A Digital Twin Application: Climate Extremes Detection and Characterization using Deep Learning

A novel approach for a Digital Twin to explore future climate extremes to assess impacts Christian Pagé, CERFACS Use Case Lead Anne Durif, CERFACS Research Engineer EGI2024 Conference, Sep 30 –Oct 2, 2024 Lecce, Italia 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 3 The Challenge: Climate Extremes ●Urgent needs of impact assessments ●Identify mitigation solutions ●Extreme events attribution ●Multiple domains: infrastructures, urban, agriculture, transportation, etc. ●Flexible tools needed for very diverse users 2020 Hurricane Delta causes damage to Louisiana's Gulf Coast 2021 Germany Erftstadt, southwest of Cologne 4 Climate Indices and Indicators - Intra-period extreme temperature range [°C] - ETR - Warm days (days with mean temperature > 90th percentile of daily mean temperature) - TG90p - Summer days (days with max temperature > 25 °C) - SU -… Temperature indices Cold indices Drought indices Rain indices Snow indices Humidity indices Compound indices Standard Indices ECA&D, ET-SCI, ETCCDI, etc. Heat indices Wind indices icclim python package https://github.com/cerfacs-globc/icclim 5 Extreme Workflow: the User Perspective 6 Why use a AI based method? ●To analyze a very large database of climate scenarios with a good performance ●Use efficiently new architectures (GPUs) ●Scalability in cloud-based environments ●Extreme Events spatial structures are similar to avalanches ○Variational Autoencoder: Deep Learning Technique Sinha, Saumya & Giffard-Roisin, Sophie & Karbou, Fatima & Deschatres, Michael & Karas, Anna & Eckert, Nicolas & Coléou, Cécile & Monteleoni, Claire. (2020). Variational Autoencoder Anomaly-Detection of Avalanche Deposits in Satellite SAR Imagery. 113-119. 10.1145/3429309.3429326. Deep Learning Method xtclim python package https://github.com/cerfacs-globc/xtclim 8 Data… 9 Fast comparison between GES