Impacts of Scale and Heterogeneity in Dynamic Global Vegetation Models

Dynamic Global Vegetation Models (DGVMs) are crucial components in coupled land-climate simulations for representing the feedbacks between the atmosphere and terrestrial biosphere. Accurate predictions of the impact of climate change depend on well-functioning DGVMs. Construction of such complex models necessitates the encoding of a large number of assumptions, both implicit and explicit. One particular assumption that is generally implicit in this class of models is the homogeneity of land cover within a grid cell. The spatial distribution of land cover, and in particular Plant Functional Types (PFTs), is often not considered. For some model processes this is equivalent to assuming that vegetation is homogeneous within the grid cell, which is typically in the order of 1 by 1 degree in size. This is clearly not the case for large areas of the surface of the Earth. A key process that will be impacted by this assumption is the net accumulation of soil carbon in a grid cell. Carbon enters soil in the form of litter from vegetation and is removed in the form of carbon dioxide respired by soil micro-organisms. Typically soil carbon in a DGVM grid cell is represented by a single horizontal box with a number of vertical layers. Each PFT drops litter into the same soil pool. Consequently the build-up of soil carbon under a grassland, for example, may be influenced by the presence of a deciduous forest in the same grid cell even though they may be physically separated by a significant distance. Clearly the model predictions, at least in terms of soil carbon and heterotrophic respiration, are scale dependent under these assumptions. Running the DGVM at a different resolution would potentially change the nature of PFTs contributing to a given area of soil. The other limit, which is not generally considered, would be to model each PFT with its own independent soil pool. However, this is also an unrealistic description of some areas of the land surface which may exist as a highly heterogeneous matrix. This paper introduces a simple index of heterogeneity that is generated from satellite derived land cover information (the GLC2000 data set) across the global land surface. Prior to transformation the data is mapped into notional PFTs. The resulting information can then be used examine the impacts of assumptions about heterogeneity. The resolution of the GLC2000 data is 1km and so there is an implicit assumption that this scale of data is sufficient to represent the heterogeneity of processes at the 1 by 1 degree scale. The Sheffield Dynamic Global Vegetation Model is used to predict global Net Ecosystem Productivity (NEP, the net accumulation of carbon that is the balance between photosynthesis and respiration) under both limits: that the PFTs within a grid cell all share the same resources and that they are independent. The heterogeneity map is then used to recombine the model outputs by weighting the contribution of each scenario based on the local heterogeneity in vegetation cover. Results, presented for the global land surface, show regions in which the implicit assumptions of homogeneity and scaling have a large impact in model predictions in NEP.