ESTIMATION OF FLOOD QUANTILFS FROM NON-STATIONARY FLOOD SERIES

Abstract

In flood frequency analysis , the given annual flood data is assumed to be random, stationary and representative of population. But in many cases the annual flood series are found to be non- stationary due to various reasons. To test whether the given annual flood series is. stationary or not, tests like Median crossing test, Turning point test, Kendall's rank correlation test, Run test, Hurst coefficient test etc. are applied to given annual flood series. Annual flood series of 76 river gauging stations, annual total rainfall series and annual daily maximum rainfall series of 205 raingauge stations of zone-3 are analyzed for non-stationarity. 17.10% of annual flood series, 11.22% of annual daily maximum rainfall and 23.90% of annual total rainfall series are found to have long-term dependence at 5% significance level. For the estimation of flood quantiles from non-stationary series , the theory based on synthetic record generation was studied . For generation of flood data , mixed-noise model (Booy and Lye, 1989) was used . Mixed-noise model is best suited for generation of data preserving characteristics of original series including non-stationarity , defined in terms of Hurst's coefficient . The mixed-noise model is applied to annual flood series of 75 river gauging stations and annual daily maximum rainfall series of 198 raingauge stations in zone-3 to generate 1000 samples of long-term series and 1000 samples of independent series, each of 100 years length.