Numerical Weather Prediction

NUMERICAL WEATHER PREDICTION (NWP) uses MATH and COMPUTERS to forecast the weather. Scientists divide the atmosphere into a 3D grid of millions of small cubes. For each cube, they have current measurements (temperature, pressure, humidity, wind). Equations describe how these will change over the next minute. The computer calculates forward — minute by minute — for hours, days, or weeks ahead.

Why it's hard. The atmosphere is CHAOTIC. Tiny errors in current measurements grow over time (this is the "butterfly effect" — Edward Lorenz discovered it studying weather). Beyond ~10 days, predictions become unreliable. ENSEMBLE FORECASTING runs the model many times with slight variations and looks at how predictions diverge. If most runs agree, confidence is high. If they spread wildly, the forecast is uncertain.

Modern NWP. The GFS (Global Forecast System, NOAA) and ECMWF (European Centre) are two leading global models. They run on supercomputers handling quadrillions of operations per second. Today's 5-day forecasts are about as good as 1-day forecasts were in the 1980s — a remarkable improvement. AI/ML models (like Google's GraphCast) are now competing with traditional physics-based models, and increasingly winning.

Numerical weather prediction is one of computational science's greatest success stories. Lives saved, billions of dollars in protected property, and countless smarter daily decisions — all from running atmospheric physics on big computers.