Scaling ecosystem models from watersheds to regions: Tradeoffs between model complexity and accuracy

Large-scale estimates of net primary production (NPP) derived from ecosystem models are based on several assumptions whose impact on NPP estimates are largely unknown. In this dissertation, I examine how estimates of NPP are affected by the use of several key assumptions associated with scaling ecosystem models from watersheds to regions.; Large-scale ecosystem models assume that canopy photosynthetic properties, such as specific leaf area (SLA) and leaf nitrogen (N), are constant within a biome. However, ignoring intra-biome variations of SLA and leaf N can lead to errors in simulated NPP ranging from 30-60%. SLA is closely related to LAI in both coniferous (R{dollar}sp2{dollar} = 0.93) and Eucalyptus (R{dollar}sp2{dollar} = 0.78) forests. Leaf N per unit area is highly correlated to canopy transmittance (R{dollar}sp2{dollar} = 0.94) in conifer forests. Because SLA and leaf N are both closely related to LAI, satellite images of LAI can be used to estimate the variability in canopy properties across biomes.; Models which estimate NPP based on the spatial distribution of absorbed solar radiation (APAR) are potentially useful for making unbiased, high-resolution estimates of NPP at large scales. An APAR model relying on elevation to predict the dry matter yield of energy {dollar}(epsilon){dollar} can accurately predict the spatial variations in NPP simulated using a computationally-intensive ecosystem model (R{dollar}sp2{dollar} = 0.90). The APAR model proved to be 36 times more computationally efficient than the ecosystem model, allowing unbiased, high-resolution estimates of NPP over large spatial scales.; Global-scale ecosystem models utilize datasets with cell sizes of {dollar}1spcirctimes1spcirc,{dollar} within which sub-cell land surface variations are averaged. The landscape averaging associated with these datasets have an unknown impact on estimates of NPP. Averaging complex landscapes can contribute error to NPP estimates, depending upon: (a) the temporal resolution for which NPP is being simulated, and (b) the complexity of the land surface relative to the size of the cell or partition being used to represent that land surface. Averaging sub-grid cell landscape variations can result in as much as a 30% error in NPP estimates. Careful partitioning of complex landscapes can greatly reduce the magnitude of this error.