Hybrid variational quantum-classical data assimilation for numerical weather prediction using Lorenz system...

Area of Science:

• Quantum Computing
• Computational Fluid Dynamics
• Atmospheric Science

Background:

• Numerical Weather Prediction (NWP) faces computational bottlenecks in data assimilation.
• Variational cost functions in NWP are high-dimensional and computationally expensive.
• Existing methods struggle with scaling computational costs alongside model resolution.

Purpose of the Study:

• To empirically evaluate a hybrid quantum-classical framework for Four-Dimensional Variational Data Assimilation (4DVAR).
• To benchmark this framework against classical optimization and deep learning methods.
• To assess the performance and reliability of quantum-classical approaches in atmospheric modeling.

Main Methods:

• Developed and tested an Adaptive Hybrid Variational Quantum-Classical (VQE) framework.
• Benchmarked against Classical BFGS optimization and deep learning models (LSTM, DA-LSTM, Neural ODE).
• Utilized PennyLane quantum simulation on Lorenz-1963 and Lorenz-1996 atmospheric models with rigorous statistical validation.

Main Results:

• The Hybrid QC-HEA achieved near-classical accuracy on L63 and L96 models, with statistically insignificant differences from classical 4DVAR.
• Hybrid QC-HEA significantly outperformed all deep learning baselines on the L96 model.
• The hybrid approach demonstrated superior operational reliability and tighter cost distribution compared to deep learning methods.

Conclusions:

• Hybrid quantum-classical data assimilation is a viable proof-of-concept for realistic atmospheric models.
• The classical refinement step is crucial for practical quantum-classical data assimilation.
• This framework provides a reproducible benchmark for future research on near-term quantum devices.