Becky Kerns , Pacific Northwest Research Station and David W. Peterson , Pacific Northwest Research Station,
Climate change is expected to alter plant species distributions; plant community composition and diversity; vegetation structure (e.g., biomass, leaf area); and ecosystem processing of carbon, nutrients, and water. These vegetation responses to climate change will be the result of many lower-level vegetation responses, including changes in net plant carbon uptake, plant water use, plant growth and biomass allocation, competitive interactions, and responses to disturbances.
The complexity of climatic influences on vegetation and the time it takes for the responses to become clear makes it difficult to project potential vegetation responses to future climatic changes based solely on theory or on laboratory and field experiments. Computer simulation models are often used to integrate theory and experimental results to project vegetation responses to changing CO2 (carbon dioxide) and climate. The ideal vegetation model for use in developing climate change adaptation strategies would simulate the full range of climatic and other environmental conditions under which plant species could establish, grow, reproduce, and persist (1). It would also account for the effects of biotic interactions and disturbances on plant species distributions and plant performance. In projecting vegetation changes over time, an ideal model would not only describe the end result, but also indicate the rates of change, intermediate conditions, and the mechanisms producing change. Such ideal vegetation models do not currently exist, though some models are approaching this ideal (e.g. a coupled BGC – species level model). Instead, most current vegetation models focus on one or more – but not all - aspects of the overall problem. They describe environmental niches, simulate competition, or examine spatial processes while greatly simplifying other aspects of the problem through assumptions, theory-based approximations, or simply purposefully ignoring them in order to focus on the necessary and sufficient parts of the problem in the model. Many types of vegetation models can and have been used to assess potential impacts of climate change on terrestrial vegetation. Many different types of models have been developed, and continue to be developed, to examine how vegetation may change with future climate change, but their utility is often not made clear to decision makers.
Types of Models
Vegetation models are generally either process based and simulate some or many of the underlying physiological processes, ecosystem processes, disturbance processes, and climatic and biotic interactions that drive changes in vegetation, or they are statistically based and quantify relationships between species occurrence data and corresponding environmental descriptors. Admittedly this dichotomy is a simplification, as the lines between models are often blurred, and some models do not fall easily into one category or the other (e.g. state and transition models). However, the dichotomy is useful for broadly understanding the major types of models that are used to examine climate impacts on vegetation.
Ecological process models focus on the “how” and “when” aspects of vegetation and ecosystem responses. Process models are of many different types, such as gap models, biogeochemical models, and dynamic global vegetation models. Gap models are used to examine species interactions and vegetation change at very fine spatial scales (plots the size of an individual canopy gap or small stand of trees) over daily to annual time steps (2). But simulated dynamics over many stands and cells are possible. Biogeochemical models are process-based models that simulate carbon, water, and mineral (nutrient) cycles in terrestrial ecosystems, including forests and rangelands. Biogeochemical models are commonly used to simulate ecosystem net primary productivity and carbon flux and storage, and this is their primary usage in climate change research (3). Dynamic global vegetation models project changes in vegetation properties (e.g., leaf area and phenology) at broad spatial scales (thousands of square kilometers, although more fine-scale projections are now available, e.g., 4) over annual to decadal time steps.
Statistical species distribution models (SDM), also known as niche models or bioclimatic envelope models, are used to describe the range of environmental conditions (niche) under which species occur by quantifying relationships between species occurrence data and corresponding environmental descriptors. The SDMs have arisen out of decades of ecological inquiry and analyses related to species-environment relationships (e.g., 5) and share a common philosophy and theoretical foundation with indirect gradient analysis, in that they attempt to relate observed spatial variations in species frequency to associated underlying environmental gradients. Although correlations between species occurrence and environmental conditions can suggest causal mechanisms, the models themselves are purely descriptive and do not directly include processes or mechanisms, although an approach to doing so has been proposed recently (6).
As noted above, not all models fall neatly into the blurry process vs. statistical dichotomy, and others share components of both. For example landscape models have a spatial component and incorporate aspects of either process model or statistical models or both (7). State-and-transition models are based on transition matrix methods that simulate vegetation dynamics using transition probabilities (deterministic and stochastic) between vegetation states. Yet another category of models are growth and yield models. Growth and yield models are used in forest management to estimate future growth and yield of forest stands based on site characteristics (e.g., site index) and mathematical or statistical representations of tree growth through time (8). Hybrid models that bridge categories are also increasingly being developed.
Options for Management
An often-repeated characterization of models is that they will always be wrong to some degree, but may be useful anyway. In applying vegetation models to inform management adaptations to climate change, it is important to match the spatial and temporal scale of the model with that of the question being asked. It is also important to note that models are developed based on assumptions that are usually reasonable for the original scale and purpose of the model, but may be completely unreasonable when models are applied to a different problem or at a different spatial scale.
There is really no “best” modeling approach – all approaches have different strengths and weaknesses. Therefore using an ensemble approaches is important. Ensemble approaches are commonly used when assessing climate projections with global climate models (GCMs). Such ensemble projections using multiple GCMs and multiple types of vegetation models can be used to find the projections common to multiple scenarios, which may be more likely to be closer to future outcomes (9, 10). In addition, the differences among vegetation model projections given their climatic drivers are themselves useful to frame discussions regarding uncertainty In general, agreement among models with different underlying principles and different conceptual approaches when climate inputs are similar (e.g., same scenarios), or across multiple climate models and multiple ecological models, suggests scientific agreement (if not consensus) on sensitivity of the vegetation response to climate. However, model development, lineages, and model input data are sometimes shared (e.g., sibling models), so it may be important to note that if there is agreement, the models need to be independent in their assumptions and parameterizations for potential reductions in uncertainty to hold (11). For example, process models from different broad categories (e.g. gap model vs. DGVM) tend to differ considerably in their conceptual approaches and development and be truly independent of each other. Therefore agreement among several different types of process models may be potentially more meaningful than agreement among several species distribution models. When models disagree and uncertainty is large, local knowledge, monitoring, observations of historical patterns, and knowledge from other sources may help decrease uncertainty. Even when models agree 1)there is a need for local knowledge, especially when translating the results into management actions, and 2) users must understand that many unknown variables and higher order interactions remain such that true outcomes might be radically different than even those derived from ensemble sets with great agreement.
Models may be best used to 1) alert us to the potential magnitude of the effects of climate change, and provide insight into mechanisms of those effects, 2) inform strategic decision making or land-use planning, rather than tactical stand management decisions, and 3) clarify policy concerns and begin to identify processes that are likely to be important (1). It is important to note that the role of models in this context is not to predict the future, but rather to help manage uncertainty by narrowing the possible range of futures to a subset of plausible futures that pertain directly to vulnerabilities for specific resources and management objectives (11). This is why the term “projection” is often used in reference to simulated model output about the future. A broad and comprehensive approach may be useful where multiple model output is first evaluated, other data are considered (e.g. paleoecological information, plant response studies, etc.), and then more specific relationships and potential outcomes based on these other data, local knowledge, finer-scale conditions, and expert opinion are used to evaluate potential outcomes.