Client: Philipps-Universität Marburg (Environmental Informatics group and University Hospital), with the Kilimanjaro Clinical Research Institute, Tanzania
Location: Moshi and Siha, Kilimanjaro Region, Tanzania
Year: 2021–2022
Sector: Environmental Epidemiology · Remote Sensing · Public Health
Stepping outside urban consulting entirely — building a satellite-data exposure model to help a medical research team understand what’s driving asthma and disease disparities between an urban and rural district in Tanzania.
A Marburg University medical research team studying asthma and atopy in the Kilimanjaro Region needed to know how much of the difference between an urban site and a rural site came down to environment, not just genetics or healthcare access. That required turning satellite imagery into something a clinical epidemiology team could actually use.
Part of RIPA2TAN (“Risk and Preventive factors for the development of Atopy and Asthma in Kilimanjaro Region, Tanzania”), a clinical cohort study run by Marburg University Hospital with the Kilimanjaro Clinical Research Institute, comparing urban Moshi against rural/semi-urban Siha. SPIN Unit contributed the geospatial environmental-exposure component (work packages 1 and 2) alongside the clinical fieldwork team.
Assembled a genuinely multi-source Earth-observation stack — Copernicus 100m land cover, Sentinel-2 10m land cover, Landsat NDVI, ASTER elevation, WorldClim climate data, SoilGrids, and OpenStreetMap infrastructure — and applied multi-radius spatial buffer analysis (from 30m out to 3km) around household and district locations to compute land-use composition, greenness, and diversity as candidate environmental predictors of disease.
Delivered as a 39-page analysis presentation midway through the broader clinical study, ahead of the main patient data collection phase in autumn 2022.
- Compared two genuinely contrasting sites — urban Moshi (population ~230,000) and rural Siha (population ~142,000) — to isolate environment from other factors in disease-burden differences.
- Already-available health indicators showed stark rural/urban disparities even before the environmental analysis was complete — a strong motivating baseline for the study.
- The team was explicit about its own methodological risks: spurious correlations from an arbitrarily chosen buffer radius, data-quality inconsistencies in the medical dataset, and unknown confounding variables — an unusually transparent approach to a genuinely hard causal-inference problem.
- Demonstrates a reusable spatial-exposure-modelling workflow — hex-grid buffer analysis across multiple satellite sources — applicable well beyond this specific health study.
Gave a clinical epidemiology team a rigorous, multi-source environmental dataset to test against real patient outcomes — a genuine cross-disciplinary application of SPIN Unit’s geospatial methodology outside its usual urban-planning context.