Assessing the impact of data quality on the results of spatial modeling using sensitivity analysis: A case study using Minnesota forest inventory data

摘要

This thesis examined data quality's impact on a spatial model designed to determine the suitability of a study site for pulpwood management. The goal was to gain a better understanding of data quality's influence on geographic information system (GIS)-based decision-making through the use of geographical sensitivity analysis. The analysis focused on changes in resolution and attribute error.;The resolution sensitivity analysis involved comparisons between the original quadtree level 10 suitability map and suitability maps generated at quadtree levels 9, 8, 7, and 6. Comparison maps indicated that eliminated and misclassified areas increased with increasingly coarser resolution. The 53 operable stands depicted on the level 10 map had a total suitable cut volume of 35,135 cords. The level 9 map did not differ from the level 10 map based on these criteria. However, level 7 had 49, and level 6 had 38 stands. Overall, as quadtree level decreased, the number of stands decreased, in particular, the smaller stands disappeared. The exception was level 8 in which an elongated stand separated and resulted in an overestimate of 54 stands and a total cut volume of 35,219 cords. Comparisons were also made in terms of misclassified areas that indicated cut losses or false decisions to cut stands. The percentage of area indicating cut loss or false cut increased with increasingly coarser resolution.;The attribute error analysis investigated the three management prescription variables of age, site index, and basal area per acre, and two error levels including $pm$25 and $pm$5 percent. The results indicated that differences exist among attributes in terms of their sensitivity to introduced error and the impact on the decision to cut stands. Age and site index attributes were more sensitive to error when compared with basal area per acre. Overall, the age variable displayed the most sensitivity to error based on the percentage of area impacting the cut decision and the frequency of change experienced by stands, particularly at the $pm$25 percent error level. Basal area per acre experienced the least overall frequency of change and impact in terms of the decision to clearcut stands. The explanation for the sensitivity of the age attribute is associated with two major factors: (1) the critical cutoff values of the decision trees, and (2) the predominant pulpwood species in the study site. An examination of the characteristics of each stand identified on the best and worst case maps for the six perturbation sets revealed that these stands were mostly comprised of black spruce, tamarack, and aspen. Studying the decision trees for these cover types revealed not only the cutoff values for age but also that age was used to determine the cutoff values for basal area per acre. Also, the average stand age for the study site was about 52 years while cutoff values for black spruce, tamarack, and aspen ranged between 20 to 80 years.;Several recommendations for improving GIS reliability and decision-making capabilities were discussed including the conceptual design for a visual error assessment interface. The significance of this research lies in its contribution to a better understanding of the data quality's influence on GIS-based decision-making, particularly for forest management applications.