Open-Meteo Weather

Weather is a primary driver of electricity prices through its effect on both supply (solar, wind generation) and demand (heating, cooling). EPF collects hourly weather data from 5 stations across Spain and aggregates them into a single national weather signal.

Data Provider

EPF uses Open-Meteo, an open-source weather API:

Forecast endpoint: https://api.open-meteo.com/v1/forecast — 7-day ahead hourly forecasts
Archive endpoint: https://archive-api.open-meteo.com/v1/archive — historical data for training
Cost: Free, no API key required
Resolution: Hourly

Weather Stations

Five stations are selected to represent Spain’s major climate zones, weighted by population density:

Station	City	Latitude	Longitude	Zone	Weight
Madrid-Barajas	Madrid	40.47°N	3.56°W	Center/West	0.35
Barcelona-El Prat	Barcelona	41.30°N	2.08°E	East	0.25
Valencia-Airport	Valencia	39.49°N	0.47°W	East	0.15
Bilbao-Airport	Bilbao	43.30°N	2.91°W	North	0.13
Sevilla-Airport	Sevilla	37.42°N	5.90°W	South	0.12

The population weighting reflects how demand is distributed: Madrid and Barcelona account for 60% of the weight because they drive the largest share of national electricity consumption.

Variables Collected

API Parameter	Column Name	Unit	Description
`temperature_2m`	`temp_c`	°C	Air temperature at 2m height
`wind_speed_10m`	`wind_speed_kmh`	km/h	Wind speed at 10m height
`precipitation`	`precipitation_mm`	mm	Precipitation amount
`sunshine_duration`	`sunshine_hours`	hours	Sunshine duration (converted from seconds)
`cloud_cover`	`cloud_cover_pct`	%	Total cloud cover percentage
`direct_radiation`	`direct_radiation_wm2`	W/m²	Direct solar radiation
`diffuse_radiation`	`diffuse_radiation_wm2`	W/m²	Diffuse solar radiation

Aggregation

All 5 stations are fetched independently, then combined into a weighted national average:

national_value = sum(station_value × weight) / sum(weights)

This produces a single hourly time series for each weather variable that represents national conditions.

Forecast vs. Historical

The weather pipeline serves two purposes:

Real-time forecasting: Fetches past_days=2, forecast_days=7 to provide a 2-day lookback plus 7-day forecast horizon. At forecast time, the model uses observed weather up to the present and forecast weather for target hours.
Historical training: The archive endpoint provides years of historical weather aligned with historical electricity prices for model training. Archive requests are chunked into 3-month periods with rate-limiting delays between calls.

Weather-Derived Features

Raw weather variables are transformed into interaction features that capture their effect on electricity supply and demand:

Temperature-demand interactions: heating degree days, cooling degree days, cold-demand cross-term
Wind-generation interaction: wind speed × wind generation share
Solar features: global horizontal irradiance (GHI = direct + diffuse), clear-sky index, solar elevation angle, daylight indicator
Cloud-solar interaction: cloud cover × solar share
Precipitation-hydro interaction: rainfall × hydro generation share

See the Weather Features section for computation details.