Air Quality Exposure

CMAQ output vs satellite-derived PM2.5 for cohort exposure assignment

Industrial smokestack billowing thick white smoke across a clear blue sky.

When a protocol reviewer asks whether to assign exposure from CMAQ output or from satellite-derived PM2.5, that framing skips a step. Chemical transport model output and satellite-derived PM2.5 answer different pieces of the exposure problem, and most cohort studies running a residential exposure model end up pulling from both before a land use regression or spatiotemporal model ever gets fit.

What CMAQ and other CTM output actually give you

CTM output like CMAQ runs on emissions inventories, meteorology, and atmospheric chemistry, and it produces a gridded concentration surface for the whole domain regardless of cloud cover. That's the real advantage over satellite: no missing days from clouds, no gaps over snow, and speciated PM2.5 if your hypothesis touches a specific component like black carbon or sulfate rather than total mass. The tradeoff shows up at the grid cell. Most CMAQ runs used in exposure work sit at 4 to 12 km resolution, sometimes finer with nested domains, but still coarse enough that a whole county of cohort participants can land in the same cell. If the study question is about within-city gradients, a 12 km cell flattens exactly the contrast you're trying to detect.

What satellite-derived PM2.5 adds

Satellite-derived PM2.5, built from AOD retrievals calibrated against ground monitors and meteorological covariates, resolves finer spatial structure than a raw CTM grid, which is why it shows up in nearly every hybrid model published in the exposure literature over the past decade. The catch is temporal coverage. AOD retrievals drop out under cloud, over bright surfaces, and over snow, so a pure satellite product has gap days a CTM surface doesn't. For a cohort with daily or sub-daily exposure windows, like perinatal studies assigning trimester-specific exposure, those gaps matter at the individual level, not just in aggregate.

Hybrid exposure models, and what to decide before picking one

The hybrid exposure model designs that dominate the current literature combine satellite AOD, CTM output, land use variables, and monitor data in some form of ensemble or mixed-effects framework, producing a single daily surface without CMAQ's coarse grid or satellite's cloud gaps standing alone. CTM output serves as the gap-filler and speciation layer. Satellite AOD serves as the spatial contrast layer. For a methods section, the defensible framing is that the exposure assignment gets built from both, not chosen between them.

Before locking an assignment method into a protocol, it helps to write down what the cohort actually needs: speciation or total mass, daily resolution or monthly averages, and how much within-city spatial contrast the hypothesis depends on. A study of long-term cardiovascular outcomes with annual averages tolerates CTM's coarser grid reasonably well. A study of acute respiratory events in children does not, where one cloud-gap day or one flattened grid cell can shift a participant from exposed to unexposed in the final dataset.

Where this gets practical is the data pull itself. A team building a cohort exposure model doesn't need a CTM output run for someone else's purposes, trimmed down to fit. It needs a daily retrieval over the exact geocodes in the cohort, at a resolution fine enough to separate participants a few blocks apart, without a gap structure that forces an imputation step before the file reaches the biostatistician. That's the gap Air Quality Exposure is built to close: a daily satellite retrieval downscaled to neighbourhood resolution, delivered over the study area you define, so the exposure input already matches the spatial scale your hypothesis needs instead of collapsing to the nearest CTM grid cell or the nearest fixed monitor.

If you're drafting the exposure assignment section of a protocol and want to see what that daily, cohort-area-specific file looks like first, it's worth a look before you commit to a method.

Start a pilot

← Back to the blog