Date: 9th May 2025, Friday (2:00pm – 3:00pm)
Presenter: Jing Wang (Tianjin Institute of Meteorological Sciences)
Topic: Changes, mechanisms, and future projections of extreme climate events in eastern China

Abstract:
Changing patterns, formation mechanisms, and future projections of extreme climate events in key urban clusters in eastern China are systematically examined, including the Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), and Pearl River Delta (PRD) regions. Specifically, the research focuses on the following: 1) autumn and winter haze pollution in the BTH region, 2) compound heat and drought events in North China and the YRD during the summer, and 3) extreme rainfall and flooding events in the middle and lower Yangtze River basin, as well as in the PRD urban cluster. The mechanisms underlying these extreme events are explored from the perspective of atmospheric circulation dynamics and external forcings, such as sea surface temperature (SST) anomalies and sea ice variations. Atmospheric circulation changes, including the intensification or weakening of the East Asian monsoon and the shifting position of the westerlies, are identified as critical drivers of extreme climate events. In particular, the frequency and intensity of haze events in the BTH region are influenced by stagnation of high-pressure systems, local emissions, and regional meteorological patterns. For heat and drought events in North China and the YRD, changes in the Asian monsoon, combined with reduced rainfall and increased evaporation, are pivotal factors. In the PRD and Yangtze River regions, the occurrence of extreme precipitation is largely driven by changes in the South Asian monsoon and the strengthening of low-pressure systems, often influenced by ocean-atmosphere interactions such as El Niño and La Niña phenomena. Using projections from the Coupled Model Intercomparison Project Phase 6 (CMIP6), future climate scenarios indicate that extreme events in these regions will intensify under various global warming scenarios. For the BTH region, haze pollution is expected to become more severe in the autumn and winter months due to a combination of higher temperatures, reduced rainfall, and altered wind patterns. In North China and the YRD, summer heatwaves and droughts will likely become more frequent and intense, with models suggesting a decrease in precipitation and an increase in evaporation. Overall, our studies highlight the need for targeted adaptation strategies and mitigation measures to address the increasing frequency and severity of extreme climate events in eastern China. Climate model projections underscore the urgency of understanding and preparing for future climate change impacts in this highly populated and economically significant region.

Speaker Profiles:
Dr. Jing Wang is a researcher at the Tianjin Institute of Meteorological Sciences, specializing in climate change and urban meteorology. His research focuses on the impacts of extreme climate events, such as haze pollution, heatwaves, droughts, and extreme precipitation, on urban areas in eastern China, including the Beijing-Tianjin-Hebei, Yangtze River Delta, and Pearl River Delta regions. Dr. Wang investigates the formation mechanisms of these extreme events through atmospheric circulation dynamics and external forcings like sea surface temperature anomalies and sea ice variations. He also uses CMIP6 projections to assess future climate scenarios and their effects on urban climate risks. Dr. Wang has hosted and participated in six provincial-level research projects on climate change and urban meteorology. He has published over 30 SCI-indexed papers in leading journals, contributing significantly to the understanding of climate change impacts on urban environments.

Date: 13th May 2025, Tuesday (11:00am – 12:00pm)
Presenter: Oliver Watt-Meyer (Allen Institute for Artificial Intelligence)
Topic: The Ai2 Climate Emulator (ACE): Capabilities, Challenges and Opportunities

