Answers are grounded in this paper and cite their sources.
• Ensembling adds limited value to physics-aware Machine Learning (PaML) models. • Single well-configured PaML matches ensemble performance • Classical boosting compensates low-capacity PaML learners. • Bagging provides limited additional predictive benefit. • GrowNet-inspired boosting shows non-monotonic behavior. Ensemble machine learning methods such as bagging and boosting are widely used to improve predictive performance in hydrological modeling. However, their value when applied to physics-aware machine learning (PaML) models remains largely unexplored. This study investigates whether ensemble strategies provide tangible predictive gains over a standalone PaML model for streamflow simulation. Four controlled experiments were conducted on five representative catchments from the Catchment Attributes and Meteorology for Large-sample Studies dataset for the contiguous United States (CAMELS-US) to evaluate the influence of training epochs, base learner capacity, ensemble size, and training data length. Bagging, classical boosting, and a growth-based boosting formulation inspired by GrowNet were compared against both weak and well-configured single PaML models. Predictive performance was assessed using the Nash-Sutcliffe efficiency (NSE), the Kling-Gupta efficiency (KGE) and the normalized root mean square error (NRMSE). Results show that ensemble strategies do not yield substantial improvements over a properly configured standalone PaML model, despite requiring greater computational resources. While classical boosting can compensate for intentionally reduced model capacity by aggregating low-capacity PaML learners, the resulting gains remain marginal relative to a moderately expressive single model. Bagging provides limited additional benefits, and the growth-based boosting formulation exhibits non-monotonic and basin-dependent behavior. Overall, the findings suggest that, within the investigated framework, refining the architecture and capacity of a single PaML model is generally more effective than adopting ensemble strategies. These results contribute to clarifying the practical role of ensemble learning in physics-aware hydrological modeling.
No takes yet. Share an insight, caveat, or question.
Adoubi Vincent De Paul Adombi (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: