The London AI forge DeepMind has a knack for good PR. The news that the company was working with the UK’s National Meteorological Service to develop a new, better model for forecasting rainfall has spread widely. But how much progress is really in the “DGMR”, the “Deep Generative Model of Rainfall”?
The new tool from DeepMind cannot be compared with “AlphaFold”, the model that can calculate the three-dimensional structure of proteins based on the amino acid sequence alone – and has thus solved a central problem in biology. But even a small improvement in the rain forecast, especially heavy rain, is of decisive importance for many industries: The spectrum ranges from the organization of outdoor events to aviation and rescue services. However, DeepMind isn’t the first company to try to solve this problem.
Predicting small weather events is a tough test
Existing forecasting techniques use extensive computer simulations of atmospheric physics with resolutions of a few kilometers. They are good for making longer-term predictions. Small weather events such as rain showers or even heavy rain, however, put modern weather models to the test, as they too cannot fully map the complex physics of cloud formation. In order to improve the quality of numerical weather forecasts, meteorologists therefore use modeling ensembles, the results of which produce a statistical distribution.
But for short-term rain forecasts – for example for the next hour – it is not worth it. The computing effort and computing time would be too high. They are therefore less suitable for so-called nowcasting – the weather within the next few hours. This is where machine learning comes into play: As early as the 1990s, researchers were training various algorithms to make predictions from measured weather data – initially with limited success.
In 2019, Google researchers presented their version of a nowcast. The short-term weather forecast for a few hours uses recordings from the rain radar as training data for a neural network. Many countries regularly publish radar readings during the day that show how clouds form and move during the day. In the UK, a new reading is published every five minutes. Putting these snapshots together results in an up-to-date stop-motion video that shows how the rain patterns move across the land, similar to the forecast images on television.
Their approach would do without programmed physical laws and thus without complex weather simulations. It delivers results within a few minutes, emphasizes Google developer Jason Hickey, first author of the publication. Weather forecast has been treated as some kind of complex image recognition problem. to Information from the researchers Hickey writes that the nowcast has proven to be clearly superior to three classic forecast models that are frequently used in the USA for forecasts of “five to six” hours.
DeepRain, DeepMC and other weather projects
The Google researchers were not left alone – other research teams also showed interesting results: In Germany, for example, researchers are working on the joint project Deep rain, which uses deep, neural networks to discover undiscovered complex patterns in rain radar data and to draw conclusions about impending heavy rain events.
And Microsoft is also there: Microsoft researchers recently presented a system on behalf of DeepMC the data from local sensors and conventional weather forecasts such as those of the National Weather Service combined. Based on the local data, the deep, neural network calculates the deviations for the respective location. The company wants to make the system available to farmers, for example, but also to producers of renewable energies.
The DeepMind team also trained their AI on radar data as is reports in nature. In contrast to many of their colleagues, the Deepmind researchers trained a Deep Generative Network – a network similar to the well-known Generative Adversarial Networks, or GANs for short. This type of AI is trained to generate new data patterns that are very similar to the real world data it was trained on.
Create and continue radar images
Usually GANs are used to calculate deepfakes or even a fake Rembrandt, for example. In this case, the DGMR has learned to generate radar images and thus continue the series of actually recorded measurement images. It’s like looking at a few frames from a movie and guessing what’s next, says Shakir Mohamed, who heads the project at DeepMind. The generative network is not only used for forecasting, but also provides – as it were incidentally – a statement about the probability with which this forecast will come true.
For a practical test, the research team asked 56 Met Office meteorologists who were not involved in the research to compare the results of the DGMR with predictions from a state-of-the-art physics simulation and a competing deep learning tool. 89 percent stated that they prefer the results of the DGMR. They took into account a number of factors – including predicting the location, extent, movement, and intensity of the rain.
“Machine learning algorithms usually try to optimize their prediction for a quantity,” says Niall Robinson, Head of Partnerships and Product Innovation at the Met Office. “However, weather forecasts can be good or bad in many ways. Perhaps one forecast predicts precipitation in the right place but at the wrong intensity, or another one calculates the right mix of intensities but in the wrong places, and so on. In this research, we have put a lot of effort into making our algorithm broad. “
Solving problems in games helps DeepMind
The team worked on the project for several years, and the input from the Met Office experts “steered our model development in a different direction than if we had worked alone,” says Suman Ravuri, researcher at DeepMind. The AI researchers want to show that their AIs have a practical use. For Shakir, DGMR is part of the same story as AlphaFold: leveraging years of experience solving difficult problems in games – and now they’re starting to tackle a list of real-world science problems.
“Basically, the DeepMind paper is a very interesting piece of work with impressive results that have been well evaluated,” says Sebastian Lerch from the Karlsruhe Institute of Technology (KIT), who is also working on a neural network that corrects the systematic errors of numerical weather models and local forecasts to enhance. “It is difficult to directly compare the different works like that of Google and Deepmind because they use different data sets and benchmark models.”
“In the short term, these purely data-driven methods are definitely better than physical weather models. But the improvements become less and less the further you look into the future, ”says Lerch. Because the AI approaches have a problem: They do not automatically take into account natural laws such as conservation of energy and mass. His conclusion: “Machine learning will not replace numerical weather models in the foreseeable future.”
Disclaimer: This article is generated from the feed and is not edited by our team.