Boosting algorithms for projecting streamflow in the Lower Godavari Basin for different climate change scenarios

Area of Science:

• Hydrology and Environmental Modeling
• Machine Learning Applications in Water Resources
• Climate Change Impact Assessment

Background:

• Accurate streamflow simulation is crucial for water resource management, especially in data-scarce regions.
• Boosting algorithms offer powerful machine learning techniques for complex environmental modeling tasks.
• Understanding climate change impacts on river systems requires robust simulation tools.

Purpose of the Study:

• To assess the performance of five boosting algorithms (AdaBoost, CatBoost, LGBoost, NGBoost, XGBoost) in simulating streamflow.
• To project future streamflow under various climate change scenarios (SSPs) using the best-performing algorithms.
• To evaluate the impact of ensembling multiple algorithms on streamflow projection accuracy.

Main Methods:

• Utilized monthly rainfall, temperature, and streamflow data (1982-2020) for training and testing.
• Employed Kling-Gupta Efficiency (KGE) to evaluate algorithm performance.
• Applied selected algorithms for short-term (2025-2050) and long-term (2051-2075) streamflow projections across four Shared Socioeconomic Pathways (SSPs).

Main Results:

• All five boosting algorithms demonstrated good streamflow simulation capabilities, with high KGE values.
• Natural Gradient Boosting (NGBoost) and Light Gradient Boosting (LGBoost) achieved the highest KGE scores.
• NGBoost projected higher streamflow than the historical average for all SSPs, while other algorithms showed varied results.

Conclusions:

• Boosting algorithms are effective tools for streamflow simulation and climate change impact assessment.
• NGBoost shows potential for projecting increased streamflow under future climate change in the Lower Godavari Basin.
• Ensembling techniques may further enhance the reliability of streamflow projections.