Predicting small water courses’ physico-chemical status from watershed characteristics with two multivariate statistical methods

Abstract

Abstract Watershed area and a bunch of relief, land use, and wastewater characteristics for 32 upland and 33 lowland small river courses are generated. Based on these characteristics, logistic binary regression models are trained to predict if the river achieves the good physico-chemical status, and discriminant analysis models are trained to predict the physico-chemical status class on a five-class scale. Univariate models revealed that elevation (for upland rivers), the share of artificial surfaces (for lowland rivers) along with forests, and wastewater quality variables such as biochemical oxygen demand, chemical oxygen demand, and phosphorus are the most significant predictors. Discriminant analysis models performed better on upland than on lowland rivers. Achievement of good status could be predicted with an accuracy of ~90% (with 2 to 4 variable logit models), whereas the status class with an accuracy of 63/48% (with 2 to 4 variable discriminant analysis models) for upland and lowland rivers, respectively. This contribution uses Hungary as a case study.