Blog‎ > ‎

Challenges to Forest Landscape Modeling Under Climate Change: Model Interpretation

posted Dec 27, 2013, 3:16 PM by   [ updated Dec 27, 2013, 3:19 PM ]
What are the specific challenges inherent to interpreting and communicating results from forest and climate change simulations?

First, we need to proactively avoid model misuse (‘do no harm’) by clearly defining the purpose for each model and application and reminding the public and policy-makers that model purpose dictates structure, which in turn dictates the range of applications. Each model application should be ascribed a guide as to how the information gained should be used and how it should not be used. We must clearly convey the purpose and limitations of each model and each model application. For example, does the model application provide operational forecasting of landscape change (predictions)? Should the model results be used in conjunction with scenarios to provide decision support (projections)? Or, at the other end of the spectrum, is the model most useful for clarifying relationships? These relationships may be among processes (be they ecological or primarily human-driven) or among scales - relationships change when examined at different extents or durations. These challenges are not unique to simulation models. As an analogy, statistical models can be optimized to maximize predictive power or towards understanding the relative importance among several independent variables - but they cannot be optimised for both purposes!
Effects of climate change, fire, and bark beetles of forest C in the Lake Tahoe Basin

We also need to more clearly communicate the uncertainty in our results. We often fall into the trap of recycling graphs from our publications - a set of boxplots or a time series with 95/5 confidence intervals. But these depictions of uncertainty are generally not accessible to the public or policymakers. We are already asking them
to absorb enormous amounts of complex information. To follow that up with dense graphics and (generally inappropriate) descriptive statistics of the results is a good way to lose half your audience. Over the last 10 years, my experience suggests that the following approaches have merit: 1) Animations of data, pairing time series with map data: Time series data alone can be too abstract and animated maps too pixelated. The combination of time series and map - using a coordinated color scheme - appeals to a broad audience. See Matthew Duveneck's research for an excellent example. 2) Create time series with colored envelopes showing percentiles. For example, a light blue shaded envelope the encompasses the 95/5 percentiles relays the same general information. Avoid confidence intervals! Simulation results rarely follow normal distributions and the concept of ‘sample’ is flawed if you have the capacity to radically increase your sample size by running more simulations (while enjoying your favorite hike, ideally). See the figure. 3) Don’t ignore spatial uncertainty. Local land managers often mis-interpret the results for areas for which they are intimately familiar. You can pair a ‘sample’ map - drawn from a single replicate - with a map showing relative uncertainty among scenarios. Or you can add spatial uncertainty as a third dimension in a contour map. Either approach reveals hidden information (What areas are most resilient to climate change?) and emphasizes the stochastic underpinnings of simulation models. In general, we need to invest much more into understanding how best to depict uncertainty and the tools to rapidly generate such depictions.

Finally, we need to overcome the divide between ”believers” and ”opponents” regarding models. Modeling is not a specialty that a few computer enthusiasts are doing, but modeling (in the sense of abstraction and simplification) is a necessary component of the scientific process; doing so is even part of everyday life, as we constantly develop mental models that are a simplified picture of reality. In this sense, the development, implementation and use of forest landscape models using computers is just a special case of a very general phenomenon. We are all modelers. Nevertheless, more education is needed, particularly at the undergraduate level before students are indoctrinated into a particular viewpoint towards models (either frivolous distractions or miraculous soothsayers). Fortunately, in my experience, models are rapidly becoming broadly accepted as critical tools in the toolbox. Any lingering mistrust is our fault for not better communicating purpose and uncertainty.

- Robert Scheller