Multi-Method Sediment Source Study In Small Watershed Of Loess Plateau Based On Fingerprint Recognition

Because of its special geographical location and climate environment,as well as strong human activities,the Loess Plateau is one of the most serious areas of soil erosion in China.In order to control the serious soil erosion,a series of soil and water conservation measures have been implemented in the Loess Plateau,and the soil erosion has been effectively controlled.Under the background of the implementation of increasingly sophisticated soil and water conservation measures,more accurate and rapid identification of sediment sources at the watershed scale has become an increasingly important part of the study of soil erosion in the Loess Plateau.In this paper,taking the Loess Plateau and Tengzigou small watershed as the research object,the fingerprint factors,the use frequency of discriminant models and the evolution of sediment sources in the study of sediment sources in the Loess Plateau are analyzed by using the integrated analysis method;On the basis of soil data collected in Tengzigou small watershed,the fingerprint factor screening method and discriminant model were optimized through artificial mixed samples and sediment data collected in a single rainfall event,and the proportion of sediment sources in a single rainfall event was calculated;Based on the historical siltation information of the damland,the different methods for screening the fingerprint factors of historical sediment sources are evaluated,and the evolution characteristics of the historical sediment sources of the damland are analyzed;With the support of measured data and calculation results,a basin sediment source analysis software including multi-fingerprint factor screening method and quantitative discrimination model of sediment source is designed and developed in order to provide scientific basis for evaluating the benefits of soil and water conservation control in the Loess Plateau,optimizing ecological restoration measures and studying the driving mechanism of water and sediment changes in the Yellow River.It also provides a convenient and fast sediment source quantification software for sediment source research,and reduces the cost of scientific research.The main conclusions are as follows:（1）There were 56 fingerprint factors used in the studies of sediment sources in the Loess Plateau,among which heavy metals were the most widely used,accounting for 53.5%of the total number of species and 60.0%of the total number of studies.Soil magnetism,¹³⁷Cs,TN and TP.Among the eight models,the basic Walling-Collins model has the highest use frequency of 65.5%in the Loess Plateau.The erosion evolution process of the Loess Plateau is divided into three stages:the first stage is before 2000,before the conversion of farmland to forest and grassland;The second stage is from 2000 to 2009,which is the implementation stage of returning farmland to forest and grassland;the third stage is after 2010,which is the consolidation stage of returning farmland to forest and grassland.Compared with the second stage,the contribution rate of sediment in the first stage increased significantly（P<0.05）,with a change rate of 34.22%,and the change rate of sediment contribution ratio in the first stage was more than 50%,and the average sediment contribution ratio of the three stages was about 2.7:7.3,3.6:6.4 and 4:6,respectively.（2）Divide the sediment source area of the small watershed in the study area,the three source areas are divided into terrace,slope and gully,and the two source areas are divided into gully land and gully along the gully line.The screening methods of fingerprint factors were evaluated by using three source fingerprint factor data sets,two rainfall event data sets and three source artificial mixed sample data sets.The screening method composed of conservative test,KruskalWallis H test and random forest had the best discrimination effect,and the correct discrimination rate was more than 90%;Walling-Collins model and Bayesian model were used to calculate two rainfall events respectively,and the screening method composed of conservative test,KruskalWallis H test and random forest performed best in the two models.（3）evaluate that fingerprint factor screening method,the Walling-Collins model and the Bayesian model by use artificial mixed samples,The screening method composed of conservation test,Kruskal-Wallis H test and stepwise discriminant analysis,and the screening method composed of conservation test,Kruskal-Wallis H test and random forest all performed well in sediment source discrimination.The conservative test,the Kruskal-Wallis H test and the screening method of random forest composition are slightly better than the former;The Bayesian model performed better,and the conservation test,Kruskal-Wallis H test and random forest screening method and Bayesian model were used to analyze the sediment sources of the two sources,and the proportion of sediment contribution of the two sources was 18.5%,81.5%and 8.2%,91.8%respectively.（4）The screening of historical sediment source fingerprint factors set two modes:overall fingerprint factor screening and multi-group fingerprint factor screening.The correct discrimination rates of single group fingerprint factor combination obtained by overall fingerprint factor screening were 60.3%and 77.8%,respectively,and the discrimination effect was poor.Multiple fingerprint factor combinations corresponding to sedimentary cycles are obtained by screening multiple fingerprint factor combinations,and the correct discrimination rate of most of them is about 90%.The sediment contribution ratio of terraced fields and sloping fields is higher in the calculation results of single group of fingerprint factors,and the results are opposite to those of multiple groups of fingerprint factors in the analysis of historical sediment source evolution characteristics of dam land.