From representation to foresight? How 3D urban simulations and deep learning are reshaping theoretical and quantitative geography

The advent of high-resolution 3D urban simulations and transformative deep learning (DL) architectures is precipitating a paradigm shift in theoretical and quantitative geography (TQG). The field lacks a comprehensive synthesis examining how these tools are reconfiguring spatial reasoning, predictive modeling, and the theory–practice interface of geography. This paper addresses this gap by articulating a conceptual and empirical investigation into the transformative interplay between 3D simulation and machine learning. Using a structured analytical framework grounded in material screening, we identify three major findings: (1) the shift from visual representation to experimental spatial laboratories, where simulations become operational epistemic instruments; (2) the emergence of a new predictive geography driven by deep learning’s capacity to learn high-dimensional spatial dependencies; and (3) an ongoing epistemological reconfiguration in which traditional geographical theory is challenged, and sometimes displaced, by data-driven inference, opaque models, and algorithmic urban foresight. We further discuss critical issues related to model interpretability, reproducibility, ethical risks, and epistemic justice in the governance of AI-based spatial knowledge. By situating 3D urban simulation and deep learning within the broader trajectory of TQG, this paper offers a synthesized perspective on the future of the field. It argues that geography now stands at a turning point: either integrate these computational paradigms into a renewed theoretical framework or risk a widening gap between technological innovation and disciplinary meaning-making.