Scientific scope and citation =============================== GeoPrior-v3 is a physics-guided AI framework for geohazard modelling. In the current documentation, its most fully developed scientific application is **urban land subsidence forecasting** through ``GeoPriorSubsNet``. Why this project exists ----------------------- Urban land subsidence is a widespread geohazard that threatens infrastructure, groundwater security, and coastal adaptation. It is well documented in rapidly urbanising and groundwater-stressed regions, including Jakarta, central Mexico, Iran, and many vulnerable coastal settings worldwide :cite:p:`Abidinetal2011,Chaussardetal2014, Motaghetal2008,GallowayBurbey2011,Shirzaeietal2021, Nichollsetal2021,Fangetal2022`. The scientific challenge is not only to detect subsidence, but also to understand and forecast it under multiple interacting pressures. Groundwater extraction, shallow anthropogenic loading, stratigraphic heterogeneity, and coastal exposure can all interact in ways that are difficult to separate in practice :cite:p:`CandelaKoster2022`. The modelling gap ----------------- Traditional physics-based simulators can provide strong mechanistic detail, but they are expensive to parameterise and difficult to deploy consistently over dense, city-scale geospatial archives. Purely data-driven machine-learning models can capture correlations well, but they often lack explicit physical accountability and may extrapolate in ways that are difficult to audit for hazard management :cite:p:`Liuetal2024,Limetal2021`. GeoPrior-v3 is designed to address this gap: it aims to support forecasting workflows that are not only predictive, but also physically interpretable, diagnostically auditable, and usable for risk-aware decision support. Why Nansha and Zhongshan matter ------------------------------- The current land-subsidence application is built around two contrasting urban regions in the Pearl River Delta: **Nansha** and **Zhongshan**. This contrast is scientifically useful because the two settings do not represent the same deformation regime. In the current manuscript, Nansha is treated as a reclaimed coastal delta with broader groundwater variability and more heterogeneous compressible sediments, whereas Zhongshan is presented as a more competent stratigraphic setting with clearer spatial organisation in effective hydrogeological structure. The two-city design therefore does more than provide two examples. It tests whether one physics-guided framework can remain useful across distinct subsidence regimes rather than only within a single local basin :cite:p:`kouadio_geopriorsubsnet_nature_2025`. What GeoPrior-v3 is ------------------- GeoPrior-v3 is the broader software and documentation umbrella for physics-guided AI workflows in geohazards. It is broader than a single urban land-subsidence problem. The present documentation emphasises reproducible workflows, diagnostics, uncertainty analysis, transfer evaluation, and interpretable physics-aware modelling. What GeoPriorSubsNet brings --------------------------- ``GeoPriorSubsNet`` is the current application-specific model for urban land subsidence. It combines an attentive spatio-temporal forecasting backbone with reduced groundwater-flow and consolidation constraints, while regularising the emergent relaxation timescale through a closure based on effective hydraulic conductivity, specific storage, and drainage thickness. The goal is not only strong multi-horizon prediction. The framework is also intended to deliver: - physically auditable forecasts, - calibrated uncertainty, - interpretable effective fields, - transfer-aware deployment diagnostics, - and hotspot-oriented risk interpretation :cite:p:`kouadio_geopriorsubsnet_nature_2025`. In that sense, GeoPrior is not only a model. It is a workflow for linking geospatial data, reduced physics, uncertainty analysis, diagnostics, and decision-facing interpretation in a single reproducible framework :cite:p:`kouadio_geopriorsubsnet_nature_2025`. How to cite the current land-subsidence application --------------------------------------------------- If you use GeoPrior's current land-subsidence framework, the GeoPriorSubsNet scientific concepts, or the Nansha/Zhongshan case study, please cite the manuscript below. .. admonition:: Preferred manuscript citation :class: note Kouadio, K. L., Liu, R., Jiang, S., Liu, Z., Kouamelan, S., Liu, W., Qing, Z., and Zheng, Z. (2025). *Physics-Informed Deep Learning Reveals Divergent Urban Land Subsidence Regimes*. Unpublished manuscript. Submitted to *Nature Communications*. BibTeX ------ .. code-block:: bibtex @unpublished{kouadio_geopriorsubsnet_nature_2025, author = {Kouadio, Kouao Laurent and Liu, Rong and Jiang, Shiyu and Liu, Zhuo and Kouamelan, Serge and Liu, Wenxiang and Qing, Zhanhui and Zheng, Zhiwen}, title = {Physics-Informed Deep Learning Reveals Divergent Urban Land Subsidence Regimes}, note = {Unpublished manuscript. Submitted to Nature Communications}, year = {2025} } Related background in this documentation bibliography ----------------------------------------------------- Readers who want a wider scientific background may find the following themes especially useful: - global and coastal subsidence risk :cite:p:`Shirzaeietal2021,Nichollsetal2021,Fangetal2022` - groundwater extraction and regional subsidence :cite:p:`GallowayBurbey2011,Motaghetal2008,Chaussardetal2014, Abidinetal2011` - anthropogenic and multi-driver subsidence interpretation :cite:p:`CandelaKoster2022` - forecasting-model background :cite:p:`Liuetal2024,Limetal2021` See also -------- .. seealso:: :doc:`references` Shared bibliography used across the documentation.