Global and local patterns of landscape change accuracy

Abstract

It is well known that land cover errors are typically spatially and temporally auto-correlated and can have a distinct spatial distribution. There may be serious consequences in land cover change analysis when the pattern and degree of spatial variation of change accuracy is not understood. This paper explores accuracy measures extracted from the transition matrices for a rural and urban setting. Intensity analysis is used to describe the minimum hypothetical error in imperfect land cover change maps at landscape level. Total land cover change is partitioned into quantity, exchange and shift disagreement based on size and intensity. Multiple resolution analysis quantifies the distances over which exchange and shift disagreement can be measured. Local, geographically weighted transition matrices are constructed to generate spatially explicit measures of change and error. The rural study area exhibited 21% change for each of two transition periods with hypothetical map error from gains (losses) amounting to 9% (5%) for the first and 8% (5%) for the second transition. Thirty two percent change was measured over a single transition in the urban setting. Hypothetical error from gain and loss was found to be 15% and 12% respectively. Allocation disagreement, based on the sum of exchange and shift components, comprised more than 50% of the measured change, predominantly due to high gain commission and omission errors involving a large dormant class. Spatially explicit allocation disagreement from the urban geographically weighted change budget modelled 18% allocation disagreement with probability higher than 0.6, in agreement with the global figure of 20%. Local disagreement in the rural catchment with a probability of greater than 0.6 agreed with the global statistics derived. Local allocation disagreement with a probability of 0.9 and higher only occurred in 1% of the study areas. It was found that at coarser resolutions, quantity disagreement provides the total change in the landscape. The study highlights the potential of local transition measures to complement global statistics extracted from the transition matrix in determining allocation as well as map error in land cover maps.