One of the hardest variables to predict in weather is rainfall, which is the main source of uncertainty in flood prediction. However, this new AI system from Google skips the rainfall prediction entirely and instead tries to predict floods from everything else that is known.

The key challenge is that the countries that need these flood predictions the most are often the places that don't have enough training data for the learning algorithm, as they lack the means to collect it. This new technique helps reuse data from the USA and other countries and apply it to less fortunate regions.

Remarkably, this new AI system is more accurate than the Global Flood Awareness Systems used around the world. This is an incredible advancement that can greatly help non-advanced nations, not just their governments but citizens as well. Some of these flood prediction models are even cheap enough to run on a smartphone in your pocket.

Google has developed a novel approach to weather prediction that outperforms traditional physics-based simulations. Instead of relying on computationally expensive weather models, they use diffusion-based AI models to generate plausible weather data from historical observations.

The key insight is that diffusion models, which are used to create images from text prompts, can also be applied to weather data. These models start from random noise and gradually transform it into realistic weather patterns, learning the intricacies of extreme weather events in the process.

This approach has several advantages over traditional methods. First, it requires far less computational power, as it does not need to run complex weather simulations. Second, it can leverage a much larger dataset of historical weather observations, allowing the AI to learn more comprehensive patterns.

The result is a weather prediction system that is more accurate than previous techniques, including NVIDIA's FourCastNet and DeepMind's GraphCast. This breakthrough has the potential to significantly improve weather forecasting, especially in regions with limited data and computational resources, ultimately saving many lives by better predicting extreme weather events.

Aircraft exhaust lines, known as contrails, can have a measurable impact on the planet's temperature. While only a small percentage of contrails (around 5%) have a long-lasting effect, this impact is significant. To address this issue, researchers have developed an AI-based simulation that can predict which planes on which routes are likely to create these persistent contrails.

The challenge lies in accurately distinguishing contrails from similar-looking cirrus clouds. However, the researchers have found a reliable way to differentiate between the two, making the problem learnable for the AI system.

To test the effectiveness of this approach, the researchers conducted a trial with American Airlines, where they proposed minor route adjustments based on the AI's predictions. The result was a 54% reduction in the creation of heat-trapping contrails, with only a 0.3% increase in fuel usage. This net effect is 20 times better for the environment than the current approach.

By intentionally avoiding the regions where persistent contrails are likely to form, airlines can significantly reduce their impact on the planet's temperature, with minimal impact on their operations. This innovative use of AI technology has the potential to make air travel more sustainable and contribute to the fight against climate change.