Google's AI-Powered Weather Model Revolutionizes Forecasting Accuracy

Google AI predicts long-term climate trends and weather -- in minutes

A computer model that combines conventional weather-forecasting technology with machine learning has outperformed other artificial intelligence (AI)-based tools at predicting weather scenarios and long-term climate trends. The tool, described in Nature on 22 July, is the first machine-learning model to generate accurate ensemble weather forecasts -- ones which present a range of scenarios. Its development opens the door for forecasting that is faster and less energy-intensive than existing tools, and more detailed than approaches based solely on AI. "Traditional climate models need to be run on supercomputers. This is a model you can run in minutes," says study co-author Stephan Hoyer, who studies deep learning at Google Research in Mountain View, California. Current forecasting systems typically rely on general circulation models (GCMs), programmes that draw on the laws of physics to simulate processes in Earth's oceans and atmosphere and predict how they might affect the weather and climate. But GCMs require a lot of computing power, and advances in machine learning are starting to provide a more efficient alternative. "We have terabytes or petabytes (one million times larger than a gigabyte) of historical weather data" says Hoyer. "By learning from those patterns, we can build better models." There are already some machine-learning forecasting models available, such as Pangu-Weather, built by the technology conglomerate Huawei, based in Shenzhen, China, and GraphCast by DeepMind, headquartered in London. These models have similar accuracy levels to typical GCMs for deterministic forecasting -- an approach that generates a single weather forecast. But GCMs aren't as reliable for ensemble forecasting, or for long-term climate predictions. "The issue with pure machine-learning approaches is that you're only ever training it on data it's already seen" says Scott Hosking, who researches AI and environmental data at the Alan Turing Institute in London. "The climate is continuously changing, we're going into the unknown, so our machine-learning models have to extrapolate into that unknown future. By bringing physics into the model, we can ensure that our models are physically constrained and cannot do anything unrealistic." Hoyer and his team developed and trained NeuralGCM, a model that combines "aspects from a traditional physics-based atmospheric solver with some AI components", says Hoyer. They used the model to produce short- and long-term weather forecasts, as well as climate projections. To assess NeuralGCM's accuracy, the researchers compared its predictions with real-world data, as well as outputs from other models, including GCMs and those based purely on machine learning. Like current machine-learning models, NeuralGCM could produce accurate short-term, deterministic weather forecasts -- between one and three days in advance -- while consuming a fraction of the power required by GCMs. But it made much fewer errors than other machine-learning models when producing long-term forecasts beyond seven days. In fact, NeuralGCM's long-term forecasts were similar to predictions made by the European Centre for Medium-Range Weather Forecast's ensemble model (ECMWF-ENS), a GCM that is widely regarded as the gold standard for weather forecasting. The team also tested how well the model could forecast different weather phenomena, such as tropical cyclones. They found that many of the pure machine-learning models produced inconsistent and inaccurate forecasts compared with both NeuralGCM and ECMWF-ENS. The researchers even compared NeuralGCM with ultra-high-resolution climate models known as global storm-resolving models. NeuralGCM could produce more-realistic tropical-cyclone counts and trajectories in a shorter time. Being able to predict such events is "so important for improving decision-making abilities and preparedness strategies", says Hosking. Hoyer and his colleagues are keen to further refine and adapt NeuralGCM. "We've been working on the atmospheric component of modelling the Earth's system ... It's perhaps the part that most directly affects day-to-day weather," Hoyer says. He adds that the team wants to incorporate more aspects of Earth science into future versions, to further improve the model's accuracy.