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. ACE is trained on either a global atmospheric model (AGCM)’s output or on observational reanalysis. It has a 1° horizontal grid with eight vertical layers and 6-hourly temporal resolution. The choice of variables predicted assists in climate interpretability and enables the enforcement of mass and energy conservation constraints. The most recent version, ACE2, simulates about 1500 years per day on a single NVIDIA H100 GPU. When forced by realistic insolation, atmospheric CO2 concentration and specified sea-surface temperature, ACE2 accurately emulates climate trends and ENSO-related interannual variability over the period 1940-2020. When coupled to a slab ocean model, ACE2 accurately emulates the equilibrium climate sensitivity of a similarly coupled AGCM to CO2 change, including spatial patterns of surface temperature and precipitation response. However, comparing ACE2 with out-of-sample transient climate change simulations exposes remaining challenges with radiative forcing and energy conservation. In addition to ACE’s utility as a fast emulator of existing climate models, it is envisioned to also provide predictions at the km-scale necessary for stakeholders. Initial work has trained diffusion-based downscaling models for surface precipitation rate on a global 3-km resolution simulation with NOAA’s X-SHiELD model. Due to ACE’s speed, it will be feasible to combine global climate simulation with downscaling over a user’s region of interest in a single workflow.

Speaker Profiles:
Dr. Oliver Watt-Meyer is a Lead Research Scientist in the Climate Modeling group at the Allen Institute for Artificial Intelligence (Ai2). His research interests are in atmospheric and climate dynamics as well as the application of machine learning to climate prediction. He actively works on developing fast, accurate and easy-to-use climate model emulators using AI. In 2016 he received his PhD from the University of Toronto, focused on the topic of stratosphere-troposphere coupling. From 2016 to 2019, he was a NOAA Climate and Global Change and NSERC Postdoctoral Fellow at the University of Washington Department of Atmospheric Sciences.

Date: 20th May 2025, Tuesday (11:00am – 12:00pm)
Presenter: Yilin Zhang (NTU)
Topic: Water Stable Isotopes as Tracers for Hydroclimate Processes in Southeast Asia

Abstract:
Stable isotopes in precipitation are powerful tracers for studying the hydrological cycle and climate variability. However, the physical processes controlling precipitation isotopes, especially in the tropics, remain insufficiently understood, limiting their use as climate proxies. Our research investigates the control mechanisms of precipitation stable isotopes in Southeast Asia, focusing on δ¹⁸O, d-excess, and ¹⁷O-excess, to improve their application as tracers for hydrological and climatic processes. High-resolution rain sampling during Northeast monsoon cold surge events in Singapore reveals that upstream convection and cloud microphysical processes, such as rain evaporation, significantly influence precipitation isotope compositions. These processes challenge the interpretation of ¹⁷O-excess as a direct tracer of moisture source conditions. Moreover, using the isotope-enabled GISS-E2.1 GCM, we evaluate its skill in simulating global and regional precipitation isotopes. While the model captures the spatial-temporal patterns of δ¹⁸O and d-excess well, it shows limitations in simulating ¹⁷O-excess, reflecting the complexity of convective processes and the need for more observational constraints. A regional-scale analysis combining observations and model simulations across Southeast Asia shows that δ¹⁸O variability is primarily driven by monsoonal circulation, upstream convection, and large-scale climate modes such as ENSO and IOD. Overall, this work highlights the importance of microphysical and convective processes in shaping tropical precipitation isotopes and advances their use in climate model evaluation and paleoclimate reconstruction.

Speaker Profiles:
Yilin Zhang is a PhD candidate at the Asian School of the Environment and Earth Observatory of Singapore, Nanyang Technological University. Her research focuses on the control mechanisms of precipitation stable isotopes in Southeast Asia and isotope-enabled general circulation model (GCM) simulations. Yilin received her BSc in Environmental Earth Systems Science (highest distinction), with a minor in Computing and Data Analysis, from Nanyang Technological University in 2021. She has published first-author papers in scientific journals including Journal of Geophysical Research: Atmospheres and Journal of Advances in Modeling Earth Systems.

