Analysing the spatial context of the altimetric error pattern of a digital elevation model using multiscale geographically weighted regression

Many freely available Digital Elevation Models (DEM) have increasingly been used worldwide due to the difficulty in acquiring accurate elevation data in some regions, emphasizing the need to investigate their accuracy and the factors that may influence their uncertainties. We performed an accuracy analysis of the Topodata DEM in the hydrographic region of Uruguay (Brazil) assuming that its vertical accuracy may be related to terrain characteristics. Multiscale Geographically Weighted Regression (MGWR) was applied to investigate the spatial scales over which terrain characteristics affect local variations in altimetric errors. MGWR outperformed Ordinary Least Squares (OLS) and Geographically Weighted Regression (GWR). MGWR results also showed that aspect, curvature, and artificial areas operate at much smaller scales than elevation and have a higher influence in areas with high positive altimetric errors. The model explains about 41% of the total variation of the altimetric error of the Topodata DEM in the study area. Our findings enrich the understanding of the global and local processes affecting the accuracy of the Topodata DEM and shed light on the importance of local terrain characteristics in effective DEM product development. HIGHLIGHTS DEM products provide fundamental information for several research areas. OLS, GWR and MGWR were applied to identify the factors explaining the altimetric error of a DEM. MGWR investigated the spatial scales over which terrain characteristics affect local variations in altimetric errors. MGWR outperformed OLS and GWR proving that terrain characteristics operate at different scales.