A novel-approach for identifying sources of fluvial DOM using fluorescence spectroscopy and machine learning model

Abstract

Rivers are well known as one of the most threatened aquatic environments, whose structure and water quality can be deeply impacted by intensive anthropogenic activities. Despite the fact that anthropogenic influences on river ecosystems could indeed be deduced from the composition and chemistry of fluvial dissolved organic matter (DOM), sources of anthropogenic loading to DOM are still poorly explored. Here, by uniting fluorescence excitation-emission matrices (EEM) and principal component absolute coefficient, four sources of DOM from seventeen rivers in major drainage basins of China could be identified, i.e., originating from municipal sewage, domestic wastewater, livestock wastewater, and natural origins, and thus being defined as MS-DOM, DW-DOM, LW-DOM, NO-DOM, respectively. Based on the random forest model, special nodes in EEM could be traced from four sources, respectively. According to parallel factor analysis, DOM mainly contained protein-like, microbial humic-like, and fulvic-like fluorescence substances, among which protein-like was dominant in MS-DOM and DW-DOM, microbial humic-like in LW-DOM, and fulvic-like in NO-DOM. Based on key peaks and essential nodes in EEM, the identifying source indices were first proposed, which could be introduced to simply distinguish the different anthropogenic-derived sources of fluvial DOM. It was associated with intensity ratios of the key peaks and the essential nodes of EEM spectra from four sources, i.e., municipal sewage (MS-SI: Ex/Em = 280/(335, 410) nm), domestic wastewater (DW-SI: Ex/Em = 280/(340, 410) nm), livestock wastewater (LW-SI: Ex/Em = 235/(345, 380) nm), and natural origins (NO-SI: Ex/Em = 260/(380, 430) nm). By statistical analysis, the high identifying source indices of municipal sewage (>0.5) and natural origins (>0.4) values could be related to MS-DOM and NO-DOM, respectively. The identifying source indices of domestic wastewater with 0.1–0.3 might be linked to DW-DOM and the identifying source indices of livestock wastewater with 0.3–0.4 to LW-DOM. Compared with conventional optical indices, the novel identifying source indices showed remarkable discrimination for the sources of fluvial DOM with different forms of anthropogenic disturbances. Hence the innovative approach could be relatively convenient and accurate to evaluate water quality or pollution risk in river ecosystems.