Modelling the spatial distribution of DEM error with geographically weighted regression: An experimental study

Digital elevation models are digital representations of topography and are the principal digital data source for many disciplines. The accuracy of a digital elevation model is particularly important with its intended use. The assessment of accuracy is usually undertaken by deriving some measures of indices. These indices are usually global indices and summarise the accuracy. In this study, a more detailed accuracy is discussed by examining the magnitudes and spatial patterning of elevation errors. Many researches have demonstrated the effects of interpolation methods and the nature of errors in digital elevation models for small-scale areas. On the contrary, this study investigated the size and spatial patterning of errors in a large-scale area, by comparing three different interpolation algorithms. This research also examined the accuracy related with data density and surface characteristics at the field scale. Finally, ordinary least squares and geographically weighted regression models were used to model the relationship between error and surface characteristics. Geographically weighted regression modelling gave better results than ordinary least-squares regression modelling.