欢迎访问中国科学院大学学报,今天是

中国科学院大学学报 ›› 2025, Vol. 42 ›› Issue (2): 209-220.DOI: 10.7523/j.ucas.2023.065

• 地质与地球科学 • 上一篇    

S型和I型锆石的机器学习划分及其在超大陆演化中的应用

孙之晗, 张毅刚   

  1. 中国科学院大学地球与行星科学学院 中国科学院计算地球动力学重点实验室, 北京 100049
  • 收稿日期:2023-03-06 修回日期:2023-05-25 发布日期:2023-05-25
  • 通讯作者: 张毅刚,E-mail:zhangyigang@ucas.ac.cn
  • 基金资助:
    国家重点研发计划(Y919017)资助

Classification of S- and I-type detrital zircon by machine learning and its application to supercontinental evolution

SUN Zhihan, ZHANG Yigang   

  1. CAS Key Laboratory of Computational Geodynamics, College of Earth and Planetary Sciences, University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2023-03-06 Revised:2023-05-25 Published:2023-05-25

摘要: 使用堆叠思想和框架融合常见的8种机器学习方法,并采用曲线下面积和正确率指标, 建立碎屑锆石的S(sedimentary)型和I(igneous)型分类模型。将该模型应用于碎屑锆石数据集,获得S型和I型锆石随时间的分布。与古地磁和地质记录对比显示,S型锆石的年龄峰同时对应着上一个超大陆裂解的终点和下一个超大陆聚合的起点,S型锆石的年龄谷(也是I型锆石的小年龄峰)对应超大陆最聚合的状态和其裂解的起始。根据S型锆石年龄峰与整体锆石大年龄峰,以及S型锆石的年龄谷与整体锆石小年龄峰的对应关系,提出整体锆石随时间分布图上的大年龄峰代表板块比较离散的状态,这时岩浆活动多,I型和S型花岗岩均有产出,板块移动速度快;而小年龄峰代表板块比较聚合的状态,这时板块比较稳定,岩浆活动少,产出以I型为主,板块移动速度慢。最后,给出一个更高准确率的判断S型和I型的决策函数和分类图解,可直接应用于相关研究工作。

关键词: 机器学习, S型和I型碎屑锆石, 超大陆演化, 堆叠, 主成分分析

Abstract: Supercontinent evolution and distribution of detrital zircon with time is a long-term hot research topic. By using the stacking framework involving eight different machine learning methods and the area under curve (AUC) and accuracy proxy, a model is established to classify S- and I-type zircons. Applying the model to global detrital zircon dataset gives the distribution of S- and I-type zircon with time. After comparing the distribution with paleomagnetism and geological records, it is found that the S-type zircon distribution peak corresponds to the end of a supercontinent breakup and the start of assembly of the next supercontinent, and that the S-type zircon distribution valley (also the small peak of I-type zircon) is related to the maximum packing of a supercontinent and the start of its breakup. Based on the correlation of S-type zircon peak with global zircon big peak and the valley of S-type zircon with the global zircon small peak, it is proposed that big peaks of global zircon distribution with time represent a dispersive state of continents, during which magmatic activity is high producing both I- and S-type granites with also a high velocity of continent movement. By comparison, the small peaks in global zircon distribution represent a packing state of continents during which the supercontinent is stable with low magmatic activity producing mainly I-type granites and with a low velocity of continent movement. Finally, a high-accuracy decision function is provided to judge S- and I-type zircons and can be applied in related studies.

Key words: machine learning, S- and I-type detrital zircon, supercontinent cycle, stacking, primary component analysis

中图分类号: