Applying Machine Learning to DEM Raster Images

Abstract

Geospatial data analysis can be improved by using data-driven algorithms and techniques from the machine learning field. The aim of our research is to discover interrelationships among topographical data to support the decision-making process. In this paper, we extracted topographical geospatial data from digital elevation model (DEM) raster images, and we discovered hidden patterns among this data based on the K-means clustering algorithm, to uncover relationships and find clusters of elevation values for the area of Jordan. We introduce a method for querying and clustering geospatial data and we built an interactive map accordingly. The method discovers hidden patterns and uncovers relationships in given large datasets. We demonstrate the applicability of the method using the Jordan map and we report on geospatial data analysis and retrieval improvements. The results show that the optimal decision is in favor of four clusters (classes). The first class includes the high elevation values, the second class includes the very low elevation values, the third class includes the medium-high elevation values, and the fourth class includes the very high elevation values.

Follow the shade: detection of optimally shaded pedestrian paths within the historic city center of Thessaloniki

“…Even if an urban DSM is available, several post-processing steps may be required to remove artifacts and extract building footprints, vegetation coverage and terrain features. Although there is a significant effort to automate this process with Machine Learning techniques [28], there is also a simpler alternative, which is to generate the required raster data from vector geospatial datasets. This approach is easily facilitated through QGIS' native tools and UMEP's dedicated pre-processing tools, such as the DSM and Tree generators.…”