Artificial intelligence (AI) has made great progress in the world of weather forecasting…European Forecasting Center makes its AI models work fully

Paul Dorian

Overview

Artificial intelligence (AI) is a series of technologies that allow computers to perform tasks that usually require human intelligence and increasingly affect the world of weather forecasts. The European Center for Medium-Scope Forecasting (ECMWF) has made great progress with its Artificial Intelligence Forecasting System (AIFS), as it has recently been fully operational and now runs side by side with the traditional integrated physics-based prediction system (IF). According to ECMWF, AIFS performs better than physics-based models and is used in many measures, including tropical cyclone orbits. In addition to ECMWF AIFS, there are at least four other known “AI-trained” weather models, including NOAA/Google Graphcast, Microsoft’s Aurora, Nvidia’s Fourcast and Huawei’s Pangu-Weather.

On Friday morning, April 4 (Friday morning) the conventional European predictive model of “total snowfall” on the continental United States (left) and European AI version (right) side by side. Map provides ECMWF, critical weather

discuss

The traditional weather forecasting method uses numerical weather forecasting (NWP), which relies on current conditions, physics-based models, and solutions to complex equations of powerful supercomputers to output parameters such as temperature, pressure, wind and precipitation in future times. Artificial intelligence (AI) models, especially machine learning, are increasingly used to improve weather forecasts by learning from large data sets to identify patterns and trends. AI models can process data faster and determine complex patterns and can lead to faster and more accurate predictions. An increasingly important role for AI in weather forecasting will be to complement and enhance traditional NWP models.

The European Center for Mid-range Forecasting (ECMWF) makes its Artificial Intelligence Forecasting System (AIFS) the first fully operational weather forecasting model using machine learning and artificial intelligence. Making such a system operation means it is publicly available and has 24/7 support for the meteorological community. The AIF can generate a wide range of output parameters including wind, temperature, and details of the type of precipitation from snow to rain. AIFS currently has a grid spacing of 28 km, and according to ECMWF, it can outperform its physics-based counterpart by up to 20% in certain measures.

At 8 a.m. Sunday, March 30, the conventional European predictive model (left) and European AI version (right) of the continent were compared. Map provides ECMWF, critical weather

AIFS uses the same initial atmospheric conditions as IFS for its predictions. These are a combination of previous short-term forecasts and about 60 million quality control observations from satellites and many other streams, including aircraft, ships, offshore buoys and many other Earth-based measurement stations. Every six hours, these initial conditions feed on AIF. The machine learning model trains how the weather has evolved in the past and it evaluates how initial conditions will affect the weather in the next few days. In contrast, IFS uses physics-based capabilities to draw predictions and performs 9km grid spacing around the world and integrates the laws of physics into its computer code.

Meteorologists rarely place complete confidence in a single model, but instead look at a set of models to predict. Most global models calculate the prospect of 10 days every six hours, and one way to predict the reliability of the Scale A model is to check its consistency from run to run. The European racing model performed well in the tests on the Hurricanes in September 2024 as it predicted a landfall in southern Louisiana, from 96 hours to a landfall near Morgan, Los Angeles. Mapping provided by Foxweather.com

The first operational version is called AIFS “single”. It makes a single prediction at once, called deterministic prediction. However, ECMWF is pushing the model to create a collection of 50 different predictions and there are slight changes at any given time to provide a variety of possible scenarios. This is called ensemble modeling, a technology developed and implemented by ECMWF more than thirty years ago. According to ECMWF, the launch of AIFS “single-person” as an operational service is the first step in upgrading its artificial intelligence prediction capabilities. The next step will be to use AI and extended range (seasonal) forecasts to conduct ensemble forecasts, and we will continue to monitor Arcfield Weather’s progress.

Meteorologist Paul Dorian
Akfield
arcfieldweather.com

Follow us on Facebook twitterYoutube