Using QSPR models to predict the removal of micropollutants

AquaPriori expanded to include models that cut experimentation time and costs

Drinking water sources increasingly contain more and more organic micropollutants (OMPs). Water utilities use a range of treatment methods to remove them. With new models that have now been included in the online tool AquaPriori, better predictions can be made about which method will perform best for the removal of specific OMPs, saving on the money and time required for experiments to identify the best treatment method. The models are based on Quantitative Structure Property Relationships that establish a link between the molecular structure of pollutants and treatment processes. For treatment with advanced oxidation and membrane technology, these models now deliver more than 80% accuracy. For treatment with activated carbon, the effects of the natural organic matter present still has to be integrated better in the models.

Because more and more organic micropollutants (OMPs) are found in drinking water sources, additional treatment methods such as membrane filtration, adsorption to activated carbon or advanced oxidation processes are often required. But how do you know which of the available methods works best for a micropollutant that has just been identified? Testing every method for each micropollutant is expensive and time-consuming. In response to this question, the collective research programme of the drinking water utilities, BTO, has developed models that can predict which treatment method is likely to be most effective.

Training models

Models were therefore trained using data from ten years of pilot experiments at KWR and from the literature. Information about the removal of 100 to 500 compounds was used for each treatment method. Quantitative Structure Property Relationships (QSPRs) were established for these compounds to establish a statistical relationship between reaction parameters (such as the quantum yield and various reaction constants) and their molecular structure. The specific parameters of a compound were used in models of the processes in the different treatment methods to predict how well that particular compound would be removed using advanced oxidation, membrane filtration and activated carbon treatment.

Predicting OMP removal on the basis of process and statistical models (QSAR/QSPRs).

Selection of OMPs for validation

To validate the models, a representative selection of OMPs was made on the basis of a mathematical selection method. This resulted in a set of thirty OMPs that have been used in pilot installations with all three treatment methods.

Advanced oxidation

For the oxidation methods, the process models were combined with QSPRs for specific kinetic parameters such as reaction rate constants for O3 and OH radicals and quantum yield. In this way, models were created that can predict the oxidation and/or photolysis of OMPs during UV treatment, ozonisation, treatment with ozone O3 and hydrogen peroxide H2O2, treatment with UV and hydrogen peroxide, and treatment with UV and ozone. In a comparison with the measured values for the set of 30 OMPs, the models were found to be more than 80% accurate.

Membrane filtration

Models were also developed to predict the retention of OMPs with membrane techniques: spiral-wound nanofiltration (NF) and reverse osmosis (RO) membranes and capillary NF membranes. QSPRs were used to predict membrane permeability for OMPs. Once again, an accuracy of more than 80% was achieved.

The retention of organic micropollutants during drinking water treatment with reverse osmosis (RO). Predicted retention is shown in orange, measured retention in green. Blue is the average retention for other types of membrane.

Treatment with activated carbon

Finally, a model was developed to predict when OMPs ‘break through’ (i.e: are no longer removed by the carbon) during activated carbon treatment. In pure water, which does not contain natural organic materials (NOMs), this can be accurately predicted. But water usually contains NOM, and that must then be included in the activated carbon treatment process because NOM competes with OMPs in the adsorption or the pores in the activated carbon, where the adsorption of OMPs takes place. A better model is currently being developed to correctly incorporate the effects of different fractions of natural organic materials on adsorption to activated carbon for different OMPs.


Using the models mentioned above, an efficient method was developed to predict which type of treatment method may work best with a new organic micropollutant. These models have been included in the online tool AquaPriori that water companies and laboratories can use to establish an estimate quickly of the removal of organic micropollutants. After that, of course, experimental studies will be needed to verify these predictions. However, because the models provide a basis for the direction of research, those experiments will require less time and funding.