ECMWF's new AI-powered weather prediction model, AIFS, is revolutionizing forecasting with greater accuracy, speed, and energy efficiency. This breakthrough could transform weather prediction for industries from renewable energy to disaster preparedness.
Image Credit: Piyaset / Shutterstock
A newly operational model, known as the Artificial Intelligence Forecasting System (AIFS), has been launched by the European Centre for Medium-Range Weather Forecasts (ECMWF), an intergovernmental center and leader in numerical weather prediction. For many measures, including tropical cyclone tracks, the AIFS outperforms state-of-the-art physics-based models, with gains of up to 20%. This high-accuracy model complements the portfolio of ECMWF's physics-based models, advancing numerical weather prediction, and leverages the opportunities made available by machine learning (ML) and artificial intelligence (AI), such as increased speed and a reduction of approximately 1,000 times in energy use for making a forecast.
Amongst the available AI models, the AIFS provides the greatest granularity sought by its user community. On top of vital fields for users, like wind and temperature, and details on precipitation types from snow to rain, ECMWF says this new service is the first fully operational weather prediction open model using machine learning with the widest range of parameters. The AIFS has been designed holistically with all users in mind. For example, in the renewable energy sector, it will help with predictions such as surface solar radiation levels or wind speeds at turbine levels to maximize operations.
The availability of this operational machine learning model, in conjunction with ECMWF's other services, will positively impact how national weather services in ECMWF's 35 Member and Cooperating States and beyond can make their predictions. Similarly, it could potentially help industries where forecasts for the medium range (days to weeks) can affect decision-making, such as the energy sector for pricing forecasts, insurance, security, and shipping sectors.
Dr Florence Rabier, Director-General of ECMWF, comments:
"This milestone will transform weather science and predictions. It showcases our dedication to delivering a machine learning forecasting model that pushes the boundaries of efficiency and accuracy, and it underscores our commitment to harnessing the power of machine learning for the weather forecasting community.
"At ECMWF, we have some of the world's leading computing models for weather prediction, the world's largest catalogue of meteorological data sets on which machine learning models are trained, and a team of experts from scientists to engineers who are driving the science and technology forward. It is credit to their hard work that we have achieved this today, making the AIFS operational producing the widest range of parameters using machine learning available to date. But it does not stop here as the roadmap for improvements of the models we have is a top priority. It is not only us who are innovating as it is important to remember that with ECMWF, 35 nations are working together to advance weather science to improve global predictions. This is to help national meteorological agencies in their work to contribute to a safe and thriving society. This will also trigger new services and products to benefit those who do not have access to meteorological capabilities, for example in developing countries."
How does the model work?
Florence continues explaining how the model works: "Imagine 800 million observations processed on a daily basis, from more than 100 different satellite data and other streams including planes, boats, sea buoys, and many other Earth-based measurement stations. These observations contain information about, for example, the Earth's atmosphere, pressure, moisture, temperature, and wind. Scientists next select around 60 million quality-controlled observations from those daily observations, which are then ingested into our Integrated Forecasting System (IFS). These then form what we call the initial conditions, the starting point of the next stage to deliver forecasts. Our set of initial conditions data, as well as our historical meteorological dataset archive, are widely used worldwide by big tech companies and small start-ups to help them innovate.
"Every 6 hours, these initial conditions feed into the newly operational Artificial Intelligence Forecasting System (AIFS), where the machine learning model, using special mathematical rules, assesses how the current meteorological conditions will influence the entire weather system on the Earth for the coming days. Today, we are launching the first fully operational model of this kind based on machine learning."
This model joins a series of other ECMWF services, one of which is the delivery of global weather forecasts by the Integrated Forecasting System (IFS), which not only provides the ability to gather the initial conditions but also produces forecasts at a world-leading resolution of 9 km over the globe. It uses physics-based capabilities to reach this, integrating the laws of physics into its computer code.
The first operational version is called AIFS-single. It runs one forecast at a time, known as a deterministic forecast. However, ECMWF is pushing this model to create a collection of 50 different forecasts with slight variations at any given time to provide the full range of possible scenarios. This is known as ensemble modeling, a technique developed and implemented by ECMWF more than thirty years ago.
ECMWF says its roadmap for its Artificial Intelligence Forecasting System is clear, and the launch of this AIFS-single 1.0 model as an operational service is only the first step. The next step will be making ensemble forecasts available following the same path. The potential to hybridize the two approaches, data-driven and physics-based, will also be a field of research over the coming years to explore this potential further.
Dr Florian Pappenberger, Director of Forecasts and Services at ECMWF, adds:
"This is a huge endeavour that ensures the models are running in a stable and reliable way. At the moment, the resolution of the AIFS is less than that of our model (IFS), which achieves 9 km resolution using a physics-based approach. We see the AIFS and IFS as complementary, and part of providing a range of products to our user community, who decide what best suits their needs."
ECMWF's weather predictions are focused on the medium range (3 days to 15 days), sub-seasonal, and seasonal (up to a year ahead). These are critical for helping national weather services plan for extreme events. The earlier an event is known to occur, the easier it is for governments and relevant agencies to prepare for it.
Florian concludes:
"ECMWF's AIFS was an experimental model for some months whilst we enhanced its capabilities by interacting with our Member States and users to refine it. We have brought it to an operational state for the benefit first and foremost of our Member and Cooperating States as well as many industry sectors, such as energy. Making such a system operational means that it is openly available and has 24/7 support for our meteorological community. As always, we have to ramp up this service to full maturity, and we look forward to engaging directly with our users to ensure all needs are covered where possible."
While the ECMWF concentrates on predictions from days to months ahead globally, the national meteorological services also focus on nationally and regionally relevant forecasts. They oversee issuing weather warnings for their country in collaboration with other agencies. This year marks 50 years since the creation of the ECMWF, and the complementarity between the ECMWF and the meteorological centers across its Member States is essential to fostering trust, excellence, and resilience in services and advancing science for the benefit of everyone.