A network approach for evaluating and communicating forest change models.
Knowledge of forest change is often formalized in state-and-transition models (STMs). These models generate forecasts of forest condition that are widely used for forest management planning. Common techniques for evaluating such models are complex, requiring specialized skills not available to non-modellers. Consequently, model transparency can be limited, hampering collaborative resource modelling that otherwise may increase the chances of management success. We demonstrate evaluation of STMs through network visualization that produces intuitively accessible results, comparable to results of more commonly applied, complex techniques. To evaluate this approach, we statistically test model similarities with empirical data. As examples, we use STMs of forest change, alternately parameterized with information from experts and literature, and compare them to our empirical reference information. Graph theoretical analyses revealed differences in structure and dynamics between alternate STMs. For example, compared to empirical STMs, expert STMs were less complex while literature STMs were more complex. Overall, expert STMs were less similar to empirical STMs than were literature STMs, suggesting information in the expert STMs may deviate more strongly from empirical reference data. We used several techniques that provided complementary information, which produced a comprehensive view of network similarity. We speculate that differences between expert and empirical STMs result from lower complexity of mental models compared to empirical data. While we illustrated our approach using a simple matrix model, it could be adapted for more complex STMs. Improvements of the proposed approach could involve representation of forest change rates with waiting times depicted by multigraphs. Synthesis and applications. Common evaluations of forest change STMs involve complex techniques not easily accessible to non-modellers. Approaching such models as networks makes their evaluation and statistical testing intuitively accessible to many audiences. Benefits of this approach to modellers include improved communication about models with non-modellers, while benefits to stakeholders and decision makers include enhanced understanding of models. This may aid a collaborative resource modelling process and should improve the chances of successful resource management plan implementation. While we used an example from boreal forests, our approach could be applied to many other vegetation types globally.