Modeling restricted bivariate censored lowflow data

Abstract

Environmental studies often result in censored data. In this article, the lowflow quantiles Q*7,2 and Q*7,10 below a limit are treated as censored data. These streamflow quantiles are important for water resources planning and management. Our partial all-subsets censored regression procedure identifies a few important explanatory variables, such as drainage area, basin slope, soil-infiltration index, rainfall index, and some combinations of them. The proposed maximum likelihood estimation method incorporates the restriction Q*7,2≥Q*7,10 and the bivariate probability distribution of the quantiles to improve model quality. Analyses of the lowflow quantiles obtained from streams in West-Central Florida show that our procedure is more appropriate than the commonly used univariate main-effects models in predicting quantiles. Copyright © 1999 John Wiley & Sons, Ltd.