Dissertation Title: Community-based sensing: Generating Granular Urban Information for Dynamic and Informal Environments

Abstract:

Informal settlements are expected to absorb a large share of future urban growth, yet many of the household-level processes that shape infrastructure access, resource scarcity, and environmental risk remain poorly measured. Conventional sources such as censuses, surveys, administrative records, and remote sensing provide important views of these communities, but they are often infrequent, spatially coarse, expensive to repeat, or unable to observe what happens inside homes. This thesis argues that community-based sensing can complement these methods by combining participatory deployment practices with low-cost, distributed sensing systems that produce granular and context-sensitive data.

The thesis develops and evaluates this approach through Axol, a water-sensing platform designed for informal household water systems. Axol measures stored water quantity, conductivity-based water-quality proxies, and bucket-use activity through low-power sensors connected to a local HomeHub and backend data infrastructure. The platform was developed through iterative laboratory tests, home deployments, institutional rainwater-harvesting systems, and fieldwork in Lomas del Centinela, an informal settlement in the Guadalajara Metropolitan Area. These deployments show both the promise and difficulty of sensing in such environments: devices must contend with humidity, heat, dust, intermittent connectivity, installation variability, user interaction, and substantial missing or noisy data.

The work makes four main contributions. First, it presents the design and field evolution of a low-cost water-sensing platform, including hardware, enclosures, installation workflows, and deployment lessons. Second, it characterizes sensor performance and evaluates data-processing methods for converting raw signals into more reliable household metrics and language-model inputs, including calibration procedures, environmental validation for optical tank sensing, anomaly filtering, neural-network-based consumption-pattern classification, confidence-aware dynamic filtering, and retrieval-augmented prompting. Third, it examines how processed sensor insights can be communicated and acted upon through human-centered interfaces, including dashboard, symbolic, ambient, and character-based interaction paradigms. Fourth, it analyzes how the sensing platform moved beyond individual deployments through institutional rainwater systems, government engagement, and multi-stakeholder coordination around pathways from household data to broader urban action.

Together, these contributions show that sensing infrastructure for informal settlements cannot be treated as a purely technical problem. Reliable insight requires a complete socio-technical pipeline: participatory problem definition, robust field hardware, careful data cleaning, interpretable interfaces, and governance pathways that preserve community control while enabling household action and institutional learning. By focusing on water in Lomas del Centinela, this thesis demonstrates one implementation of community-based sensing and outlines how similar approaches could support more equitable and sustainable resource management in other underserved urban and environmental contexts.

Committee members:

Kent Larson
Principal Research Scientist
MIT Media Lab

Joe Paradiso
Alexander W Dreyfoos (1954) Professor
MIT Media Lab

Gesa Ziemer
Director City Science Lab Hamburg
HafenCity University - UNITAC