In the past year, something of a revolution has hit the world of weather forecasting as artificial intelligence-based weather forecasts have come to the fore. Traditional weather forecasting methods rely on creating a digital three-dimensional grid that replicates as closely as possible the state of the atmosphere at the start of the forecast.
Once this “initialised state” is determined, complex equations are used to predict how the state of the atmosphere will evolve in the hours and days ahead. For decades, much research has gone into improving these forecasts, focusing on getting the starting point right, increasing vertical and horizontal resolution of these grids, and, of course, making refinements to the equations.
The new generation of AI weather forecasts take a completely different approach, learning from analysing years of initialised data instead of using equations. AI tools are statistical models, so they hunt for patterns in initialised data over the last few decades and then use this to make predictions. Despite the lack of physical equations, they are remarkably accurate and can be run in a fraction of the time of the traditional methods.
In the world of commodity trading, where accurate forecasting is a key determinant for speculating on the price of food, energy or raw materials, the agility these new models offer in tailoring forecast horizons, and speeding up the time it takes to create a forecast have been embraced by traders and analysts looking to gain an advantage.
