Canadian A.I. Removal Rate Estimator (CAIRRE): an artificial intelligence model to predict the removal of chemicals in secondary wastewater treatment plants

Understanding the removal of a chemical in a wastewater treatment plant (WWTP) is important when performing chemical risk assessments. Chemicals undergoing assessment often have limited experimental measurements of physicochemical properties, biodegradation rates, and WWTP removal efficiencies. Models available to risk assessors to predict WWTP removal efficiencies are best used with high-quality input data and knowledge of plant conditions, information often unavailable when performing chemical risk assessments. In this work, we outline the development of the Canadian A.I. Removal Rate Estimator (CAIRRE), an artificial intelligence (A.I.) model suite designed to estimate removal efficiencies from secondary WWTPs. CAIRRE was trained on median experimental removal efficiencies for 161 chemicals across 59 secondary WWTPs in Canada, the USA (California), and various other locations curated from literature. The CAIRRE regression model has a validation Pearson R2 of 0.81 based on leave-one-out-validation (LOOV) results. When used to predict effluent concentrations for a test set containing 53 chemicals not seen during model training, CAIRRE was able to reproduce experimental observations with a Pearson R2 of 0.91. The CAIRRE model outperformed existing mechanistic and fugacity WWTP models which rely on physical-chemistry and biodegradation data provided by the user. This work demonstrates that the A.I. modeling approach taken in the development of CAIRRE is a promising strategy for predicting removal efficiencies of chemicals from secondary WWTPs.