Google DeepMind's GenCast predicts weather 15 days in advance more accurately than existing models with a 97.2% probability, providing early warnings for risks like heatwaves and hurricanes through over 50 scenarios.
Imagine this. Suppose you are planning a precious family wedding outdoors two weeks from now, or a large-scale regional festival where tens of thousands of people will gather. Your biggest worry will undoubtedly be the ‘weather.’ You might worry, “What if it rains? It would be dangerous if a sudden heatwave hits…” Until now, the accuracy of weather forecasts has dropped sharply after just one week, making the weather two weeks away essentially a matter of ‘fate.’ But now, Google DeepMind’s new AI weatherman, GenCast, has arrived to ease our concerns.
Why is this important?
Weather forecasting goes beyond simply deciding whether to take an umbrella in the morning. This is because unannounced hurricanes, record-breaking heatwaves, or sudden strong winds cause numerous casualties and enormous economic losses. GenCast has the ability to predict the risks of these extreme weather phenomena with great precision as early as 15 days in advance [Source Title].
In particular, if national or local governments can know the path of a hurricane even a few days earlier, they can secure ‘golden time’ to issue evacuation orders or install protective barriers [Source Title]. Additionally, renewable energy such as solar and wind power is absolutely affected by the weather; planning energy supply through GenCast enables much more efficient operation [Source Title]. In fact, GenCast has been proven to consistently provide higher economic value than existing systems when predicting extreme temperature changes or strong winds [Source Title].
Easy Understanding: GenCast’s Two Secrets
How does GenCast get the weather right so much smarter than previous models? The secret lies in two major innovative technologies.
1. “The wisdom of 50 is better than one dogmatic expert” (Ensemble Model)
Existing weather forecasting models have mainly been deterministic. Simply put, after calculating vast amounts of data, they output a single result, such as “Exactly 0.5mm of rain will fall at 2 PM tomorrow.” It’s similar to a single meteorologist providing an answer while blindly trusting their own calculations [Source Title].
On the other hand, GenCast uses a probabilistic model. This is called an ‘ensemble’ (a method of running multiple prediction models simultaneously and comparing the results), and GenCast simulates more than 50 different scenarios at once for a single forecast [Source Title].
To use an analogy, it’s like asking 50 experts to each predict the future and then synthesizing their opinions. If 45 out of 50 say “There is a high possibility of a typhoon in two weeks,” we get much more multi-dimensional and actionable information. We can even get a glimpse of the ‘worst-case scenario’ that could occur.
2. “Turning a landscape in the fog into a clear high-definition photo” (Diffusion-based Model)
GenCast uses a unique technology called a diffusion-based model [Source Title]. The principle is the same as the technology used by the latest image generation AIs like ‘Midjourney’ or ‘DALL-E.’
Initially, it starts from a state of blurry noise where data is clumped together (an uncertain atmospheric state), but based on the vast amount of historical weather data the AI has learned, it sophisticatedly clears this noise step by step. The process proceeds as if a sculptor were carving a delicate statue from a rough block of stone, eventually completing a precise weather map with a very dense resolution of 0.25 degrees [Source Title].
Current Situation: Surpassing the ‘Ultimate Boss’ of Weather Forecasting
GenCast’s performance has already been officially recognized by being published in the academic journal Nature, the pinnacle of the scientific community [Source Title].
Even more surprising are the actual test results. GenCast went head-to-head with the ENS system of the European Centre for Medium-Range Weather Forecasts (ECMWF), which is currently evaluated as the most accurate in the world, and won by decision [Source Title]. Looking at specific indicators, in 15-day forecast tests, GenCast showed superior accuracy in a whopping 97.2% of cases compared to traditional methods [Source Title].
In essence, artificial intelligence has brilliantly overcome the limitations of ‘Numerical Weather Prediction’ (a method of solving physics laws through computer simulation), which had been the standard for meteorology [Source Title].
What Happens Next?
Instead of complaining that “the weather is so fickle you never know,” we will now live in an era where we ask, “Which of the 50 scenarios presented by AI is the most likely risk?”
Our future changed by GenCast:
- Faster and more accurate warnings: We can drastically reduce casualties by identifying the path of typhoons or hurricanes more than a week earlier [Source Title].
- Innovation in industrial sites: Astronomical costs will be saved in weather-sensitive industries, such as determining the sowing time for crops, optimizing ship routes, and stably operating the power grid [Source Title].
- Systematic risk management: We can protect the safety of citizens through specific probability indicators, such as ‘There is an 85% chance of a critical heatwave occurring,’ rather than just ‘It will be hot’ [Source Title].
GenCast is opening a new horizon in weather forecasting. Now, the weather 15 days later is no longer an area left to ‘luck,’ but an area we can prepare for and manage in advance.
AI Perspective
Weather forecasting is like solving a complex riddle between data and physical laws. In that process, GenCast has fully demonstrated the powerful statistical analysis capabilities of AI that go beyond human limitations. I hope that this technology, which resolves uncertainty through the wise collaborative method of ‘ensembles,’ will become the most reliable shield protecting humanity as we face the major challenge of climate change.
References
- GenCast predicts weather and the risks of extreme conditions with state …
- Probabilistic weather forecasting with machine learning - Nature
- GenCast predicts weather and the risks of extreme conditions with state …
- GenCast: Diffusion-based ensemble forecasting for medium-range weather
- Inside Google’s GenCast: Learn About AI in Weather Forecasting
- Google AI boosts weather accuracy - LinkedIn
- GenCast: New AI model is redefining weather forecasting
- GenCastpredictsweatherandtherisksofextremeconditions…
- GenCastfrom Google DeepMind provides betterweatherforecasts
-
[GoogleGenCast: A New Era in AIWeatherForecasting Communeify](https://www.communeify.com/en/blog/google-gencast-ai-weather-prediction-revolution/) - Google’s DeepMind redefinesweatherforecasting with… - The Watchers
- GenCastpredictsweatherandtherisksofextremecondi…
- Generative Artificial Intelligence and Its Implications forWeatherand…
- Google unveilsGenCastWeatherAI - The Story Thailand
FACT-CHECK SUMMARY
- Claims checked: 19
- Claims verified: 18
- Verdict: PASS
- It provides only one best estimate
- It provides over 50 different prediction scenarios (ensembles)
- It only tells you whether it will rain or not
- 3 days
- 7 days
- 15 days
- It is more accurate about 50% of the time
- It performed better in approximately 97.2% of cases
- Its accuracy is lower than existing models