Abstract
Integration of spatial biophysical data to drive terrestrial ecosystem models requires not only that the data layers themselves be spatially compatible, but also that they adequately resolve landscape attributes that the models require. We have integrated soils, topographic, vegetation, and meteorological data to simulate carbon, nitrogen, and water dynamics at watershed (7 km) to basin (63,000 km) scales within the South Platte River Basin in Colorado, Wyoming, and Nebraska, using varying data resolutions and levels of topographic aggregation. Simulations indicate that spatial correlation of input variables with synoptic weather patterns is critical in determining dynamic processes such as hydrologic outflow and evapotranspiration, yet is not always possible across large regions or heterogeneous terrain. We examine the flow of spatial biophysical data as input to multiple-scale ecosystem process simulations and discuss how spatial data requirements, compatibility, and availability are affected across scales.
Simulation of ecological processes at regional to global scales typically requires both intensive and extensive biophysical parameterization of the earth's surface and atmosphere. Unfortunately for ecological modelers, critical processes are continuously interacting at spatial scales from local (e.g. watershed hydrologic partitioning) to regional (e.g. synoptic meteorological events), and at temporal scales that range from instantaneous (e.g. vapor flux densities) to seasonal (e.g. vegetation phenology), and are thus not easily incorporated into ecosystem models that require multiple data of fixed spatial or temporal resolution as input. In order for ecological models to adequately describe or predict physical processes such as net primary productivity or streamflow dynamics, input data resolution must be compatible not only with the scale of actual ecological functioning, but also with the resolution at which interpretation of results is critical.
The compilation and implementation of spatial biophysical data into ecosystem process simulation has at least four basic considerations. 1) What are the minimal biophysical data required to parameterize all relevant processes in the model? 2) What are the maximum spatial and temporal resolutions of these data that will render meaningful simulations? That is, what are the spatial and temporal levels at which the processes are critical in the real, not simulated, world? 3) Do these data exist, or can they be formulated from existing data? 4) Once required or optimal data resolutions are determined or hypothesized, are multiple data "layers" compatible in their resolutions to be combined in a geographic information system?
As we move toward the application of ecosystem models to continental and global scales, it is essential to aggregate small-scale ecologic functioning to scales that allow current computational hardware to store and process the data without insurmountable time and costs involved. In aggregating spatial data to these scales, however, critical ecological, climatic, or topographic detail may be lost, and large scale simulation results may become an aggregation of errors rather than broad representation of earth processes.
For example, through modeling studies that utilize suites of spatial data to simulate landscape ecological processes, we have tested the effects of partitioning landscapes from sub-watershed to regional scales on hydrologic and ecosystem process model simulations . Results showed that as we decrease the scale of topographic resolution to match coarse resolution biophysical data input to the models (e.g. 1 km AVHRR data for vegetation characteristics), key topographic relationships with primary production, meteorology, and hydrologic partitioning are lost, and simulation errors dramatically increase across regional and larger scales. Lammers, et al. (1995) state, "Initial attempts to extend these models over larger land areas implicitly assumed that the surface could be treated as a spatially exhaustive set of homogeneous areas, acting independently and in parallel. However, recent work has demonstrated that surface heterogeneity is both strongly expressed at all scales and is not simply averaged in a functional sense (Avissar, 1992; Band, 1993). The nonlinear response of water and carbon flux processes to available soil water, meteorological variables and certain vegetation canopy attributes commonly results in significant bias when computing areal averaged flux using mean or average surface conditions... In scaling or aggregating a biophysical model over progressively larger areas, a key problem is estimation of this distributional information as direct sampling becomes infeasible."
In one study (Lammers et al., 1995) simulations of forest ecosystem processes over a 3000 km2 watershed were used to investigate scale effects on sampling and representing land surface attributes. Specifically, the development and control of bias in simulated carbon and water exchange processes as both scale and resolution of the landscape changed was investigated. This study showed that an order of magnitude resolution change of the original land data sets for topography and vegetation cover can produce similar results in the carbon and water flux processes as long as a joint distribution function describing significant surface attributes is preserved. It may be possible to define this distribution function by partitioning surface heterogeneity as variance between the spatial units used for simulation and variance within units. This scheme implies, however, that biophysical attribute data are available at a common, minimum resolution, and that within-unit variance may be tracked across aggregate scales. When coupling multiple "layers" in a GIS, data are typically gathered from disparate sources, and common scale must be initially reconciled, which in itself, ideally, requires tracking of variance-to-scale relationships. In another study (Band et al., 1995), the effects of digital elevation model (DEM) resolution were investigated for the computation of hydrological terrain- soils indices for input to hydrological models. This study found that DEM resolution appears to have regular impacts on simulated hydrographs, with greatest sensitivity in grid based (raster) methods and the least sensitivity in contour and slope line based (vector) methods. Again, this problem may be confounded when varying degrees of topographic representation must be coupled with other biophysical data such as vegetation quantification (leaf area index, specific leaf area, etc.). Finding mutual distribution functions among various layers in a GIS may be a formidable task, but one that would ultimately define the interaction of topography-vegetation-soils-hydrology.
Whether we are simulating local or global scales, physical processes must be described at the scale at which they are critical to ecological functioning. As a scientific community, however, we are currently severely limited in our ability to adapt input databases to address specific ecological questions because of inherent problems and variability of biophysical information across scales. Rather we must spend our time adapting arbitrarily-scaled data sets to one another, all the while trying to evaluate what we are missing in the first place with coarse resolution data, and what we are losing in the process of scaling fine-resolution data to coarser scales. Quite often databases are created with resolutions borne of hardware limitations and processing times, rather than regard to relevant physical scales. Because it is always better to scale fine resolution data up to larger scale applications than the converse (i.e. loss of feature or process resolution can be monitored and evaluated), databases should be created at the finest resolution possible, from the initial hardware (e.g. satellites) to archiving and distribution, so that end users of the data can objectively evaluate the level of aggregation that is suitable to the specific application.
The objective of this study is to illustrate the need for spatially compatible biophysical data sets by presenting examples in which scale is shown to be a critical parameter in simulation results over a given landscape. By tracing the flow of data through coupled hydrologic-ecosystem process models, we identify points at which data layer resolution incompatibility becomes a problem in addressing the ecological questions the models are designed to answer.
References
Lammers, R.B., L.E. Band and C.L. Tague (1995) Scaling Behavior of
Watershed Processes. In P. van Gardingen, G. Foody and P. Curran
(eds.) Scaling-up. Final revision submitted, Cambridge University
Press.
Avissar, R. (1992) Conceptual aspects of a statistical dynamical approach to represent landscape subgrid-scale heterogeneities in atmospheric models. Journal of Geophysical Research 97, 2729-2742.
Band, L.E. (1993) Effect of land surface representation on forest water and carbon budgets. Journal of Hydrology 150, 749-772.
Band, L.E., R. Vertessy and R.B. Lammers (1995) The Effect of
Different Terrain Representations and Resolution on Simulated
Watershed Processes, Zeitschrift fur Geomorphologie, vol. 101,
November, pp. 187-199.
Robert G. Kremer
Natural Resource Ecology Laboratory
Colorado State University
Fort Collins, CO 80523
970/491-1966
FAX:970/491-1965
rkremer@nrel.colostate.edu
Fill in your name and email address here