AI helps to produce breakthrough in weather and climate forecasting

Artificial intelligence has helped to make a breakthrough in accurate long-range weather and climate predictions, according to research that promises advances in both forecasting and the wider use of machine learning. Using a hybrid of machine learning and existing forecasting tools, a model led by Google called NeuralGCM successfully harnessed AI to conventional atmospheric physics models to track decades-long climate trends and extreme weather events such as cyclones, a team of scientists found. This combination of machine learning with established techniques could provide a template for refining the use of AI in other fields from materials discovery to engineering design, the researchers suggest. NeuralGCM was much faster than traditional weather and climate forecasting and better than AI-only models at longer-term predictions, they said. "NeuralGCM shows that when we combine AI with physics-based models, we can dramatically improve the accuracy and speed of atmospheric climate simulations," said Stephan Hoyer, senior staff engineer at Google Research and a co-author of a paper on the work published in Nature. The paper said NeuralGCM proved faster, more accurate and used less computing power in tests against a current forecasting model based on atmospheric physics tools called X-SHiELD, which is being developed by an arm of the US National Oceanic and Atmospheric Administration. In one trial, NeuralGCM identified almost the same number of tropical cyclones as conventional extreme weather trackers did, and twice the number of X-SHiELD. In another test based on temperature and humidity levels during 2020, the error rate was between 15 and 50 per cent less. NeuralGCM's calculations were able to generate 70,000 simulation days in 24 hours using one of Google's customised AI tensor processing units, the paper says. By contrast, for comparable calculations, X-SHiELD generated only 19 simulation days, and needed 13,824 computer units to do it. Google collaborated on the development of NeuralGCM with the inter-governmental European Centre for Medium-Range Weather Forecasts (ECMWF). The European group made its model publicly available in June, and Google has made the code for NeuralGCM open access. It uses 80 years of ECMWF observational data and reanalysis for machine learning. Google's DeepMind unit last year unveiled an AI-only weather forecasting model called GraphCast, which outperformed conventional methods for periods up to 10 days ahead. Established forecasting agencies such as the UK Met Office also have projects to integrate machine learning into their work. Peter Dueben, head of the ECMWF's earth system modelling and a co-author of the latest paper, said AI-only models were "often viewed sceptically" by experts because they were not based on mathematical equations devised from physics. The combination of the physics-based model with the deep learning model "seems to get the best of both worlds", he said, adding that the approach was a "big step towards climate modelling with machine learning". There was still more "work to do", such as to enable NeuralGCM to estimate the impact of CO₂ increases on global surface temperatures, Dueben said. Other areas in which the model needed to be better included its capacity to simulate unprecedented climates, the paper said. An expert not involved in the work, Cédric M. John, head of data science for the environment and sustainability at Queen Mary University of London, said there was "compelling evidence" that NeuralGCM was more accurate than machine learning alone and faster than the "full-physics" model. While there was still "room for improvement", the possibility of error should be measurable and upgrades should be possible, he suggested. "Importantly, this hybrid model does well at capturing an ensemble of predictions, and the practical implication of this is that an estimate of the uncertainty of the prediction can be derived," said John. Google has become involved in a growing number of environmental surveillance initiatives. It provides technological support for a satellite mission to track planet-warming emissions of methane and partners Nasa, the US space agency, to help local governments monitor air quality.

A new weather prediction model from Google combines AI with traditional physics

The model, called NeuralGCN and described in a paper in Nature today, bridges a divide that's grown among weather prediction experts in the last several years. While new machine-learning techniques that predict weather by learning from years of past data are extremely fast and efficient, they can struggle with long-term predictions. General circulation models, on the other hand, which have dominated weather prediction for the last 50 years, use complex equations to model changes in the atmosphere and give accurate projections, but they are exceedingly slow and expensive to run. Experts are divided on which tool will be most reliable going forward. But the new model from Google instead attempts to combine the two. "It's not sort of physics versus AI. It's really physics and AI together," says Stephan Hoyer, an AI researcher at Google DeepMind and a coauthor of the paper. The system still uses a conventional model to work out some of the large atmospheric changes required to make a prediction. It then incorporates AI, which tends to do well where those larger models fall flat -- typically for predictions on scales smaller than about 25 kilometers, like those dealing with cloud formations or regional microclimates (San Francisco's fog, for example). "That's where we inject AI very selectively to correct the errors that accumulate on small scales," Hoyer says. The result, the researchers say, is a model that can produce quality predictions faster with less computational power. They say NeuralGCM is as accurate as one-to-15-day forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF), which is a partner organization in the research. But the real promise of technology like this is not in better weather predictions for your local area, says Aaron Hill, an assistant professor at the School of Meteorology at the University of Oklahoma, who was not involved in this research. Instead, it's in larger-scale climate events that are prohibitively expensive to model with conventional techniques. The possibilities could range from predicting tropical cyclones with more notice to modeling more complex climate changes that are years away. "It's so computationally intensive to simulate the globe over and over again or for long periods of time," Hill says. That means the best climate models are hamstrung by the high costs of computing power, which presents a real bottleneck to research. AI-based models are indeed more compact. Once trained, typically on 40 years of historical weather data from ECMWF, a machine-learning model like Google's GraphCast can run on less than 5,500 lines of code, compared with the nearly 377,000 lines required for the model from the National Oceanic and Atmospheric Administration, according to the paper. NeuralGCM, according to Hill, seems to make a strong case that AI can be brought in for particular elements of weather modeling to make things faster, while still keeping the strengths of conventional systems.

