AbstractProduced water (PW) generated by Australian offshore oil and gas activities is typically discharged to the ocean after treatment. These complex mixtures of organic and inorganic compounds can pose significant environmental risk to receiving waters, if not managed appropriately. Oil and gas operators in Australia are required to demonstrate that environmental impacts of their activity are managed to levels that are as low as reasonably practicable, for example, through risk assessments comparing predicted no‐effect concentrations (PNECs) with predicted environmental concentrations of PW. Probabilistic species sensitivity distribution (SSD) approaches are increasingly being used to derive PW PNECs and subsequently calculating dilutions of PW (termed “safe” dilutions) required to protect a nominated percentage of species in the receiving environment (e.g., 95% and 99% or PC95 and PC99, respectively). Limitations associated with SSDs include fitting a single model to small (six to eight species) data sets, resulting in large uncertainty (very wide 95% confidence limits) in the region associated with PC99 and PC95 results. Recent advances in SSD methodology, in the form of model averaging, claim to overcome some of these limitations by applying the average model fit of multiple models to a data set. We assessed the advantages and limitations of four different SSD software packages for determining PNECs for five PWs from a gas and condensate platform off the North West Shelf of Australia. Model averaging reduced occurrences of extreme uncertainty around PC95 and PC99 values compared with single model fitting and was less prone to the derivation of overly conservative PC99 and PC95 values that resulted from lack of fit to single models. Our results support the use of model averaging for improved robustness in derived PNEC and subsequent “safe” dilution values for PW discharge management and risk assessment. In addition, we present and discuss the toxicity of PW considering the paucity of such information in peer‐reviewed literature. Integr Environ Assess Manag 2024;20:498–517. © 2023 Commonwealth Scientific and Industrial Research Organisation. Integrated Environmental Assessment and Management published by Wiley Periodicals LLC on behalf of Society of Environmental Toxicology & Chemistry (SETAC).

Key Points For produced water (PW) toxicity data sets with eight species, model averaging was advantageous over single model fitting by reducing occurrences of very wide 95% confidence limits around PC95 and PC99 values and were less prone to derivation of overly conservative PC99 and PC95 values. The two model averaging software packages that we assessed were competitive contenders for use as alternatives to Burrlioz in the Australian context, and either would be appropriate for use with the case study PW data sets. Our findings strongly support the use of model averaging over single‐distribution model fitting to derive predicted no‐effect concentration values for PW discharges.