Anthropocene Stage 8b enrichment plan

Documentation index · Implemented Stage 8 notes · Generation guide · Snapshot assets

Status and recommendation

Stage 8b is now an incremental implementation with two accepted standard families: complete-2025/partial-2026 particulate observation atlases and complete-2025/partial-2026 NOAA CPC temperature fields. All four pass IDs are year-bearing and default-generated. The implementation also provides explicit global H3 coverage/missing semantics and a hard global-FIRMS refresh gate. CAMS and ocean products remain planned. PurpleAir now has a credential-free, synthetic interface experiment; a real community-sensor adapter remains planned and separately permission-gated.

The completed and planned sequence is:

1. Implemented: require global NASA FIRMS fire input instead of silently publishing a regional-only refresh; source-date and world-region audits prevent promotion without actual global rows. A checked global fire snapshot still requires a maintainer-provided free map key.
2. Implemented: add NOAA CPC gridded daily maximum and minimum temperature as the primary non-sparse temperature field while retaining the year-qualified particulate observation atlas as a separate product.
3. Planned: add globally complete CAMS analysis fields for modeled PM2.5 and smoke context without relabeling either one as a ground observation;
4. Planned: add NOAA OISST marine-heatwave and Coral Reef Watch reef-heat-stress products as a separate ocean theme; and
5. Interface implemented; observed data planned: keep PurpleAir in a separately generated, permission-gated community-sensor product. The current synthetic overlay tests style and lifecycle only.

The pass should not solve sparse observations by treating missing cells as zero or by blending station, model, satellite, and community-sensor values into one score. It should show a globally complete field tier underneath source-separated observations.

Why the current map is North-America-heavy

The checked anthropocene-particulate-2026.geojson and the rendered anthropocene-particulate-2026 plate are behaving consistently with their inputs. The imbalance is an acquisition and evidence-class problem, not a projection or marker-layout bug.

There is also a time-boundary issue. The image is titled 2026 partial year. Fires that occurred in Siberia, Australia, or elsewhere during 2025 cannot appear in a calendar-2026 product even after geographic coverage is repaired. The 2025 and 2026 products must therefore be generated separately; prior-year events must not be inserted into the 2026 snapshot.

Evidence tiers and products

Stage 8b should make the evidence class visible in the schema, layer IDs, legend, and filenames:

A single overloaded PNG cannot legibly encode several full global fields. Treat one year’s Anthropocene pass as a bundle of source-separated themes:

• anthropocene-particulate-YEAR-PROJECTION retains and expands the existing event/station atlas;
• anthropocene-temperature-YEAR-PROJECTION combines the non-sparse temperature field with station markers;
• anthropocene-fire-air-YEAR-PROJECTION combines global fire detections, modeled smoke context, modeled PM2.5, and distinct regulatory monitors;
• anthropocene-ocean-YEAR-PROJECTION contains marine heatwaves and reef-only bleaching heat stress; and
• anthropocene-purpleair-YEAR-PROJECTION is an optional community-sensor variant of the air-quality theme.

The SVG can retain independently toggleable semantic groups. Each default PNG should show only the theme named in its filename. This preserves the current no-composite-score boundary.

Recommended sources

Global core

NASA FIRMS is the direct fix for missing Russian, Australian, African, Asian, and South American fire detections. Use one canonical FIRMS VIIRS union for the metric and demote CWFIS to Canadian QA/fallback so the same satellite fire is not presented as two independent observations. Acquire standard-processed records when available and near-real-time records only for the unresolved tail; record sensor, processing class, collection version, acquisition time, and raw hash. NASA Earth-science data are generally open, with dataset citation strongly encouraged; see NASA’s data-use and citation guidance.

No equivalently mature, downloadable global analyst-smoke-polygon feed was identified. The honest replacement for blank global smoke coverage is a separate CAMS modeled plume field, with HMS retained as North American analyst observations. Surface PM2.5, vertically integrated smoke burden, and fire-source emissions remain three different metrics.

Regional observation enrichment

The global core prevents blank regions. Regional feeds can then improve the observation tier without becoming hard dependencies for worldwide coverage.

E-OBS, CMA, and AMeDAS are valuable, but stitching national station networks together is not the non-sparse solution. Station identifiers, duplicates, local-day definitions, quality flags, and reuse terms differ. CPC supplies one globally consistent field while those sources validate or enrich individual regions.

