Research Progress and Prospects of Data-Knowledge Coupled Spatiotemporal Prediction Modeling in Geosciences

Objectives: Spatiotemporal prediction plays a crucial role in Earth system science, focusing on forecasting the multiscale future evolution of geospatial processes. With the development of big data, artificial intelligence (AI), and Earth science, spatiotemporal prediction has transitioned from a model-driven approach to a data-driven paradigm and, ultimately, to a data-knowledge coupled paradigm. This paper reviews the development of spatiotemporal prediction models, including statistical, physical, AI, and data-knowledge coupled models, while analyzing their limitations and challenges. This paper aims to review current progress of data-knowledge coupled spatiotemporal prediction modeling in geosciences, highlight the important challenges and provide insights on future development. Methods: This paper provides a systematic review of the evolution of spatiotemporal prediction models in Earth system science, with particular emphasis on the transition from statistical, physical, and AI-based models to data-knowledge coupled models. The paper further elaborates on the basic concepts, theoretical framework, and key technologies involved in data-knowledge coupling. Results: (1) This article proposes a four-in-one data-knowledge coupled spatiotemporal prediction framework of "data empowerment, knowledge guidance, joint drive, and intelligent computing solutions". (2) The review identifies the limitations of traditional spatiotemporal prediction models and emphasizes the advantages of data-knowledge coupled models that integrate both data-driven and knowledge-driven approaches. Key challenges in this field include the logical depth of data-knowledge coupling, the accurate representation of geoscientific processes, and overcoming professional barriers in spatiotemporal prediction modeling. (3) This paper highlights three promising development directions of spatiotemporal prediction in geosciences, including the data-knowledge coupling in the era of large models, the universality of data- knowledge coupled spatiotemporal prediction models, the intelligence of data- knowledge coupled spatiotemporal prediction models. Conclusions: Data-knowledge coupled spatiotemporal prediction models offer significant potential for advancing Earth sciences by enhancing prediction accuracy, interpretability, and robustness. In the future, the development of data- knowledge coupled geoscience spatiotemporal prediction will be more versatile and intelligent, with self-learning, self-evolution, self-adaptation, and self-generation capabilities, forming a highly intelligent prediction system that will demonstrate excellent adaptability and versatility in multiple fields, tasks, and scenarios.