Next webinar
Date and Time: Wednesday, 28th of May at 16h BST.
Title: Towards Safe and Fearless Lossy Compression of Weather and Climate Model Data
Presenter: Juniper Tyree, INAR, University of Helsinki
Abstract:
While the output volumes from high-resolution weather and climate models are increasing exponentially, data storage, access, and analysis methods have not kept up. Data size reduction is needed to avoid data storage cost concerns eventually restricting model development and use. Data compression is a vital tool for data size reduction. As lossless compression is no longer sufficient to produce the required compression ratios, lossy compression methods should be applied instead. However, as information loss can be scary - what if important details are lost - lossy compression has not yet been adopted by the weather and climate community.
As part of the EuroHPC ESiWACE, Phase 3, Centre of Excellence, we are working to make lossy compression safe and fearless by bridging the gap between advancements in lossy compression and safety concerns in weather and climate science. In this presentation, Juniper Tyree will provide an overview of the ongoing work on data compression that she leads in ESiWACE3: investigating and collecting safety requirements for lossy compression, benchmarking existing compression methods for performance and safety, and guaranteeing that user-provided safety requirements are upheld by lossy compression.
We believe that trusting lossy data compression requires convincing yourself by trying it out, on your own data, with your own analyses. We have thus developed an open science online laboratory in which anyone can conveniently explore data compression without any setup. We provide this online laboratory as a service to the community to share computational examples, host interactive code documentation and research results, compression-related or not. We will provide a short live demonstration of this laboratory during the presentation.
Past webinars
Title: The Ai2 Climate Emulator: Capabilities, Challenges and Opportunities
Presenter: Oliver Watt-Meyer, AllenAI
Abstract:
Ai2 Climate Emulator (ACE) is a fast machine learning model that simulates global atmospheric variability in a changing climate over time scales ranging from hours to centuries. It generates atmospheric phenomena such as the Madden Julian Oscillation, sudden stratospheric warmings, and tropical cyclones while also accurately simulating 20th century climate trends and El Niño-related interannual variability. Its efficient running 1500 simulated years per day on an NVIDIA H100 will make climate models more accessible and enable the generation of very large ensembles.
Title: Exploring Dataflow Architectures for Improved Efficiency in Earth System Models
Presenter: Justs Zarins, EPCC
Abstract:
Earth system models are crucial for simulating environmental processes but demand significant computational resources and energy. In this presentation we will explore the potential of dataflow architectures to enhance both computational and energy efficiency of ESMs. We will primarily discuss the Cerebras Wafer Scale Engine, examining its capabilities and evaluating its suitability for the shallow water equation.