AI-powered weather and climate models are set to change the future of forecasting

UNSW Sydney provides funding as a member of The Conversation AU. A new system for forecasting weather and predicting future climate uses artificial intelligence (AI) to achieve results comparable with the best existing models while using much less computer power, according to its creators. In a paper published in Nature today, a team of researchers from Google, MIT, Harvard and the European Centre for Medium-Range Weather Forecasts say their model offers enormous "computational savings" and can "enhance the large-scale physical simulations that are essential for understanding and predicting the Earth system". The NeuralGCM model is the latest in a steady stream of research models that use advances in machine learning to make weather and climate predictions faster and cheaper. What is NeuralGCM? The NeuralGCM model aims to combine the best features of traditional models with a machine-learning approach. At its core, NeuralGCM is what is called a "general circulation model". It contains a mathematical description of the physical state of Earth's atmosphere, and it solves complicated equations to predict what will happen in the future. However, NeuralGCM also uses machine learning - a process of searching out patterns and regularities in vast troves of data - for some less well-understood physical processes, such as cloud formation. The hybrid approach makes sure that the output of the machine learning modules will be consistent with the laws of physics. The resulting model can then be used for making forecasts of weather days and weeks in advance, as well as looking months and years ahead for climate predictions. The researchers compared NeuralGCM against other models using a standardised set of forecasting tests called WeatherBench 2. For three- and five-day forecasts, NeuralGCM did about as well as other machine-learning weather models such as Pangu and GraphCast. For longer-range forecasts, over ten and 15 days, NeuralGCM was about as accurate as the best existing traditional models. NeuralGCM was also quite successful in forecasting less-common weather phenomena, such as tropical cyclones and atmospheric rivers. Why machine learning? Machine learning models are based on algorithms that learn patterns in the data they are fed with, then use this learning to make predictions. Because climate and weather systems are highly complex, machine learning models require vast amounts of historical observations and satellite data for training. The training process is very expensive and requires a lot of computer power. However, after a model is trained, using it to make predictions is fast and cheap. This is a large part of their appeal for weather forecasting. The high cost of training and low cost of use is similar to other kinds of machine learning models. GPT-4, for example, reportedly took several months to train at a cost of more than US$100 million, but can respond to a query in moments. A weakness of machine learning models is that they often struggle in unfamiliar situations - or in this case, extreme or unprecedented weather conditions. To do this, a model needs to be able to generalise, or extrapolate beyond the data it was trained on. NeuralGCM appears to be better at this than other machine learning models, because its physics-based core provides some grounding in reality. As Earth's climate changes, unprecedented weather conditions will become more common, and we don't know how well machine learning models will keep up. Nobody is actually using machine learning-based weather models for day-to-day forecasting yet. However, it is a very active area of research - and one way or another, we can be confident that the forecasts of the future will involve machine learning.

AI weather and climate forecasting advances with new model, study shows

Why it matters: Its creators say the new model, dubbed "NeuralGCM," has proven to be more accurate than other purely machine learning-based models for one- to 10-day weather forecasts, along with the top extended-range models in use today. Zoom in: The findings demonstrate how quickly the field of AI-based weather and climate forecasting are advancing. How it works: The new model, from scientists at Google Research, Google DeepMind, MIT, Harvard University and the European Center for Medium-Range weather forecasts, uses machine learning and a neural network. Aaron Hill, an assistant meteorology professor at the University of Oklahoma, told Axios that one of the biggest "novelties" of the new model is how it keeps some of the large-scale physics and replaces some parts of the modeling with AI. Between the lines: Hill, who wasn't involved in the new study, said AI and machine learning techniques are rapidly being adopted in the weather and climate research communities. What they're saying: Hoyer, of Google Research, said public sector agencies have come to see that they need to more fully invest in AI systems, which are developing quickly and showing promise, but not to replace their traditional weather and climate models just yet.

