基于多元统计分析的岩溶洼地小流域泥沙来源分析——以重庆市沙坪坝青木关岩溶槽谷区为例

Sediment source in karst depression basin based on multivariate statistical analysis:A case study of Qingmuguan karst trough area in Shapingba,Chongqing

本研究以青木关岩溶洼地小流域为研究区域,采集林地、草地、坡耕地3类泥沙源地土壤样品和洼地沉积物样品,测定磁化率、TP、TC、TN、^(137)Cs和地球化学元素等26个指标,采用3种统计分析方法选择3组不同的最佳指纹因子组,分别为Kruskal-Wallis H检验(KW-H)与判别函数分析(DFA)相结合(KW-H+DFA)、主成分分析(PCA)与DFA相结合(PCA+DFA)、KW-H与PCA相结合(KW-H+PCA),然后利用多元混合模型对3个潜在源地的相对贡献进行估算。结果表明:KW-H+DFA估算林地、草地、坡耕地的相对贡献分别为25.17%、32.19%、42.64%;PCA+DFA估算的相对贡献分别为26.72%、29.14%、44.14%;KW-H+PCA估算的相对贡献分别为23.67%、27.36%、48.97%。基于不同统计组合的GOF、RMSE、MAE分别为94.47%、9.85、5.71;91.32%、19.95、10.08;90.11%、20.76、12.12。研究结果表明,不同的统计方法会对沉积物来源的判别分析及精确度产生影响,强调在对泥沙来源进行研究时,应重视统计方法的选择,以便更准确地识别泥沙来源,进而更好地实施相应的水土保持措施。

In order to understand the influence of different statistical methods on sediment source identification,and then select a reasonable statistical method,so as to more accurately identify the sediment source contribution of different sources in karst depression small watershed,so as to formulate more effective sediment control measures.In this study,the Qingmuguan karst depression basin was selected as the research area.Soil samples from 3 sediment sources,forest land,grassland and sloping land,and sediment samples from depression were collected,and 26 indexes including magnetic susceptibility,TP,TC,TN,^(137)Cs and geochemical elements were determined.Three groups of optimal fingerprint factor groups were selected by 3 statistical analysis methods.The results are found:Kruskal-Wallis H test(KW-H)combined with discriminant function analysis(DFA),Principal component analysis(PCA)combined with DFA,and KW-H combined with PCA were used to estimate the relative contributions of the 3 potential sources.The results showed that the relative contribution of forest land,grassland and slope land was 25.17%,32.19%and 42.64%,respectively,estimated by KWH+DFA.The relative contributions of PCA and DFA were 26.72%,29.14%and 44.14%,respectively.The relative contributions estimated by KW-H+PCA were 23.67%,27.36%and 48.97%,respectively.The GOF,RMSE and MAE of different statistical combinations were 94.47%,9.85 and 5.71,respectively.RMSE of different statistical combinations were 91.32%,19.95,10.08;The MAE of different statistical combinations were 90.11%,20.76,12.12.The results show that different statistical methods will affect the discriminant analysis and accuracy of sediment sources.It is emphasized that the selection of statistical methods should be paid attention to in the study of sediment sources,so as to identify sediment sources more accurately and facilitate the implementation of corresponding soil and water conservation measures.