Turbidity, which is the cloudiness or haziness of water caused by large numbers of individual particles, is a critical factor for drinking water producers. High turbidity can harbour harmful microorganisms and reduce the effectiveness of disinfection processes, making it essential to monitor and manage to ensure safe drinking water.This study was conducted at the Rökebo water production plant, which produces drinking water for Sandviken and nearby areas, serving around 29,000 people. The treatment process includes several steps, such as chemical precipitation, filtration, UV treatment, and chlorination, to ensure the water is safe to drink.
A new plant is being constructed to use only lake water and will include additional treatment steps to remove natural organic matter.Lake Öjaren is moderately sized, covering 21 square kilometres with an average depth of 4.66 meters, which means it is a shallow lake. The depth and shape of the lake influence how it responds to wind and weather, which can stir up sediments and affect water clarity. The catchment area of Lake Öjaren consists mainly of forest and moraine but has 5.5% clear-cuts, which contribute to higher turbidity levels in combination with heavy precipitation.
Climate change is expected to bring warmer temperatures and more rain to Sweden, affecting Lake Öjaren’s water quality and availability. Projections indicate that runoff to the lake will increase by about 15%, which is more than the average for the area. Less precipitation will fall as snow, leading to more water flowing into the lake during winter. These changes will likely increase the levels of nutrients and organic matter in the lake, increasing turbidity and calling for an adaptation strategy at the drinking water plant. We tested several machine learning models to predict water turbidity, including ElasticNet Regression, RandomForestRegressor, GradientBoostingRegressor, and XGBoost.
These models helped us understand which factors most affect turbidity. For example, the RandomForestRegressor model performed well, showing that air temperature, wind speed, and rainfall from the past few days were important predictors. The XGBoost model also provided valuable insights, particularly emphasising the impact of rainfall from four days prior. Despite using general meteorological data, our models successfully predicted local conditions in Lake Öjaren, demonstrating their robustness. However, capturing extreme turbidity events remains challenging. High-quality data and advanced techniques are crucial for improving predictive accuracy. Future work should focus on collecting more detailed data and refining models to support effective water management and mitigate climate change impacts on the drinking water production. This ongoing research is vital for ensuring a reliable drinking water supply, even under changing environmental conditions.