Phosphate source apportionment across the agriculture-urban gradient in Asia's longest river: Combining machine learning and multi-isotope techniques

Abstract

Excessive phosphorus (P) can have serious impacts on water quality and ecosystems, and accurately identifying P sources is crucial for preventing and controlling eutrophication in watersheds. Traditional source-tracing models are limited in their ability to identify the driving factors influencing P dynamics, leading to the accuracy of source apportionment being influenced by potential driving factors. Therefore, this study introduces machine learning (ML) methods, combined with various receptor models and isotope techniques, to quantitatively analyze the sources and driving factors of phosphate in the Yangtze River Basin (YRB). The ecological condition in the YRB is relatively favorable, with an average concentration of soluble reactive phosphorus (SRP) at 0.076 mg/L and phosphate saturation levels ranging from 18 % to 95 %. The results of δ¹⁸O_(PO4) indicate that agricultural discharges, livestock discharges, phosphate rock, and mixed sources predominantly composed of sewage discharges are the main P sources in the YRB. In the upstream, phosphate rock (54.7 %) and livestock sources (33.9 %) are the primary phosphate sources, whereas in the midstream, agricultural sources (66.1 %) dominate phosphate sources. In the downstream, agricultural sources (48.3 %) and mixed sources (33.3 %) are the main contributors to phosphate. Consequently, the application of ML provides an effective approach for the analysis of P pollution sources and the identification of driving factors, offering important scientific evidence for the management of eutrophication in large river basins across the agriculture-urban gradient.