Land transition estimates from erroneous maps

Abstract

If a scientist overlays two perfectly accurate maps of land categories of the same place from two points in time, then the differences indicate land change. A land transition matrix summarizes the amount of land that changes from each category at the initial time to each category at the subsequent time. This article proposes methods to compute the land transition matrix in a manner that accounts for errors in the maps, where confusion matrices estimate those errors. If empirical confusion matrices are not available, sensitivity analysis can show the effect of possible errors. The proposed methods produce maps that show the probability of any land transition, given the maps and their confusion matrices. Additional techniques show how possible errors in the maps influence the total land change in terms of two components of quantity and allocation. This article illustrates the methods using data from 1971 to 1999 in Massachusetts, USA.