Site-level PFAS fingerprint dataset across eight major river basins of China (1,639 sampling sites, 2010-2023)
Description
This dataset provides the site-level per- and polyfluoroalkyl substance (PFAS) concentrations underlying a national-scale study of compositional fingerprints and ecological risk in Chinese surface waters. Concentrations of ten core congeners (PFOA, PFOS, PFHxA, PFHpA, PFBS, PFBA, PFNA, PFPeA, PFHxS, PFDA) were compiled from 49 peer-reviewed studies published between 2010 and 2023 for 1,639 surface-water sampling sites, which were assigned to the eight major river basin divisions of China. Non-detects were handled by a three-track rule: values reported below the limit of detection were set to LOD/2; values reported as not detected without an LOD were set to 0; and congeners not measured at a site were set to 0 in the main-line treatment, with the sensitivity of this rule reported in the associated manuscript. The dataset also contains the self-organizing map (SOM, k = 4) fingerprint-cluster assignments for the same sites, together with the derived compositional percentages and the driver-feature fields used in the analysis. Two tabular files are provided: site-level concentrations with per-value imputation provenance, and cluster assignments with basin, grid coordinates and mass percentages. Full column descriptions, imputation rules and provenance are given in README.md. All values are traceable to the source publications listed in the accompanying manuscript.
Files
Steps to reproduce
Site-level concentrations were extracted from the 49 peer-reviewed studies listed in Table S1 of the associated manuscript; every record in pfas_1639sites_core10_filled.csv carries the source DOI in the column 论文doi. Values reported below the limit of detection were set to LOD/2; values reported as not detected without an LOD were set to 0; congeners not measured at a site were set to 0 (main-line treatment). Per-value provenance is recorded in the <CONGENER>_fill_method columns. Compositional percentages were computed after normalizing the sum of the 10 core congeners to 100. SOM clustering was performed with k = 4 on the compositional matrix; the resulting cluster labels, grid coordinates and derived percentages are provided in som_cluster_assignments_1639.csv. Column names in the first file are in Chinese; see README.md for the English column dictionary.
Institutions
- Tianjin UniversityTianjin, Tianjin