Modeling Global Biosphere-Atmosphere Interactions
Elena Shevliakova
Carnegie Mellon University
Pittsburgh, Pennsylvania
Modeling Global Biosphere-Atmosphere Interactions
Elena Shevliakova
Carnegie Mellon University
Pittsburgh, Pennsylvania
Modeling interactions between the atmosphere and biosphere continues to pose a significant difficulty, partly because of the tightly coupled flow of water, energy, and carbon between terrestrial ecosystems and the atmosphere. In recent years the climate-change community has taken a number of different approaches toward understanding physical, ecological and biogeochemical aspects of vegetation-climate interactions. Studies have been performed on different spatial and temporal scales, from an individual leaf to the entire globe (IPCC 1995).
Modeling biosphere-atmosphere interactions on the global scale
Past studies of the global-scale biosphere-atmosphere have been performed in one of two diagnostic ways: (1) impacts of climate change on the biosphere, or (2) biosphere feedbacks to the climate system. Global impacts studies have been performed with biogeography and/or biochemistry models (VEMAP, 1995). Biosphere feedbacks to the climate system have been explored with land-surface models in GCMs. A third, newly emergent approach is the integrated study of biosphere and climate interactions (Foley, et al., 1996).
Biogeography models attempt to simulate global equilibrium distributions of vegetation under current and future climates. Most of these models are correlative in nature and rely on observed associations between climate and vegetation. Most have been developed with a rigid environmental envelope approach in which a few climatic constraints determine the pattern of vegetation cover. More recent models such as BIOME-1 (Prentice, et al. 1992), MAPSS (Neilson, et al., 1992), and BIOME 3.0 (Haxeltine and Prentice, 1996), have combined elements of correlative and mechanistic approaches, but focus primarily on modeling potential vegetation types. Despite recent advances in understanding the factors that control vegetation distribution, existing models are not able to simulate prevalence of different vegetation types equally well.
Biogeochemistry models focus on the simulation of nutrient flows among vegetation, soil and the atmosphere. Typically these models deal only with changes in carbon and nitrogen flows and net primary productivity (NPP) due to climate change. These models assume constant distributions of vegetation and soil types. The VEMAP (1995) study compared the sensitivity of different biogeochemistry models to assumptions about different vegetation distributions, carbon fertilization effects and climate change.
Modeling interactions between the atmosphere and biosphere continues to pose a significant difficulty, partly because of the tightly coupled flow of water, energy, and carbon between terrestrial ecosystems and the atmosphere.
Land-surface models aim to represent the exchange of energy and moisture between the atmosphere and the Earth's surface for prescribed distributions of vegetation and soil. The joint WGNE-GCIP Project for Intercomparison of Land Surface Parameterization Schemes (PILPS) has considered 22 land-surface models to understand their capabilities and potential applications. At the land surface, the available radiative energy is partitioned into latent and sensible heat fluxes depending on the type of vegetation. The earlier land-surface models used very simple parameterizations for such partitioning. In later models such as the Simple Biosphere (SiB) model of Sellers et al. (1985), the Biosphere-Atmosphere Transfer Scheme (BATS) of Dickinson et al. (1993), and the Land Surface Exchange (LSX) of Pollard and Thompson (1995) plant-soil-atmosphere interactions are explicitly parameterized to represent vegetation physiological processes such as evapotranspiration and seasonal changes in foliage. As a rule, detailed physical descriptions of canopy and soil processes in these models lead to an explosion in the number of parameters and needed computational resources. The latter is of particular concern in long-term climate change simulations.
Typically, the vegetation distribution in land-surface models is assumed to be constant over time. There are two general approaches to describing vegetation at the grid scale of an atmospheric GCM (~ 300 km x 300 km). In the first, more widely used approach (e. g., SiB), the entire area of each grid cell is assumed to be covered by a homogeneous mixture of vegetation. In the second approach of Mosaic LSM, different types of vegetation are assumed to occupy different areas of the cell and to interact independently with the atmosphere (Koster and Suarez, 1992).
