Abstract
With the rapid advancement of artificial intelligence (AI), research in physical geography is undergoing a profound paradigm shift from traditional empirical statistics toward intelligent prediction and process-based modeling. Leveraging cutting-edge techniques such as deep learning, physics-informed neural networks (PINNs), and multimodal data fusion, substantial progress has been achieved in characterizing complex geographical processes, enhancing the physical consistency of models, and enabling cross-scale system coupling. Meanwhile, the synergistic integration of AI and physical models has promoted a transition from a purely data-driven paradigm to a mechanism-driven framework, ushering geoscience intelligence into a new stage. Nevertheless, challenges, including data scarcity, limited model interpretability, interdisciplinary integration barriers, and ethical and sustainability concerns, continue to constrain further development. This paper systematically reviews the key applications and theoretical evolution of AI in physical geography, critically examines current technical bottlenecks and future directions, and aims to provide theoretical guidance and practical insights for advancing intelligent geographical research and for building more robust and scientifically sound Earth system modeling frameworks.
