Bay Area Air Quality Episode Finder

Sentinel-5P measures a tropospheric vertical column, not the air at ground level. Column amount and the concentration a person breathes are different physical quantities. Nothing here is an AQI value, a health category, or an episode declaration.

Run parameters

Indicator

PM2.5 is out of scope. Any future PM2.5 work would be an estimate workflow, never a direct satellite observation.

Target period

Local dates, America/Los_Angeles. End date is exclusive.

Map layer

Historical years

Same calendar month, prior years only. The target year and future dates are never used.

Analysis endpoints are not built yet. This control is inactive until the production API exists.

Map

Display processing only, never feeds the statistics

Basemap only — no data layer is loaded.

Signed anomaly below median

above median Colour is a display stretch, not a threshold

Readouts

Awaiting a run

Regional mean column

—

TROPOMI → L3 grid → daily composite → area-weighted mean not a surface concentration

Signed anomaly

—

target mean − same-month median, prior 3 years not an episode declaration

Percentile in history

—

share of historical days at or below target descriptive reference only

Valid-area coverage

—

valid pixel area ÷ region area, per day reported always, never used to exclude

Daily series against baseline

Gaps are missing days, never interpolated

No series loaded

The daily series renders here once the production API exists.

Episode assessment

Not implemented

No feature in this project classifies air-quality episodes. Doing so requires persistence and spatial-extent criteria, corroborating ground-monitor evidence, and meteorological context, none of which is defined yet. Until then this dashboard reports satellite-column anomalies and the coverage behind them, and stops there.

Known limits of this measurement

Column, not ground level

The instrument measures NO₂ through the depth of the troposphere. That is a different physical quantity from what a monitor at street level records.

Grid is finer than the footprint

The 0.01° product grid is oversampled relative to the native footprint of roughly 3.5 × 5.5 km. Neighbouring cells are not independent kilometre-scale measurements.

Coverage varies by day

Cloud and retrieval quality decide how much of the region is seen. Identical means from 90% and 15% coverage describe very different amounts of evidence, so coverage travels with every number.

The record is not uniform

Processor and algorithm versions change through the archive. This is a rolling comparison against recent same-month history, not a homogeneous long-term trend.

Spatial coincidence is not cause

A pattern that lines up with a highway or a facility is geography, not attribution. This product cannot assign column values to a road segment or a site.

No surface validation yet

Comparison against BAAQMD and EPA surface monitors, overpass timing, and meteorological context remains future work.