Development of global dynamic models of vegetation remains a high priority task in predicting realistic impacts of future climate changes on vegetation. Shevliakova discussed the attempts she is aware of to address this issue. These attempts to simulate vegetation dynamics fall under three general strategies. The first strategy, represented by the work of Woodward et al. (1995), aims to model such global vegetation characteristics as Leaf Area Index (LAI), Net Primary Productivity (NPP) and stomatal conductance. These are calculated with a plant productivity and phyto geography model. The second strategy, advanced by Foley et al. (1996), is based on the concept that vegetation system dynamics are shaped to maximize NPP. Thus, for a given climate, a plant functional type (PFT) can be identified which can survive given the local conditions (soil and climatic) and which maximizes NPP. The third strategy is incorporated in the process-based model Hybrid of Friend, et al. (1997). It is based on a stand modeling approach and is able to project spatially, temporally and biologically detailed responses to climate change.
Development of global dynamic models of vegetation remains a high priority task in predicting realistic impacts of future climate changes on vegetation.
An integrated approach to modeling biosphere-atmosphere interaction was implemented in the Integrated Biosphere Simulator (IBIS) model (Foley, et al., 1996). The first version, IBIS 1.1 consists of four different modules that describe vegetation phenology, biogeochemistry, land-surface-atmosphere interaction, and dynamic changes in specific plant types. A combination of different PFTs provides the vegetation cover. In a manner similar to biogeography models, this model applies a set of climatic constraints to define which PFTs can potentially exist within each grid cell. Then, using the approach of Haxeltine and Prentice (1996) from BIOME 3.0, potential PFTs are ranked according to their NPP and the dominant PFTs are defined to prevail. The information about vegetation structure is used in a land-surface module based on the LSX land-surface model (Pollard and Thompson, 1995). The land surface module simulates fluxes of energy, moisture, momentum and CO2. The IBIS model can be directly incorporated into a GCM and allows study of atmosphere-biosphere interactions in an internally consistent manner.
Scale and Vegetation Description
The choice of vegetation description is scale dependent. At a smaller scale it is possible to describe vegetation in fine taxonomic details ( e. g., species). At a larger scale, description in terms of general physiognomic or environmental features has been often used (e. g., biome). A number of different classification schemes have been proposed (Table 1.31). Until recently, environmental schemes have been frequently used in modeling biosphere-atmosphere interactions. Köppen (1936) developed a bioclimatic classification scheme based on physiological vegetation classification. This scheme depends on mean monthly temperatures and seasonal precipitation expressed through mean annual precipitation. The Köppen scheme distinguishes 12 different bioclimates. Another popular and widely used bioclimatic scheme is the Holdridge Life Zone Classification (1947), based on the three climate parameters: annual biotemperature (ABT), annual potential evapotranspiration (APT), and average total annual precipitation. Annual biotemperature is calculated by averaging only positive temperatures over the course of a year.
The choice of vegetation description is scale dependent.
|
|
|
|
|
|
Physiognomic |
|
structural type or growth form (e.g., grass, broad-leaved deciduous tree) |
formation biome |
|
Environmental |
|
bioclimatic zones |
|
|
Many-factor |
|
microlandscape or nature-complex, biogeocoenose |
landscape units |
|
Biotic areas |
|
vegetation girdle ~ geographic areas of species |
biocoenoce-type biotic provinces |
|
Zones and series |
|
ecological series ~ ordination along env. gradients |
belts or zones formation series |
|
Species dominance |
|
'association' |
|
|
Vegetation dynamics |
|
succession sequence of ecological series |
associations climax communities |
|
Numerical |
|
measurements of relative similarity of species distribution |
nodum |
The IPCC impacts study criticized correlative biogeography models for lacking causal relationships between climate and plant physiology, and thus their presumed ability to predict only the current distribution of biomes. In the later mechanistic models such as BIOME 3.0 and IBIS 1.1, the representational sophistication of the vegetation increased further to describe vegetation not in terms of biomes but in terms of plant functional types (PFTs). The notion of PFTs is not completely new in vegetation classification and is related to physiognomic approach (Table 1.31).
The notion of growth form goes back to Waring's "Oecology of Plants" (1909). According to Waring, plant formations (or biome types) are characterized by their physiognomy and structure (e. g., dominant growth forms) and by some environmental characteristics (e. g., location or temperature range). For example, consider different grass formations: grassland (savanna), temperate grassland (steppe), alpine meadows, and salt marshes. All these formations are dominated by the same growth form: grasses and grass-like plants. Some researchers used classifications similar to growth form such as life-form, physiognomic plant types, and PFTs. Many different schemes were developed to classify vegetation formation from the physiognomic point of view, with different kinds and numbers of growth forms and consideration for environmental factors.
