Ecosystem Processes at the Watershed Scale: Scaling from Stand to Region
Larry Band
University of Toronto
Toronto, Canada
Ecosystem Processes at the Watershed Scale: Scaling from Stand to Region
Larry Band
University of Toronto
Toronto, Canada
The flux of energy, carbon and water between the land surface and atmosphere are primary regulators of ecosystem form and function. Landscape to regional level ecosystem productivity, watershed runoff quantity and quality, soil erosion and nutrient cycling and export are influenced by these vertical exchange processes and by the distribution and connectivity of land surface stands or patches. As we scale models or analyses of land surface processes from the level of individual patches to regions, the influence of higher resolution patterns often cannot be ignored due to strongly nonlinear local effects of surface state (within patch effects), and the effects of land surface lateral "circulation" of water (between patch effects). Both of these effects can lead to bias in estimation of mean mass and energy storage and flux variables as we progressively aggregate surface features and lose the effects of both heterogeneity and pattern.
A set of general approaches to scaling or aggregating estimates of surface behavior drawn from small scale models to larger scales have been outlined (Ratstetter et al., 1992, Band et al., 1991):
(1) Aggregation and averaging of surface characteristics with the same model structure (simple aggregation);
(2) Scaling distributions of surface characteristics to larger areas with the same model structure;
(3) Calibrating simpler model structures for larger areas, using selected, small area simulations that are more physically based and verifiable.
Band addressed several key questions regarding these approaches to scaling water and carbon flux and stores from local to regional extent:
(1) When is simple aggregation sufficient for scaling?
(2) Are aspatial probability distribution functions of key surface parameters sufficient for scaling (implicitly assuming key processes are independent)?
(3) To what extent do specific patterns (surface connectivity) of key parameters need to be specified (parameters not independent)?
(4) Are key processes and models formulated and observed at plot scales applicable at larger scales?
Not all of these questions are fully answered at the current time, but they are at the core of the overall uncertainties in applying models developed and tested in small research plots or catchments to large terrestrial surfaces.
As we scale from the level of individual patches to regions, the influence of higher resolution patterns often cannot be ignored.
Band addressed these questions in reverse order for the case of hydrologic cycling. The current state of the science of hydrology has grown out of two areas of application. The first is an agriculturist tradition where the major driving questions centered on the need for knowing magnitude and timing of irrigation water, thereby requiring a knowledge of soil water budgets. The control volume has been a soil column, typically conceptualized as a one dimensional system with water entering from the soil surface by infiltration or from below by capillarity or ground water rise, and exiting by deep percolation or evapotranspiration. Within the time scale of rainfall or irrigation application, and over topographically flat, homogeneous research plots, vertical head gradients are considered dominant over lateral gradients, such that lateral, within-soil flux could be neglected. Runoff is considered an out put, but typically only in terms of runoff produced by infiltration excess from the surface.
Generalizing, concepts and models have been applied as local, one-dimensional (1-d) point processes over time. The basic equation used one that underlies most soil hydrology schemes in land surface process models can only be considered an approximation, as important terms affecting the movement of water through soil are generally poorly known, highly nonlinear, and show substantial hysteresis which is not incorporated. In addition, layering of soils and the well expressed lateral variability of water conductivity through soil (which may vary over orders of magnitude within distances of meters), indicates that simple aggregation may not be reliable, and there are strong uncertainties in both parameter estimation for this modeling approach and in being able to validate the model at any scales larger than a soil column.
Water resources engineering has focussed on the generation and routing of river discharge from watersheds, for the impacts on flooding or water supply. This tradition has not required detailed understanding of soil water processes, and has also traditionally considered runoff generation solely as a surface partitioning process. Also generalizing, concepts and models have been applied to storm events, lumping entire catchments into average, or representative states, and often adapting the agricultural soil 1-d model, if a soil model is used at all.
Important terms affecting the movement of water through soil are generally poorly known, highly nonlinear, and show substantial hysteresis.
Over the past two decades, more attention has been put into hydrologic problems that require more detailed consideration of the processes by which water enters, moves through and exits hillslope systems. These studies have been motivated by the need to assess the biogeochemistry of terrestrial ecosystems, particularly in response to chronic inputs of natural and anthropogenically-derived pollutants. Advances in soil water measurement techniques has also aided in substantially revising conceptual models of land surface hydrology to account for a wider range of flowpath and flux processes, with the demonstration that a substantial amount of subsurface flow and redistribution may occur both vertically and laterally by non-Darcian flow. It is now standard for watershed models used to estimate and assess runoff generation and surface ecosystem dynamics to make use of the variable source area model, and attempt to account for substantial heterogeneity and redistribution of soil moisture during and between storm events. In this regard, it is interesting that, to date, land surface processes (LSP) models used in conjunction with atmospheric models are still reliant on lumped, 1-d models, although the ability to build in the impacts of small (hydrological) scale redistribution into the resolution of atmospheric models has become a major research focus.
