Revolutionising water quality monitoring in the information age
Automated low-cost sensors distributed in the environment have the potential to revolutionise the way we monitor and manage air, water and soil. This project aims to develop novel statistical methods to detect anomalies in the data generated from these in situ sensors with computationally efficient modelling on river networks through space and time, with the applied goals of automating anomaly detection in water-quality data and generating predictions of sediment and nutrient concentrations throughout river networks in near-real time. This will represent a fundamental increase in scientific knowledge, which will be immediately useful in the domains of aquatic science, environmental monitoring, and statistics.
The Queensland Department of Environment and Sciences manages the Great Barrier Reef Catchment Loads Monitoring Program which monitors the concentration of sediments, nutrients, and pesticides in multiple sites along the river networks on the east coast of Queensland. The data from the in situ sensors located in these sites, provide high-frequency measurements for several water-quality variables that we will use in our research.
Monash University researchers:
Other chief investigators:
- Kerrie Mengersen (Queensland University of Technology)
- Erin Peterson (Queensland University of Technology)
- Catherine Leigh (RMIT University)
- James McGree (Queensland University of Technology)
We are also working with researchers from Queensland University of Technology, Queensland Department of Environment & Science, Southeast Queensland Healthy Land & Water, University of Pau & Pays de L’Adour (France), University of Alaska Fairbanks (USA), and University of Moratuwa (Sri Lanka).
It is important to identify statistical relationships between water-quality variables through space and time in order to distinguish real water-quality events from anomalies. These relationships depend on the local climate — the relationship between water-quality variables in temperate regions will be different from that of tropical or sub-tropical regions. We aim to develop statistical modelling methods to analyse these relationships between variables and to compare them among different study regions.
The data produced by the sensors are prone to errors due to mis-calibration, bio-fouling, weakening battery charge, and other technical failures. Hence it is essential to develop techniques to identify these anomalies to enhance the reliability and the quality of data produced by sensors. We aim to develop advanced near-real-time anomaly detection algorithms, using data from neighbouring sensors, taking account of the spatial and temporal correlations.
We aim to develop spatio-temporal statistical models to predict sediment and nutrient concentration from anomaly-corrected data produced by in situ sensors. Computationally efficient techniques will be developed that extend existing geostatistical models to the spatio-temporal case.
Finally, we aim to develop adaptive sampling designs to place sensors across the river network, optimising the quality of data and the cost-effectiveness of the placement.