A Principal Component Regression Method for Estimating Low Flow Index

Abstract

Low flow indices are very important for water resources planning, pollution control, conservation and even recreational use. Determining these indices depends on having access to daily flow discharges. However, in some cases, such data are either insufficient or are not available at all. Hence, in these cases, estimation of the indices requires the use of data in catchments for which streamflow data have been collected. In this paper, it was attempted to estimate the low flow index (7Q10), the 7-day, 10-year lowflow, using principal component regression (PCR) based on physiographic and hydrologic variables. To do so, a two-step procedure was followed. In the first step, ranking method was applied to determine the best fitted distributions on yearly minimum discharges in each gauging station according to distribution suitability for fitting on extremes; the better the distribution fits the data, the higher number is given as ranking. Adding the ranking numbers dedicated to each gauging station, it was revealed that Gamma distribution with two parameters got the highest value and therefore was chosen as the representative distribution in the region. Using Gamma distribution in gauging stations, 7Q10 was estimated in all gauging stations in the basin. In the second step, a PCR was developed due to existence of high-correlated independent variables. To choose the influential components for use in PCR, eigenvector analysis and factor analysis were performed. The results show that the components chosen through the two approaches correspond to each other well. To evaluate the efficiency of the developed PCR in modeling 7Q10, calibration and verification were pursued. The results approve the efficiency of model in predicting 7Q10 in the region under study.