Who settles where? Simulating urban growth and socioeconomic level using cellular automata and random forest regression

Anahi Molar-Cruz, Lukas D. Pöhler, Thomas Hamacher, Klaus Diepold

Research output: Contribution to journalArticlepeer-review

2 Scopus citations


Cities in developing countries share a pattern of accelerated and largely unplanned urbanization that results in internal socioeconomic inequalities. The modeling of urban growth with spatial distribution of socioeconomic groups has been studied only to a limited extent. This paper proposes a method to simulate both urban growth and the socioeconomic group that is likely to settle in a particular location as a function of local environmental characteristics. Using a cellular automata model, newly urbanized cells are identified and then, as a post-processing step, distributed among five socioeconomic groups in a preferential settlement selection process. A land value map learned with a random forest regressor is used for this purpose. Our case study is Greater Mexico City during the period 1997–2010. We identified that the main features influencing the location of socioeconomic groups are the closest socioeconomic group, the distance to water bodies, and the distance to the urban center. This suggests that the newly urbanized cells are likely to settle in neighborhoods of similar socioeconomic levels. Moreover, the increasing distance from the urban center results in a generally decreasing land value. However, regions with a high land value were also found in remote areas where environmental features that improve the ecosystem services are present.

Original languageEnglish
Pages (from-to)1697-1714
Number of pages18
JournalEnvironment and Planning B: Urban Analytics and City Science
Issue number6
StatePublished - Jul 2022


  • Greater Mexico City
  • Urbanization
  • cellular automata
  • land value map
  • random forest regression


Dive into the research topics of 'Who settles where? Simulating urban growth and socioeconomic level using cellular automata and random forest regression'. Together they form a unique fingerprint.

Cite this