Don't Waste Data: Transfer Learning to Leverage All Data for Machine-Learnt Climate Model Emulation

2022/10/08 by Raghul Parthipan, Parthipan, Raghul, Damon Wischik +1
Earth and Planetary Sciences · Environmental Science · #Chaotic Dynamics (nlin.CD) #Climate variability and models #FOS: Computer and information sciences #FOS: Physical sciences #Hydrological Forecasting Using AI #Machine Learning (cs.LG) #Meteorological Phenomena and Simulations

paper · pdf · doi:10.48550/arxiv.2210.04001

openalex publication_date 2022/10/08 · openalex created_date 2022/10/12 · openalex updated_date 2026/07/28

Abstract

How can we learn from all available data when training machine-learnt climate models, without incurring any extra cost at simulation time? Typically, the training data comprises coarse-grained high-resolution data. But only keeping this coarse-grained data means the rest of the high-resolution data is thrown out. We use a transfer learning approach, which can be applied to a range of machine learning models, to leverage all the high-resolution data. We use three chaotic systems to show it stabilises training, gives improved generalisation performance and results in better forecasting skill. Our code is at https://github.com/raghul-parthipan/dont_waste_data

Related

• Transfer Learning for Emulating Ocean Climate Variability across CO₂ forcing
• Transferring climate change physical knowledge
• Machine Learning Workflows in Climate Modeling: Design Patterns and Insights from Case Studies
• Physics-informed machine learning: case studies for weather and climate modelling
• Statistics of extreme events in coarse-scale climate simulations via machine learning correction operators trained on nudged datasets