Ensemble Forecasting
The standard method in numerical weather prediction for addressing chaotic uncertainty. A set of forecast members is generated by perturbing initial conditions or model physics, and the dispersion among members is used to estimate forecast uncertainty.
Timeline (last 90 days)
Machine learning methods can automatically extract from massive historical samples relationships that human forecasters find difficult to model explicitly, and their inference is fast and their computational cost far lower than traditional ensemble forecasting.
On the subseasonal scale, a single deterministic forecast is almost meaningless, and operational forecasting usually adopts the ensemble forecasting method, generating a set of forecasts (dozens of members) by perturbing initial conditions or model parameters, and using the dispersion of the members to characterize forecast uncertainty.
For machine learning models, how to generate physically meaningful ensemble members and how to calibrate ensemble dispersion so that it matches actual error remain an active research direction.
All Facts (3)
Machine learning methods can automatically extract from massive historical samples relationships that human forecasters find difficult to model explicitly, and their inference is fast and their computational cost far lower than traditional ensemble forecasting.
50%UnverifiedOn the subseasonal scale, a single deterministic forecast is almost meaningless, and operational forecasting usually adopts the ensemble forecasting method, generating a set of forecasts (dozens of members) by perturbing initial conditions or model parameters, and using the dispersion of the members to characterize forecast uncertainty.
50%UnverifiedFor machine learning models, how to generate physically meaningful ensemble members and how to calibrate ensemble dispersion so that it matches actual error remain an active research direction.
50%