01

Database scope

The harmonized database contains 24,599 concentration records for 14 PFAS from 45 countries and regions between 2002 and 2024.

Concentration records

—

Period

—

Countries / regions

—

PFAS

—

Release date

—

14 PFAS, including:

02

Literature search and screening

Web of Science and PubMed were searched for peer-reviewed studies reporting PFAS concentrations in drinking water. Search terms combined drinking water terms, including “drinking water” and “tap water,” with PFAS terms, including “per- and polyfluoroalkyl substances,” “fluorocarbons,” and 15 target PFAS compounds.

The searches identified 2,902 records in Web of Science and 577 in PubMed. After 542 duplicates were removed, 2,937 records were screened. Review articles, conference abstracts, and reports were excluded at this stage (n = 698), leaving 2,239 articles for eligibility assessment. Eligibility assessment excluded 2,091 records that examined non-drinking water matrices, reported only non-target PFAS, lacked identifiable sampling locations or years, or did not provide extractable concentration data. Screening retained 148 studies after consensus, and reference screening identified one additional study, resulting in 149 studies and 24,599 concentration records.

Studies were eligible if they reported extractable concentrations of at least one target PFAS in drinking water, tap water, or water explicitly used for drinking, together with sampling location, sampling year, and LOD/LOQ information. Three researchers independently screened titles and abstracts, followed by the full texts and supplementary materials of potentially eligible studies. Discrepancies were resolved through discussion until consensus was reached. The three researchers then independently extracted and cross-checked the required data.

03

Data cleaning and harmonization

Raw records were cleaned and harmonized before statistical analysis and model development. Sampling locations were assigned to the smallest consistently identifiable administrative unit. Where data coverage was sufficient, records from China and the United States were further assigned to provincial and state levels, respectively. Country names and Chinese province/U.S. state names were connected to stable geographic identifiers through an explicit crosswalk.

All concentration data were converted to ng/L. Median concentrations were used as representative values because environmental concentration data are commonly right-skewed; when medians were unavailable, mean values were used instead. Non-detect concentrations were replaced with LOD/√2. When multiple records were available for the same spatial unit, sampling year, and PFAS compound, they were aggregated using the median concentration.

The final harmonized database contains 24,599 concentration records from 45 countries and regions between 2002 and 2024.

04

Website calculations

Each displayed row represents one geographic unit × sampling year × PFAS combination. Median is the harmonized median concentration in ng/L, and Total_Samples is the supporting sample count.

05

Model predictions

The prediction layer presents 2015 results for PFOA, PFOS, PFHxS, and PFNA. We used spatial machine-learning models to predict the median levels of these four PFAS in drinking water at the global scale. The displayed values are Monte Carlo median predictions expressed in ng/L.

The model applicability domain is determined using the Mahalanobis distance in model feature space. The prediction controls can display either all locations or within-domain locations only.

06

Geographic resolution

China is shown at province level, the United States at state level, and all other covered locations at country level, following the same geographic structure as the observed database.

07

Limitations

• Geographic coverage of the observed data is uneven, with relatively sparse data in Africa, South America, and parts of Asia.
• Sampling intensity and sampling years vary across regions.
• The website displays aggregated values, not individual sample measurements.
• Grey areas indicate no available data under the active filters.