Land use regression vs. satellite downscaling for PM2.5 exposure assignment

An LUR exposure model trained on a sparse monitor network has little local data to anchor to, and most cohort study areas don't sit inside a dense one. Satellite downscaling gets pitched as the fix. It's worth being precise about what each approach is doing before you pick one for a new exposure assignment.
What land use regression estimates
A land use regression model regresses observed PM2.5 at reference monitors against predictor variables, road density, traffic counts, population density, land cover, elevation, sometimes NDVI, then uses that fitted surface to predict concentrations at unmonitored locations, including your participants' residential addresses. It's a workhorse approach for a reason: it's interpretable, it doesn't require remote sensing expertise in-house, and a well-specified model captures the kind of small-area variation that matters for a near-roadway contrast.
The limitation shows up fast in cohort work. LUR needs a training set of monitors dense enough to resolve the gradient you're trying to measure, and most PM2.5 networks are built for regulatory compliance, not epidemiologic contrast. If your cohort spans a rural county with one regulatory monitor eighty kilometers from half your participants, the regression has nothing local to anchor to. Temporally, classic LUR also tends to produce an annual or seasonal average surface rather than a daily one, which is fine for a chronic-exposure design and a real constraint if you're looking at acute effects, hospitalization windows, or trimester-specific prenatal exposure.
What satellite downscaling adds, and what it costs you
Satellite-based exposure assessment starts somewhere different: aerosol optical depth or similar retrievals from a satellite platform, calibrated and downscaled to a finer resolution using ancillary imagery. The appeal for a cohort study is coverage. The retrieval exists over the study area whether or not a monitor is nearby, and because the source imagery comes in on a regular overpass schedule, you can build a daily surface instead of a single static average.
That coverage comes with its own constraints. AOD-to-PM2.5 conversion is sensitive to atmospheric conditions, cloud cover knocks out retrievals on a meaningful fraction of days, and the downscaling step still needs calibration logic to translate a coarse-resolution column measurement into a neighbourhood-scale surface. The satellite product is a modeled estimate built from spectral data and calibrated against whatever ground monitors exist in the region. It is not a direct reading at the participant's doorstep, and the downscaling inherits whatever error sits upstream in that calibration.
Picking between them for a specific cohort
For a cohort distributed across sparse monitoring territory, satellite downscaling usually wins on coverage alone: it gives you a value at every residential address on every valid retrieval day, which LUR structurally cannot do without monitors nearby. For a tightly clustered urban cohort sitting inside a dense regulatory network with strong traffic and land-use covariates already assembled, a well-built LUR model can resolve fine within-city contrast that a downscaled satellite product, bounded by its base resolution before downscaling, may smooth over.
In practice, a lot of exposure assignment protocols end up pragmatic rather than purist: LUR where the monitor density supports it, a satellite-derived surface to fill the gaps, sometimes both as parallel estimates for a sensitivity analysis. What matters for your protocol writeup is that you can say, address by address and day by day, where the number came from and what it's standing in for.
If the gap in your protocol is specifically coverage, participants outside monitor range, a need for daily rather than annual values, a defined study area rather than a patchwork of regional products, that's the problem Air Quality Exposure is built around: daily downscaled retrievals delivered over exactly the boundary you draw for your cohort.
Get in touch if you want to talk through whether a daily downscaled surface fits your exposure assignment plan.