ClimaX:
A foundation model for weather and climate

Source: arXiv


In this publication, researchers Tung Nguyen, Johannes Brandstetter, Ashish Kapoor, Jayesh K. Gupta, Aditya Grover have developed a weather modeling program called ClimaX. ClimaX is based on a deep neural network architecture. 

 

Using Deep Learning, ClimaX algorithms are trained to predict atmospheric variables with heterogeneous datasets (CMPIP6 and ERA5) covering different variables, spatio-temporal coverage and physical underpinnings. 

 

ClimaX aims to replace existing models (general circulation models - GCM), which are based on physical laws. GCMs use a system of differential equations relating to the flow of energy and matter in the atmosphere, land and oceans, which can be integrated over time to obtain forecasts for the relevant atmospheric variables.

ClimaX is built as a foundation model for any weather and climate modeling task.

Figure 1 - ClimaX is built as a foundation model for any weather and climate modeling task.

The ClimaX model is a versatile model, capable of adapting to the needs of researchers.

The complete work carried out by researchers Tung Nguyen, Johannes Brandstetter, Ashish Kapoor, Jayesh K. Gupta, Aditya Grover to create the model is available in a PDF file (ClimaX: A foundation model for weather and climate.pdf).

The researchers have made the source code of their application available.

Finetuning pipeline for ClimateBench. A different set of input and output variables requires different embedding layers and prediction heads. Attention layers can be frozen or finetuned.

Figure 2 - Finetuning pipeline for ClimateBench.
A different set of input and output variables requires different embedding layers and prediction heads. Attention layers can be frozen or finetuned.


Previous
Previous

Detection of Bow Echoes in Kilometer-Scale Forecasts Using a Convolutional Neural Network

Next
Next

A better way to study ocean currents