Four graphs from the paper.
The proposed model (blue line) accurately predicts the contaminant concentration in three different core samples. More importantly, the model can learn the retardation factor as a function of contaminant concentration, which could not be captured by a conventional, calibrated diffusion-sorption model (orange). With an uncertainty quantification method, the range of the model prediction shown as blue shade, and the the model predictions successfully cover the noisy measurement data. Credit: Praditia et al. [2022], Figure 11
Editors’ Highlights are summaries of recent papers by AGU’s journal editors.
Source: Water Resources Research

Machine learning (ML) approaches are highly useful in leveraging observation data to discover new insights about a natural system. However, they require a significant amount of data, which are often expensive and difficult to acquire.

When data availability is sparse, the finite volume neural network (FINN) model proposed by Praditia et al. [2022] offers a better solution to effectively exploit and learn from the data. FINN demonstrates the ability of ML models to represent physical processes as expressed in partial differential equations (PDEs). The approach also shows a significant potential for generalization to various conditions and for incorporating uncertainty quantification. Additionally, FINN performs exceedingly well compared to other ML approaches, and it is able to quantify the uncertainty of “missing pieces” in the differential equation models that were learned by FINN.

Citation: Praditia, T., Karlbauer, M., Otte, S., Oladyshkin, S., Butz, M. V., & Nowak, W. (2022). Learning groundwater contaminant diffusion-sorption processes with a finite volume neural network. Water Resources Research, 58, e2022WR033149.

—Xavier Sanchez-Vila, Editor, Water Resources Research

Text © 2023. The authors. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.