Quantifying Very Extreme Precipitation and Temperature Using Huge Ensembles Generated by Machine Learning-based Climate Model Emulators

2025-10-09

AI Summary

Short observational and climate model records hinder reliable quantification of extremely rare extreme precipitation (e.g., Probable Maximum Precipitation, PMP) and heat events. Method: We propose a machine learning–driven climate simulation augmentation framework: an advanced simulator, ACE2, trained on ERA5 reanalysis, generates a high-resolution 10,560-year meteorological dataset; combined with Peaks-over-Threshold (POT) extreme value theory, it enables robust probabilistic and intensity estimation of centennial- to millennial-scale extremes. Contribution/Results: This work is the first to empirically validate the reliability of ML-based simulators under strong extrapolation—far beyond their training distribution—and demonstrates robustness across seasonal cycles and storm types. The approach substantially reduces statistical uncertainty in extreme-value estimation, delivering a practical, physically informed extreme-value statistics framework for risk assessment of critical infrastructure.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior

Application Category

• Web Mining and Content Analysis: Web data generation and simulation
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Weather extremes produce major impacts on society and ecosystems and are likely to change in likelihood and magnitude with climate change. However, very low probability events are hard to characterize statistically using observations or climate model output because of short records/runs. For precipitation, consideration of such events arises in quantifying Probable Maximum Precipitation (PMP), namely estimating extreme precipitation magnitudes for designing and assessing critical infrastructure. A recent National Academies report on modernizing PMP estimation proposed using huge climate model-based ensembles to estimate extreme quantiles, possibly through machine learning-based ensemble boosting. Here we assess such an approach for the contiguous United States using a huge ensemble (10560 years) from a state-of-the-art emulator (ACE2) trained on ERA5 reanalysis. The results indicate that one can practically estimate very extreme precipitation and temperature quantiles using appropriate statistical extreme value techniques. More specifically, the results provide evidence for (1) the use of threshold-exceedance methods with a sufficiently high threshold for reliable estimation (necessary for precipitation), (2) the robustness of results to variations in extremes by season and storm type, and (3) well-constrained statistical uncertainty. Our results also show that the emulator produces extremes outside the range of the ERA5 training data. While this suggests that such emulators have potential for quantifying the climatology of extremes, we do not extensively investigate if this particular emulator is fit for purpose. Our focus is on how to use huge ensembles to estimate very extreme statistics, and we expect the results to be relevant for future improved emulators.

Problem

• Estimating extreme precipitation and temperature quantiles using machine learning emulators
• Addressing statistical challenges in characterizing very low probability weather extremes
• Providing reliable methods for infrastructure design through huge climate ensembles

Innovation

• Machine learning emulators generate huge climate ensembles
• Threshold-exceedance methods estimate extreme precipitation quantiles
• Statistical techniques constrain uncertainty in extreme value analysis