Development of a low-cost dissolved oxygen estimation model using calibrated arduino-based sensors

Abstract

Conventional river water quality monitoring is often constrained by high operational costs and maintenance-intensive sensors. To address these limitations, this study proposes a low-cost framework for estimating dissolved oxygen (DO) concentrations without direct DO probes by deploying an Arduino-based multi-sensor array (pH, temperature, and turbidity) coupled with multivariate regression modeling. The baseline model was calibrated using laboratory experimental data and subsequently validated under field conditions in the Kızılırmak River, Turkey (). The proposed framework demonstrated high predictive accuracy, yielding a coefficient of determination () of 0.842, a Leave-One-Out Cross-Validation () score of 0.798, a Mean Absolute Percentage Error (MAPE) of 4.91%, and a Mean Squared Error (MSE) of 1.64. The primary novelty lies in substituting expensive optical DO probes with robust proxy sensors integrated into a computationally lightweight model suitable for edge-computing IoT nodes. Serving as a foundational proof-of-concept, the developed regression coefficients are site-specific and require localized re-calibration (15–20 paired samples) prior to deployment in distinct catchment basins. Overall, this framework offers a scalable, cost-effective solution for continuous environmental impact assessment and sustainable water resource management.