Abstract
The data of annual peak flows in river system is not straight forward but rather complex function of hydrology and geography of the area. In such composite situations, models based on soft computing approaches such Artificial Neural Network (ANN) may increase understanding of the hidden hydrological processes. The primary objective of this study is to determine the best approach among Traditional Artificial Neural Network (T-ANN) and Wavelet Artificial Neural Network (W-ANN) using Annual Peak flows data of Mangla site on River Jhelum, Pakistan. The results reveal the better performance of W-ANN approach under db2 scheme. Further, to justify the authenticity of W-ANN we compared it with the frequency analysis by fitting best probability distribution and non-parametric kernel density estimation (KDE) using Gaussian kernel approach for the same data. Results reveal that probability distribution modeling approach by fitting the Generalized Logistic (GLO) distribution being the most suitable model provides comparable results with W-ANN-db2 approach but outperforms than non-parametric KDE. Overall, the performance of W-ANN-db2 is better in the present study. The findings of the study are useful for better policy implications in water resources management.