Sadaghat, Behnam, Behzadian, Kourosh ORCID: https://orcid.org/0000-0002-1459-8408 and Najjar-Ghabel, Saeid (2025) Towards sustainable WWTPs: explainable modeling and scenario analysis of N₂O emissions. Journal of Water Process Engineering, 79. ISSN 2214-7144

Preview

PDF Sadaghat et al 2025-JWPE.pdf - Published Version Available under License Creative Commons Attribution. Download (17MB) | Preview

Abstract

Accurate prediction of nitrous oxide (N₂O) emissions from wastewater treatment plants (WWTPs) has been hindered by the dynamic, non-linear nature of the processes involved and by a lack of uncertainty treatment and scenario analysis in past studies. While machine learning (ML) has been shown to improve accuracy, limitations have been found in standardizing preprocessing, dual optimizer usage, post-hoc interpretability, and decision-focused validation. In this study, N₂O emission rates were predicted using Elastic Net (EN), Quantile Regression (QR), Gaussian Process Regression (GPR), and Gradient Boosting Regression (GBR). Data preprocessing was conducted by daily averaging of WWTP N2O emission rate, Z-score filtering, VIF-based multicollinearity check and feature selection, and normalization. The models were evaluated in a 5-fold cross-validation with five iterations. Hyperparameter optimization was performed using the Puma Optimizer (PO) and Bayesian Optimization (BO). Post-hoc sensitivity (as here specifically defined as feature-attribution interpretability) was achieved using SHAP (global effects) and LIME (local effects). Uncertainty was approximated using a Monte Carlo procedure (noisy inputs, 1000 iterations).
Numerically, baseline GPR reached R2 (test) = 0.958; (RMSE (test) = 0.020; MAE (test) = 0.012). With optimization, GPR + BO (GPBO) achieved the highest performance with R2 (test) = 0.986, RMSE (test) = 0.012, and MAE (test) = 0.007, while GPR + PUMA (GPPO) performed as high as R2 (test) = 0.980; (RMSE (test) = 0.014; MAE (test) = 0.009). SHAP and LIME analysis further established dissolved N₂O as the most significant predictor, with conditional contributions from ammonium (NH₄+), dissolved oxygen (DO), nitrate (NO₃−), and temperature. Scenario analysis with CGAN (N₂O + 20 %; NH₄+ − 15 % with +5 °C; and multi-variate perturbations) was conducted with the top model to demonstrate decision-oriented responsiveness.
The study contribution is an end-to-end, decision-ready pipeline that combines standardized preprocessing, dual optimizers, SHAP and LIME interpretability, Monte Carlo uncertainty, and CGAN-based what-if testing, capabilities that have been limited or scattered in prior work. Relevance is demonstrated by actionable environmental and economic co-benefits: peak N₂O abatement potential (minimized GHG footprint, alignment with IPCC-style reporting) and aeration energy and chemical dosing cost minimization via predictive control, thereby facilitating sustainable and low-cost WWTP operation.

Abbreviations
N₂O, nitrous oxide; NH₄+, ammonium; DO, dissolved oxygen; NO₃−, nitrate; WWTPs, wastewater treatment plants; ML, machine learning; EN, Elastic Net; QR, Quantile Regression; GPR, Gaussian Process Regression; GBR, Gradient Boosting Regression; PO, `Puma Optimizer; BO, Bayesian Optimization; XAI, Explainable Artificial Intelligence; SHAP, Shapley Additive explanation; LIME, Local Interpretable Model-Agnostic Explanations; CGAN, Conditional Generative Adversarial Network; R2, coefficient of determination (RMSE), (MAE), (RSE), (MAPE), and (SMAPE); RMSE, root mean square error; RSE, relative squared error; MAE, mean absolute error; SMAPE, symmetric mean absolute percentage error; MAPE, mean absolute percentage error

Item Type:	Article
Identifier:	10.1016/j.jwpe.2025.108818
Keywords:	Nitrous oxide; Water treatment plants; Regression methods; Sensitivity techniques; Prediction
Subjects:	Construction and engineering > Civil and environmental engineering
Date Deposited:	29 Sep 2026
Dates:	Date Publication status 22 September 2025 Accepted 9 October 2025 Published Online
School, department or research centre:	School of Computing and Engineering
Keywords:	Nitrous oxide; Water treatment plants; Regression methods; Sensitivity techniques; Prediction
URI:	https://repository.uwl.ac.uk/id/eprint/15453
Sustainable Development Goals:	Goal 6: Clean Water and Sanitation	Sustainable Development Goals:	Goal 9: Industry, Innovation, and Infrastructure	Sustainable Development Goals:	Goal 11: Sustainable Cities and Communities	Sustainable Development Goals:	Goal 13: Climate Action

Downloads

Actions (admin access)