Extraction of 3D building information from DSM and use for population disaggregation

Kurzfassung

The lack of 3D building data has been a limiting factor in the important domain of population mapping. Several studies already confirmed the benefits of using 3D data over 2D data in population estimation. Lidar is the most reliable source of 3D data. However, because of high associated costs, its availability is limited. In such cases remote sensing offers a choice alternative. In this study, the feasibility of using building height model derived from Pleiades DSM for disaggregating population to fine scales in Kinshasa, DRC is tested. First, two DTM filtering algorithms, multi directional slope dependent (MSD) filter and geodesic morphological reconstruction, used to derive nDSM are compared regarding their ability to extract buildings. It was found out that MSD extracted DTM with RMSE of 0.132 compared to 0.394 of geodesic morphological construction. Although, the absolute accuracy of the extracted buildings could not be determined due to lack of reference data, building heights extracted from geodesic morphological reconstruction were consistently lower compared to those extracted by MSD. Second, the suitability of coarse resolution DSM to extract building heights was assessed on Pleiades DSM down sampled to 1m, 5m and 10m spatial resolution in addition to 30m ALOS DSM. There was consistent increase in DTM RMSE with reducing spatial resolution. Finally, the extracted building heights were used for fine scale disaggregation in a simple volume-based weighting approach. The best-case scenario (using 0.5m dataset) achieved a total relative estimation error of 24.17 % which deteriorated to 28.01 % with the use of 10m dataset.