Improved Understanding of Dissolved Organic Matter Processing in Freshwater Using Complementary Experimental and Machine Learning Approaches

Abstract Dissolved organic matter plays an important role in aquatic ecosystems and poses a major problem for drinking water production. However, our understanding of DOM reactivity in natural systems is hampered by its complex molecular composition. Here, we used Fourier-transform ion cyclotron resonance mass spectrometry (FT-ICR-MS) and data from two independent studies to disentangle DOM reactivity based on photochemical and microbial-induced transformations. Robust correlations of FT-ICR-MS peak intensities with chlorophyll a and solar irradiation were used to define 9 reactivity classes for 1277 common molecular formulas. Germany’s largest drinking water reservoir was sampled for 1 year, and DOM processing in stratified surface waters could be attributed to photochemical transformations during summer months. Microbial DOM alterations could be distinguished based on correlation coefficients with chlorophyll a and often shared molecular features (elemental ratios and mass) with photoreactive compounds. In particular, many photoproducts and some microbial products were identified as potential precursors of disinfection byproducts. Molecular DOM features were used to further predict molecular reactivity for the remaining compounds in the data set based on a random forest model. Our method offers an expandable classification approach to integrate the reactivity of DOM from specific environments and link it to molecular properties and chemistry.