Methodology

We developed 3 modelling workflows to produce fine-grained maps of near-surface air temperature, thermal susceptibility, and population distribution.

AIR TEMPERATURE
Air temperature was measured in the settlements during the warm season with low-cost loggers and GNSS receivers, carried by community teams walking planned routes. A temporal correction model was applied to adjust measurements for diurnal decline, using data recorded at a reference station. Predictors such as ECOSTRESS LST, Sentinel-2 indices, and building morphometrics were used in machine-learning models to predict near-surface air temperature across settlements.

Photo: Air temperature survey — Near-surface air temperature measurement campaign, Nairobi. Photo by Ishaïa Odudus

THERMAL SUSCEPTIBILIY
Household micro-surveys on thermal susceptibility were co-designed with community leaders and implemented by settlement-based CBO/NGO teams using a mobile app. A geospatial stratified sampling scheme was developed to ensure representativeness. Survey data were aggregated into a composite Living Environment Thermal Susceptibility Index (LETSI), which was modelled wall-to-wall with Sentinel-2 and geodata covariates using machine-learning models.

POPULATION
Population distribution was estimated using micro-survey data collected alongside the susceptibility data, a bottom-up extrapolation method, and deep learning modelling (ResNet-18) applied to PlanetScope imagery. Integration with LETSI and air temperature enables the estimation of highly exposed and susceptible populations.

Photo: Susceptibility survey — Thermal susceptibility & population survey, Buenos Aires.