The study of deep geodynamic processes and their surface expressions—such as earthquakes and volcanic activity——is critical for both scientific advancement and societal safety. While traditional geodynamic models focus on fundamental Earth dynamics with limited direct ties to observations, the emergence of data-driven computational models now enables the integration of real-world data into physics-based simulations. By leveraging techniques such as data assimilation and artificial intelligence,these advanced models can invert observations to constrain model parameters and initial conditions, turning vast and diverse datasets into predictive insights. This talk will explore how such computational approaches are transforming our ability to reconstruct past events,analyze present conditions,and forecast future scenarios in geodynamics. Case studies willinclude mantle-lithosphere dynamics, earthquake mechanics, and lava flow behavior. Each example demonstrates how combining numerical simulation with geological, geophysical, and geodetic data enhances our understanding of planetary dynamics while providing practicalsolutions to challenges in geohazard assessment and georesource management. Through these applications, computationalgeodynamics is proving essential for building a more resilient and resource-aware society.

The Precambrian geologic record is characterized by a "long-tailed" data problem: vast volumes of information exist,yet it is often fragmented, sparse, and confined to specific spatiotemporal and disciplinary boundaries.This limits our ability to extract overarching knowledge about early Earth's tectonic regimes and habitability conditions. Big data science, coupled with machine learning (ML)， offers a transformative approach to overcome this challenge by enabling the integration and conjointanalysis ofdisparate,large-scaledatasets.This presentationdemonstratesthis paradigm through a comprehensive case study of the North China Craton (NCC). We integrate extensive rock and mineral geochemistry data and apply methods including local singularity analysisand wavelet analysis to zircon records.Thisreveals an-800-500 Myr periodicity since 3.5 Ga,suggesting NCC's co-evolution with the global supercontinent cycle since the Archean.To investigate the NCC's formation mechanism, we employed an ML-based reconstruction of crustal thickness evolution. The analysis identifies subduction-related arc formation as the primary driver,with secondary contributions from mantle activities. Further spatiotemporal analysis of magmatic intensity and cHf(t) values supports a model of prolonged accretion of arc massifs onto the Eastern Block. Our work exemplifies how big data and ML can unlock the latent potential in fragmented geologic records, providing a new, data-driven framework for interpreting the complex processes that shaped the early Earth.

报告系统阐述了基于数字建模和物理过程的地震灾害预评估所涉及的关键科学问题与数值模拟技术。基于地震动力学破裂相图，构建了一个新的地震分类体系，将地震划分为安全性较高的“自停止型”地震（可进一步分为“慢自停止型”和“扩展停止型”）与破坏性更强的“特征型”地震（包括亚剪切与超剪切地震）。该分类体系为快速识别地震类型、服务震害防御提供了理论依据。研究进一步通过破裂相图，明确了不同类型地震的震级范围及其主控因素（如成核区参数与断层几何构造），揭示了地震破裂的动力学特征。在数值模拟方面，研究以弹性动力学方程与断层本构关系为基础，采用有限差分、边界积分方程及有限元等方法，对震源破裂过程进行模拟，深化了对地震物理机制的理解。上述工作为发展物理意义明确的地震灾害评估方法与推动防震减灾实践提供了重要的科学支撑。

系统梳理了国内外学者的重要贡献，重点介绍了我国自主研发的陆面模式 CoLM（theCommonLand Model)从早期版本到2024版的迭代升级过程。2024版CoLM实现了耦合人类活动的陆面过程精细化建模，具备了全球高分辨率模拟能力，在数值天气预报、气候预测、水文水资源等领域展现出了广泛的应用价值。新版CoLM不仅显著提升了模拟精度，其多尺度适配特性更为相关研究提供了强有力的技术支撑，目前已成功应用于国家天气、气候预测研究与业务实践。

人工智能（AI）改变了天气预报的方式，谷歌、华为和英伟达等科技公司训练的AI模型可在1分钟做出10天的预测，准确性能媲美传统数值模型且计算成本更低。这些“深度学习”AI模型不是求解大气动力方程本身，而是基于过去40年欧洲中期天气预报中心再分析数据的训练。AI模型预报的最大缺陷是很难预测极端天气。事实上极端天气过程都是高度非线性的突变过程，AI模型能很好预测突变前和后的大气状况，却无法精确捕捉和预测大气的突变过程，而大气动力模式却可以较好地预测。未来，精准天气预报需要深度融合人工智能与动力模式的优势。本报告将探讨如何融合动力模式与人工智能（DAI）构建地球系统预测模型，并给出了具体的应用。