Abstract

Forest and grassland regions in Canada are classified into specific fuel types by scientists for wildfire spread models. These fuel types are considered "hard" classifications as any region is given one type, whereas regions may be mixtures of two or more fuel types. In this talk, we discuss fuel type classification in Canada from the data-driven perspective of probabilistic or "soft" clustering. Fuzzy clustering algorithms, such as the popular fuzzy C-means algorithm, extend hard clustering to allow for a degree of uncertainty in the cluster assignments through a fuzzifying parameter in the objective function. However, challenges remain in selecting an optimal fuzzy parameter, incorporating mixed continuous and categorical (mixed-type) data of forests and grasslands, and balancing variables based on variable importance to clusters. To allow for mixed-type data in fuzzy clustering, we discuss various extensions to the fuzzy C-means algorithm, including a kernel weighting approach in the objective function to allow for nonlinear cluster structures and mitigate the sensitivity of selecting the optimal fuzzy parameter. We investigate the results of applying fuzzy clustering methods for estimating fuel types using an Alberta wildland fuels dataset and compare fuzzy cluster assignments with existing scientific fuel classifications.