Numerical Weather Prediction (NWP)
The traditional physics-based approach to weather forecasting, which solves mathematical equations representing atmospheric dynamics on a grid — the method AI weather models are increasingly being benchmarked against.
What it means
Numerical weather prediction (NWP) is the dominant method in operational meteorology: it simulates the atmosphere by dividing it into a three-dimensional grid and numerically solving the equations of fluid dynamics and thermodynamics that govern atmospheric motion. At each time step, the model computes how temperature, pressure, wind, and moisture evolve at every grid point.
Modern NWP has improved dramatically over 50 years and is the method behind the forecasts issued by major weather agencies. The European Centre for Medium-Range Weather Forecasts (ECMWF) runs the two NWP systems that have become the gold standard benchmarks for AI weather models:
- HRES (High Resolution): deterministic single forecast; used as the benchmark for GraphCast and Pangu-Weather
- ENS (Ensemble Prediction System): probabilistic ensemble of 51 forecasts; used as the benchmark for GenCast
Why AI models are being compared to NWP
Since 2022, several AI-based weather models have achieved forecast accuracy comparable to or exceeding ECMWF HRES on standard verification metrics, while running in minutes on GPU hardware rather than hours on supercomputer clusters. This comparison — AI model vs. ECMWF — has become the standard way to report the performance of new AI weather models.
Key comparison points:
| NWP (ECMWF HRES) | AI models (e.g., GraphCast) | |
|---|---|---|
| Physics | Explicit equations of atmospheric dynamics | Learned from ERA5 reanalysis data |
| Compute | Hours on large supercomputer | Minutes on GPU |
| Interpretability | Equations have physical meaning | Black box — learned patterns |
| Extreme events | Better at rare events outside training distribution | Weaker on events not well-represented in training data |
| Ensemble | ENS: 51 members (expensive) | GenCast: can generate many members cheaply |
What NWP does that AI models don’t (yet)
- Physical constraints: NWP explicitly conserves mass, energy, and momentum. AI models learn statistical approximations — they may produce physically inconsistent outputs, especially in rare conditions
- Extreme events: AI models trained on historical data underperform on events outside their training distribution (very strong hurricanes, unprecedented heat extremes). NWP physics-based approach has no such distributional limitation in principle
- Coupled systems: Full earth-system models couple atmosphere, ocean, land, and ice. Current AI weather models primarily operate on atmospheric variables from ERA5