A Self-Adaptive Multi-Physics PINN Framework for Bidirectional Coupling of Aerosol Transport, Surface Evolution, and Atmospheric Dynamics

Understanding how airborne particles interact with the environment remains a major challenge in atmospheric science. Most existing models treat aerosol transport, surface deposition, and material degradation as separate processes, often neglecting the dynamic feedback that occurs between them. In reality, however, these processes are strongly interconnected: particles not only move through the atmosphere but also deposit on surfaces, alter their properties, and subsequently influence future transport behavior. In this work, we present a self-adaptive multi-physics framework based on physics-informed neural networks (PINNs) that unifies these traditionally disconnected processes into a single predictive model. The proposed approach simultaneously captures aerosol transport in the atmosphere, particle deposition onto surfaces, and the time-dependent evolution of surface characteristics such as roughness and reactivity. Unlike conventional models, the framework introduces a bidirectional coupling mechanism, where surface changes dynamically influence atmospheric transport, and vice versa. A key novelty of this study lies in the development of a self-adaptive learning strategy that allows the PINN to automatically adjust its internal weighting and constraints based on evolving environmental conditions. This enables the model to remain stable and accurate under highly nonlinear and transient scenarios, such as fluctuating pollution levels and thermal gradients. Furthermore, the integration of surface evolution physics into atmospheric modeling represents a significant conceptual advancement, moving beyond static boundary assumptions toward a fully interactive system representation. The results demonstrate that the proposed framework can capture complex aerosol–surface–atmosphere interactions that are not accessible through conventional modeling approaches. By explicitly accounting for feedback mechanisms and multi-scale coupling, this work provides a new perspective on air quality dynamics and environmental degradation. Overall, this study establishes a novel interdisciplinary pathway that connects atmospheric science, surface engineering, and physics-guided artificial intelligence, with potential applications in pollution forecasting, urban environmental management, and the design of resilient material systems exposed to atmospheric conditions.