Anthropocene Stage 13 source expansion review
Documentation index · Anthropocene implementation · Enrichment plan · Generation methods
Decision
The source-separated particulate observation atlas remains scientifically useful, including its large polar smoke geometry, but its uneven source coverage should be improved without combining measurements, analyses, forecasts, and attribution into one score. Stage 13 re-evaluated five practical source families.
The recommended order is:
- expand climate-record stations from the GSN subset to full GHCN-Daily;
- add an independently labeled global surface-PM field from OpenAQ’s public archive after provider/unit/coverage QA;
- prototype CAMS or MAIAC as a separate global aerosol analysis/remote-sensing
product, never as
observed-smoke-days; and - use PurpleAir only as a supplemental, explicitly authorized source when a bounded sensor set or provider-arranged bulk extract is available.
No Stage 13 source is silently merged into either checked particulate atlas. The complete-2025 and partial-2026 counts and digests remain reproducible, and each new family must pass a source-specific preparation and promotion review first.
Source findings
| Source | What it can add | Access and scale | Stage 13 classification |
|---|---|---|---|
| NOAA GHCN-Daily | Many more global TMAX, TMIN, and precipitation stations than the current GSN subset | Public station/year files; the complete archive is several GB compressed and requires the existing 30-year/183-day eligibility calculation over much more history | First implementation candidate for climate-record coverage |
| OpenAQ Open Data on AWS | Global provider-reported surface PM2.5 and other pollutants, partitioned by location/year/month/day | Anonymous gzip CSV archive; measurements remain provider-specific and need units, averaging, location, instrument, and duplicate QA | First implementation candidate for a separate air-quality-exposure field |
| CAMS global EAC4 reanalysis | Globally complete PM2.5, black/organic carbon, and aerosol optical-depth estimates | 0.75° gridded analysis, every three hours, 2003 onward; data assimilation combines observations with an atmospheric model | Separate modeled/reanalysis pass, not an observation count |
| NASA MODIS MAIAC MCD19 | Global aerosol optical depth and aerosol-type context, including biomass-burning classification | High-volume gridded remote sensing with cloud/retrieval QA and Earthdata acquisition requirements | Separate satellite aerosol pass after a bounded tile/time design |
| PurpleAir API | Dense low-cost PM2.5 observations where sensors exist | API key required; historical queries are one sensor at a time, and PurpleAir asks large historical users to arrange a bulk request rather than crawl thousands of sensors | Optional supplemental source, not the default global acquisition path |
CAMS GFAS fire emissions is also valuable for global biomass-burning flux and estimated species emissions. It is derived from satellite fire radiative power, however, so it belongs beside FIRMS as a fire/emissions product rather than being relabeled as an analyst-observed smoke plume.
Why OpenAQ precedes PurpleAir
OpenAQ publishes day files for all sensors at a location in an anonymous public object archive. That structure is compatible with bounded, resumable year/location acquisition and makes a global coverage inventory possible before downloading the complete measurement body. A preparer can retain source provider, sensor, unit, averaging period, coordinates, and observation time before deriving any daily threshold.
PurpleAir is still useful, especially where regulatory and OpenAQ monitors are sparse. It is not a good unattended global-history default: official guidance says the historical endpoint handles one sensor per request, API consumption uses account keys/points, and large historical requests should be coordinated with PurpleAir. A production integration therefore needs all of:
- an explicit
PURPLEAIR_API_READ_KEYauthorization boundary; - a pinned sensor manifest or provider-supplied bulk extract;
- attribution and use-term review;
- channel agreement, uptime, humidity/correction, indoor/outdoor, and location QA; and
- an independent metric such as
purpleair-pm25-exceedance-days, never an assertion of smoke attribution.
Stage 15 now provides an exploration-only synthetic interface overlay to test default visibility, 60% opacity, projection behavior, and claim language. Its anchors are explicitly not sensors and contain no PM values. Until the access and QA requirements above are available, PurpleAir observations remain absent rather than being simulated with partial or rate-limit-hostile acquisition.
Promotion gates
Full GHCN-Daily
- pin archive/version, station inventory, and complete SHA-256 manifest;
- rerun the existing quality-flag rejection and 30 prior years × 183 valid days eligibility rules;
- report eligible stations and H3 cells by world region against the GSN baseline;
- retain strict record comparisons and 1991–2020 wet-day p95 semantics; and
- establish a workstation disk/time budget before making it standard.
OpenAQ surface PM
- inventory public locations and PM2.5 sensors before selecting files;
- accept only convertible mass-concentration units and record every rejected unit/provider combination;
- normalize timestamps to the source’s explicit UTC/local semantics without inventing missing time zones;
- deduplicate sensor/provider/location/day while preserving raw provenance;
- publish geographic and reporting-day coverage gates; and
- label the product surface air-quality exposure, not smoke.
CAMS or MAIAC aerosol
- keep analysis/model and satellite-retrieval layers separate;
- select one variable, level, temporal aggregation, version, and QA contract;
- expose missing/cloud/retrieval failure instead of treating it as zero;
- validate multiple world regions and high latitudes; and
- use a new output stem so the legacy HMS smoke layer remains identifiable.
Implementation boundary
This Stage 13 review added no hidden downloader, credential, or generated metric. The later synthetic Stage 15 interface experiment also performs no acquisition. The next bounded implementation should begin with an OpenAQ inventory fixture or a full-GHCN development slice, then add a deterministic preparer and coverage report before any default artifact is changed. PurpleAir should be implemented only after an operator supplies a valid key plus a bounded acquisition scope or bulk-data arrangement.