Mapping Singapore by Pléiades Stereo Data: Carbon Reporting and more
Publication from Digital
36th EARSeL Symposium, 20-24 June 2016 University of Bonn, Center for Remote Sensing of Land Surfaces \& Department of Geography, Germany , 1/2016
Singapore launched a project for the carbon accounting covering both the historic development over the last 20 years and also the development of a system for mapping and reporting in the upcoming years. In the frame of this project, land use, land use change and forestry (LULUCF) is one important issue, which is covered by remote sensing technology. While the historic mapping is performed using SPOT and LANDSAT data, the hereby presented current and future mapping is based on Pléiades stereo data.
From such Pléiades stereo pairs, several products are generated: digital surface models (DSMs), digital terrain models (DTMs), normalized digital surface models (nDSMs) and pansharpened true orthophotos. These products serve as input for land use change mapping. Change mapping is especially challenging since in urban regions skyscrapers tilt to different directions in the stereo images obscuring the neighbouring areas. Thus, for the DSM generation, a two-step interpolation scheme was developed that fill holes in the matching result. First, small areas are filled by an area-based linear interpolation. Second, larger gaps that most frequently occur due to occlusions are filled towards the ground height level. Thus, instead of artificially introducing smooth transitions, the novel method yield sharp edges in the matching result and therefore also in the final DSM. For gathering a DTM from a given DSM an improved method has been developed by the authors  and has been applied on the Pléiades data. This method proposes a simple filtering that can be applied to DSMs in order to extract DTMs. From both DSM and DTM, an nDSM is generated including the heights of buildings and vegetation. While building heights are irrelevant for carbon accounting, the height of the vegetation gives an important parameter of the carbon content of the vegetation. In the true orthophoto generation the previously extracted DSM is used within the ortho-rectification step together with a direct rectification procedure. In contrast to the standard indirect ortho-rectification method this processing does not yield duplication of elevated structures (e.g. buildings). This ensures that tall buildings are correctly located and not tilted and occluded areas are specifically marked as nodata (void) areas. This processing is done for both Pléiades stereo images, so occluded areas in one image can be filled by using the other. The same procedure is also applied for clouded areas. Then, the nDSM together with the spectral information of the true orthophoto are the basis for the classification of land cover and then follow-up land use categories. These categories are ultimately combined with a set of permanent carbon sample plots, which are measured in the field, to estimate the carbon stock and carbon emissions from LULUCF for the whole area of Singapore.