This work investigated the use of microbial fuel cells (MFCs) for the degradation of polyethylene terephthalate (PET) and the simultaneous generation of electricity. The study implemented two separate single-chamber MFCs, one with a co-culture of Ideonella sakaiensis and Geobacter sulfurreducens (I.S-G.S) and the other with Ideonella sakaiensis and activated sludge (I.S-AS). The effectiveness of microplastic (MP) degradation was assessed based on the electroactivity of the anodic biofilm, the reduction in particle size, and the decrease in PET mass. Both systems achieved a significant reduction in MP size and mass, with the I.S-AS system notably surpassing the I.S-G.S in terms of efficiency and electricity generation. The I.S-AS system achieved a 30% mass reduction and 80% size reduction, along with a peak voltage of 222 mV. The study concludes that MFCs, particularly with the activated sludge co-culture, offer a viable and more environmentally friendly alternative for MP degradation and energy recovery. These findings suggest a promising direction for improving waste management practices and advancing the capabilities of bio-electrochemical systems in addressing plastic pollution. Further research is recommended to optimize the operational conditions and to test a broader range of MP sizes for enhanced degradation efficacy.

Prediction of pavement roughness is central to network-level planning, yet environmental stressors remain underrepresented in data-driven models. This study develops an input variable selection and ensemble-learning workflow to predict the International Roughness Index (IRI) and support network-level identification of deterioration risk. Segment-level roadway data for Beaumont, Texas (2019–2023; ≈11,900 segments) integrate pavement condition surveys, design and geometry attributes, maintenance and rehabilitation (M&R) records, and environmental layers describing flood inundation depth and velocity, flood susceptibility, topographic convergence, and ground deformation. Edgeworth-approximation Conditional Mutual Information (EA-CMI) and SHapley Additive exPlanations (SHAP) are used to screen and interpret predictors, and Random Forest and Extreme Gradient Boosting (XGBoost) are evaluated against a linear baseline. XGBoost achieves the strongest performance (R² ≈ 0.79 in 2019; R² ≈ 0.73 in 2023), while a SHAP-guided reduced model retains near-parity accuracy with fewer inputs, lowering data demands for practical deployment. Inter-year transfer of the model trained on the 2019 data to the 2023 observations reveals temporal drift, with larger errors concentrated among the most distressed segments. EA-CMI prioritizes structural and distress-related variables while mitigating redundancy before model fitting, and SHAP corroborates the dominant role of distress indicators while highlighting interaction patterns associated with flooding and deformation. Spatial diagnostics indicate clustering in both roughness change and transfer residuals, with deterioration corridors more frequently co-located with high flood exposure and ground movement, whereas spatially coherent improvements are more consistent with recent M&R. Overall, the workflow supports updating and interpretation of roughness predictions for resilience-oriented pavement management in hazard-prone regions.