A generalized approach for predicting and mitigating total dissolved gas supersaturation at data-scarce dams

Area of Science:

• Environmental Science
• Hydropower Engineering
• Computational Ecology

Background:

• Total dissolved gas (TDG) supersaturation downstream of dams poses ecological risks, particularly to fish, due to gas bubble disease.
• Existing TDG prediction models are often limited by site-specificity, high data requirements, and a lack of interpretability.

Purpose of the Study:

• To develop TransTDG, a generalized, interpretable, and transferable framework for TDG forecasting and mitigation in data-scarce dam environments.
• To bridge the gap between site-dependent black-box models and complex physics-based simulations for TDG management.

Main Methods:

• Coupling transfer learning (TL) with symbolic regression to create an adaptable and explicit TDG modeling framework.
• Introducing two TL strategies to enhance model adaptability for diverse dam conditions.
• Validating the model on 12 data-limited dams along the lower Columbia and Snake Rivers.

Main Results:

• Achieved a mean R² of 0.93 and RMSE of 1.74% with only 1% local data.
• Maintained high accuracy (R² = 0.80, RMSE = 2.98%) even without upstream TDG measurements, with a mean relative error of 1.0%.
• Demonstrated operational relevance through flow optimization, yielding a 4.7% TDG reduction and 8.5% power output increase.

Conclusions:

• TransTDG offers a transferable and transparent solution for TDG modeling in data-constrained settings, advancing ecological management.
• The framework proves effective across various river systems, with over 95% of predictions showing <5% relative error on additional test dams.
• This approach supports sustainable hydropower operations by enabling effective TDG mitigation and improved energy generation.