Analysis of Precision Changes of Otho-Normal Series to Annual Maximum Discharge Sample Size

Flood risk analysis studies carry out based on estimation of the probability density function (PDF) of flood peak. Conventional approaches to estimate the probability density function distributions are parametric methods that have constraints such as sensitivity of fitness precision to reduction in sample size. Non-parametric methods to estimate the PDF have fewer restrictions than the parametric distributions. In this study, Exponential, LogNormal, Gamma and Gumbel as parametric distributions, Ortho-Normal Series method as a non-parametric method, and Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC) and Mean Square Error (MSE) as precision criteria were applied to investigate the precision of mentioned methods and their sensitivity to reduction of annual maximum peak flow series. The precision and change in precision of fitness of parametric and non-parametric distributions to sample size reduction were investigated based on five hydrometry station (with 38 to 48 years recorded datasets) in Golestan province. The results show that the precision of nonparametric ortho-normal series method is considerably higher than parametric distributions. In addition, orthonormal series method is less sensitive to sample size reduction than parametric distribution that makes it as suitable option where the recorded data sets are belonging to relatively short time period.