A Singh-Based Fuzzy Logic Framework for Short-Term Discharge and Flood Peak Prediction Under Data-Scarce Hydrological Conditions

Short-term discharge forecasting remains challenging under highly variable, nonlinear, and data-constrained hydrological conditions, where conventional machine learning and deep learning approaches require large datasets and extensive parameterisation, and offer limited interpretability. This study proposes a Singh-based fuzzy logic framework for one-hour-ahead discharge and flood peak prediction using hourly streamflow from the Schwarzbach catchment (135 km2, Hesse, Germany). Four storm-driven flood events were analysed under contrasting hydrological conditions. Temporal dependence was characterised using autocorrelation function (ACF), partial autocorrelation function (PACF), and fuzzy-cluster transition analysis to link event-specific discharge dynamics with predictive behaviour. The results revealed substantial heterogeneity in hydrological memory, autocorrelation decay, and state-transition complexity across events, ranging from smooth memory-dominated responses to oscillatory multi-peak behaviour. Despite these differences, the framework reproduced all flood events with Nash–Sutcliffe efficiency (NSE) above 0.99 while capturing peak timing and magnitude, including complex double-peak dynamics. Because NSE is referenced to the variance of the observed series, forecasts were benchmarked against a persistence predictor, which itself attains a NSE of 0.958–0.976 at this lead time and time step; relative to persistence the framework reduced the root mean squared error by 38–68%, corresponding to skill scores of 0.612–0.897, whereas an AR(1) benchmark yielded no meaningful improvement. Prediction uncertainty increased with hydrological complexity but remained free of systematic bias. The results demonstrate that fuzzy logic provides an interpretable and computationally efficient alternative for short-term discharge forecasting that improves measurably on a naïve benchmark, offering potential for flood early warning systems and water resource management in data-scarce catchments.