Impact/context

Climate change has intensified the frequency and severity of extreme weather events, increasing risks to communities and critical infrastructure worldwide. Among these, typhoons are particularly destructive: their strong winds, heavy rainfall, and resulting floods and landslides can cause widespread human and economic losses. In regions like Taiwan, with its steep and mountainous terrain, the impact is especially severe. The island experiences on average around 3.5 typhoons and dozens of torrential rainstorms each year, leading to an estimated 374.3 million euros in annual economic losses due to infrastructure damage, agricultural losses, and disruptions to economic activity. These realities highlight the urgent need for more accurate and efficient forecasting models to support disaster preparedness and resilience.

How quantum could help

Traditional classical methods need extensive computational resources to process vast amounts of meteorological data, where current numerical weather prediction (NWP) models rely on large-scale supercomputing infrastructure, which is costly and energy-intensive. As the scale and complexity of these models grow, alternative solutions are necessary. Machine learning methods also suffer from computationally demanding resources due to the complexity of atmospheric dynamics and the heavy resource requirements for model training and extensive parameter tuning[1].

A quantum computing solution could help through a Quantum-Enhanced Parameter-Efficient Framework[2,3,4] — a hybrid quantum-classical approach that uses quantum neural networks (QNNs) to generate trainable parameters only during training—avoiding lengthy processes at inference time. Building on the framework’s success across multiple domains, including image classification[5], time-series forecasting[6], flood prediction[7], and large language model (LLM) fine-tuning[3], this application introduces Quantum Parameter Adaptation (QPA) for efficient typhoon forecasting model training[8]. QPA enables parameter-efficient learning while maintaining predictive accuracy, offering a scalable and energy-efficient approach to climate modeling through hybrid quantum-classical methods.

References

1] Ho, Kin Tung Michael, et al. “Quantum computing for climate resilience and sustainability challenges.” 2024 IEEE International Conference on Quantum Computing and Engineering (QCE). Vol. 2. IEEE, 2024.
[2]. Liu, Chen-Yu, Chu-Hsuan Abraham Lin, and Kuan-Cheng Chen. “Quantum-train with tensor network mapping model and distributed circuit ansatz.” ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025.
[3] Liu, Chen-Yu, et al. “A quantum circuit-based compression perspective for parameter-efficient learning.” International Conference on Learning Representations (ICLR) 2024.
[4] Cerezo, Marco, et al. “Variational quantum algorithms.” Nature Reviews Physics 3.9 (2021): 625-644.
[5] Liu, Chen-Yu, et al. “Quantum-train: Rethinking hybrid quantum-classical machine learning in the model compression perspective.” Quantum Machine Intelligence 7.2 (2025): 80.
[6] Hsu, Yu-Chao, et al. “Quantum kernel-based long short-term memory for climate time-series forecasting.” 2025 International Conference on Quantum Communications, Networking, and Computing (QCNC). IEEE, 2025.
[7] Lin, Chu-Hsuan Abraham, Chen-Yu Liu, and Kuan-Cheng Chen. “Quantum-train long short-term memory: Application on flood prediction problem.” 2024 IEEE International Conference on Quantum Computing and Engineering (QCE). Vol. 2. IEEE, 2024.
[8] Liu, Chen-Yu, et al. “Quantum-Enhanced Parameter-Efficient Learning for Typhoon Trajectory Forecasting.” 2025 IEEE International Conference on Quantum Computing and Engineering (QCE). Vol. 1. IEEE, 2025.