Handling uncertainty in GIS and environmental models : An application in forest management

The study of uncertainty in Geographic Information Systems (GIS) and environmental models has received increasing attention in recent years, due in part to the widespread use of GIS for resource management. This study used GIS-based techniques in order to compare several different forest inventory and forest cover datasets. These datasets pertain to an area of the boreal mixedwood forest in northeastern Alberta which covers roughly 73,000 km2, and which has recently been approved for logging. The datasets include two forest inventories based on aerial photographs, and a forest cover classification based on remotely sensed satellite data. Simple logical operations were used to transform the datasets to a form suitable for comparison. Standard GIS overlay techniques were used to compare the agreement among different datasets. Visualization techniques were used to display patterns of agreement in attribute space (contingency tables), and in geographic space (maps of uncertainty). Agreement between the two forest inventories was about 50% (Percent Correctly Classified), with a Kappa value of 0.4, for a classification based on species composition. In general, much of the misclassification was between ecologically similar types, particularly between different combinations of aspen and white spruce. Comparison of the forest inventories with the classified satellite image was done using a simplified land cover classification with five categories. Agreement was about 55% (Percent Correctly Classified), with a Kappa value of 0.3. Possible sources of discrepancy among datasets include change over time, differences in spatial scale, differences in category definitions, positional inaccuracy, boundary effects and misclassification. Analyses were conducted to characterize the effect of each of these sources of disagreement. The agreement was strongly affected by the distance to boundary, indicating a boundary effect extending to more than 100 meters. Differences in spatial scale accounted for a small proportion of discrepancy. None of the other possible sources had a measurable effect on the discrepancy. It was therefore inferred that misclassification accounted for a large proportion of the discrepancy. Estimated levels of uncertainty were propagated through models including simple growth and yield tables and a more complex harvest scheduling model. It was found that uncertainty in model outputs was strongly affected by uncertainty in inventory data, uncertainty in volume yield curves, and perhaps most importantly, by a poor understanding of disturbance and forest dynamics in the region. The results of the analysis show that these uncertainties may have significant economic and ecological implications.Arts, Faculty ofGeography, Department ofGraduat