Ensemble surrogate modeling of advective-dispersive transport with intraparticle diffusion model for column-leaching...

Area of Science:

• Environmental Science
• Geochemistry
• Computational Modeling

Background:

• Column-leaching tests assess environmental risks of contaminated soils and waste materials reused in construction.
• Contaminant transport involves complex solute dynamics and mass transfer, often requiring computationally intensive numerical models.
• Inverse modeling and sensitivity analysis are time-consuming, especially with detailed sorption kinetics like intraparticle diffusion.

Purpose of the Study:

• To develop computationally efficient surrogate models for column-leaching tests.
• To accurately emulate complex solute transport and mass transfer dynamics.
• To enable faster inverse modeling and parameter estimation for environmental risk assessment.

Main Methods:

• Developed two ensemble surrogate models: random forest stacking and inverse-distance weighted interpolation.
• Utilized Extremely randomized Trees (ExtraTrees) as base surrogate models.
• Employed adaptive sampling with criteria for exploitation and exploration to optimize models.
• Applied Simulation-Based Inference (Neural Posterior Estimation) for parameter distribution estimation.

Main Results:

• The ensemble surrogate model accurately emulates the original numerical model with a relative root mean squared error of 0.09.
• The model successfully estimated the complete posterior parameter distribution using Simulation-Based Inference.
• Posterior distribution samples aligned perfectly with observed copper-leaching data for both surrogate and original models.

Conclusions:

• Ensemble surrogate models offer a computationally efficient alternative to traditional numerical models for column-leaching tests.
• These models accurately predict contaminant leaching behavior and environmental risks.
• The approach facilitates robust parameter estimation, enhancing the safe reuse of contaminated soils and waste materials in construction.