Non-sparse global-field option

Use a second normalized dataset rather than injecting modeled values into the existing positive-observation GeoJSON:

anthropocene-observations-2025.geojson
anthropocene-fields-2025.geojson
anthropocene-observations-2026.geojson
anthropocene-fields-2026.geojson

The field dataset should use H3 resolution 3 by default. Its cell scale is a reasonable match for CPC’s 0.5 degree, CAMS’s 0.4 degree, and OISST’s 0.25 degree grids, gives a nearly uniform-area display, and avoids producing hundreds of thousands of tiny SVG elements. Resolution remains profile-configurable.

For every metric/cell, store both the value and its denominator or coverage:

{
  "cpc_temperature_record_high_days": 4,
  "cpc_temperature_valid_days": 365,
  "cams_pm25_high_days": 12,
  "cams_pm25_valid_days": 365
}

A zero with positive valid_days is an observed/analyzed zero; absence or zero valid days is unknown. Render covered zero cells in a quiet neutral tone and missing cells as gaps or a missing-data hatch. This makes coverage genuinely non-sparse without drawing false events.

Recommended field contracts are:

• sample the CPC TMAX/TMIN field consistently to H3 cells, then count strict calendar-day records against all prior valid years from 1979 through the preceding year; require at least 30 prior years and keep its 06Z reporting-day definition in metadata;
• compute CAMS daily mean surface PM2.5 in physical units, retaining both days at or above 15 micrograms per cubic metre (the WHO 2021 24-hour guideline) and high days at or above 35.5 micrograms per cubic metre (the first U.S. AQI category above 100); make the displayed threshold a profile choice;
• summarize CAMS organic-matter aerosol optical depth as a continuous annual or year-to-date burden until a smoke-day threshold is validated against HMS and AERONET; do not invent an unvalidated binary smoke count;
• count OISST marine-heatwave days against a fixed 1982–2011 climatology, using the documented seasonally varying 90th-percentile threshold and five-consecutive-day event definition; and
• count CRW reef-pixel days at Bleaching Alert Level 1 or higher, preserving Degree Heating Week and alert-level metadata.

The normalized schema consequently needs typed measures (day-count, mean, maximum, cumulative, or category), floating-point values, valid_days, source/evidence class, grid version, temporal convention, and quality flags. The existing observation schema can remain v1 while the new field file starts at v2.

PurpleAir experiment and optional observed-data pass

Stage 15 implements a 12-plate, default-visible PurpleAir interface experiment for two years and six projections. Its 12 geographic anchors are synthetic rendering fixtures, not sensors; it contains no PM values, measurements, or observation claim. The purpleair-experiment layer is capped at 60% opacity, remains exploration-only, and is built only by its explicit target or make all-experiments.

Assuming redistribution and display permission are resolved, a separate observed-data pass remains worth implementing as an opt-in community-sensor artifact. It adds useful local density in places with participating owners, but it is not globally uniform and will not by itself repair Siberian, Chinese, African, or oceanic coverage. Obtain a read key at PurpleAir Develop and keep it only in the acquisition environment; the current synthetic experiment neither needs nor accepts a key.

Acquisition should use a written bulk-data arrangement or a supplied annual snapshot. PurpleAir’s own API guidance asks clients not to collect several thousand sensors through the per-sensor historical endpoint. Pin the public outdoor-sensor inventory and historical measurements separately for 2025 and 2026, including sensor index, coordinates, location type, channel A/B cf_1, humidity, sampling interval, and timestamps.

A defensible first QA contract is:

1. include only public outdoor sensors and preserve relocation/archival boundaries as new sensor epochs;
2. reject non-finite/negative readings, relative humidity above 95%, missing humidity, either channel above 3,000 micrograms per cubic metre, and samples where the A/B absolute difference exceeds 5 micrograms per cubic metre and relative difference exceeds 70%;
3. require at least 27 of 30 expected two-minute samples for an hourly mean and at least 18 valid hours for a local-day mean;
4. apply the U.S. EPA correction to the mean of A and B cf_1 values, 0.524 * PM - 0.0862 * RH + 5.75, clipping negative corrected values to zero; and
5. aggregate valid sensor-day means to the H3 cell median before thresholding, so dense neighborhoods do not receive more day-count opportunities merely because they contain more sensors.

