PFASorptionML

Large Scale Assessment

PFAS compound

Select a compound

2D structure

Pick a compound to load its molecular properties.

Ambient pH 7.0

Acid–base speciation

Species distribution as a function of pH

Select a PFAS with a known pKa to see its acid–base speciation across pH.

Log Kd [L/kg]

.........

Soil property

………

Local Assessment

PFAS Compound Lookup

Enter a PFAS abbreviation, CAS number, or PubChem CID.

Search a compound to view its properties...

Enter Soil Properties

Provide soil parameters for K_d estimation.

Organic Content (%)

pH [-]

CEC [cmol(+)/kg]

Silt (%)

The hybrid model uses organic carbon, pH, CEC and silt. Sand and clay are not model inputs. The molecular descriptors (q_head, P_ES, f_polar, n_CF₂) come from the selected PFAS, evaluated at your soil pH.

The previous model uses MW, logK_ow, net charge, pH, CEC, sand, silt and clay. Charge density is filled from the database or computed from pK_a. No molecular charge descriptors and no uncertainty band.

Upload Local Data (OPTIONAL)

Upload your local dataset (.xlsx, .csv, or .txt).

Your file must include these columns:


      lon, lat, oc, cec, ph, silt

No data uploaded yet.

About PFASorptionML

PFASorptionML predicts how strongly soils hold PFAS. You pick a compound and a place. The model returns the sorption coefficient K_d, with its uncertainty.

The model is a hybrid of soil chemistry and machine learning, trained on lab measurements. Inputs are four soil properties (organic carbon, CEC, pH, silt) and molecular descriptors of the compound, including its charge at the ambient pH.

What you can do here:

Read predicted log K_d anywhere on the global map, for 60+ PFAS.
See how a compound speciates with pH, and its DFT charge surface in 3D.
Compare the K_d map with topsoil properties, side by side.
Enter your own soil data for a local prediction.

Version 2.0 (Hybrid) is submitted to Environmental Science & Technology Letters. Version 1.0 was published in Environmental Science & Technology (2025).

Predictions are model estimates, not measurements.