The Ipswich watershed in northeastern Massachusetts, USA, is experiencing important land-use changes, which are contributing to severe environmental problems such as eutrophication, ground water depletion and loss of wildlife. The objective of this paper is to model deforestation between 1971, 1985 and 1991 in the watershed of the Ipswich River in Massachusetts, USA, where most of the forest loss is attributable to new residential development. The maps of suitability for deforestation are calibrated with maps of real change between 1971 and 1985 by using logistic regression, multi-criteria analysis and spatial filters. The maps of 1971 and 1985 serve also as the basis to extrapolate the quantity of predicted future deforestation. Then, the calibrated suitability maps and extrapolated quantities predict the location of deforestation between 1985 and 1991. The predicted deforestation maps are validated with the map of real forest loss of 1985-1991. relative operating characteristic (ROC) and variations of the Kappa index of agreement (Kno, Klocation and Kquantity) measure the validation. For most simulation runs, Kno = 93%, Klocation = 8% and Kquantity = 100%. The best predictor of quantity of deforestation from 1985 to 1991 is linear extrapolation forward in time of the deforestation that occurred from 1971 to 1985. It is difficult to predict the exact locations of deforestation in the watershed because only 2% of the watershed is deforested from 1971 to 1991, the patches of deforestation are scattered evenly across the landscape, and the some of the most important variables are not readily available in digital form. Nevertheless, the best predictor of location of deforestation (ROC = 70%) is a suitability map that uses a spatial filter and multi-criteria evaluation of elevation, slope, and proximity to existing residential areas. The locations that are most threatened are those that are unprotected, near existing residential development and in towns where the demand for new residential development is high. (C) 2001 Elsevier Science B.V. All rights reserved.
