Examining potential impacts of map errors on land change detection.
Nicholas Cuba and Robert Gilmore Pontius Jr
Post-classification comparison of categorical maps from two points in time is a commontechnique to detect land cover and land usechange, yet few practical applications accountfor errors in the maps. Our research presents amethod that calculates the potential forcategory-level classification error to explainobserved map differences. The method isapplied to land cover change detection for theRío Piedras River Watershed, San Juan, PR, toderive minimum thresholds of the randomly-allocated map error intensity that potentiallyexplain each empirically observed categoricaltransition from 1999 to 2003. To investigateeach transition from category i at time 1 tocategory j time 2, we consider four possibleexplanations that relate to four types ofrandomly-allocated, category-level errorintensities: commission and omission errors foreach of the two categories involved. If theminimum error thresholds for a categoricaltransition are greater than 100% then randomlyallocated error cannot completely explain thedifference between the maps. If the minimumerror thresholds for a categorical transition areless than the error suspected to be in the map,then the empirically observed transition couldbe fully explained by randomly allocated maperror and potentially does not represent trueland cover change.