Correlative model – This type of model uses statistical methods to determine an association between a subject of interest (abundance of a bird species, for example) and a factor that might affect it (such as rainfall or food supply). Though we may suspect a causal link behind the association, the correlative model does not speak to that: its value is to indicate an association and its strength.
Population dynamic model – Here, changes in the size or structure of a population are modeled over time as a function of the survival and reproductive success of individuals. The model can then be used to project future changes in population size with regard to changes in the environment or management practices. This kind of model can also evaluate uncertainty (see below), predicting for instance the probability that a population will dip below a critical size over the next 50 years.
Spatial scale and resolution – Spatial scale generally refers to the size of an area being evaluated. For a bird species, the relevant scale may range from less than a hectare (2.5 acres) for an individual foraging during the day or defending a territory over a breeding season, to an entire continent or hemisphere when considering the species' migratory movements. Spatial resolution is the level of detail that can be inferred or discriminated: a model using information at a meter-square scale has a much higher spatial resolution than one using information at a kilometers-square scale. In general, different ecological phenomena require models at different spatial scales and resolution.
Uncertainty – Scientists and managers often face the kind of uncertainty that derives from environmental unpredictability or from imperfect information about the organism or the biological relationships that affect it. Models help to assess the effects of these sources of uncertainty and determine where the uncertainty lies—whether it is environmental in origin or stems from our lack of knowledge regarding the species.
Validation – In this important phase of model development, scientists test the model's performance by comparing its predictions to already known results. Validating a model that predicts future scenarios presents an obvious challenge, since known results from the future are not yet available. One approach here is to use the model to "predict" the past. While such comparisons can be useful, a model's success in predicting the past is no assurance that it can also predict the futur.