Using Expected Moments Algorithms To Estimate Parameters Of Flood Frequency Distribution

Flood is a pervasive natural hazard that routinely impacts society. Affected by globalclimate changing and human activities, the loss that caused by flood disaster will becomemore and more serious. The estimation precision of design flood is directly related to theinvestment and security of water resouces projects. So, the estimation precision of designflood have become an important problem.Flood frequency analysis is a key technology ofestimation of flood design values under given design frequency (return period). Reasonabledesign flood can provide the basis for flood management, spillway design for dams andplanning decision for hydraulic engineering.This dissertation summarized the present researches about flood frequency analysis bothat home and abroad, and explored in the applications of common and new parameterestimation methods in simple and non-simple sample. The main contents of this dissertationare as follows:（1）Research on the application of new parameter estimation method——ExpectedMoments Algorithm and common parameter estimation methods（moment method, weightfunction method, probability weighted moment method, L-moment method）in non-simplesample. Given P-III distribution, these methods were applied to annual maximum flood peakdischarge with historical flood information from ten hydrological stations in shannxiprovince.（2）Research on the application of methods that partial L-moments method andhigher-order L-moments method in simple sample， GEV distribution was chosen forapplication on annual maximum flood peak discharge from ten hydrological stations inshannxi province, and the fitting results were evaluated.（3）Research on the Statistical properties of parameter estimation methods. MonteCarlo simulation was used to analyze the statistical properties of parameter estimationmethods. Simple sample and non-simple sample with threshold model and given-numbermodel were built. The results of all samples were calculated by Matlab2009a. Finally, Basedon the standard of quantitation and graph, the Statistical performance of all methods wereanalyzed and evaluated.The main conclusions are shown as follows: （1）Different Parameter estimation methods had different effects on fitting results.Innon-simple sample, the EMA estimator performed better than other methods in fitting longhistorical flood information samples, probability weighted moments method was close toL-moments method, and the length of historical flood information has little influence on thelast two methods in fitting, weight function method performed better than the moment method.In simple sample, both r partial L-moments method and higher-order L-moments method canimprove the fitting effect of larger value of the annual maximum flood peak discharge, so itcan be expected to provide the basis for the design flood calculation.（2）Monte Carlo simulation was employed to analyze the statistical properties ofparameter estimation methods. Under different censoring levels and orders, the statisticalproperties of partial L-moments method and higher-order L-moments method performed well.With censoring level and order increasing, parameter estimator and design flood have lowerbias. However, when return period increased, parameter estimator and design flood have moreerrors. Otherwise, the parameters of distribution have great influence on statistical propertiesof design flood. In non-simple sample, the results of threshold model sample andgiven-number model sample were close. The statistical properties of all methods could beimproved when the length of observed information and historical flood information increased.The EMA estimator performed better than weight function method and the moment method,Comparing weight founction, moment method, probability weighted moment method andL-moment method show their stable properities.