The EPA correction and cleaning study is documented by Barkjohn, Gantt, and Clements and the operational equation appears in the EPA PM2.5 sensor QAPP. It was developed from U.S. collocations. Outside its validation domain, either require a named regional collocation correction (strict mode) or mark the EPA correction explicitly as an extrapolation (exploratory mode). Never merge PurpleAir counts into EPA, EEA, or other regulatory totals, and never infer smoke from a PM2.5 exceedance alone.

Ocean-distress implementation

Coral heat stress is now practical enough to promote from an indefinite future phase, provided the reef-mask rights are resolved. CRW v3.1 supplies daily 5 km SST, HotSpot, Degree Heating Week, and Bleaching Alert Area files from 1985 to the present. Implement ocean-heat:coral-bleaching-heat-stress-days only in reef-containing cells and describe it as satellite-derived heat stress associated with bleaching risk. It is not an observation that coral actually bleached.

The production mask should be CRW’s reef-containing-pixel mask or another explicitly licensed global reef extent. The checked Natural Earth reef vectors are suitable for a small test fixture, not proof of complete global reef coverage. UNEP-WCMC and Allen Coral Atlas candidates require a separate redistribution/derived-work review before their mask is checked into this repository.

Add ocean-heat:marine-heatwave-days from OISST in the same phase. Marine heatwaves provide globally complete ocean context outside coral latitudes and avoid making reef locations stand in for all ocean distress. Use a fixed climatology and a published consecutive-day definition, record the OISST error and ice masks, and keep marine-heatwave and coral-stress layers independently toggleable because one is global ocean temperature stress and the other is a reef-specific ecological risk indicator.

A later cryosphere theme can use the NOAA/NSIDC Sea Ice Index v4 for Arctic and Antarctic extent/concentration anomalies. Monthly products are more appropriate for trend context than turning every short-term daily variation into a local distress event. Ocean acidification, deoxygenation, harmful algal blooms, and field-observed bleaching remain deferred: their current global datasets are sparse, lagged, survey-dependent, or do not share a defensible annual cell-day contract.

Two-year artifact contract

Each year is an independent reproducibility unit:

Use year-bearing profiles and filenames; do not let two runs overwrite the same unqualified path:

assets.static/anthropocene/anthropocene-particulate-2025-profile.json
assets.static/anthropocene/anthropocene-particulate-2025.geojson
assets.static/anthropocene/anthropocene-particulate-2026-profile.json
assets.static/anthropocene/anthropocene-particulate-2026.geojson
assets.static/anthropocene/anthropocene-temperature-2025-profile.json
assets.static/anthropocene/anthropocene-temperature-2025.geojson
assets.static/anthropocene/anthropocene-temperature-2026-profile.json
assets.static/anthropocene/anthropocene-temperature-2026.geojson
assets.generated/cahill-keyes/png/anthropocene-temperature-2025-ck-44-22.png
assets.generated/cahill-keyes/png/anthropocene-temperature-2026-ck-44-22.png

Implemented temperature-field targets are:

make generate-anthropocene-2025
make generate-anthropocene-2026
make generate-anthropocene-years
make generate-anthropocene-year-artifacts

The aliases generate both implemented standard themes across all six projections. The synthetic PurpleAir and bounded water-debris experiments stay outside these aliases. Future fire/air and ocean themes can join only after an explicit lifecycle decision. For visual year-to-year comparisons, a later product should also support a 2025 window truncated to the same month/day as the 2026 snapshot or render rates per 100 valid days. The canonical 2025 artifact remains complete and the comparison window must be labelled separately.

Implementation sequence

1. Dual-year and schema plumbing — implemented for particulate and CPC

• Split the single profile path into explicit 2025 and 2026 profiles.
• Add year-bearing normalized and generated filenames plus per-year checksum manifests.
• Generalize source statistics instead of hard-coding one field per Stage 8 source.
• Add evidence class, typed field measures, valid-day denominators, source versions, local/UTC reporting-day rules, and partial/complete status.
• Remove the ambiguous unqualified artifact after both year-bearing particulate snapshots and compatibility Make aliases are checked.

