Scientists Create Earth's Most Accurate Digital Twin To Date, Could Revolutionise Climate Modelling

A recent scientific paper has described a near 1-kilometer resolution digital twin of Earth, developed by researchers led by Daniel Klocke of the Max Planck Institute in Germany. The model, which combines weather forecasting and climate modelling, is considered a significant breakthrough in the field. According to Science Alert, the model operates at a 1.25-kilometer resolution, a massive improvement over traditional models that typically have a 40-kilometer resolution.

It uses 672 million calculated cells covering the land, sea, and atmosphere, allowing for unprecedented detail in tracking both rapid weather systems and slower climate processes.

The primary goal is to improve the prediction of environmental changes and their local impacts, such as extreme weather events, floods, and droughts, which was difficult with lower-resolution models.

How did they do it?

This advancement was made possible by utilising immense computational power, specifically 20,480 GH200 superchips, to handle the nearly 1 trillion "degrees of freedom" or values that the model had to calculate.

The original model, written in Fortran, had long been a challenge for anyone attempting to modernise legacy code from before 1990. Over time, it had accumulated layers of add-ons and outdated structures, making it increasingly incompatible with modern computational systems. To address this, the authors turned to a framework called Data-Centric Parallel Programming (DaCe), which restructures data handling to align with the needs and efficiency of today's computing architectures.

Several other organisations are also developing digital twins of Earth or its systems:

Destination Earth (DestinE): A flagship initiative by the European Commission, implemented by the European Space Agency (ESA), the European Centre for Medium-Range Weather Forecasts (ECMWF), and the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), which aims for a full digital replica of Earth by 2030. It has already produced initial high-resolution models focusing on weather extremes and climate change adaptation.

NVIDIA Earth-2: This project, using an AI-driven cloud platform, aims to simulate and visualise climate predictions at a global scale with high precision and speed. It uses generative AI to produce high-resolution forecasts efficiently.

NASA: As part of its long-term observation strategy, NASA is working on its own Earth System Digital Twin to explore various "what-if" scenarios regarding climate and human activity.