Enhancing Hydrological Forecasting in the Sabarmati Basin through Hybrid Approaches Integrating Physically-Based and Machine Learning Models

The impacts of climate change on hydrological extremes such as floods and droughts pose significant challenges for sustainable water resource management in semi-arid regions like the Sabarmati Basin, India. Accurate forecasting of these extremes is crucial for improving resilience and supporting decision-making in water resource and emergency management. While physically-based hydrological models have long been instrumental in simulating water cycle processes, their limitations in capturing nonlinearities and biases in observational and climatic data necessitate innovative solutions. In this context, hybrid modeling approaches, which integrate machine learning techniques with physically-based models, present a promising avenue for enhancing hydrological forecasting.

This study investigates the potential of hybrid models to improve the forecasting of hydrological extremes in the Sabarmati Basin across different temporal scales. We integrate outputs from a calibrated Soil and Water Assessment Tool (SWAT) model with advanced machine learning algorithms, such as Long Short-Term Memory (LSTM) networks and Gradient Boosting Machines (GBM). The hybrid framework leverages the strengths of physically-based models in simulating water cycle dynamics and the adaptability of machine learning models in capturing complex, nonlinear relationships and correcting biases.

Key contributions include: (1) the development of a hybrid framework capable of forecasting floods and droughts by combining real-time climate inputs with hydrological outputs, (2) scenario-based assessments using CMIP6 projections to evaluate future hydrological risks under different SSPs, and (3) the analysis of uncertainties and insights into the physical and human-induced processes driving hydrological extremes. Preliminary results demonstrate that the hybrid model improves predictive accuracy for flood peaks and drought indices, reducing forecasting errors compared to standalone models. These advancements have direct implications for operational water resource management and climate adaptation planning in the Sabarmati Basin.

This work contributes to ongoing efforts in hydrological forecasting by highlighting the effectiveness of hybrid approaches in addressing challenges associated with scale, predictability, and uncertainty, while offering a scalable framework for application in similar semi-arid regions globally.