Flood Impact Assessment, LSTM-Based Forecasting, and Revenue Loss Analysis of the Melamchi Water Supply Project

The Melamchi Water Supply Project (MWSP) was launched to address water shortages in Kathmandu Valley but faced significant disruptions due to the catastrophic floods of June 2021. These floods caused 503 non-functional days and revenue losses of NRs. 823.10 million, underscoring systemic vulnerabilities, inadequate disaster preparedness, and insufficient contingency planning. This study evaluates MWSP’s performance by analyzing water discharge trends, forecasting missing data, and estimating financial impacts. A Long Short-Term Memory (LSTM) model was employed to predict water discharge using the recorded data at Sundarijal Treatment Plant by Kathmandu Upatyaka Khanepani Limited (KUKL), achieving a Mean Absolute Percentage Error (MAPE) of 6.82% and a coefficient of determination (R²) of 0.63. However, the model struggled with abrupt changes caused by disasters, leaving 37% of variance unexplained. The findings highlight the need for enhanced disaster resilience, diversified water supply strategies, proactive maintenance, and hybrid forecasting models that integrate machine learning with domain-specific knowledge and external variables. These measures can improve MWSP’s reliability and sustainability, ensuring consistent water supply amidst urbanization and climate-induced risks. This research provides actionable insights for stakeholders to strengthen water management frameworks and support Kathmandu Valley’s long-term water security.