Interpretable kinetic modeling of natural organic matter transformation and attenuation during ozonation and catalytic ozonation
Journal of Environmental Chemical Engineering, cilt.14, sa.6, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 14 Sayı: 6
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.jece.2026.124950
- Dergi Adı: Journal of Environmental Chemical Engineering
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Chemical Abstracts Core, Compendex, INSPEC
- Anahtar Kelimeler: Catalytic ozonation, Kinetic modeling, Natural organic matter, Ozonation, Symbolic regression, Water treatment
- Ondokuz Mayıs Üniversitesi Adresli: Evet
Özet
Natural organic matter (NOM) transformation and attenuation during adsorption, ozonation, and catalytic ozonation is commonly described using simplified kinetic models that struggle to represent nonlinear, condition‑dependent behavior. In this work, NOM transformation was comparatively evaluated using humic acid as an aromatic surrogate, monitored by DOC, UV254, and SUVA across five configurations: adsorption on granular activated carbon (GAC) and iron‑coated zeolite (ICZ), sole ozonation, and catalytic ozonation with GAC or ICZ. Adsorption alone afforded limited NOM removal; even at 60 min, DOC reductions remained modest (≤17.7% with ICZ, ≤13.6% with GAC), indicating that sorption mainly redistributes NOM fractions rather than driving substantive structural change. Sole ozonation preferentially suppressed UV‑absorbing chromophores, yielding marked UV254 attenuation but only moderate DOC removal. Within the experimental window, catalytic ozonation delivered the strongest overall performance, with ICZ‑assisted ozonation achieving up to 66.4% DOC removal and more than 97.5% UV254 attenuation. To describe the resulting decay patterns, symbolic regression was integrated with classical kinetic analysis. Symbolic equations captured multivariable, nonlinear behavior with strong full-data descriptive performance relative to fixed-form models (PySR R2≥0.96 in most full-data fits). Because full‑data fit can overstate predictive reliability, grouped, leave‑condition‑out cross‑validation was additionally performed to estimate generalization to unseen operating conditions. Grouped validation showed the most consistent performance for catalytic ozonation, with pooled test R2 values of 0.862–0.962 across DOC, UV254, and SUVA. In contrast, adsorption SUVA and sole-ozonation SUVA showed weak and variable held-out performance; therefore, these equations should be interpreted as descriptive, within-domain models.