Sensitivity Analysis of ANN and Generalized Neuron Models for Hydrological Modeling

Seema Narain, Ashu Jain

Abstract: Artificial Neural Networks (ANNs) have gained significant attention for hydrological modeling due to their ability to handle complex, nonlinear relationships. Traditionally, the Multi-Layer Perceptron (MLP) model has been the most commonly employed ANN structure, although it lacks systematic guidelines for architecture selection and often requires extensive trial and error. To address these limitations, this study explores the application of a Generalized Neuron (GN) model, which offers a simplified architecture and improved flexibility. Three distinct GN configurations were tested using daily rainfall and streamflow data from the Kentucky River basin. The models were assessed for performance under varying initial weights and limited training data. Results indicate that GN models demonstrate higher resilience to initialization, improved generalization, and competitive accuracy with fewer parameters compared to MLPs. This suggests GN models are a promising alternative for efficient rainfall-runoff modeling.

Keywords: Artificial neural networks, hydrologic models, generalized neurons, water resources, rainfall runoff process, hydrology