Abstract
This study presents the development and evaluation of an independent Artificial Neural Network, (ANN) model tailored for daily streamflow forecasting within the upper Mahanadi River basin, spanning a lead time of 1 to 3 days. Utilizing hydro- meteorological data from 2001 to 2011, including daily discharge measurements from five stations, inflow data from the Hirakud Reservoir, and spatial average rainfall, temperature data, the ANN model is meticulously calibrated and evaluated. The methodology involves a meticulous selection of input time steps (lags) for each variable, employing both correlation-based and constant lag methodologies. Performance evaluation system of measurement, such as Nash-Sutcliffe Efficiency (NSE), Ratio of root mean square error to standard deviation of measured data, (RSR), Mean Absolute Error, (MAE), and Error in volume, (Evol), are working to assess the model's predictive capability.
The discoveries indicate that the ANN model consistently demonstrates commendable performance, particularly for lead times up to 2 days. However, its accuracy moderates for a 3-day lead time forecast. Analysis of input variables reveals that incorporating discharge, rainfall, temperature, and inflow data, either individually or in combination, yields the best outcomes for runoff forecasting up to two days lead time. Furthermore, input time step selection significantly influences the model's predictive accuracy, with correlation-based lag times generally enhancing performance compared to constant lag times. Based on these findings, leveraging correlation-based lag times is recommended for developing ANN-based daily streamflow forecasting models, providing valuable insights for flood prediction applications.