Which Model is Right for the Job?
Spatial and Temporal Scales
Model Limitations
Use of Specific Fire Models in CRAFT
Use of Specific Vegetation Change Models in CRAFT

WHICH MODEL IS RIGHT FOR THE JOB?

Forest managers have broad array of modeling tools to choose from, but they have little guidance for when and where to use them. Fire behavior and vegetation change models may not have been designed for the temporal or spatial scales that managers are interested in. For example, in order to address a broad scale fire question, models may generalize the distribution of fire behavior, watershed concerns, or habitat conditions too much to be broadly applicable for interdisciplinary analysis. Effects analyses in CRAFT often require that modeled results can be translated across disciplines.

Before deciding on a fire or vegetation change model, managers must clearly identify their objectives and have a conceptual model of how alternatives will affect their environment. In other words, information needs must precede selection of the model and how it is used. The vegetation model used should also be capable of meaningfully capturing the relevant successional processes of the project area. This step is critical for four reasons:

(1) The processes that govern vegetation dynamics are highly variable among vegetation types, among regions and even across watersheds. Models differ in their ability to simulate these differences in how vegetation changes with disturbance. If fire effects are required output, vegetation models should be flexible enough to output a range of possible conditions under different fuel and weather scenarios.

(2) Vegetation models also differ in their ability to address management questions across spatial and temporal scales. For example, if a management problem requires spatially-explicit answers about wildlife habitat quality, models that predict future burn acreage with no fine spatial sense of relative effects are inappropriate.

(3) The type of successional question that is asked leads to fundamental differences in how belief nets are structured (see discussion of belief network templates). In other words, when a management problem is fundamentally one of vegetation change, the belief network should be built around that question. In other instances, changes in vegetation and fuel may be one of several factors that surround a broader management question of, say, fire risk, and the model may be relatively insensitive to different ways of modeling vegetation.

(4) Models differ in how readily they can be adapted for rigorous probabilistic analysis. If a model's software code is proprietary or obscure, it may be difficult to apply. Therefore, the selection of a model may depend on the technological capabilities of the management team.

Scientists and managers have been modeling vegetation dynamics for decades, but many mainstream models are not capable of considering multiple successional pathways or incorporating varied disturbance effects. Most existing models are much better at modeling simple seral transitions over time, and they have no or only limited capability to capture the more complex dynamics of uneven-aged forests that thrive with low severity disturbance. In most low and mid elevation forests of the West, low severity disturbance can sustain overstory vegetation for an extended period of time. In these forests, it is the lack of frequent fire that has increased levels of hazardous fuel conditions above pre-existing levels. For many forest management questions, successional models must meaningfully capture the effects of often subtle management alternatives.

Back to Top

SPATIAL AND TEMPORAL SCALES

To a large degree, the model that is selected and how it is used depends on the spatial and temporal scale of the management question. For example, fire behavior models are often designed with a very specific scale in mind. Questions that relate to fuel break viability, landscape connectivity or variable fire effects within a landscape may be best addressed using FARSITE or FlamMap. On the other hand, local-scale questions that relate to fire behavior within a stand can be addressed with BehavePlus, FOFEM, Consume or Nexus. Nonetheless, some belief network model structures can allow stand level models to be used to address landscape scale questions.

Among the principles of building belief nets is that belief network models can flow from broad scales to fine scales, but not the reverse. Although stand level models, such as FVS, were designed to address local scale issues, multiple plots from a vegetation type can be run in FVS and analyzed probabilistically for a broad landscape. In many instances, it may be easiest to "down-scale" within a belief network when this approach is used rather than using two fundamentally different vegetation models.

Addressing long term risks can be extremely difficult, as both fire behavior and vegetation dynamics may be contingent on changes that happen at broad spatial scales. There are abundant ways that temporal dynamics can be incorporated into a belief network (see our discussion of belief network templates). Modelers should first decide on what process or management question they are interested in, then determine which vegetation model is appropriate based on the needs of their belief network.

Back to Top

MODEL LIMITATIONS

Existing fire and vegetation change models were not designed to be used for the probability-based computations that are needed in CRAFT. This means that in the near term, managers will need to work within the constraints of what is available.

Most existing models are rigidly deterministic. For example, fire models have been designed to calculate some fire behavior output, such as flame length, rate of spread, or crowning potential, based on a single set of input parameters. This approach masks the inherent uncertainty of the modeling effort and can be overcome through greater use of probability. By calculating a wider range of input parameter values and then considering their probability of occurrence, modelers can perform more meaningful effects analysis.

DecNet is software designed to perform multiple runs of complex belief network models in FVS. Other programs can be modified to accommodate probability-based analysis as well.

Back to Top