09.05.2025

An exciting week in Earth system modeling begins on May 12th, 2025 as researchers from around the globe converge, virtually and physically, for the inaugural Digital Earths Global km-Scale Hackathon, a coordinated effort to accelerate research using regional and global climate simulations at unprecedented kilometer-scale resolution.

Km-scale models (horizontal resolutions < 10km) represent a pivotal change in the way scientists simulate the Earth system. Km-scale models begin to explicitly capture critical phenomena such as moist convection, topographic effects, and coastal interactions. This enables far more physical and accurate simulations of high-impact weather and climate features like tropical cyclones, atmospheric rivers, and extreme precipitation, especially in regions with complex terrain.

But with this leap in fidelity comes a new generation of challenges. The transition to km-scale modeling fundamentally changes not only the science, but the entire ecosystem of climate research. From computation to data access and workflows, we need a full rethinking of how we build and use models. In model development, processes like land hydrology behave differently at fine scales, allowing for new and more physical approaches to coupling between land, atmosphere, ocean, and coasts. Computationally, running these models, especially at global scale, demands resources at the forefront of computing. Even regional simulations produce data at a scale that pushes current infrastructure. In terms of data management, the volume of output makes storage and sharing a major constraint, sometimes more costly than the simulations themselves. Traditional workflows neither work nor flow at this scale, with memory limitations requiring new tools and in-situ analysis methods.

The km-Scale Hackathon addresses these challenges head-on. Hosted across a network of research centers on 5 continents, the hackathon brings over 500 researchers directly to the data, enabling real-time collaboration, parallel analysis, and technology testing. Using shared Jupyter-based platforms and common libraries, teams will analyze 1-year-long global simulations and extended regional runs from the year 2020, chosen for its neutral climate patterns. In addition to the 10 local nodes, special projects sponsored by NVIDIA, ESA, and Destination Earth will be carried out by satellite nodes. These will enable targeted exploration of themes such as AI-enhanced analysis with NVIDIA, EarthCARE data utilization with ESA, and the use of DestinE datasets with ECMWF. 

Research topics proposed by participants include the energetics of tropical rainbelts, precipitation over ice-sheets, triggering of deep convection, EarthCARE-Curtains, MCS tracking, shallow circulations in the ITCZ, mesoscale structure of stratocumulus, global representation of local extremes, Tropical Cyclones, realism of GSRM clouds, air-sea interaction in the tropics, land-atmosphere interactions, urban processes.

The hackathon also aims to test cutting-edge solutions:

  • Advanced Computation: GPU acceleration, AI/ML model emulation, and regionally refined models

  • Data Strategies: In-situ analysis to reduce data movement, hierarchical data formats, and lightweight output strategies

  • Workflow Innovations: Bringing analysis platforms to the data, new parallel tools, and adaptive use of structured and unstructured grids

The broader goal is to advance the capabilities of regional and global modeling groups and help align efforts across institutions, as well as expand the user base of these models. By testing workflows and tools in a shared setting, the event will contribute to more coordinated and scalable approaches to working with km-scale climate data.

Follow the live-blog throughout the week 12-16 May 2025.