Siwei Gu

The structural diversity and complex transport behavior of per- and polyfluoroalkyl substances (PFAS) complicate a universal characterization of their removal in membrane systems. This study compiles 2353 data points from the literature on PFAS rejection by nanofiltration and reverse osmosis membranes, spanning a broad range of PFAS, membranes, feedwater compositions, and operating conditions. Using machine learning, this data set is modeled to evaluate how solute, membrane, and solution properties jointly influence PFAS removal. Of the 13 experimental system descriptors analyzed, membrane water permeance and PFAS molecular volume demonstrated the strongest main effect on rejection, emphasizing the significance of steric exclusion. The effects of background ions and dissolved organic matter were highly condition-dependent, exhibiting nonmonotonic behavior governed by competing mechanisms. Low concentrations of organic matter and ions enhanced PFAS rejection, consistent with complexation that increases apparent solute size, while higher concentrations reduced PFAS rejection, indicating that charge shielding and concentration polarization increasingly drive transport behavior. Overall, this analysis provides a unified, data-driven framework for interpreting previously inconsistent findings across studies, identifies critical gaps in existing experimental data, and highlights opportunities to guide targeted membrane design and treatment strategies.