PS 43-154 - Optimizing a fuzzy logic model of forest resilience in the Sierra Nevada, California

Wednesday, August 9, 2017
Exhibit Hall, Oregon Convention Center
Tim Sheehan1, Heather L. Romsos1 and Wayne D. Spencer2, (1)Conservation Biology Institute, Corvallis, OR, (2)Conservation Biology Institute, San Diego, CA

Mixed conifer forests of the Sierra Nevada are home to rare species, including the California spotted owl and Pacific fisher. Past fire suppression, forest management, and climate change have caused forest stress and increased risk of severe fire. Recent severe drought has resulted in a massive tree die off in many areas. Resilience in Sierra Nevada forests is important for maintaining habitat for rare species and ecosystem services.

Fuzzy logic modeling using the Environmental Evaluation Modeling System (EEMS) has proven valuable in evaluating wildland conditions and guiding management decisions. A fuzzy logic model consists of a logic tree whose structure is guided by variable interactions and informed by expert opinion. Genetic algorithms use genetic and Darwinian principles -- testing, survival of the fittest, recombining aspects of candidate solutions -- to find an optimal solution. We have developed an optimization method using a genetic algorithm to optimize fuzzy logic model operators and parameters in a model with a specified structure. We applied this method to several proposed model structures to determine if it improved model results compared to a model informed by expert opinion and whether this method can be used to discern regional changes in the relative influence of variables.


We developed a genetic algorithm software framework to optimize fuzzy logic operators and parameters for logic models constructed using EEMS for forest resilience in the Sierra Nevada. We found that for each candidate model structure, the genetic algorithm was able to improve model performance and identify the model structure producing the best results. This algorithm resulted in some variation in logic operators and parameters compared with the original model. In some cases, small changes to the optimal model produced a simpler model with little impact to the accuracy of the model.

We conclude genetic algorithms are a viable technique for optimizing fuzzy logic models.