Artificial Intelligence and Machine Learning for Enhanced Representation of Processes and Extremes in Earth System Models (AI4PEX)

全球变暖继续以惊人的速度持续，给社会带来了前所未有的挑战，迫切需要以科学为主导的缓解和适应。地球系统模型(ESM)是预测气候变化的重要工具，为决策者提供重要信息。然而，对预测气候变化的信心受到一些不确定性的影响；(I)对于大气二氧化碳(CO2)的给定增加(地球的平衡气候敏感度)地球将变暖多少，ESM意见不一；(Ii)排放的二氧化碳将留在大气中以使地球变暖(人类排放的二氧化碳的一半被陆地和海洋吸收)；(Iii)地球系统中有多少多余的热量将进入海洋内部，延迟表面变暖(地球系统中约90%的热量进入海洋)。这些不确定性的核心是对地球系统的反馈认识不足和建模不当，特别是云反馈、碳循环反馈和海洋热量吸收。对这些现象的糟糕描述降低了ESM预测的准确性，并对预测未来的气候极端情况和社会影响产生了影响。我们的目标是改善这些反馈在ESM中的表现，减少全球变暖预测中的不确定性。我们提出了一种多学科的方法，重点是通过将观察与先进的机器学习(ML)和人工智能(AI)相融合，“学习”如何准确描述支撑这些反馈的过程。这些数据和方法受物理定律的约束，将使地球系统模型的精度发生阶段性变化。AI4PEX将使欧洲处于地球系统建模革命的前沿，从而提高气候变化预测的准确性，并为实施《巴黎气候协议》和《欧洲绿色协议》提供更好的支持。

Global warming continues at an alarming rate, presenting unprecedented challenges to society that require urgent, science-led mitigation and adaptation. Earth system models (ESMs) are essential tools for projecting climate change, providing important information to decision makers. However, confidence in predicted climate change is undermined by a number of uncertainties; (i) ESMs disagree on how much the Earth will warm for a given increase in atmospheric carbon dioxide (CO2) (Earth’s equilibrium climate sensitivity); (ii) how much emitted CO2 will stay in the atmosphere to warm the planet (half the CO2 emitted by humans has been absorbed by the land and ocean) and (iii) how much excess heat in the Earth system will enter the ocean interior, delaying surface warming (~90 % of the heat in the Earth system goes into the ocean). Central to these uncertainties are poorly understood, and poorly modelled, Earth system feedbacks, in particular cloud feedbacks, carbon cycle feedbacks and ocean heat uptake. Poor representation of these phenomena degrades the accuracy of ESM projections, with implications for anticipating future climate extremes and societal impacts. We aim to improve the representation of these feedbacks in ESMs, reducing uncertainty in global warming projections. We propose a multidisciplinary approach, focused on “learning” how to accurately describe processes underpinning these feedbacks, through a fusion of observations with advanced machine learning (ML) and artificial intelligence (AI). Such data and approaches, constrained by the laws of physics, will deliver a step change in the accuracy of Earth system models. AI4PEX will place Europe at the forefront of a revolution in Earth system modelling, leading to increased accuracy of climate change projections and superior support for implementation of the Paris Climate Agreement and the European Green Deal.