Data model (high level)¶

Schemas are created by docker/postgres/initdb and extended by sql/migrations/ + sql/views/. This page summarizes major relations — not every column.

Schemas¶

Schema	Purpose
`raw`	Vendor payloads (JSONB), gridded weather staging.
`normalized`	Fire detections, AQ measurements, wind observations, weather grid cells.
`geo`	Census county and tract geometries (`geoid`, `geom`).
`analytics`	Risk scores, plume/dispersion exposures, calibration, alerts, lag/DLQ bookkeeping.

Core normalized tables¶

Table	Purpose
`normalized.fire_detections`	FIRMS-derived points with optional `county_geoid` / `tract_geoid`.
`normalized.air_quality_measurements`	OpenAQ-derived PM and related measurements.
`normalized.wind_observations`	Wind stations / observations (`wind FROM` convention).
`normalized.weather_grid_cells`	Gridded weather cells when grid ingest is enabled.

Analytics highlights¶

Table	Purpose
`analytics.smoke_risk_scores`	County/tract risk scores and bands per model version.
`analytics.smoke_plume_exposures`	Corridor plume heuristic exposures.
`analytics.smoke_dispersion_exposures`	Gaussian proxy dispersion-style exposures (`gaussian_v0`).
`analytics.dispersion_aq_comparisons`	Lag-window AQ vs dispersion summaries + evidence labels.
`analytics.risk_observations`	Optional labeled observations for evaluation.
`analytics.risk_model_evaluations`	Batch evaluation metrics vs observations.
`analytics.parse_errors`	Quarantined normalization failures.
`analytics.alert_events`	Materialized alert incidents (when materialization is run).

Views¶

Stable analytics.v_* views power Grafana and ad hoc queries. See Tables and views for a longer list.

DDL source of truth: sql/migrations/, sql/views/, docker/postgres/initdb/ (fresh volumes).