Researchers from the Warsaw University of Technology and the Institute of Environmental Protection – National Research Institute have developed an innovative method employing Machine Learning to assess the status of unmonitored waters. According to Andrzej Martyszunis, Prof. Małgorzata Loga, and Dr Karol Przeździecki, this method enables precise, mathematical analysis of results and can serve as an excellent complement to expert evaluations.
They described their innovative method in a Nature article titled, “Using machine learning for the assessment of ecological status of unmonitored waters in Poland.”
Systematic water monitoring in Poland
Water pollution monitoring in Poland began to be systematically addressed and widely discussed in the 1960s. During this period, new legal regulations were introduced, and in subsequent years, specialised institutions responsible for environmental control were established.
Currently, overseeing water protection in Poland have been largely consolidated within the Chief Inspectorate of Environmental Protection (GIOŚ) and the Polish Waters National Water Management Authority. Water status monitoring is now based on national and EU regulations, particularly the Water Framework Directive.
Approximately 93% of the uniform bodies of surface water are subject to regular, cyclical monitoring, although this does not mean full coverage of all waters in the country. The Chief Inspectorate of Environmental Protection (Główny Inspektorat Ochrony Środowiska) conducts regular monitoring activities, with the results systematically published on its official website.
The most common assessment methods
The assessment of the ecological and chemical status of water is a critical step in the water management cycle.
The most commonly used methods include physico-chemical monitoring, which involves the systematic assessment of water quality through the analysis of physical parameters (e.g., temperature, conductivity, turbidity) and chemical parameters (e.g., oxygen content, concentrations of nitrates, phosphates, heavy metals, organic substances); biological monitoring, which evaluates the condition of aquatic ecosystems based on the analysis of organisms living in the water (e.g., phytoplankton, macrophytes, macroinvertebrates); and hydromorphological monitoring, which includes the analysis of water flows, riverbed structure, and hydrotechnical modifications.
For unmonitored water bodies or those with no or incomplete data, the assessment of water status primarily relies on expert judgement. In such cases, specialists, based on their knowledge, experience, and the limited data available, provide an approximate evaluation of the water’s condition.
A new alternative?
The issue Polish scientists have decided to tackle concerns unmonitored water bodies. These are often waters that lack sufficient monitoring data or have none at all, but still require assessment. The researchers aimed to find an alternative to the most commonly used method in that case, which involves assessing the status of unmonitored waters through expert judgment.
“Without appropriate data, the expert method is justified because it relies on the knowledge and experience of local specialists who are familiar with the streams and lakes they oversee. This gives them the best understanding of their current status. The problem arises when such experts are unavailable, or the number of unmonitored waters becomes too large for detailed analysis,” explains Andrzej Martyszunis, co-author of the article.
According to the researchers, the use of machine learning enables the processing of available data to assess unmonitored waters with relatively high accuracy. This also allows them to utilise information not derived from monitoring, such as catchment characteristics, anthropogenic pressures on a given river, and similar factors, and, through mathematical models, identify correlations between these factors and the actual status of the river.
What does Machine Learning offer in water monitoring?
“This approach ensures that the assessment is no longer subjective or immeasurable but mathematically calculated, allowing us to provide an exact level of confidence in the results,” adds Martyszunis.
According to the researchers, this method is both quick and accurate, making it theoretically useful for preliminary assessments of water status. Their results could then inform the planning phase of field research, which is both economically and logistically demanding.
As Martyszunis emphasises, collaboration with experts would be invaluable in implementing this method. Their critical insights could refine the model further to meet the expectations of potential users.In terms of technical aspects, the current solution is a Python-based program that essentially only requires a user interface to be fully operational.
“We were particularly pleased that our chosen approach, machine learning, allowed us to create a model that delivered fairly decent results.” concludes Martyszunis.
More:
- Jarosz, Polska Ludowa 1944-1956: T. 15. Warszawa: Instytut Historii Polskiej Akademii Nauk 2017, https://rcin.org.pl/ihpan/Content/64916/PDF/WA303_84239_B155-Polska-T-15-2017_Jarosz.pdf
- Główny Inspektorat Ochrony Środowiska, Monitoring jakości wód, https://www.gov.pl/web/gios/monitoring-jakosci-wod
