Predicting time-since-fire from forest inventory data in Saskatchewan, Canada

Time-since-fire data are used to describe wildfire disturbances, the major disturbance type in the Boreal forest, over a landscape. These data can be used to calculate various parameters about wildfire disturbances, such as size, shape and severity. Collecting time-since-fire data is expensive and time consuming; the ability to derive it from existing forest inventory data would result in availability of fire data over larger areas. The objective of this thesis was to explore the use of forest inventory information for the prediction of time-since-fire data in the mixedwood boreal forests of Saskatchewan. Regression models were used to predict time-since-fire from forest inventory variables for each inventory polygon with a stand age. Non-water polygons with no stand age value were assigned values from neighbouring polygons, after splitting long polygons that potentially crossed many historic fire boundaries. This procedure filled gaps that prevented polygons from being grouped together in latter analysis. The predicted time-since-fire ages were used to generate wildfire parameters such as age-class distributions and fire cycle. Three methods were examined to group forest inventory polygons together to predict fire event polygons: simple partitions, hierarchical clustering, and spatially constrained clustering. The predicted fire event polygons were used to generate polygon size distribution wildfire metrics. I found that there was a relationship between time-since-fire and forest inventory variables at this study site, although the relationship was not strong. As expected, the strongest relationship was between the age of trees in a stand as indicated by the inventory and the time-since-fire. This relationship was moderately improved by including tree species composition, harvest modification value, and the ages of the surrounding polygons. Assigning no-age polygons neighbouring values and grouping the forest inventory polygons improved the predicted time-since-fire results when compared spatially to the observed time-since-fire data. However, a satisfactory method of comparing polygon shapes was not found, and the map outputs were highly dependent on the grouping method and parameters used. Overall it was found that forest inventory data did not have sufficient detail and accuracy to be used to derive high quality time-since-fire information.Forestry, Faculty ofGraduat