Treatment wetlands experience significant seasonal temperature variations that affect
biological treatment processes through complex interactions between plants, microorganisms,
and environmental conditions. This project employs a dual methodological approach to
optimize cold-climate treatment wetland design and operation. First, we developed
interpretable machine learning models using data from 27 published studies and 118
treatment wetlands to predict effluent temperature (RMSE = 0.7°C) and ammonia (RMSE 2.9
– 9.3 mg/L) and organic matter (RMSE 11 – 44 mg/L) concentrations in different wetland
configurations. Model interpretation revealed that influent temperature is the dominant
predictor of effluent temperature followed by air temperature, with hydraulic loading rate
modifying this relationship. For contaminant removal, we found that plant species' coldtemperature benefits are most pronounced in saturated systems, while their impact is
marginal in unsaturated configurations.
Building on these insights, we conducted controlled microcosm experiments with Carex
utriculata in batch-operated wetlands across temperature conditions (23°C and 5°C) and after
harvesting. C. utriculata maintained superior organic matter removal in cold conditions
(86±5% versus 73±9% for unplanted systems) and continued to outperform unplanted
systems even after harvesting. Evapotranspiration-driven water level reduction in warm
conditions improved porosity maintenance by 2-3% compared to unplanted systems. Our
integrated approach demonstrates that C. utriculata provides resilient cold-temperature
treatment benefits while offering operational advantages through reduced clogging potential
and maintained performance post-harvest, addressing key challenges identified in machine
learning analysis of published literature.
