Population Mapping Using Satellite Imagery and Machine Learning: A Case Study of Mexico City

Abstract

This thesis presents a comprehensive methodology for identifying and analyzing urban builtup areas (CA) and constructed housing areas (CHA) using satellite imagery and advanced machine learning techniques. The study is divided into two phases: Phase 1 focuses on detecting
CA using a Random Forest (RF) model, while Phase 2 aims at identifying CHA with a deep
learning model based on the InceptionV3 architecture. The methodology was applied to a case
study in Mexico City (CDMX), leveraging high-resolution satellite images and ancillary data.
In Phase 1, the RF model demonstrated robust performance with high accuracy, precision,
recall, and F1-score in identifying built-up areas. However, the resulting population distribution, based on CA, revealed a median percentage error of 45.67, suggesting the need for further
refinement in distinguishing residential from non-residential areas. Phase 2 improved upon
these results by employing the InceptionV3 model to identify CHA, leading to a more accurate population distribution with a median percentage error of 36.31. This phase successfully
addressed the limitations observed in Phase 1, providing a more precise allocation of population within urban areas.
The thesis also highlights the practical application of the proposed methodology in identifying irregular settlements in CDMX, which are often challenging to detect through conventional methods. The findings underscore the methodology’s potential in supporting datadriven urban planning and population density analysis.
Key contributions of this research include the development of a robust workflow for urban
analysis using satellite imagery and the exploration of future work avenues, such as automating
labeling processes and estimating CHA density without relying on proprietary building data.
This work demonstrates the value of integrating machine learning with geospatial data to enhance urban analytics, offering significant implications for urban planners, policymakers, and
researchers in the field of sustainable urban development.
Keywords: UrbanAnalytics,Machine Learning, Satellite Imagery, Population Density, Constructed Housing Areas (CHA)

Date of Award	2024
Original language	English (American)
Supervisor	Stephan Dreiseitl (Supervisor) & Francesco Calimeri (Supervisor)