Environmental monitoring of aquatic systems is the key requirement for sustainable environmental protection and future drinking water supply. The quality of water resources depends on the effectiveness of water treatment plants to reduce chemical pollutants, such as nitrates, pharmaceuticals, or microplastics. Changes in water quality can vary rapidly and must be monitored in real-time, enabling immediate action. In this study, we test the feasibility of a deep UV Raman spectrometer for the detection of nitrate/nitrite, selected pharmaceuticals and the most widespread microplastic polymers. Software utilizing artificial intelligence, such as a convolutional neural network, is trained for recognizing typical spectral patterns of individual pollutants, once processed by mathematical filters and machine learning algorithms. The results of an initial experimental study show that nitrates and nitrites can be detected and quantified. The detection of nitrates poses some challenges due to the noise-to-signal ratio and background and related noise due to water or other materials. Selected pharmaceutical substances could be detected via Raman spectroscopy, but not at concentrations in the µg/l or ng/l range. Microplastic particles are non-soluble substances and can be detected and identified, but the measurements suffer from the heterogeneous distribution of the microparticles in flow experiments.

The entire water cycle is contaminated with largely undetected micropollutants, thus jeopardizing wastewater treatment. Currently, monitoring methods that are used by wastewater treatment plants (WWTP) are not able to detect these micropollutants, causing negative effects on aquatic ecosystems and human health. In our case study, we took collective samples around different treatment stages (aeration tank, membrane bioreactor, ozonation) of a WWTP and analyzed them via Deep-UV laser-induced Raman and fluorescence spectroscopy (LIRFS) in combination with a CNN-based AI support. This process allowed us to perform the spectra recognition of selected micropollutants and thus analyze their reliability. The results indicated that the combination of sensitive fluorescence measurements with very specific Raman measurements, supplemented with an artificial intelligence, lead to a high information gain for utilizing it as a monitoring purpose. Laser-induced Raman spectroscopy reaches detections limits of alert pharmaceuticals (carbamazepine, naproxen, tryptophan) in the range of a few µg/L; naproxen is detectable down to 1 × 10−4 mg/g. Furthermore, the monitoring of nitrate after biological treatment using Raman measurements and AI support showed a reliable assignment rate of over 95%. Applying the fluorescence technique seems to be a promising method in observing DOC changes in wastewater, leading to a correlation coefficient of R2 = 0.74 for all samples throughout the purification processes. The results also showed the influence of different extraction points in a cleaning stage; therefore, it would not be sensible to investigate them separately. Nevertheless, the interpretation suffers when many substances interact with one another and influence their optical behavior. In conclusion, the results that are presented in our paper elucidate the use of LIRFS in combination with AI support for online monitoring.

No abstract