A manifold learning perspective on surrogate modeling of nitrate concentration in the Kansas River

Abstract

A non-linear surrogate model of nitrate concentration in the Kansas River (USA) is described. The model is an (almost) Piece-wise Linear response surface that provides a mean field approximation to the dynamics of the measured data for nitrate plus nitrite (target product) correlations to turbidity and chlorophyll-a concentrations (input variables). The method extends the United States Geological Survey’s linear procedures for surrogate data modeling allowing for better approximations for river systems exhibiting algal blooms due to nutrient-rich source waters. The model and visualization procedures illustrated in the Kansas River example should be generally applicable to many medium-size rivers in agricultural regions. HIGHLIGHT Non-linear surrogate modeling of nutrients is illustrate in the Kansas River that uses commonly available sensors, and initial nitrate plus nitrate calibration data, to enable future estimations of nutrients using only the turbidity and chlorophyll-a sensors.;

Keywords

• algal bloom
• kansas river
• nitrate
• non-linear
• surrogate modeling
• water quality