Big Data in Earth system science and progress towards a digital twin

Xin Li,Min Feng,Youhua Ran,Yang Su,Feng Liu,Chunlin Huang,Huanfeng Shen,Qing Xiao,Jianbin Su,Shiwei Yuan,Huadong Guo

DOI: https://doi.org/10.1038/s43017-023-00409-w

IF: 37.214

2023-05-03

Nature Reviews Earth & Environment

Abstract:The concept of a digital twin of Earth envisages the convergence of Big Earth Data with physics-based models in an interactive computational framework that enables monitoring and prediction of environmental and social perturbations for use in sustainable governance. Although computational advances are rapidly progressing, digital twins of Earth have not yet been produced. In this Review, we summarize the methodological and cyberinfrastructure advances in Big Data that have advanced the progress towards a digital Earth twin. Data assimilation provides the framework for incorporation of high-resolution observations into Earth system models but lacks the decision-making interface and learning ability needed for the digital twin. Machine learning (and particularly deep learning) in Earth system science is now more capable of reaching the high dimensionality, complexity and nonlinearity of real-life Earth systems and is expanding the learning ability from Big Data. Progress in causal inference and reinforcement learning are, respectively, increasing the interpretability of Big Data and the ability of simulations to solve sequential decision-making problems. Social sensing data could provide inputs for multiagent deep reinforcement learning via feedback loops between agents and the environment, enabling large-scale applications in human system modelling. Future research must focus on finding the optimal way to integrate these individual methodologies to achieve digital twins.

English Else