Regional ANN model for estimating missing daily suspended sediment load in complex, heterogeneous catchments

To reduce the costs associated with monitoring suspended sediment load (SSL) in rivers, creating more cost-effective and easily measurable indirect estimation methodologies that rely on interactions with other variables is necessary. This work introduces a new method to assess the capability of regional models to expand ungauged SSL to gauging locations in a diverse region with little in situ data and complex hydrography. The estimates were derived using discharge (Q) data, utilizing models based on artificial neural network (ANN) feedforward backpropagation (FFBP) techniques for the Ramganga River Basin. This work confirms the practical capacity and utility of ANN for simulating intricate nonlinear dynamics in natural river systems in the context of the Himalayas. The modelling method is based on the daily Q and SSL data collected from 2007 to 2009. The initial phase involved developing and training the ANN utilizing the FFBP Algorithm within the Matlab (MATLAB R2015a). The networks were optimized utilizing the process of enumeration. The optimal network was subsequently employed to forecast the SSL values, which are ungauged parameters, at the Moradabad (MBD) gauging site. The second stage involves validating the predicted (ungauged) SSL values of MBD by utilizing them to estimate the SSL values of Dabri (DBI) gauged sites. The predicted values obtained from the model are contrasted with the authentic observed values of SSL at DBI. The R² value for the optimal network was found to be 0.998, accompanied by an MSE of 0.00112. The study offers valuable insights into the modelling of ANN and emphasizes the significance of comprehending a river basin and its influencing components to simulate the SSL effectively. This study demonstrates that the proposed methodology allows for highly efficient regional streamflow estimation in ungauged basins, even in diverse geographical areas.