Date: 26th May 2025, Monday (11:00am – 12:00pm)
Presenter: Marion Mittermaier (UK Met Office)
Topic: Met Office K-Scale and ML modelling activities: an invitation

Abstract:
This presentation will cover three distinct topics: the K-Scale hierarchy of models, the 5 km trailblazer and how these can feed into the AI4NWP activities in the future. Met Office groups across Science are engaging in activities to explore the benefits of K-scale over global and near-global domains across a range of activities. Met Office is participating in the DYAMOND3 initiative which aims to produce year-long simulations using the DYAMOND protocol. See The DYAMOND Model intercomparison(s) — easy.gems documentation and DYAMOND Initiative — ESiWACE. The Met Office K-Scale hierarchy includes global and tropical channel output from the GAL9 and RAL3.3 configurations, which will also include newer versions of the CoMorph convection scheme. The year-long, free-running simulations will provide unparalleled ability to examine the behaviour of convection, for example, in great detail and depth. Two of the year-long simulations: CoMA9_TBv1p2 and RAL3p3 are now (almost) complete. The second activity relates to the running of 100 15-day 5-km global (and current operational N1280) forecasts as a 5-km trailblazer activity. These are following the DIMOSIC protocol (DIfferent MOdel Same Initial Condition) experiments, where the cases are initialised 3 days apart. These cases span the period September 2020, also for 1 year, and overlap with the DYAMOND runs. Here the focus is specifically on how the synoptic evolution is changed with resolution and how, or if the error growth differs. These case studies have now been kicked off and should complete by mid-June. Thirdly, whilst the evaluation associated with both of these activities, especially the trailblazer, is to answer an operational benefit question, the secondary (and becoming increasingly prominent) question relates to whether these models can be used for creating ML training datasets. The talk is primarily aimed at extending an invitation to interested parties to participate in the evaluation activities. An outline of the evaluation plans will be provided. All the data will be accessible via JASMIN.

Speaker Profiles:
Dr. Marion Mittermaier is Scientist at the UK Met Office. Marion’s work focuses on bridging the gap in diagnostics between regional and global NWP and across time scales, and evaluating the impact of observation uncertainty on verification results. She develops tools to identify systematic errors and maximize forecast usefulness. Marion joined the Met Office in 2004, and has a PhD from the University of Reading, and started her career at the South African Weather Service. In 2015 she was the first women to be awarded the prestigious Met Office LG Groves prize that dates back to WW2, and was co-chair of the WMO Joint Working Group on Forecast Verification Research for the maximum two terms until 2020.

Date: 27th May 2025, Tuesday (11:00am – 12:00pm)
Presenter: Tennessee Leeuwenburg (BoM)
Topic: PyEarthTools: A Framework for Machine Learning in Earth System Science

Abstract:
The software ecosystem for machine learning can be quite confusing due to the many tools, languages and frameworks. Anemoi (from the ECMWF) and PyEarthTools (PET) are two Python frameworks which cater specifically to machine learning for Earth system science. This talk will focus on PET, which has been developed by the Bureau of Meteorology (Australia), with the support of NIWA (New Zealand), the Met Office (United Kingdom) and ACCESS-NRI (Australia). It will review where PET sits in the ecosystem and demonstrate how it can be used to train machine learning models. PET offers a repeatable approach to loading data, converting data into a normal form suitable for machine learning, defining and training neural networks, and then running inference and validation. Multiple tutorial notebooks are available demonstrating its core functionality. PET seeks to support a wide range of model types, not restricted to any particular neural architecture, and can work with a variety of underlying machine learning engines such as PyTorch, TensorFlow and JAX. PET has been tested at the Met Office, at NIWA, and at the NCI (Australia) research facility. It is still a new package, having been first released this year (2025), and is still undergoing significant development. The development team are extending the number of pre-defined data sources and data normalisation routines to address this often-challenging aspect of a machine learning project. More information on the project can be found at https://pyearthtools.readthedocs.io/.

Speaker Profiles:
Tennessee Leeuwenburg is the Team Leader for Data Science and Emerging Technologies at the Bureau of Meteorology (Australia). He is the maintainer of the “scores” Python package (https://scores.readthedocs.io/ ) and the PyEarthTools Python package (https://pyearthtools.readthedocs.io/ ). He is currently conducting research into machine learning applied to gridded nowcasting.