Coupling Ensemble Smoother and Deep Learning for Flow and Transport Data Assimilation

Author: Jichao Bao, Department of Geology and Geological Engineering

Mentor: Dr. Liangping Li, Department of Geology and Geological Engineering

It is of significance to understand the groundwater flow and solute transport systems for water resources management and aquifer remediation. Data assimilation approaches such as Ensemble Smoother with Multiple Data Assimilation (ES-MDA) can assimilate dynamic data (e.g., hydraulic head and concentration data) into groundwater models to improve the model predictive ability. However, the ES-MDA has an optimal solution only when the aquifer parameters follow a multi-Gaussian distribution. To deal with non-Gaussian conditions such as channelized aquifers and improve the predictive ability of flow and transport simulation, we propose to couple ES-MDA and deep learning. Specifically, Variational Autoencoder (VAE), a deep learning method, is used to reparameterize channelized aquifers with low-dimensional latent variables. The ES-MDA is then used to update the latent variables by assimilating hydraulic head or concentration data into the groundwater model. Synthetic cases are presented to show the performance of the proposed approach. The results demonstrate that the integration of ES-MDA and VAE can reconstruct the channel structures and reduce the uncertainty of flow and transport prediction.