Air

Airsheds on five continents — Delhi, Beijing, Lagos, Mexico City, Los Angeles, Kraków, Santiago — each scored against the same seven forecast centres. Every number on this page is computed from receipts we hold locally; nothing here waits on an upstream call.

receipts held

1,000

resolved against truth

0%

sites

7

forecast centres

14

Forecast horizon → error

Mean absolute error by lead time, per channel, from resolved receipts

MEASURED

No channel in this domain has resolved across enough lead times yet.

Which centre is actually right here

Best and worst measured producer per channel

MEASURED

No resolved receipts in this domain yet.

Prediction ↔ observation

Daily forecast mean against daily observed mean; the gap between the lines is the standing bias

MEASURED

No channel has enough resolved days to plot yet.

Where the air forecast fails

Mean absolute temperature error per airshed, worst first

MEASURED

No site has enough resolved receipts to rank yet.

Ventilation error per airshed

Wind-speed error — the other half of an air-quality forecast

MEASURED

No site has enough resolved receipts to rank yet.

Models trained on air receipts

Stored weights, their measured error and their skill against persistence

MEASURED
residual·air:pm2_5 +6h0.80-0.04n=526,280
residual·air:pm2_5 +24h0.81-0.03n=504,275
residual·air:pm2_5 +72h0.77+0.12n=462,313
residual·air:pm25 +6h0.14+0.73n=432
residual·air:precipitation +6h0.77+0.04n=31,648
residual·air:precipitation +24h0.55+0.35n=24,138
residual·air:precipitation +72h0.42+0.32n=6,307
residual·air:surface_pressure +6h0.78+0.33n=31,636
residual·air:surface_pressure +24h0.62+0.33n=24,136
residual·air:surface_pressure +72h1.05+0.04n=6,286
residual·air:temperature_2m +6h0.66+0.35n=526,559
residual·air:temperature_2m +24h0.64-0.10n=504,537
residual·air:temperature_2m +72h0.62+0.10n=462,589
residual·air:wind_speed_10m +6h0.90-0.03n=526,415
residual·air:wind_speed_10m +24h0.82+0.15n=504,377
residual·air:wind_speed_10m +72h0.86+0.25n=462,441

columns · MAE · skill vs persistence · training samples