Comparison Between L-moments,Maximum Likelihood And Conventional Moment For Rainfall Frequency Analysis In Fujian Region

Rainfall frequency analysis which based on hydrological statistics and hydrology theory is to analyze the relationship between rainfall and their occurrence probability,provide scientific basis for planning and design of water conservancy and hydropower projects,urban drainage,flood control planning and construction of sponge city.In the frequency analysis of rainfall,the accuracy of parameter estimation in the frequency estimation and the accuracy of the frequency estimation are not ideal.In order to reduce the unbiased property of the parameter estimation process in the rainfall frequency analysis and improve the robustness to the mega value simulation.In this paper,three parameter estimation methods of linear moment method,maximum likelihood method and conventional moment method are compared for the first time with a large number of measured rainfall data under the Pearson type III distribution curve.Fujian area' s historical rainfall data as an example to estimate the parameters of 217 site data in this area.At the same time,Monte Carlo method is used to simulate the annual maximum daily rainfall value of 217 stations,compare and analyze the bias of the estimation process and the robustness of super large value,and the comparison between the design values of annual precipitation calculated by different parameter estimation methods in time and space is also studied.The results show that:1)In the unbiased aspect,the linear moment method's linear deviation coefficient and the linear partial state coefficient can all fall on the 45 degree line of the X-Y coordinate measuring the degree of deviation,it is basically unbiased.The simulation results of the deviation coefficient and the partial coefficient of the maximum likelihood method deviate from 45 degrees,the simulation results of the deviation coefficient and the partial coefficient of conventional moments deviate from 45 degrees.2)In terms of robustness,for sites with maximum values of site 71623500,the coefficient of deviation and the coefficient of skewness obtained by the linear moment method fall on the 45 degree line,which indicates that the robustness to outsize is good.The maximum value simulated by the maximum likelihood method deviates from the 45 degree line,the conventional moment method completely deviates from the 45 degree line for the simulation of the maximum value.3)On the temporal distribution map of rainfall estimation,there are differences in rainfall estimation at different recurrence periods among linear moment method,maximum likelihood method and conventional moment method.When the reappearance period is less than 10 years,there is little difference between the estimated value of rainfall obtained by the three parameter estimation methods,the degree of fitting with measured rainfall data is good.When the reappearance period is more than 10 years,there are differences in rainfall estimated by three parameter estimation methods,and with the increase of the reappearance period,the difference between the estimated rainfall of different parameter estimation methods is also increased.The estimated value of rainfall estimated comared with the measured data by the linear moment method is small,the maximum likelihood method is the next,and the conventional moment method is the largest.Therefore,the error of the calculated value can be reduced by using the linear moment method relative to other methods.4)On the spatial distribution map of rainfall estimation,there are differences in rainfall estimation at different recurrence periods among linear moment method,maximum likelihood method and conventional moment method.When the recurrence period is less than 10 years,the spatial distribution map of rainfall estimated by the linear moment method,the spatial distribution map of rainfall estimated by the conventional moment method and the rainfall spatial distribution map estimated by the maximum likelihood method are very close,but with the increase of the recurrence period,there are obvious differences.The spatial distribution map of rainfall estimated by the linear moment method is more intuitive,accurate,accurate and comprehensive,while the effect of the rainfall spatial distribution estimated by the conventional moment method is second,and the estimated value is smaller than the linear moment method,and the accuracy is a little worse.The effect of the rainfall space distribution map estimated by the maximum likelihood method is poor,which is due to the fewer sites,the lower samples and the applicable strips.It is not consistent with the spatial distribution pattern of actual rainstormTherefore,from the spatial distribution map of rainfall,we can see that the linear moment method is better than the conventional moment method,and the maximum likelihood method is the worst.In summary,linear moment method is superior to two other commonly used methods in terms of unbiases and robustness in parameter estimation,in the estimation of the estimated value of the rainfall frequency,the estimated value of rainfall estimated by the linear moment method is also better than that estimated by the maximum likelihood method and the conventional moment method.Therefore,the linear moment method is the best parameter estimation method in hydrological frequency analysis.