Technical details
This page explains which forecast model we use, which data it was checked against, and how to interpret its errors without losing the technical traceability.
Forecast model
NO2 model v1
active modelHorizon
24h
Training window
2025-01-08 → 2025-12-31
Test window
2026-01-01 → 2026-04-29
Current operational read
Cloudflare D1 + ingestion
The API is already reading from Cloudflare D1, although the last valid observation is still 28 Sept 2026, 02:00. The next point to review is the protected ingestion cycle, not database connectivity.
Environment
production
reported by the API
Observation source
Cloudflare D1
active read path
Latest valid observation
28 Sept 2026, 02:00
over 12h old
Remote ingest
cron ready
D1 yes · jobs yes
MAE
7.85 µg/m³
Average expected gap between the forecast and the real measurement.
RMSE
11.96 µg/m³
How large the misses can become when conditions shift more sharply.
R²
0.43
How much of the real-world variation the model is able to explain.
NO2 · 24h
compared metrics
persistence
n=78077
same_hour_yesterday
n=78077
hist_gradient_boosting_v1
n=78077
This is an indicative forecast based on recent patterns. It does not replace official public warnings.
NO2 · error chart by model
MAE & RMSE compared
Model features
01
Supervised learning
Trained on real readings from the official network, not on synthetic or simulated data.
02
24-hour horizon
Produces hourly forecasts for the next 24 hours from the moment of generation.
03
Temporal variables and station context
Includes time of day, day of week, month, weekend flag, and station encoding as the model's explanatory base.
04
Comparison against simple baselines
Each run is evaluated against the recent trend, the same hour yesterday, and the 24-hour average to validate genuine improvement.
Model v1 limits
01
No validated extreme episodes
Training data does not cover severe pollution episodes or exceptional meteorological events.
02
Coverage limited to the official network
The model produces station-level forecasts for the official network, but it does not interpolate beyond those points or generate a continuous surface across Madrid.
03
No real-time updates
Forecasts are regenerated on scheduled cycles, not with each new official reading.
04
Indicative, not regulatory
This forecast is an internal technical reference. It does not replace official air quality warnings.