HEALTH IMPACT

Health Impact Assessment Platform

Python · GEE · geospatial

Global trends in urban PM2.5 concentrations and attributable mortality, 2000 to 2019
Global trends in urban PM2.5 concentrations and attributable mortality, 2000 to 2019.Southerland VA, Brauer M, Mohegh A, et al. The Lancet Planetary Health, 2022. CC BY-NC-ND 4.0.

The Health Impact Assessment Platform answers a deceptively hard question: how much disease is attributable to the air people breathe? How many deaths and illnesses trace back to fine particulate matter or nitrogen dioxide, how that burden is distributed across a population, and how much of it would lift under a cleaner scenario?

The science behind that question is well established. The tooling around it is not. The methods have historically lived in desktop software that is hard to install, centered on United States data, or reachable only by people who can write code. If you are a researcher in Mexico City or an analyst at an advocacy organization, the path from “I have exposure data” to “here is the attributable health burden, with a credible uncertainty range” is far longer than it should be.

The problem

A health impact assessment has to bring four things into the same calculation: where pollution concentrations are and where they would be under a counterfactual, how many people are exposed and at what ages, the underlying rates of death and disease, and the epidemiological evidence linking exposure to outcome. Assembling those by hand, for a region that existing tools never prioritized, is slow, error-prone, and difficult for a reviewer to retrace. The result is that the people closest to the most polluted places are often the least served by the analysis tools.

How it works

The platform walks the analyst through the assessment one step at a time: define the study area and pollutant, bring in the exposure surface and an optional cleaner target, specify the exposed population by age group, choose baseline health rates, and select a concentration-response function from the published literature. It then combines them into an attributable burden. Economic valuation is available as a final, optional step.

Methods you can defend

Rigor is the whole point, so the methods are explicit and auditable. Exposure is population-weighted: high-resolution satellite particulate data is combined with gridded population on Google Earth Engine so that a concentration reflects where people actually live, not just the average over a polygon. The platform carries several concentration-response frameworks side by side, from the standard regulatory approach to the Global Burden of Disease splines and the GEMM and Fusion functions, so a result can be tested against more than one model of the science. The data curation, the grid alignment, and the weighting assumptions are written down in runbooks, which means a reviewer can trace every number back to its source.

Built to reach people

The tool runs in a browser and installs entirely in user space, with no administrator rights, no Docker, and no cloud account required to get started, so it works in exactly the lower-resource settings that need it most. It pairs planetary-scale datasets with local boundaries, resolving results down to census tracts, municipios, or districts, so global data answers a neighborhood question. The architecture is cloud-ready by configuration rather than by rewrite, and an embedded assistant guides dataset and method choices at each step, putting the methodological judgment usually locked in a manual directly beside the decision.

Where it stands

The platform has carried real work: population-weighted particulate and nitrogen dioxide analyses across Mexico, fine particulate burden estimates for Amazonian cities in Brazil, national United States estimates against regulatory rollback scenarios, and census-tract environmental justice screening that links exposure to community demographics. Multi-year analysis is built in, so the question is not only how large the burden is today but whether it is improving.

What travels

The patterns underneath are not specific to air quality. Every analysis exports as a configuration a collaborator can re-run, so the work is reproducible rather than a one-off. And the same approach, bringing large-scale data down to a resolution where it answers a local question, applies to any analysis that has to hold up to scrutiny and reach the people who will act on it.

← All work