Small settlements coalesce into larger cities

How the OpenMOLE platform and High-Throughput Compute helped to validate a long-held theory in geography

Ever since the first urban centres appeared thousands of years ago, cities are a key feature of human development. They have a life of their own, governed by interactions between people, resources and other settlements. These balances are delicate and can change: what looked like a humble village many centuries ago can now be a huge metropolis, while some of the great cities of the past are now nothing more than ruins.

Understanding the dynamics of cities is key to plan the future. It is predicted that, by the end of the century, 80 percent of the human population will be urban. And for that we need accurate models, able to take into account all the complexities of an ever-changing system. But how good are these models?

Denise Pumain and her team from the ERC Geodivercity have focused on SimpopLocal – a computer model based on the assumption that organisation of cities as a system relies on their spatial interactions framing the diffusion of innovations.

SimpopLocal simulates the evolution of agriculture-based villages under strong environmental constraints that may, or may not, be overcome by technology. The model considers six parameters to account for population growth, resource consumption and the emergence of innovations. Each simulation starts with 100 small settlements with a random number of inhabitants between 38 and 133 and covers the equivalent of 4,000 years.

The problem is that the historical record is not complete. We don’t have precise economic or demographic data covering the last 4,000 years of a city, not even for Rome or Jerusalem. An alternative had to be found , to validate the SimpopLocal model in a non-empiric way.

They chose to simulate the evolution/dynamic of the system by giving 10 different values to each of the six parameters, covering the range of possible scenarios. Then, they hit a bottleneck, because the number of combinations quickly escalated to millions and manual checking became impossible.

The team of Geodivercity used the EGI High-Throughput Compute service to handle the lion share of the work. The compute resources were made available through the complex-systems virtual organisation, supported by 10 federated data centres based in France and Greece. The workflow of the calculations was managed through the OpenMOLE platform.

The team’s strategy was to submit 5,000 jobs in parallel and to use the results to generate the next set of parameter values. In total, they submitted about 200,000 jobs for each of the ten calibrations and consumed around 50 million CPU hours.

The work validates the SimpopLocal model by showing that SimpopLocal is able to produce realistic patterns of gradual hierarchisation of system of cities and concludes that the coalescing small settlements theory is a good framework to understand the development of cities. The team published their findings as part of their 2017 book Urban Dynamics and Simulation Models.

HTC usage

The team completed 10 model calibrations. For each, they submitted about 200,000 jobs of one hour. In total they consumed around 50 million CPU hours provided by the complex-systems virtual organisation.

The complex-systems virtual organisation is supported by 10 EGI federated data centres from France and Greece:

  • GR-01-AUTH
  • GRIF-LAL
  • GRIF-LLR
  • GRIF-LPNHE
  • HG-02-IASA
  • HG-03-AUTH
  • HG-05-FORTH
  • HG-08-Okeanos
  • IN2P3-IRES
  • M3PEC
Reference

D Pumain et al. Urban Dynamics and Simulation Models. Springer. ISBN 978-3-319-46495-4. (abstract)