 400 to 1100 mm), and the tree coverage reaches its peak for P ann = 1500 mm or higher, all in good agreement with MODIS data.