AI for meteorology and climate adaptation

What is AI?

Artificial Intelligence (AI) enables a machine to perform intelligent tasks such as planning, reasoning, predicting, and classifying.

Machine Learning (ML) is a branch of AI in which a computer program learns from existing data to make predictions or perform tasks. For example, an ML program can recognize animals in photos by learning from existing images.

Since 2012, AI has made a huge leap forward thanks to developments in deep learning (artificial neural networks with many layers), along with more powerful computers. With the power of deep learning, technologies such as self-driving cars, virtual assistants (Siri, Google Assistant, Alexa, …), and services based on large language models, such as chatGPT, are getting better and better. AI has more and more practical applications, also in science.

AlphaGo, an AI developed by DeepMind, beats Lee Sedol, the world champion of Go

AI for meteorology

Predicting the weather is challenging because many different time and spatial scales are involved. Numerical weather models solve the physical equations of the atmosphere to make a prediction. They require a lot of computing power and become less reliable for predictions far into the future.

The finer the models, the more accurate the local predictions are, as shown in the figure below. Unfortunately, this also requires much more computing power. An alternative is to refine the models with AI.

The effect of resolution: Refining a temperature forecast from a lower to a higher resolution. In the right figure, you can see much more clearly the local impact on the temperature.

Calculating the urban climate

Cities have a unique climate due to the many buildings and streets. During the day, they store sunlight as energy, which is emitted as heat at night. This urban heat island effect means that cities are warmer at night than the surrounding rural areas. In addition, human activity locally produces extra heat (e.g., heating, cars, etc.).

Predicting the urban climate is complex because traditional weather models usually work on larger scales. The fine urban landscape creates microclimates that can vary greatly within a city, due to local effects such as air pollution and differences in building density. This makes urban predictions challenging