🤖 AI Summary
Weather forecasting faces persistent challenges in accuracy and robustness due to the chaotic and dynamically nonstationary nature of atmospheric systems. To address this, we propose a quantum neural network (QNN)-based approach for meteorological time-series forecasting—specifically for short- to medium-term wind speed and temperature prediction. Our method employs parameterized quantum circuits to instantiate the QNN, trained end-to-end on the NASA POWER real-world meteorological dataset and validated via classical simulation. Experimental results demonstrate that the QNN achieves a 12.7% reduction in mean absolute error (MAE) over classical RNNs for wind speed prediction, accelerates convergence by 3.2× for temperature forecasting, and exhibits superior resilience and stability against abrupt data perturbations. This work establishes a reproducible paradigm and empirical foundation for deploying quantum machine learning in Earth system science.
📝 Abstract
Weather forecasting plays a crucial role in supporting strategic decisions across various sectors, including agriculture, renewable energy production, and disaster management. However, the inherently dynamic and chaotic behavior of the atmosphere presents significant challenges to conventional predictive models. On the other hand, introducing quantum computing simulation techniques to the forecasting problems constitutes a promising alternative to overcome these challenges. In this context, this work explores the emerging intersection between quantum machine learning (QML) and climate forecasting. We present the implementation of a Quantum Neural Network (QNN) trained on real meteorological data from NASA's Prediction of Worldwide Energy Resources (POWER) database. The results show that QNN has the potential to outperform a classical Recurrent Neural Network (RNN) in terms of accuracy and adaptability to abrupt data shifts, particularly in wind speed prediction. Despite observed nonlinearities and architectural sensitivities, the QNN demonstrated robustness in handling temporal variability and faster convergence in temperature prediction. These findings highlight the potential of quantum models in short and medium term climate prediction, while also revealing key challenges and future directions for optimization and broader applicability.