Multiple regression analysis for predicting few water quality parameters at unmonitored sub-watershed outlets in the St. Joseph River basin, USA

Abstract

In this study, six multiple regression models were tested for predicting water quality during spring and fall seasons at unmonitored sites within St. Joseph River basin, USA. A relationship between a total of 28 independent features that were derived from land use, morphology and water balance parameters was established with the known water quality at the specified monitoring sites along the River. Each model was tested, trained and cross validated for their prediction efficacy. The results indicated that ridge regressor best predicted the nonpoint water quality parameters during both the seasons. The results were validated for one sub-watershed outlet. A relative error was found to be low but relatively higher during fall season compared to spring season. The usefulness of this study lies in populating river monitoring program with water quality data from unmonitored sites, and thus, be made available for modelling and developing management strategies.

Keywords:

• Nonpoint pollution
• geomorphology
• SWAT model
• multiple regression techniques
• cross validation

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data required for the present study mainly included the discharge data and water quality data, and their sources are mentioned in the manuscript.

Additional information

Funding

The corresponding author is grateful to the Department of Science and Technology (DST), Govt. of India and the Indo-U.S. Science & Technology Forum (IUSSTF) for providing WISTEMM Women Overseas fellowship to conduct research at Purdue University, IN, USA.