Quantum Machine Learning for Hydrological Simulation: A Perspective on Emerging Opportunities and Challenges

ABSTRACT

In recent years, classical machine learning (CML) has significantly advanced hydrologic prediction; however, the statistical foundations of these models impose structural limitations for hydrologic time series simulation. These limitations arise from their reliance on empirical risk minimization, which prioritizes statistical fit over physical constraints. As a result, CML models may exhibit mass conservation violations, sensitivity to hydrologic regime shifts, overfitting, and weak uncertainty quantification. Quantum machine learning (QML) has recently emerged as a promising research direction that can augment CML simulations while leveraging the scalability, data representation, and integration. By leveraging superposition, entanglement, and interference, QML can provide new approaches for representing high‐dimensional hydrologic features, addressing complex optimization problems, and enhancing uncertainty quantification through quantum‐enhanced sampling. Proof‐of‐concept simulations using Quantum Long Short‐Term Memory (QLSTM), Variational Quantum Circuits, and hybrid quantum‐classical approaches at multiple USGS gauging stations demonstrate improved daily streamflow predictions across the continental US (CONUS) relative to classical LSTM. However, these results should be interpreted cautiously because practical deployment remains constrained by noisy intermediate‐scale quantum hardware, finite‐shot effects, data encoding bottlenecks, and limited interpretability. Rather than replacing CML, near‐term progress is likely to depend on physics‐guided hybrid quantum–classical machine learning approaches, in which quantum components augment CML through enhanced feature representations, optimization, and uncertainty quantification while preserving physical consistency. This paper examines the opportunities and current limitations of CML and QML for hydrological simulation, proposes physics‐guided hybrid quantum–classical machine learning as a practical pathway toward quantum‐driven hydrological simulation, and presents a research agenda to guide future research.