Weather, behaviour, and demand methodology

Public research model

Weather forecast → comparison with local 1991–2020 climatology → weather stress and extremeness → versioned research response curves → behaviour indices. An index of 115 means a metric above its normal-weather behavioural reference for that place and season; it does not mean 15% sales growth.

Commercial model

Historical sales are matched to ERA5-Land reanalysis and calendar factors. Two models are trained: a no-weather baseline and a weather-aware model. They are compared on expanding time-series folds. The future period uses ProjectEOL’s published GFS forecast. A reanalysis backtest is not a complete measure of operational GFS error; the architecture permits future archived-GFS backtesting.

Forecasts are statistical estimates, not guarantees of sales, revenue, or profit.