The growth-form based approach is the best suited to the analysis of ecosystem impacts and feedbacks to the climate system on the global scale.
The growth-form based approach is the best suited to the analysis of ecosystem impacts and feedbacks to the climate system on the global scale because it provides the means to: 1) explore changes in vegetation structure; 2) reflect on important characteristics of the land surface such as the amount of respiring phytomass ( e. g., trees vs. grasses), the amount and configuration of photosynthetic and transpiring surfaces (e. g., needle leave vs. broadleaf), and seasonal variations ( e. g., deciduous vs. evergreen); 3) describe only the general characteristic of plants that are essential in modeling atmosphere-biosphere interactions and are computationally efficient.
Analysis of Current Global-Scale Climate-Vegetation Relationships
There is a widespread acknowledgment that climatic factors such as ambient air temperature, incident solar radiation and water availability play an important role in the distribution and functioning of vegetation. These climatic factors are complex and can, therefore, be described by alternate ensembles of variables. Climate factors can be included in a vegetation model in a number of ways. First, they can enter as threshold constraints in phenology modules. Second, they can be external drivers of physiological functions (e. g., photosynthesis and respiration). Third, they can be used as scalars in representing the likelihood of different events in the life of vegetation (e. g., the probability of a disturbance). Fourth, climate variables can be a factor in vegetation classification itself.
A probabilistic approach has been applied to the analysis of relationships between global vegetation and climate (Siegel, et al., 1995a; Shevliakova 1996). This approach is applicable to different types of vegetation characterization and to ensembles of physiologically relevant climate variables. It is also computationally efficient . The probabilistic approach to the analysis of vegetation distribution was originally proposed in the 1960s by Richard Goodall (1970), but had only limited use due to the lack of computational power and large-scale information on climate and vegetation at that time. This approach has been used on a small scale to simulate current and potential future distributions of plant species in central European mountain forests (Kienast, et al. 1996).
In their earlier work, Shevliakova and colleagues found that non-parametric methods provide computationally efficient means to explore highly nonlinear relationships between different sets of climatic variables and different vegetation types. These methods can be easily used with different types of vegetation classifications. Non-parametric density-estimation methods are well understood and have increasingly been used for both univariate and multivariate analysis. Shevliakova and colleagues applied a multivariate non-parametric density estimation approach to estimate the probability density functions of different Olson vegetation types over North America. Table 1.32 shows the fewest-variable combinations that are necessary to achieve an excellent degree of agreement between the predicted and observed distributions for each vegetation type.
Non-parametric methods provide computationally efficient means to explore highly nonlinear relationships between different sets of climatic variables and different vegetation types.
Vegetation types are ordered according to the mean value of the latitude at which they occur. The first column shows mean values for latitudes at which each Olson vegetation type occurs. Columns 3 through 11 show the combination of different variables necessary to describe the prevalence of that type. The column titled "other" is included to show that variables other than those used in columns 3 through 11 may be needed to achieve excellence in prediction. For two vegetation types, wooded tundra margin and deserts, no combination of explanatory variables was found to provide excellent agreement between predictions and observations. Table 1.32 shows that in high latitudes, seasonal characteristics such as minimum and maximum temperatures and available moisture during the warm period play an important role. This table also indicates that the required number of variables increases in the area below wooded tundra margin (a proxy for tree-line), where different kinds of forests are currently observed.
Table 1.32
Summary of key variables for describing different vegetation types
Although many models of vegetation distribution have similar principles of plant functioning at their foundation, they choose very different sets of climatic variables to represent these principles or use different methodologies to estimate the variables . For example, consider the representation of cold tolerance of plants. Woodward, in his earlier studies, uses the absolute minimum temperature, obtained by regression of the monthly minima from available meteorological stations. In BIOME 1.0, Prentice and colleagues use the temperature of the coldest month, estimated from the 12 monthly averages from Leemans and Cramer climate database. In the BIOME 3.0 model of Haxeltine and Prentice, thresholds for absolute minimum temperatures are obtained by looking over available data from meteorological stations from Muller database. As Katherine Prentice notes, "Any combination of terms has potential, and yet there is no perfect index from a biological point of view."
