Glossary

Ensemble Forecasting

Running many slightly different versions of a forecast model to produce a range of possible outcomes rather than a single prediction — the standard approach for quantifying forecast uncertainty in meteorology.


Ensemble forecasting runs many parallel simulations of a forecast model, each starting from slightly different initial conditions or using slightly different model parameters. Instead of a single “best guess” forecast, the output is a distribution of possible futures — a spread of trajectories that reflects genuine uncertainty in the initial state and model formulation.

Why ensembles matter:

The atmosphere is chaotic — small differences in initial conditions grow over time (the “butterfly effect”). A single deterministic forecast cannot capture this uncertainty. An ensemble communicates not just what is most likely to happen, but how confident we are, and what the range of plausible outcomes looks like. This is essential for applications like extreme event risk assessment, where the probability of a rare outcome matters as much as the most likely one.

AI ensemble forecasting:

Traditional ensemble approaches (like ECMWF’s ENS, which runs 51 members) are extremely computationally expensive. AI models are enabling cheaper ensembles:

  • GenCast (Google DeepMind) generates probabilistic forecasts using a diffusion model, producing ensemble-like uncertainty estimates without running dozens of full model copies
  • NOAA AIGEFS — NOAA’s AI-based ensemble system that has shown extended forecast skill compared to traditional ensemble methods

Deterministic vs. ensemble:

Deterministic Ensemble
Output Single forecast Distribution of forecasts
Uncertainty None communicated Explicitly represented
Cost Lower Higher (traditional); improving with AI
Use case Day-to-day forecasting Risk assessment, extreme events

Related terms: Reanalysis Data, Numerical Weather Prediction

Related guide: Climate & Earth Science