In a study, researcher Nourani from Near East University, in collaboration with other researchers, delved into the assessment of groundwater quality in the vicinity of a waste landfill site in Tychy-Urbanowice, southern Poland. The study focused on the impact of various environmental factors, such as air and water temperature, groundwater table position, and precipitation, on the electrical conductivity of groundwater.
Using data from sensors strategically placed in boreholes, the research employed twenty neural network architectures of a Multilayer Perceptron Model (MLP) to analyze the influence of individual factors. Ultimately, three MLP models with seven to nine neurons in the hidden layer were selected based on their superior learning quality and minimal validation and test errors.
The maximum test quality recorded was 0.8369, with the smallest test error at 0.0011, showcasing the effectiveness of the chosen MLP models. A global sensitivity analysis underscored the crucial role of water temperature and groundwater table position in determining the electrical conductivity value.
The article highlighted the applicability of artificial neural networks in forecasting water quality parameters by continuously monitoring selected groundwater parameters. Specifically, the research focused on the P19 piezometer at the Tychy landfill site.
The study’s findings suggested that artificial neural networks offer a reliable approach for assessing changes in water quality parameters over time. The sensitivity analysis revealed that all input variables, including temperature, precipitation, and groundwater table position, significantly influenced the neural network’s operation.
In conclusion, the research underscored the importance of employing artificial neural networks for accurate and dynamic evaluations of groundwater quality. The obtained results contribute to the broader understanding of the intricate interplay between environmental variables and groundwater conductivity, emphasizing the potential for further analysis and application of this approach in similar contexts.
More Information:
https://sciendo.com/article/10.2478/environ-2023-0013