The choice of climate variables is an important part of the specification of the land-cover model. The analysis outlined above is able to provide insights and help in choosing the necessary set of climate variables for adequate representation of climate-vegetation relationships. This choice depends not only on ecophysiological relevance, but also on the availability and accuracy of information ( e. g., soil nutrient characteristics are less readily available than soil types). Often it has been known from the small-scale studies (e. g., individual plants or communities) that variables such as vapor-pressure deficit, photosynthetically active radiation (PAR), snowpack, and wind speed are important factors in plant functioning, but information about these variables is not available on a global scale. In such cases it may be feasible to use the information from the 1xCO2 equilibrium simulations by different GCMs.
Aggregation of processes in individual tree models: a Case Study of TREGRO
In global vegetation models (e. g., IBIS, Hybrid) different types of vegetation compete for resources. Their growth and development are simulated through modeling carbon accumulation in different compartments: leaves, branches, stems, and roots. Representation of carbon accumulation and partitioning in global vegetation dynamics models is similar to individual plant models ( e. g., TREGRO of Weinstein, et al., 1991), but individual plant models have more plant compartments, finer temporal resolution and are parameterized for a particular species. Parameter estimation for the individual plant models has been facilitated through a sustained effort in data gathering on tree responses to the prevailing environment and imposed stresses.
In order to understand the increased sophistication and complexity of individual plant models, sensitivity and uncertainty analyses must be performed. Siegel et al. (1995b) performed such analyses for the TREGRO model and explore effect compartment and temporal aggregation on the model's predictions. Model reduction techniques have been used in different fields such as chemical engineering, economics and ecology. Reduced form models offer computational efficiency and often require specification of fewer input parameters. The disadvantage of reduced form models is that they may be less accurate, and be unable to fully reproduce the behavior of the parent model. Because the accuracy and reliability of the simplified model depends upon the simplifying assumptions employed, model reduction can be viewed as a sensitivity study of model predictions to the nature of the simplifying assumptions.
In the case study of TREGRO, simulations and analyses were performed for a red spruce tree using meteorological data collected near Ithaca, New York. In the full TREGRO model, a red spruce tree is represented by 12 compartments ( i. e., branch, stem, 4 leaf (needle) age classes, and coarse and fine root classes in 3 soil layers). Carbon stored in each of these compartments is divided into three types: living structure; dead structure or wood; and total non-structural carbon (TNC). The total carbon balance in a plant is calculated on a daily time step. The key to time aggregation is estimation of photosynthesis for time steps greater than one hour. All other parameters in TREGRO are calculated on a daily time step.
Model reduction can be viewed as a sensitivity study of model predictions to the nature of the simplifying assumptions.
In order to explore the sensitivity of model predictions to uncertainty in the empirical model parameters, a subset of 40 parameters was treated probabilistically: respiration and growth rates for different compartments, ozone stress-related modifiers, fractions of leaves of each class in shade and sun through the course of a day, and initial masses of tissues in the compartments. The probability distributions and interrelationships for these parameters are subjective judgments of the model developers. The relationships between inputs and outputs were examined through partial rank correlation analysis.
The case study results indicate that the most compact model is over an order of magnitude faster than the full TREGRO model. When using a time aggregated TREGRO, leaf TNC and biomass typically provide better results than the woody parts. When both time aggregation and compartment aggregation are employed the total biomass results are closer to the full TREGRO results (see Figure 1.33). This increase in computational efficiency is achieved at a small loss in accuracy of total biomass predictions. However, simulations of TNC in the aggregated versions of TREGRO are different from TNC values in the full version of the model. This suggests that this model reduction approach may be suitable for use in analyses where accurate TNC results are not needed.
Figure 1.33
Comparison of the mean values of total tree biomass from the full version of TREGRO (12 compartment, hourly time step) and three versions of µ-TREGRO (24 hr time step, compartments and environmental inputs aggregation). In all four cases, the means were derived from 100 simulations with different combinations of 40 parameters describing physiological and growth processes.
This case study of the individual plant growth model TREGRO provides insight into the key factors controlling the simulated dynamics, and guides the development of an acceptable reduced form model in which longer time steps and fewer compartments are used in the simulation. The development of such reduced form models offers computational efficiency gains and requires fewer parameters for larger scale models. Analysis of model uncertainties with different compartment and time step aggregation are needed both to permit quantification of the impact of the myriad uncertainties in model inputs, empirical parameters and underlying structure, and also to guide future model development on different scales.
