In 1950, a pioneering team led by meteorologist Jule Charney, alongside Ragnar Fjørtoft and mathematician John von Neumann, embarked on a groundbreaking endeavour at the Aberdeen Proving Ground in Maryland, USA. For 33 relentless days and nights, they harnessed the ENIAC, the world's first programmable electronic digital computer, to produce the inaugural numerical weather prediction. This feat demanded a full 24 hours of computation to generate a mere 24-hour forecast, marking the dawn of computational meteorology.
The ENIAC forecasts, detailed in the seminal 1950 paper "Numerical Integration of the Barotropic Vorticity Equation" published in “Tellus”, represented a seismic shift from manual calculations to machine-driven prediction. Charney's team simplified atmospheric models to the barotropic vorticity equation, dividing the atmosphere into grid cells and solving differential equations via finite differences. Though rudimentary, these predictions over North America proved viable, paving the way for operational numerical weather prediction within five years.
Yet the limitations were stark: ENIAC's vacuum tubes and punched cards constrained scope and speed, with forecasts covering vast 700 km grids at three-hour intervals. This era underscored computing's potential, and its bottlenecks, in tackling weather's chaos.
Numerical weather prediction (NWP) has advanced dramatically since ENIAC. By the 1960s, models added atmospheric layers and leveraged satellites for better storm tracking. The 1970s–1980s introduced primitive equations, finer resolutions, and regional models. Today, NWP provides the core inputs for solar and wind forecasts, simulating irradiance and winds via physical equations.
Modern AI enhances NWP through neural networks, probabilistic models, and physics-informed machine learning that correct biases and handle uncertainty. Today, 76 years later, AI can generate equivalent forecasts in seconds.
The true revolution, however, lies not in AI replacing physics, but in their intelligent fusion. This hybrid approach is epitomised by Renewcast.
At Renewcast we turn renewable uncertainty into measurable profit by delivering superior wind and solar forecasting solutions based on a hybrid physics + AI platform. For wind, we use an AI-native Digital Twin with attention mechanisms and real-time retraining. For solar, we apply a two-step hybrid model (detailed physical plant modelling + machine-learning correction), including stow-aware features. Probabilistic ice forecasting is already available with 83 % average accuracy.
From ENIAC’s marathon computations to today’s hybrid AI systems, forecasting has evolved into a powerful tool for the energy transition. The pinnacle of traditional Numerical Weather Prediction has been reached. The next chapter is being written now.
References