Numerical modelling is a useful tool for groundwater resource management. However, many groundwater models are subject to large uncertainty. This is partly due to an inadequate understanding of the subsurface, as traditional hydrogeological techniques generally can only provide point-scale information, while groundwater models require spatially continuous data.
Electromagnetics (EM), a geophysical technique that estimates electrical conductivity (EC), is commonly used to compliment groundwater modelling due to its relatively large spatial coverage. EC can be related to hydraulic conductivity (K) through a petrophysical relationship as they are mutually dependent on pore surface area and volume. Petrophysical relationship often contains empirical constants that require calibration. In practice this is often achieved by comparing with aquifer tests, which may be suboptimal due to the scale difference between EM and aquifer tests.
Surface Nuclear Magnetic Resonance (SNMR) is a geophysical technique that is gaining traction in hydrogeology due to its ability to estimate K, water content and porosity. However, in practice SNMR data are more difficult and expensive to obtain than EM, hence they are generally not as spatially continuous. Nevertheless, SNMR can provide an additional set of K estimates at a similar scale to EM, which can constrain the petrophysical relationship between EC and K.
The objective of this research is to evaluate the benefits of including EM and SNMR in groundwater modelling using a synthetic approach. The methodology includes developing a range of reference models, representing the synthetic truths, to cover various hydrogeological conditions that may affect the effectiveness of EM and SNMR, including clay condition, groundwater salinity and level of hydrogeological heterogeneity. Synthetic data are sampled from the reference models and used to create ensembles of groundwater models using a Markov-Chain Monte Carlo (MCMC) approach. The performance of the model ensembles is evaluated by comparing their predictions with the reference models. The results show that including EM and SNMR in a homogeneous, conductive ground environment may have the potential to distort the model predictions due to the geophysical inversion uncertainty and noise. In contrast, model predictions in a heterogeneous environment can be improved by the spatial pattern information from EM and the additional K constraints from SNMR, where the amount of improvement depends on the ground conductivity. In addition to demonstrating the benefits of EM and SNMR to groundwater modelling, this work also presents a workflow that couples EM, SNMR and MODFLOW in a MCMC framework.