References
Dickinson, R. E., A. Henderson-Sellers and P. J. Kennedy, 1993. Biosphere-Atmosphere Transfer Scheme (BATS) Version 1e, as Coupled to the NCAR Community Climate Model, NCAR Technical Note NCAR/TN-387+STR, NCAR Climate and Global Dynamics Division, Boulder, Colorado.
Foley, J. A., I. C. Prentice, N. Ramankutty, S. Levis, D. Pollard, S. Sitch and A. Haxeltine, 1996. An integrated biosphere model of land surface processes, terrestrial carbon balance, and vegetation dynamics. Global Biogeochemical Cycles 10(4): 603-628.
Friend, A. D., A. K. Stivens, R. G. Knox and M. G. R. Cannel, 1997. A process-based, terrestrial biosphere model of ecosystem dynamics (Hybrid v3.0). Ecological Modeling 95: 247-287.
Goodall, D. W., 1970. Statistical plant ecology. Annual review of Ecology and Systematics 1 : 99-124.
Haxeltine, A. and I. C. Prentice, 1996. BIOME3: An equilibrium terrestrial biosphere model based on ecophysiological constraints, resource availability, and competition among plant functional types. Global Biogeochemical Cycles 10(4): 693-709.
Holdridge, L. R., 1947. Determination of world formulations from simple climatic data. Science 105: 367-368.
IPCC, 1995. Climate Change 1995: Impacts, Adaptations and Mitigation of Climate Change: Scientific-Technical Analyses. Contribution of Working Group II to the Second Assessment Report of the Intergovernmental Panel on Climate Change (IPCC). Cambridge University Press.
The most compact model is over an order of magnitude faster than the full TREGRO model.
Kienast, F., B. Brzeziecki and O. Wildi, 1996. Long-term adaptation potential of Central European mountain forests to climate change: a GIS-assisted sensitivity assessment. Forest Ecology and Management 80: 133-153.
Köppen, W., 1936. Das geographische System der Klimate. Handbuch der Klimatologie. Berlin, Gebruder Borntrager.
Koster, R. and M. Suarez, 1992. Modeling the Land Surface Boundary in Climate Models as a Composite of Independent Vegetation Stands. Journal of Geophysical Research 97(D3): 2697-2715.
Neilson, R. P., G. A. King and G. Koerper, 1992. Toward a rule-based model. Landscape Ecology 7: 27-43.
Pollard, D. and S. L. Thompson, 1995. Use of a land-surface-transfer scheme (LSX) in a global climate model: the response to doubling stomatal resistance. Global and Planetary Change 10: 129-161.
Prentice, I. C., W. Cramer, S. P. Harrison, R. Leemans, R. A. Monserud and A. Solomon, 1992. A global biome model based on plant physiology and dominance, soil properties and climate. Journal of Biogeography 19: 117-134.
Sellers, P. J., 1985. Canopy reflectance, photosynthesis, and transpiration. International Journal of Remote Sensing 6 (8): 1335-1372.
Shevliakova, E., 1996. Application of Statistical Methods for Modeling Impacts of Climate Change on Terrestrial Distribution of Vegetation. Pittsburgh, Carnegie Mellon University.
Siegel, E., H. Dowlatabadi and M. J. Small, 1995a. A Probabilistic Model of Ecosystem Prevalence. Journal of Biogeography 22: 875-879.
Siegel, E., H. Dowlatabadi and M. J. Small, 1995b. Sensitivity and Uncertainty Analysis of an Individual Plant Model and Performance of its Reduced Form Versions: A Case Study of TREGRO. Journal of Biogeography 22: 689-694.
VEMAP, 1995. Vegetation/Ecosystem Modeling and Analysis Project: Comparing biogeography and biogeochemistry models in a continental-scale study of terrestrial ecosystem responses to climate change and CO2 doubling. Global Biogeochemical Cycles 9(4): 407-437.
Woodward, F. I., T. M. Smith and W. R. Emanuel, 1995. A global land primary productivity and phytogeography model. Global Biogeochemical Cycles 9(4): 471-490.
Weinstein, D. A., R. M. Beloin and R. D. Yanai, 1991. Modeling Changes in Red Spruce Carbon Balance and Allocation in Response to Interacting Ozone and Nutrient Stress. Tree Physiology 9: 127-146.
The development of such reduced form models offers computational efficiency gains and requires fewer parameters for larger scale models.
Forward to Next Section//Back to Table of Contents//AGCI Homepage//Comments: agcimail@agci.org