Hydrologic regionalization for the estimation of streamflows at ungaged sites based on time series analysis and multivariate statistical analysis

The study developed a regionalization scheme for the estimation of monthly streamflows at ungaged sites using a time series model and multivariate statistical analysis. The methodology includes several procedures: (1) Estimating streamflow parameters by a time series model; (2) Developing three data sets A (streamflow parameters), B (7 watershed variables), and C (the combination of data sets A and B); (3) Classifying watersheds into regions by cluster analysis (CA) based on three sets of criteria (data sets A, B, and C); (4) Evaluating three sets of criteria; (5) Investigating the regional differences and the relationships between streamflow parameters and between watershed variables by discriminant analysis and principal component analysis; (6) Constructing the relationships between streamflow parameters and watershed variables by canonical correlation analysis and multiple regression analysis and between mean monthly streamflows (Qm) and watershed area (Aw) by simple regression analysis (SRA); and (7) Validation. The analysis results show that: (1) Sixteen streamflow parameters were estimated from the time series model based on 25 years of specific monthly streamflows (Qs) at 94 stations in the southeastern states; (2) The streamflow parameters were found to be a better set of criteria than the other sets of criteria; (3) Six clusters (obtained from CA) tested significantly different; (4) The 6 clusters in data set A were separated by physiographical boundaries especially the two main clusters; (5) Similarities and differences were found between streamflow parameters and between watershed variables in the two main clusters; (6) The streamflow parameters and Qm can be reasonably predicted by streamflow variables and Aw, respectively; (7) The statistical properties of the synthetic Qs in validation seemed to be preserved in the two main clusters except for one of the two main clusters in the wet season; (8) The synthetic specific Qm improved over the predicted specific Qm from SRA; and (9) The synthetic Qs generated from the time series model were reliable in one of the two main clusters. The proposed regionalization scheme was found to be feasible, even though only one of the two main clusters is reliable for the estimation of monthly streamflows at ungaged sites. A topographic map and the annual precipitation are the required information for the streamflow estimations. The estimations not only preserve many streamflow properties, but also the stochastic nature of streamflows