2. Highest-value geographic repairs — CPC implemented; FIRMS gated

• Implemented: make a missing/empty FIRMS capture a release error. An explicit regional-development-only override exists for pipeline debugging but cannot pass the global row/date/region audits.
• Implemented infrastructure: query FIRMS-advertised availability so standard S-NPP data and its NRT tail do not leave a multi-month hole, union detections by H3 cell and reporting day, and allow an explicit multi-sensor source list. A checked refresh remains pending the required map key.
• Replace GSN-only ingestion with full eligible GHCNd for the observation tier.
• Implemented: sample CPC to an explicit resolution-3 global H3 domain, retain valid-day denominators, distinguish covered zero from missing, and generate complete-2025 and partial-2026 temperature artifacts. Both checked products serialize 41,162 cells and contain 11,945 covered land-domain cells, with nonzero regional audits outside North America.

3. Global atmosphere

• Acquire CAMS analysis PM2.5 and organic-matter aerosol fields for both years, never future forecast lead times.
• Add valid_days, unit conversion, daily completeness, and source-cycle metadata.
• Add approved OpenAQ/EEA/direct-provider regulatory stations with an explicit provider and license allowlist.
• Validate modeled PM2.5 against regulatory monitors and modeled smoke context against HMS/AERONET without forcing them to agree.

4. Optional community sensors

• Implemented interface only: synthetic anchors, six-projection style, default visibility, 60% opacity, claim boundary, and no credential.
• Implement the real PurpleAir bulk snapshot adapter, observation QA fixture, correction modes, local-day aggregation, and separate artifacts.
• Require the permission record and raw manifest before the opt-in target can run.

5. Ocean fields

• Implement OISST marine-heatwave fixtures and full-year/partial-year aggregation.
• Resolve a production reef mask, then add CRW bleaching-heat-stress fixtures and reef-only validation.
• Render ocean fields below reef symbols and expose stress source/version in SVG metadata.

6. Coverage and release QA

Every candidate should emit a machine-readable matrix for North America, South America, Europe, Africa, northern Asia, East Asia, and Oceania with source rows, valid days, input records, positive cells, and missing cells. Release checks should require:

• nonzero global FIRMS rows and expected acquisition-date coverage;
• at least 99% of eligible land H3 field cells with CPC coverage for each included day, excluding source-declared missing dates;
• at least 99% of expected CAMS and OISST field cells or an explicit documented outage mask;
• nonzero source coverage audits for Siberia, China/Japan, Australia, Europe, Africa, and South America without requiring that every region have an event;
• partial_year: false for 2025 and partial_year: true for 2026;
• separate raw manifests, normalized hashes, source dates, and metric totals for each year and theme;
• no observation/model/community-source metric merging;
• explicit zero-versus-missing tests and projection checks on all six maps; and
• a source-rights manifest recording license, attribution, redistribution, credentials, and whether only derived H3 values may be checked in.

Deferred or rejected shortcuts

• Do not use CAL FIRE, CWFIS, HMS, EPA, or NOAA Storm Events as if any were a global feed.
• Do not keep FIRMS optional in a product advertised as global.
• Do not label TROPOMI aerosol index, total PM2.5, carbon monoxide, or all CAMS organic aerosol as direct observed wildfire smoke.
• Do not use PurpleAir as the default regulatory layer or apply a U.S.-derived correction worldwide without an extrapolation flag.
• Do not use E-OBS alone to fill China, Japan, Australia, Russia, or the 2026 near-real-time gap.
• Do not merge 2025 Siberian fires into the 2026 calendar-year map.
• Do not add coral stress to land H3 cells or describe heat-stress risk as field-observed bleaching.
• Do not calculate an Anthropocene severity score from unlike source values.

The first confirmation point is complete: two year-bearing, source-pinned particulate products and two year-bearing CPC temperature products are standard, while a global observation refresh cannot silently omit FIRMS. Promotion of a new global fire snapshot is intentionally still blocked until a maintainer supplies FIRMS_MAP_KEY and the staged rows pass date and region audits. The synthetic PurpleAir interface is implemented without pretending that community-sensor observations were acquired. This increment repairs the year and temperature ambiguity and establishes the field contract needed by CAMS, a future observed PurpleAir adapter, and ocean products without prematurely committing to their calibration choices.