Uncertainty Propagation in Predictive Modelling of Land Use Change - A Case Study

摘要

A spatial uncertainty propagation analysis was performed to evaluate the effect of uncertainties associated with input data on predictions of land use change made by the CLUE-S model (Conversion of Land Use and its Effects at Small regional extent). The case study concerned the rapidly urbanising Kuala Lumpur region in Malaysia. Elevation, slope and accessibility of road network, sawmills, important towns and secondary towns were identified as important uncertain inputs. Uncertainties in the Digital Elevation Model (DEM) were modelled using a geostatistical error model. Uncertain accessibilities were modelled assuming equal travel speed for all road segments of the same type in the road network. Uncertainty in travel speed was assumed to be normally distributed. Next, using a Monte Carlo simulation approach, 100 input map realisations were drawn from their respective distributions and fed into the CLUE-S model to predict land use and its associated uncertainty over a nine-year period. The distribution of model results was compared with a validation land use map and CLUE-S results obtained using error-ignored inputs. The analysis involved Shannon's information entropy and confusion matrices. Overall, the propagated uncertainty in the predicted land use was small. Elevation uncertainty caused substantial uncertainty in only a small part of the study area. Elsewhere, the effects of DEM uncertainty were limited and the results were similar to the model result obtained with error-ignored DEM. Uncertain accessibilities affected a larger part of the study area. A comparison of the CLUE-S model results with the validation map indicated that there are other important sources of uncertainty. Further analysis of model settings and model parameters is therefore needed to gain a better understanding of the uncertainties involved in predicting land use change in the Kuala Lumpur region.