Digital Twin for Climate Extremes Detection and Characterization using Deep Learning

Digital Twin for Climate Extremes Detection and Characterization using Deep Learning Christian Pagé (CERFACS) EGI2025 Digital Twins in Action: Scalable Models for Science 2 Generic detection algorithm: ○Intense rainfall ○Drought ○Heatwave ○Cold spell ○High wind Characterization (What-if Scenarios): ○Frequency of occurrence ○Spatial extent ○Intensity (if relevant) ○Duration Pictures: The Guardian, World Meteorological Organization, WMO, Patch, Direct Energy Different methods for one-time events (intense rainfall, high wind) vs. longer-term events (droughts,...) The Challenge: Climate Extremes Deep Learning Method xtclim python package 4 Data… 5 Extreme Workflow: the User Perspective 6 Integration with interTwin DTE •Integration with itwinai and DTE Core Components (unlocks additional functionalities) •Extensive Logging •Hyper-Parameters' Optimization •Containerization •Advanced workflow composition & execution: yaml 7 SQAaas: Software Quality Assessments Software and Data quality validation platform (1) exploiting DevOps, serving as basis for Model Validation https://sqaaas.eosc-synergy.eu (1) Software Quality Assurance as a Service and FAIR-eva, by CSIC, LIP and UPV OSCAR by UPV 8 Integration with interTwin DTE Take-Home Messages •interTwin DTE enables fast Digital Twin Application Development and Exploitation •The Convolutional Variational Auto-Encoder (CVAE) achieves Unsupervised Anomaly Detection •Events can be characterized with various indicators •Results are consistent •High Performance of this approach due to ML unlocks the ability to better quantify climate change impact uncertainties Christian Pagé [email protected] @pagechristian