A statistical approach for the estimation of water quality characteristics of ungauged streams/watersheds under stationary conditions

Abstract

Water quality and quantity data are essential for water resources management, but historical records are not always available at sites where information is required, especially for water quality (WQ) data because of the high costs of implementing and operating water quality monitoring networks. This leads to the impossibility/impracticality of monitoring all streams/watersheds in a region under monitoring. A novel statistical approach is proposed for the estimation of WQ characteristics (25th, 50th, and 75th percentiles) at ungauged streams/watersheds. The proposed approach is based on developing a WQ causes-effects functional relationship, which has not been done before. The WQ causes are represented by the watershed physiographic (natural and anthropogenic) attributes and the effects are represented by the WQ characteristics. For a target ungauged stream/watershed: (i) a canonical correlation analysis-based-neighborhood approach was applied to identify homogeneous gauged watersheds; and (ii) nonlinear regression models were then calibrated using data available at the selected gauged watersheds to develop the causes-effects functional relationship between the physiographic attributes as predictors, and WQ characteristics as dependent variables. Once the regression models were validated, the physiographic characteristics of the target ungauged stream/watershed were used to estimate its WQ characteristics. The jack-knife validation method was applied to evaluate the proposed statistical approach using 50 gauged streams from the Nile-Delta of Egypt. Results showed accurate and precise estimation of the Electric Conductivity and Sodium percentiles. For Iron, relatively less precision was obtained due to the absence of attributes that can describe the Iron concentrations.