Study On The Inconsistency Of Flood Extreme Value Sequence And Frequency Distribution Considering Covariates

In recent decades,the hydrological process has been affected by multiple factors such as extreme climate,human activities,and the underlying surface conditions of the watershed.The original consistency of the hydrological sequence has also changed,especially the flood and rainstorm data have been affected more obviously.This leads to questions and challenges based on traditional hydrological consistency theoretical research and the frequency analysis results derived therefrom.Therefore,the research of hydrological inconsistency theory has gradually become a focus of attention and attention,and attempts to apply it to flood frequency analysis.In order to explore the impact of hydrological sequence inconsistency on flood frequency analysis,this paper took five different typical watersheds in Hei Longjiang Province as the study area,and used Generalized Additive Models for Location,Scale and Shape.Shape,abbreviated as（GAMLSS）,based on two types of factors,time and precipitation,analyzed the consistent and non-uniform flood frequency of the flood variables（Q,W₁,W₃,W₇）.Analyze the fitting effect of the empirical data and the theoretical quantile curve,and determine the appropriate theoretical distribution of floods under changing conditions.The main research contents and conclusions were as follows:（1）Mann-Kendall test method and R/S analysis method were used to test the inconsistency of the flood extreme value series of 28 hydrological stations in the five river basins of the province,and analyzed the characteristics of its inter-annual changes.The study find that the flood extreme value sequence of each basin has abrupt changes and trends,most of the basin flood extreme value sequences are in a downward trend,and their continuity is strong,and their downward trend will remain in the future.In addition,due to the impact of catastrophic floods on the underlying surface conditions,the sudden changes in the extreme flood sequence of each basin were mainly concentrated in the 1960s,1990s and early 21st century.（2）In order to consider the influence of time factor on the calculation of flood frequency,the GAMLSS model was constructed to study the distribution of flood frequency under consistent and non-uniform conditions.The results show that,based on the assumption of consistency,the optimal distribution of the flood extreme value sequence in each basin is selected mainly by the lognormal distribution（LN）,followed by the generalized gamma distribution（GG）.Although the selected distributions are relatively concentrated,each basin shows different fitting results according to their respective basin characteristics;based on the assumption of inconsistency of time covariates,the optimal distribution of flood extreme value series in each basin is still selected based on LN Mainly,followed by GG distribution.Although the results of the distribution selection are similar to the consistent results,due to the participation of historical catastrophic floods in the model calculations,this makes the quantile curve of the flood theory more affected by the continuity and discontinuity of the time variable,which is more than the fitting result of the consistent condition as reasonable.（3）On the basis of historical catastrophic floods and climate factor data,the influence factors of precipitationvariables（P₁,P₃）are selected.Establish two deep learning models of Deep Neural Network（DNN）and Distributed Random Forest（DRF）to interpolate the extraordinary precipitation under the historical extraordinary floods.The results show that the simulated historical extraordinary precipitation can be used as the interpolation value of the corresponding precipitation of the historical extraordinary flood that is missing in the subsequent flood frequency analysis.（4）In order to consider the influence of precipitation factors on frequency calculation,combined with the interpolated historical extraordinary precipitation values,the GAMLSS model with precipitation as a covariate was constructed to study the inconsistent frequency distribution of flood extreme value series.The results show that the optimal distribution of flood frequency in each basin is mainly selected by LN,and other preferred distributions are gamma distribution（GA）,Weibull distribution（WEI）,and GG distribution.Due to the influence of extreme precipitation information,the optimal distribution types of flood frequency in each basin are quite different;in addition,with the increase of extreme precipitation samples,the flood theory quantile curve has more complicated changes,which indicates the extreme precipitation samples’Added to increase the instability of the model simulation results.