Quantum-enhanced feature engineering for hurricane rapid intensification prediction via classical shadow protocols

Predicting rapid intensification (RI) in tropical cyclones, defined as a wind speed increase of ≥30 knots within 24 hours, remains one of the most consequential unsolved problems in operational forecasting. RI events are rare, emerge from complex ocean-atmosphere interactions, and produce training datasets that are small, noisy, and heavily class-imbalanced. These properties make RI prediction an ideal candidate for quantum-assisted machine learning, where the goal is not to replace classical models but to extract richer predictive structure from limited data.

We present a hybrid quantum-classical framework for hurricane RI prediction applied to Atlantic storm records accessed via the HURDAT2 dataset using the tropycal Python package, focusing on storms threatening the Hampton Roads, Virginia region. Our approach is grounded in the Learning Under Quantum Privileged Information (LUQPI) framework where meteorological variables (e.g., sea surface temperature, vertical wind shear, ocean heat content, and minimum central pressure) are encoded into parameterized quantum circuits via angle and IQP encoding strategies. Classical shadow protocols are then applied to efficiently extract Pauli-string expectation values as quantum-derived features, requiring only O(log M) measurements to estimate M observables. These features are concatenated with original meteorological inputs and used to train classical classifiers (SVM, XGBoost, neural networks), producing a pipeline that requires quantum hardware only at training time and deploys fully classically.

Our approach builds directly on prior work where quantum feature engineering and shallow parameterized quantum circuits outperformed classical baselines on a similarly imbalanced tornado intensity dataset, with the most significant gains observed on rare, high-intensity events. We present preliminary results extending this framework to hurricane RI classification, discuss the impact of circuit design choices on forecast skill at operationally relevant intensity thresholds, and outline a pathway for integrating quantum-assisted forecasts into emergency management workflows for coastal communities facing disproportionate climate risk.