Modelling and Optimizing a Graph-based Settlement Delineation and Analysis Work-flow

Kurzfassung

State-of-the-art settlement pattern analysis relies on satellite imagery, mapping the outlines of residential areas. The high resolution of these images yields multiple disjoint settlements for a single urban area, distorting the analysis of settlement patterns. Therefore, a workflow was developed to aggregate and analyze urban areas modelled as a settlement graph. Clustered settlements are merged according to the edge contraction procedure, and their importance is evaluated with selected centrality measures. Through workload distribution and parallel execution of Common Workflow Language (CWL) tasks with a proprietary scheduler, the workflow can process even the most densely populated regions while respecting workflow stage inter dependencies modelled in a DAG. The procedure is ultimately deployed and executed on the HPC platform “terrabyte”. The workflow’s configuration, especially implicit parameters like the settlement density in the region of interest, significantly impacts the time required to complete the procedure. Concurrent processing is restricted on the level of connected components, leading to an unbalanced workload distribution when processing large urban areas. To combat inefficient resource utilization on the deployment platform, endeavours are made towards a more flexible CWL modelling and elastic resource allocation.