Answers are grounded in this paper and cite their sources.
Physics-informed neural networks (PINNs) and other physics-informed machine-learning techniques have proven to be a powerful way to link data-driven inference to governing equations in weather and hydrological modeling. This systematic review brings together the literature on PINN architectures, variants, application areas, governing equations, data sources, validation methods, and operational limitations. Boolean search strings were used in major scholarly databases, and records were filtered based on PRISMA-guided inclusion and exclusion criteria, along with quality-appraisal criteria. The review separates relatively well-established applications, such as groundwater-flow inversion, shallow-water flow, and Saint-Venant flow modeling, from some sparse-observation reconstruction tasks from early proof-of-concept studies in high-dimensional atmospheric forecasting and digital-twin deployment. The synthesis reveals that PINNs can have a positive impact on physical consistency and decrease the need for dense labels, but they have practical limitations due to training instability, stiffness, sensitivity to loss weighting, computational costs, scalability in high-dimensional domains, uncertainty quantification, and sparse field-scale validation. Finally, the study provides conditions under which the use of PINNs is advantageous over purely data-driven neural networks, surrogate models, and conventional numerical solvers, and highlights avenues for operational use in combination with data assimilation, forecasting systems, and hydrometeorological digital twins.
No takes yet. Share an insight, caveat, or question.
Waqas et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: