Kalender

Machine learning and the Rise of Data-Driven Models for Weather and Climate

Tidspunkt

06. mai 2025, 09:00-10:00

Sted

BCCR lecture room 4020, Jahnebakken 5

Abstract

With the modern availability of data and computing resources, machine learning is becoming an important tool in solving problems in many domains, from robotics and medicine to weather and climate. In this talk I will attempt to give an accessible introduction to core machine learning concepts, aimed at beginners in the topic. I’ll walk through ideas like supervised, unsupervised and representation learning and share some examples from my own research. I will end by looking at recent developments in large general-purpose foundation models, such as AURORA, and how they’re being applied in climate forecasting and beyond.

Speaker information

Linus Ericsson is a postdoctoral researcher at the University of Edinburgh, specialising in representation learning, neural architectures, and domain adaptation. His current research explores efficient ways of finding the right neural network structure for the right task and how to make existing networks cheaper to run. Linus has published in top venues such as NeurIPS, CVPR, and the IEEE Signal Processing Magazine. His interests span a wide array of topics, including multimodal learning and responsible AI applications in climate and healthcare.

Flere kalenderoppføringer

Se alle

23.05.25

i2B Invites Workshop on Arctic Past Climates

Local organisers: Stijn De Schepper and Petra Langebroek

19.05.25

BCCR Seminar: “Past Climate from Water Isotopes in a High-Resolution Greenland Ice Core and Speleothem Fluid Inclusions”.

BCCR Monday Seminar will be given by Prof. Ryu Uemura from Nagoya University. He will talk about “Past Climate from Water Isotopes in a High-Resolution Greenland Ice Core and Speleothem Fluid Inclusions”. Abstract This presentation will introduce two recent studies using the stable isotopes of water to investigate past climate variability. The first study presents a high-resolution ice core drilled by a Japanese team from southeastern Greenland. A multi-month age scale was developed by matching oxygen isotope data with isotope-enabled climate model outputs (1871–2020 CE). High snow accumulation (ca. 1 m yr⁻¹) preserved sub-seasonal signals, enabling monthly order dating. The chronology was validated using accumulation rates, hydrogen peroxide, and volcanic signals. The second study analyzes fluid inclusion (liquid water trapped in calcite) isotopes in a stalagmite from Okinawa (Japan) to assess hydroclimate responses to volcanic and solar forcings during the Holocene. The results show that clusters of large volcanic eruptions coincide with cooling phases, suggesting that clusters of volcanic activity during the mid-Holocene likely contributed to climate change at multi-decadal scales via a teleconnection between the subtropical Northwest Pacific and the North Atlantic regions.