Smoke over the contiguous United States

Two independent measurements of the same thing, kept side by side on purpose, and a paper that reads one against the other.

Folder	Product	Record	Source
noaa/	Smoke-covered area and gridded smoke-days, from NOAA HMS smoke polygons	2006–2025	NOAA NESDIS Hazard Mapping System
modis/	Monthly 25 km AOD and smoke-condition statistics, from MODIS MAIAC	2000-02 → 2025-07	NASA MCD19A2.061
paper/	The manuscript on what the MAIAC record shows, its figures, and the code behind them	2003–2024 window	both of the above

The HMS polygons are hand-drawn by NOAA analysts tracing visible smoke in satellite imagery. The MAIAC fields are remotely sensed values. The two fail in unrelated ways, which is the entire point: analyst practice drifted mid-record (HMS analysts shifted to fewer, continental-scale polygons around 2017–2020, which inflates recent years), and that drift has no counterpart in MAIAC. MAIAC's own biases — cloud screening, bright-surface retrieval failures, snow-covered winters, orbital sampling — have no counterpart in HMS. Where the two agree, the signal is probably robust.

---

noaa/ and modis/ are self-contained — each has its own code, data, figures and README, with paths resolved relative to itself, and neither imports from the other. paper/ is the one thing that spans them: it reads both archives and reports the MAIAC record with HMS as an independent check. Start with:

• noaa/README.md — method, outputs, and the caveats that matter most (density is unusable before 2011; the density classes are nested, not disjoint; polygon area is extremely long-tailed).
• modis/README.md — what the six variables actually mean, why mean_smoke_aod_055 is not smoke-only AOD, and why every ratio must be read next to its sample-size weight.
• paper/README.md — the headline numbers, the three decisions that shape all of them, and what the code computes that the paper no longer reports.

cd noaa  && python3 process_smoke.py && python3 plot_smoke.py
cd modis && python3 maiac/plot_month.py data/maiac/monthly/maiac_smoke_25km_2023_06.nc
cd paper && python3 scripts/make_all.py

Not in this repository. The data trees are outputs rather than source, and are large, so they are not archived here: noaa/data/, modis/data/, and the derived paper/results/. noaa/figures/ is regenerated in a minute from noaa/data/processed/ and is likewise left out; the modis/ and paper/ figures are here. The manuscript itself (paper/maiac_conus_smoke_record*.docx, including the redlined revision) is not published. The practical consequence: a fresh clone gets all the code and the paper's figures, but the first two commands above have to be run — in that order — before the third has anything to read.

Common caveats for both products: smoke aloft is not surface exposure — these are top-down views, so pair with surface PM2.5 for air-quality work — and day boundaries are UTC, so a west-coast evening plume can land on the next day.