Practical data-based modelling approach for estimating river water turbidity and total organic carbon.

Abstract

The quality of fresh water affects not only the aquatic environment and human health, but also the drinking water treatment and operation of a wide range of industrial processes. Optimal and proactive process operation requires continuous monitoring of the raw water quality. However, due to the high purchase cost and laborious maintenance, specific hardware sensors are underutilized in monitoring raw water sources such as rivers, which results in lack of crucial environmental monitoring data necessary for optimal and resource efficient operation of industrial processes or assessing the general safety of water. The research presented in this paper introduces a practical, straightforward, and cost-effective alternative approach via data-based modelling to estimate two important river water quality variables, turbidity and total organic carbon, in real-time. A single year-round multiple linear regression model with only two robustly and fast measurable input variables, river water level and water temperature, was proved to accurately estimate the water turbidity and total organic carbon during training period (R: 0.80 and R: 0.85, respectively) and with three independent testing datasets including varying conditions. The presented approach is easily parameterizable, calibratable and can be utilized for real-time river water quality monitoring in various locations enabling increased awareness on water safety and for instance proactive adjustments to water dependent processes.