Spatial Uncertainty Assessment in the Reconstruction of Presettlement Forest Patterns in Western NY, USA

A spatially continuous representation of presettlement forest patterns has broad applications, including local and landscape-scale vegetation reconstruction, landscape change, and analysis of effects of human impact on land use change. Typically, presettlement vegetation patterns are reconstructed using historical survey data in the form of Geographical Information Systems (GIS) layers using a variety methods, i.e., environmental modeling and spatial interpolation techniques [Wang and Larsen, 2006]. Given that historical data processing involves aggregation of spatial data obtained from multiple sources and types, landscape pattern reconstruction is inevitably accompanied by uncertainty originating from data transformation and representation errors. It is therefore essential to provide a model of uncertainty associated with predicted values at locations where survey data are unavailable, as well as to evaluate the impact of such uncertainty on the consequent analysis and classification results. Despite the increased interest in characterizing and quantifying uncertainty originating from aggregating spatially explicit information for landscape reconstruction, there is limited literature regarding uncertainty analysis on reconstructed surfaces. Even those limited efforts have focused only on various aspects of classification uncertainty; for example, Bolliger and Mladenoff [2005] proposed a method to quantify spatial classification uncertainty using fuzzy approaches whose main focus was on a measure of uncertainty associated with classification procedures, while taking for granted uncertainty arising in the data preprocessing procedure. In the current paper, we propose a geostatistical framework to quantify uncertainty associated with a set of tree species occurrence probability at a set of unsurveyed locations, and assess the impact of the uncertainty associated with the reconstructed map of presettlement forest patterns on the overall uncertainty regarding the classification results.