The impacts of conceptual lumping of short spatial scale heterogeneity of soil moisture dynamics on land surface carbon and water exchange have been assessed by a number of researchers (Entekhabi and Eagleson 1989, Band 1983, Famiglietti and Wood 1995, Sellers et al., 1992). Results indicate that simple aggregation may be sufficiently reliable under certain conditions (i. e., either uniformly wet or uniformly dry), but the potential for significant bias exists when the distribution of surface moisture ranges from well-watered to significantly drier conditions spatially. This state is expected to occur during a dry down when a certain portion of the land surface is maintained in a wet condition by lateral recharge (e. g., bottomland) while other portions dry by a combination of evapotranspiration and drainage (e. g., uplands). While the results of investigations have been variable, it appears that disagreement results from whether specific tests included system states that include this degree of soil moisture variability.
Simple aggregation may be sufficiently reliable under certain conditions but the potential for significant bias exists when the distribution of surface moisture ranges from well-watered to significantly drier conditions spatially.
This effect is illustrated in three examples using a common modeling framework. RHESSys (Regional HydroEcological Simulation System) incorporates a hillslope hydrologic model for lateral soil water redistribution modified from TOPMODEL (Beven and Kirkby, 1979), with a canopy model for computing surface water, carbon and nutrient budgets, adapted from BIOME-BGC (Running and Hunt, 1993). The model framework includes the ability to aggregate and disaggregate the description of the landscape, while maintaining the ability to statistically represent smaller length scale heterogeneity in key variables when scaling to larger regions (Band et al., 1993, Nemani et al., 1993).
The first example of aggregation effects is from simulation work in a forested mountain catchment in western Montana through a deep summer drought in 1988 (Band, 1993). Soup Creek is a 13 square kilometer catchment with a substantial range of conditions including steep north and south facing slopes and significant soil water gradients, particularly along the forested, south facing slopes (see Figure 1.4).
More lumped approaches tend to show more extreme behavior in the areally average flux rates as the full system either wets up or dries down uniformly, while more distributed approaches show a buffered response to both wetter, more optimal conditions, and drought conditions.
Simulation of water and carbon budgets (Figure 1.5) were carried out by hydroecological model with a range of surface representations from fully lumped (ave), to locally lumped by hillslopes (hill) or by grids approximating AVHRR footprints (avhrr), to more distributed, resolving hillslope hydrologic redistribution of soil water downslope (top). Evapotranspiration (ET) and net canopy photosynthesis (PSN) show that more lumped approaches (ave and avhrr) tend to show more extreme behavior in the areally average flux rates as the full system either wets up or dries down uniformly, while more distributed approaches show a buffered response to both wetter, more optimal conditions (near yearday 180), and drought conditions (around yearday 250). This reveals potential bias due to lumping. The more distributed approaches made use of both spatial patterns and covariance of critical surface parameters such as leaf area index (LAI), slope, aspect and soil conditions, as well as incorporating the effects of downslope translocation of water down drainage paths.
The second example is from a very different region, the Southern Old Black Spruce site of the BOREAS experiment, near Prince Albert, Saskatchewan. This site is very flat lying, with only a few meters of total relief within about a square kilometer around a flux tower, and a groundwater table close to or intersecting the surface. A variably sparse canopy of black spruce overlies a forest floor largely dominated by feather moss or sphagnum. Sphagnum occurs in wetter sites with a sparser canopy while feather moss occur in somewhat drier sites with more closed canopies. Significantly, sphagnum is capable of drawing substantial water from the groundwater table by capillarity, while feather mosses appear to be largely precipitation irrigated, and therefore shows much greater variability in moisture content over time. Similar to the Soup Creek site, the landscape has a distribution of surface wetness conditions, with areas of greater sphagnum cover maintaining higher water content and surface evaporation and carbon assimilation during dry periods. However, if 1-km block samples were taken with modal filter on surface cover, sphagnum would largely disappear as it makes up no more than 20 to 40 percent of the flux tower site. The results demonstrate that if the area was aggregated with a modal filter, the lack of the sphagnum stratum would bias the results, particularly during dry periods. Therefore, maintenance of at least aspatial distribution functions of surface conditions is required in this case to avoid bias in aggregation.
