Abstract: Artificial Intelligence (AI) technologies have been extensively applied across a diverse range of disciplinary fields. In the realm of Earth sciences research, traditional data processing methods are frequently confronted with formidable challenges due to the intricate physical processes and massive datasets inherent to the field, making AI an indispensable tool for elevating research efficiency and predictive accuracy. With the rapid advancement of deep learning and neural network technologies, AI methodologies are capable of providing feasible solutions for scenarios featuring data scarcity or high-dimensional complex systems, which remarkably enhances the predictive accuracy of Earth science models and accelerates computational processes. In subfields including seismology, geodynamics, climate modeling and mineral resource exploration, AI technologies enable scientists to extract valuable information from complex datasets and generate more precise predictions in a significantly shortened timeframe through the advanced processing of large-scale geological, geophysical and environmental data. Notably, the proposition of Physics-Informed Neural Networks (PINNs) has further propelled the application of AI in Earth sciences: by integrating data-driven approaches with fundamental physical laws, PINNs offer a novel paradigm for addressing complex problems that cannot be efficiently tackled by traditional simulation methods. Although AI technologies have yielded promising preliminary results in Earth sciences applications, numerous challenges remain to be addressed, such as high-dimensional data processing, AI model interpretability, and the modeling of multi-physics coupling problems. This paper provides a comprehensive review of the applications of AI—especially deep learning and Physics-Informed Neural Networks—in Earth sciences, discusses their respective advantages and existing challenges across various application domains, and prospects the future research directions in this field.