Sie befinden Sich nicht im Netzwerk der Universität Paderborn. Der Zugriff auf elektronische Ressourcen ist gegebenenfalls nur via VPN oder Shibboleth (DFN-AAI) möglich. mehr Informationen...
This article presents a method to estimate and to visualize the certainty of land change models as they extrapolate beyond the time interval for which empirical data exist. The method to project the certainty relies on measurements of model performance during a validation run with historic data and on the assumption that the model's accuracy approaches randomness as it predicts farther into the future. A land change model typically predicts each pixel as exactly one category for each year. This article presents a technique to convert those predictions into conditional probabilities. As an example, we use the model Geomod to extrapolate forest change over a century for the Plum Island Ecosystems, which is a Long Term Ecological Research site of the United States' National Science Foundation. Geomod uses calibration information between 1971 and 1985 in order to predict the changes from 1985 to 1999, at which point the validation procedure measures the model's predictive accuracy. Then the model is re-calibrated with information from 1985 to 1999 in order to extrapolate into the future, assuming a business as usual scenario. As time progresses, the expected accuracy approaches 0.5, which is the probability at which the model's prediction is as accurate as a random prediction, since the application involves two categories. The extrapolated accuracy of the prediction for the entire study area in the year 2097 is 68%. The method is designed to work with any number of categories so it can be used with a variety of land change models.[PUBLICATION ABSTRACT]