Ensuring water quality from source to tap requires efficient monitoring. Traditionally, water quality monitoring relied on Hardware (Hard) sensors, such as probes installed in lakes, rivers or pipes and in laboratory equipment. These technologies and methods are reliable, but are expensive to install and maintain, require frequent cleaning and calibration, require intensive labor and most importantly lack the flexibility in adapting to new monitoring requirements without physical intervention. They can only measure what is happening to the specific variable they measure, in the exact spot they were installed.
But what if we need to understand what is happening across the entire surface of a water body and not just in a specific point, or predict the bacteriological contamination in a remote alpine spring buried under snow?
Or forecast how much energy a treatment plant will need tomorrow? Traditional monitoring can provide remedies to be above cases.
On the contrary, ToDrinQ project overcomes these limitations by developing Soft (Software) Sensors for water quality monitoring, management and analysis. In Layman’s terms, soft-sensors attempt to replace hard-sensors with algorithms and computational models with predictive capacity, able to fuse large amounts of data streams, such as historical records, weather forecasts or satellite imagery, to calculate target parameters that are difficult, costly, or impossible to measure directly on-time.
ToDrinQ develops Soft-Sensor technology to solve different problems across its European demonstration sites. For the Athens Demo Case in Lake Yliki, we utilise satellite data to provide efficient monitoring and key insights regarding water quality across the lake’s surface. Monitoring a large lake like Yliki is difficult. Traditional sampling methods are slow and limited and provide only a snapshot of the quality conditions in the lake. Our soft sensors developed for the Athens Demo Case utilize satellite images, Earth Observation (EO) data, Deep Learning (DL) algorithms and hydrological models to monitor the entire lake surface at once. Our soft-sensors can estimate Chlorophyll-a, Dissolved Oxygen, pH, the overall Water Quality Index and the occurrence probability of algal bloom events at the pixel level across the entire lake surface and at frequent time steps without the need for in-situ measurements or installment of additional hard sensors. In addition to that, our Earth Observation-based approach provides an Early Warning System for nutrient run-off in the Yliki lake, giving the water utility, EYDAP, the capability to protect the reservoir proactively.
Another soft-sensor application is the monitoring of Bacteriological contamination in the Swiss Alps (Val de Bagnes). There, the Mayentzet reservoir is fed by mountain springs that are difficult to access. Snowmelt can wash bacteria into the water supply system. Instead of relying only on hardware, we developed a soft sensor that analyzes meteorological data (e.g., snowfall, temperature, precipitation) and forecasts the bacteriological contamination in the reservoir of interest using Machine Learning (ML) model.
Finally, at Waternet’s treatment facilities (Amsterdam Demo Case), we are using soft sensors to look into the future of the process to balance safety with efficiency. We developed models that predict turbidity levels, both at the plant inlet and the coagulation outlet up to six hours in advance, allowing operators to adjust chemical dosing proactively instead of reacting to spikes. In addition to that the ozone exposure soft sensor enables engineers to fine-tune disinfection levels to save energy and prevent harmful by-products at daily scale.