A Google AI model is improving climate forecasting

The news comes weeks after a report concluded that the use of AI has increased Google's emissions by 48 percent. A new Google AI tool promises to improve climate prediction. NeuralGCM, developed by Google Research, uses physics-based modelling and AI to create fast and precise simulations of Earth's atmosphere. Traditional climate models work by fragmenting the planet into large pixelated pieces, which makes it hard to accurately predict small-scale conditions like clouds, turbulence, and convection. The combination of physics-based simulations and AI seems to resolve this issue, and in 2020, NeuralGCM predicted annual temperature and humidity levels 15-50 percent more accurately than its non-AI counterpart X-SHiELD, and much faster - NeuralGCM generated those predictions in 8 minutes compared with 20 days for X-SHiELD. Google claims NeuralGCM can operate on a single AI chip, or TPU (Tensor Processing Unit), while some high-resolution atmospheric models require access to expensive supercomputers, and thousands of chips called CPUs (Central Processing Unit). This means that the model will be accessible on laptops and to researchers worldwide. Better efficiency could also reduce the model's energy consumption. The news comes weeks after Google's own environmental report concluded that the company's use of energy-devouring AI over the past five years has increased its emissions by nearly 50 percent.

Mixed AI/physics forecast model handles both weather and a bit of climate

Google/academic project is great with weather, has some limits for climate. Right now, the world's best weather forecast model is a General Circulation Model, or GCM, put together by the European Center for Medium-Range Weather Forecasts. A GCM is in part based on code that calculates the physics of various atmospheric processes that we understand well. For a lot of the rest, GCMs rely on what's termed "parameterization," which attempts to use empirically determined relationships to approximate what's going on with processes where we don't fully understand the physics. Lately, GCMs have faced some competition from machine-learning techniques, which train AI systems to recognize patterns in meteorological data and use those to predict the conditions that will result over the next few days. Their forecasts, however, tend to get a bit vague after more than a few days and can't deal with the sort of long-term factors that need to be considered when GCMs are used to study climate change. On Monday, a team from Google's AI group and the European Centre for Medium-Range Weather Forecasts are announcing NeuralGCM, a system that mixes physics-based atmospheric circulation with AI parameterization of other meteorological influences. Neural GCM is computationally efficient and performs very well in weather forecast benchmarks. Strikingly, it can also produce reasonable looking output for runs that cover decades, potentially allowing it to address some climate-relevant questions. While it can't handle a lot of what we use climate models for, there are some obvious routes for potential improvements. Meet NeuralGCM NeuralGCM is a two-part system. There's what the researchers term a "dynamical core," which handles the physics of large-scale atmospheric convection and takes into account basic physics like gravity and thermodynamics. Everything else is handled by the AI portion. "It's everything that's not in the equations of fluid dynamics," said Google's Stephan Hoyer. "So that means clouds, rainfall, solar radiation, drag across the surface of the Earth -- also all the residual terms in the equations that happen below the grid scale of about roughly 100 kilometers or so." It's what you might call a monolithic AI. Rather than training individual modules that handle a single process, such as cloud formation, the AI portion is trained to deal with everything at once. Critically, the whole system is trained concurrently, rather than training the AI separately from the physics core. Initially, performance evaluations and updates to the neural network were performed at six-hour intervals, since the system isn't very stable until at least partially trained. Over time, those are stretched out to five days. The result is a system that's competitive with the best available for forecasts running out to 10 days, often exceeding the competition depending on the precise measure used (in addition to weather forecasting benchmarks, the researchers looked at features like tropical cyclones, atmospheric rivers, and the Intertropical Convergence Zone). On the longer forecasts, it tended to produce features that were less blurry than those made by pure AI forecasters, even though it was operating at a lower resolution than they were. This lower resolution means larger grid squares -- the surface of the Earth is divided up into individual squares for computational purposes -- than most other models, which cuts down significantly on its computing requirements. Despite its success with weather, there were a couple of major caveats. One is that NeuralGCM tended to underestimate extreme events occurring in the tropics. The second is that it doesn't actually model precipitation; instead, it calculates the balance between evaporation and precipitation. But it also comes with some specific advantages over some other short-term forecast models, key among them being that it isn't actually limited to running over the short term. The researchers let it run for up to two years, and it successfully reproduced a reasonable-looking seasonal cycle, including large-scale features of the atmospheric circulation. Other long-duration runs show that it can produce appropriate counts of tropical cyclones, which go on to follow trajectories that reflect patterns seen in the real world.