Demand forecast vs climatology, next 14 days
Daily electricity demand, GWh per day. Solid: forecast weather. Dashed: climatology, the demand at the average temperature for that date (1996–2025, 30 years). Shaded: days beyond the point where the forecast has scored worse than climatology.
How the forecasts have scored
Demand error by lead day (MAPE, %)
Temperature error by lead day (MAE, °C)
Wind and solar generation, next 14 days
Daily generation, GWh, from forecast wind at 100 m and surface solar radiation at each region’s generation areas, mapped to EIA-reported output.
Solid: forecast weather. Dashed: climatology, the generation under the average wind and sunshine for that date (2015–2025, 11 years). Levels are anchored to the last 28 days of observed output; shaded days are beyond where the forecast stops scoring better than climatology.
How the wind and solar forecasts have scored
Wind: generation error by lead day (MAPE, %)
Solar: generation error by lead day (MAPE, %)
Crop stress: US corn and soybean states, next full weeks
How the crop-stress forecasts have scored
Weekly heat-stress degree days (above 30 °C): forecast error by lead (MAE)
Weekly afternoon vapour-pressure deficit (kPa): forecast error by lead (MAE)
Weekly rainfall (mm): forecast error by lead (MAE)
Weekly change in good+excellent rating (points): error vs baselines (MAE)
Gas burned in power plants, next 14 days
Five modelled regions together, GWh of gas-fired generation per day
Error by lead day (MAPE, %), forecast vs climatology, both anchored to the last four weeks
Gas storage: expected weekly change in the next EIA reports
What the forecast weather is worth in power prices
Weather-driven premium over the climatological price, % of the next-day peak price
Live experiment: do free weather forecasts, turned into decisions, make money?
Cumulative net return (log, after costs): decisions from the forecast vs the same decisions from climatology, holding the same instruments, and the market