Smart Sensor Networks for Urban Water Quality Monitoring and Forecasting

Thesis Abstract

Urban water quality is increasingly threatened by episodic contamination, climate-driven variability, and aging municipal infrastructure, necessitating timely detection, accurate forecasting, and data-driven decision support. This study addresses the gap in scalable, real-time monitoring of urban water systems, integrating smart sensor networks with predictive analytics to enable proactive management of water quality and public health risks. The aim is to develop and validate an ICT-enabled framework for continuous sensing, data fusion, and short- to medium-term forecasting of key water quality indicators in urban distribution networks. Specific objectives include (1) designing a sensor network topology and communication protocol optimized for coverage, reliability, and low power consumption; (2) developing a data processing pipeline for real-time anomaly detection and multivariate forecasting of critical parameters (e.g., turbidity, residual disinfectant, pH, dissolved organic carbon, conductivity) using machine learning models; (3) evaluating sensor calibration, drift correction, and data quality assurance under varying hydraulic conditions; (4) integrating forecasts into a decision-support dashboard for operators to trigger preventive actions; and (5) assessing the social and institutional feasibility of adoption within a municipal water authority. Methodologically, the study adopts a mixed-methods research design comprising a quantitative instrumental phase and a qualitative validation phase. The population includes municipal water distribution systems in a mid-sized city with a population of approximately 1.2 million and an existing, upgrade-ready distribution network. A stratified random sample of 60 distribution nodes is selected, with 40 nodes instrumented with multi-parameter probes and supervisory control and data acquisition (SCADA) integration, and 20 nodes serving as control to benchmark performance. Data collection instruments encompass calibrated water quality sensors measuring turbidity, residual chlorine, pH, conductivity, and turbidity-optical parameters, alongside flow meters and pressure transducers; sensor calibration logs, meteorological data, and SCADA historical records are incorporated. The data collection period spans 18 months to capture seasonal variability and extreme events, supplemented by 12 semi-structured interviews with water system operators and engineers to capture practical feasibility and decision-making processes. Analytical techniques include time-series forecasting using ARIMA, Long Short-Term Memory (LSTM) neural networks, and gradient boosting regression to compare predictive accuracy for each water quality parameter, with model selection based on cross-validated RMSE and R-squared performance. Multivariate anomaly detection employs isolation forests and hierarchical clustering, while data fusion is implemented via Kalman filtering and Bayesian network integration to combine sensor data with hydraulic models. Calibration drift and sensor reliability are examined through mixed-effects modeling to account for nested sensor and node-level variability. A theoretical lens combining Technology Acceptance Model (TAM) and Diffusion of Innovations (DOI) is applied to interpret operator acceptance and organizational uptake. The study also leverages the Water-Energy-Food Nexus framework to evaluate systemic implications of enhanced monitoring on energy use and resource efficiency. Expected findings include improved accuracy of short-term water quality forecasts (up to 72 hours) with reduced RMSE by 25–40% for key indicators, timely detection of contamination events with a false-positive rate below 5%, and demonstrable benefits in operational decision-making via the dashboard, including optimized flushing schedules and targeted residual monitoring. Sensor network performance is anticipated to reveal robust operation under fluctuating demand and hydraulic transients, with calibration drift correction extending effective sensor lifespans. The research is expected to contribute to knowledge by (i) advancing integrated ICT-driven sensing and forecasting architectures for urban water systems, (ii) providing a validated methodology for real-time data fusion and anomaly detection in distribution networks, and (iii) offering a framework for evaluating organizational readiness and policy implications for smart water management. The study concludes that a sensor-enabled forecasting framework can significantly enhance water quality resilience, enabling proactive interventions and reducing public health risk. Recommendations include standardization of data interoperability protocols, development of modular dashboards for operator training, investment in routine sensor maintenance programs, and policy guidance for governance structures that support data sharing and continuous improvement within municipal water authorities.