Fractal-Constrained Deep Learning for Turbulent Flow Reconstruction and Prediction

ABSTRACT

The reconstruction and prediction of high-resolution turbulent flows are important yet challenging. We propose a fractal-constrained deep learning (DL) framework that tackles two key tasks: (1) super-resolution from coarse flow data, and (2) forecasting the future evolution of turbulence. Fractal theory is embedded into the network architecture, enabling the model to explicitly learn the scale-invariant, self-similar structures that characterize turbulence. The proposed framework combines a DL network architecture with a fractal invariance constraint, to quantify and learn self-similar properties across spatial-temporal scales of turbulence. This design yields a physics-data combined that reduces reliance on high-resolution labels, thereby supporting few-data and even data-free learning. Applied to turbulent flows, the models accurately reconstruct fine-scale structures and physic properties, while maintaining high efficiency throughout the training and forecasting stages.

Tropical Cyclone Modelling with Wave-Coupled Processes

ABSTRACT

Wave–atmosphere coupling is increasingly recognized as a key driver of tropical cyclone (TC) intensity and structure. Sea-state-dependent surface roughness, wave-enhanced momentum and heat fluxes, and spray-mediated exchanges can all modify the boundary layer and, in turn, cyclone evolution. In this talk I will introduce a modelling framework that couples a spectral wave model with a regional atmosphere–ocean system to quantify these feedbacks across the TC life cycle. I will discuss (i) how wave age and significant wave height modulate drag and enthalpy transfer; (ii) the role of spray at high wind speeds; and (iii) impacts on storm size, and surge-relevant wave conditions. Using various experiments, I will compare uncoupled vs coupled configurations, highlight skill improvements against available observations, and explore sensitivities to physics choices. I will close with implications for hazard assessment (extreme waves) and forecast operations of TC risk.