Inflow modeling strategies designed to enhance the ecological sustainability of hydroelectric reservoir operation planning

BACKGROUND AND OBJECTIVES: The rapid global expansion of water infrastructure alters major rivers, causes severe biodiversity declines, and destabilizes aquatic ecosystems. Amid intensifying eco-hydrological pressures, accurate prediction of reservoir inflow remains imperative to safeguard downstream aquatic ecosystems, yet data-sparse tropical environments chronically impair the generalizability of deep learning architectures due to pronounced overfitting. The study objectives were to evaluate three statistical inflow modeling frameworks, representative of design-era hydrological understanding, against the observed ۱۹۹۸–۲۰۲۳ flow regime to identify the most reliable forecasting approach for supporting environmental-flow-informed reservoir operation planning under monsoonal variability.METHODS: Using a multi-decadal hydro-meteorological dataset from ۱۹۹۸ to ۲۰۲۳ (n = ۳۱۲ monthly time steps) at Karian Reservoir in Banten Province, Indonesia, three primary modeling approaches were tested: Autoregressive Integrated Moving Average, Markov Chain, and Multiple Linear Regression. To rigorously assess model diagnostics and predictive performance across historical drought and wet regimes, the Pearson correlation coefficient, Kling-Gupta efficiency, and operational trajectory metrics were integrated.FINDINGS: Deterministic multiple linear regression achieved the highest overall predictive accuracy with a Pearson correlation coefficient of ۰.۷۴, a Kling-Gupta efficiency of ۰.۶۴, and a superior operational trajectory guideline correlation of ۰.۸۵. The observed enhancement in model efficacy stems from the methodical integration of lagged streamflow predictors that effectively characterize catchment-scale moisture persistence. Conversely, the Autoregressive Integrated Moving Average model exhibited the weakest diagnostic capability (Pearson correlation coefficient = ۰.۶۲; Kling-Gupta efficiency = ۰.۶۰) due to systematic overestimation of dry-season baseflows, risking artificial inflation of safe water availability. Meanwhile, the stochastic Markov yielded a commensurate aggregate Kling-Gupta efficiency score of ۰.۶۶ and exhibited a superior proficiency in replicating long-term probability distributions.CONCLUSION: While Multiple Linear Regression shows the strongest performance for monthly operational scheduling, the Markov Chain model offers a complementary probabilistic framework for screening extended seasonal drought risk. Independent out-of-sample validation against a held-out ۲۰۱۹–۲۰۲۳ period revealed substantial performance degradation for both models (Multiple Linear Regression Nash-Sutcliffe efficiency declining from ۰.۶۱ to ۰.۰۲; Markov Chain Nash-sutcliffe efficiency declining from ۰.۲۸ to -۱.۰۱), directly evidencing non-stationary rainfall-inflow relationships. Ultimately, these results establish that the evaluated models serve as an initial decision-support tool for raw water allocation planning, though their sustained operational reliability necessitates systematic recalibration.

کلیدواژه ها:

Environmental flows ، inflow forecasting ، Karian Reservoir ، Markov Chain ، monsoonal tropics ، Multiple linear regression