Geospatial Analysis of Informal Settlements in a Large Latin-American City

Occupying a significant share of the urban landscape in Latin America and the Caribbean (LAC), informal settlements are the home to nearly one-quarter of the urban population in this region. Mapping land cover and informal settlement patterns is crucial not only for supporting urban planning efforts but also for developing a better understanding of informal settlement distribution and development. My dissertation integrates remote sensing, machine learning, and spatial analysis to achieve two major goals: (1) to advance mapping techniques for informal settlements and urban land cover from satellite imagery and (2) to explore the linkage between urban settlement patterns and crime incidents. The study area covers the municipality of Rio de Janeiro in Brazil, the third-largest South American city with a complex urban landscape, diverse demography, long-standing development of informal settlements, and crime issues. The city stands out as a challenging mapping case and a special area to explore the nexus between urban layout and crime incidents. The dissertation consists of six chapters. Chapter 1 introduced the background of informal settlement and research design. Chapters 2 to 4 discussed three informal settlement mapping methods with the use of ancillary data fusion, multiple classifier systems (MCS), and deep learning, which could be adopted for different situations in terms of data availability and computational resources. Specifically, Chapter 2 examined the relative merits of a textural metric, road density, slope, and thermal infrared information in relation to the complexity of urban land cover classes. Chapter 3 introduced the design of a concatenate MCS with a training sample refinement mechanism to enhance the urban settlement mapping accuracy without using ancillary data. Chapter 4 proposed a patch-based fully connected neural network (PB-FCN) customized for a limited number of manual-collected training data from medium-resolution satellite imagery. Chapter 5 discussed the relationship between the areal percentage of two urban built-up types and reported violent crimes, robberies, and thefts, and examined whether their relationships would be conditioned by socioeconomic factors, both spatially and aspatially. Finally, Chapter 6 discussed the significance and the broad impact of this research. The major novelties of this dissertation include: 1) revealing the role of thermal information in resolving class ambiguity, 2) examining the impact of training data for the combiner upon the MCS performance, 3) demonstrating the strength of deep learning for remote sensor data mining and addressing some critical issues in designing deep learning models for remote sensor data, and 4) examining crime issues in a Latin-American city combining both structure and environmental factors from a spatial perspective. Overall, this dissertation took the initial steps in data inventory and spatial analysis to address the housing challenges faced by one-third of the urban population. For the discipline of geography, this study improves the understanding of the distribution and coverage of informal settlements spatially and reveals the urban structure of a large Latin-American city. For crime studies, this study provides spatial insights into the link between urban patterns and criminal incidents. Transferrable mapping methods have been developed based on a global dataset for urban settlements and land cover types, which is significant for informal settlement management, resource allocation, and impact analysis of planning policies. Future work can be prospected in examining spatiotemporal patterns of informal settlements across regions and environmental inequalities among different urban settlement types.