Entropy-Guided Multivariate Groundwater Network Design for Multi-Source Data Assimilation Under Observational Uncertainty

Abstract

Observational uncertainty poses major challenges to groundwater model calibration. As the primary source of information for multi-source data assimilation, monitoring network design is critical for accurately characterizing subsurface dynamics. Under limited measurement accuracy or cost constraints, monitoring networks must remain robust to observational errors. This study develops a multivariate network design framework that quantifies the uncertainty of multicomponent responses using joint entropy and employs deep learning to accelerate computations. Case study results show that the framework reliably estimates non-Gaussian permeability fields even under high-noise observations. The calibrated reactive transport model demonstrates strong capability in reproducing historical data and predicting system responses. This work advances the understanding of multi-source data fusion and supports the development of groundwater monitoring networks under observational uncertainty. Moreover, the proposed approach can be extended to the design of geophysical survey lines that integrate geophysical data.