Maintenance of at least aspatial distribution functions of surface conditions is required in this case to avoid bias in aggregation.
The final example involves simulations of the full South Platte watershed, with an area of approximately 63,000 km2 in western Nebraska, southern Wyoming and Colorado . Most of the watershed is in the high plains region, while the headwaters are in the Rocky Mountain Front Range. While it may appear obvious to separate these two distinct physiographic and climatic provinces, at the level of resolution of most global atmospheric circulation models, this may be difficult. Figure 1.6 shows simulated areally averaged runoff production over the full basin over the year 1992, with a distribution of model simulation units ranging from one (fully lumped) to over 150 (denoted as simulation 8, or "many"). The figure is essentially the distribution of runoff hydrographs over a range of simulations with varying degrees of model distribution . The significant effect is the threshold response of runoff as the model becomes sufficiently distributed to adequately resolve the Front Range and High Plains as distinct physiographic and climatic units. While this effect may appear to be an obvious expectation, over smaller regions, similar effects can be seen by lumping together regions that are nonstationary in terms of means and covariances of important controlling surface and climate conditions.
The common behavior observed in these simulation experiments due to aggregation is to cause a bias towards underestimating areally averaged surface evapotranspiration and latent heat flux during drydowns.
Discussion
The common behavior observed in these simulation experiments due to aggregation is to cause a bias towards underestimating areally averaged surface evapotranspiration and latent heat flux during drydowns. This is partially due to the shape and steepness of soil characteristic curves relative to the distribution function of surface soil moisture, but also reflects interactions and spatial covariance of vegetative cover, soil type, and smaller scale climatic patterns. Neither the distribution function of soil moisture nor the spatially co-varying patterns of surface variables are maintained as surface characteristics are increasingly aggregated. A common limitation in all the examples is that the model approach uses prescribed meteorology rather than a coupled land surface/atmospheric model. Progress is being made in coupling the distributed land surface approach with RAMS (e. g., Walko et al., 1994) in which case the significant feedbacks may be better evaluated. However, it might be speculated that a coupled boundary layer would respond to surface that is biased to be drier by deepening, entraining more upper atmosphere air, and drying further, which, due to the lack of wet surfaces (particularly non-stomatal forest floors), may develop a positive feedback.
In summary, Band's presentation illustrated that detailed physical treatment of 1-d soil hydrology processes may not be sufficient to resolve important feedbacks between surface and atmosphere. This is due to the uncertainty of process representation within the 1-d model conceptualization, difficulty in adequately parameterizing effective average conditions over large land areas, and the inability to resolve the apparent "buffering" capacity of landscapes brought about by lateral redistribution of soil water over longer time scales that maintains more persistent wet areas through drought conditions.
Acknowledgments: Vapor flux data for the BOREAS Southern Old Black Spruce tower flux site was measured and provided by TE-1, Dr. Paul Jarvis, through the BOREAS Information System.
References
Band, L. E., 1993. Effect of land surface representation on forest water and carbon budgets. Journal of Hydrology 150:749-772.
Band, L. E., P. Patterson, R. Nemani, S.W. Running, 1993. Ecosystem processes at the watershed scale: incorporating hillslope hydrology. Forest and Agric. Meteorology 63:93-126.
Beven, K. J. and M. J. Kirkby 1979. A physically based, variable contributing model of basin hydrology. Hydrol. Sci. Bull. 24:43-69.
Entekhavi, D. and P. S. Eagleson 1989. Land surface hydrology parameterization for atmospheric general circulation models including subgrid scale spatial variability. J. Climate 2:816-831.
Famiglietti, J. and E. F. Wood 1995. Effects of spatial variability and scale on areal-average evapotranspiration. Wat. Res. Res. 31:699-712.
Nemani, R., S. W. Running and L. E. Band, 1993. Overview of a regional hydroecological simulation system (RHESSys) and application to modeling land-atmosphere interactions. In: Environmental Modeling with GIS, Oxford University Press.
Sellers, P. J., M. D. Heiser, F. G. Hall, 1992. Relations between surface conductance and spectral vegetation indices at intermediate (100 m to 15 km) length scales. J. Geophys. Res. 97(D17):19033-19059.
Detailed physical treatment of 1-d soil hydrology processes may not be sufficient to resolve important feedbacks between surface and atmosphere.
Forward to Next Section//Back to Table of Contents//AGCI Homepage//Comments: agcimail@agci.org