Chemical log removal credit assessment for a reverse osmosis barrier using a long term operational dataset

Broadscale assignment of removal credits (as log reduction values (LRVs)) for chemical contaminants in reverse osmosis (RO) systems is not as well established as equivalent LRVs for pathogens in water reuse operations. Existing studies on contaminant removal are predominantly based on laboratory and pilot-scale experiments with limited replication, and where monitoring has been undertaken at full scale, the availability of datasets is limited. This study presents an analysis of a comprehensive dataset. A further issue is that it is preferable to assign removal credits (typically used for regulatory purposes) to classes of chemicals rather than to individual chemical species, whereby a single chemical is chosen as a conservative representative of a sub-class of chemicals that share similar mechanistic removal traits. The combination of assigning removal credits using a lower-tail LRV estimate calculated from the LRV distribution measured under the most challenging operational conditions and using a chemical that is conservative in removal performance for a given sub-class of chemicals (in terms of removal mechanism) offers a highly conservative but defensible approach that makes it consistent with the allocation of removal credits for pathogens. A long-term operational dataset from a full-scale advanced water treatment plant producing purified recycled water was used to establish the criteria that allowed statistical evaluation of representative operational performance indicators of an RO water treatment barrier. Parameters including sulfate, sodium, and electrical conductivity were used in this instance, inclusive of establishing the minimum replicate data requirements for LRV estimation. The dataset (>4 years of data, >500,000 readings, >100 chemical species) was then evaluated across a range of chemical sub-classes using indicative LRV credits to validate the approach and assess the robustness of the conservative credit assignment. Application of conservative sub-class-level assumptions resulted in no cases where the presence of a molecule or ion in the RO permeate was not predicted (i.e. no false negative results). This provides confirmation that these groupings allow protective coverage of observed chemical removal behaviour through an RO barrier. These findings demonstrate that full-scale operational monitoring data can support statistically robust, risk-based assessment of RO chemical removal. The extensive operational dataset supports the derived chemical sub-classes and the process for identification of performance surrogates. There were no observed gaps in the RO chemical sub-class groupings, demonstrating that chemical sub-classes (using performance surrogates) can be used to assess individual chemical removal LRVs without undertaking challenge testing for individual chemicals.