Refining the diagnostics of non-point source metals pollution to urban lakes based on interaction normalized PMF coupled with Bayesian network.

Spatiotemporal variability complicates source apportionment of metals in urban lakes, especially when rainfall drives urban non-point source pollution. As, Cd, Cr, Pb, Hg, Ag, Co, Cu, Fe, Mn, Ni, Sb, Sr and Zn concentrations in 648 water samples collected before and after rain in 6 urban lakes of Beijing, China were determined during 2013-2015. The response of metals concentrations after rain to the interaction between rainfall and antecedent dry days was significant. Metals concentrations were normalized pursuant to the interaction effect as the input of positive matrix factorization (PMF) to develop the interaction normalized-PMF (IN-PMF). Four primary pollution sources were diagnosed. Sediment release was considered to be the main source of Fe, Co and Ni independent of rainfall. Hg, As and some Cr associated with pesticides and fertilizers were likely to come from soil erosion and runoff from green space. It is probable that road runoff was the dominant source for heavy metals related to traffic emissions, including Pb, Cd, Cu, Sb, Mn and Zn. Cr, Sr and some Cu and Zn as key elements of rooftops can be regarded as from roof runoff. The IN-PMF lowered roof and road runoff contributions and raised the contribution of soil erosion from green space, with Pb, Sb, Cu, Zn, Cd and Mn increasing by 15.9%, 10.7%, 13.1%, 12.2%, 13.3% and 16.8%. The results shed more light on the stormwater runoff pollution mitigation on impervious surfaces and metals enrichment problems in infiltration soil on green space in the low impact development (LID) setting. The Bayesian network revealed the spatial variability of transport and fate of metal elements from land surfaces to urban lakes, supplementing the secondary pollution sources from different land use. This study will provide new insights for source apportionment of non-point source pollution under the background of sponge city construction.