Uncertainty quantification of salinity in coastal aquifer (Binah Thuan province, Vietnam) using Bel1D inversion of transient electromagnetic data

Vietnam south central provinces have been facing saltwater intrusions problems for many years, particularly in Binah Thuan province. Binah Thuan has a hyper-arid climate and during the dry season, seawater penetrates the river estuaries and contaminates the fresh groundwater reservoirs. Recent hydrogeological and geophysical data have shown that saltwater intrusion was not limited to river but was likely the results of a complex situation involving fossil saline water heterogeneous sediments and anthropogenic activities. To determine the state of salinity transient electromagnetic (TDEM) soundings were collected along the Luy River, where the situation is most problematic. TDEM data can reveal the vertical variation in electrical conductivity. An inversion was carried out with SimPEG, an open-source Python package for solving the electromagnetic forward and inverse problem. It revealed the conductivity transition of different layers from the surface until about 40 meters depth, with a gradual increase in electrical conductivity showing the encroachment of salt-water in the shallow aquifer. Deeper, a decrease in conductivity shows the presence of bedrock. However, SimPEG’s inverse solution is deterministic and, thus, provides only one possible solution among many others. For an uncertainty quantification of the salinity, we apply a new stochastic inversion approach named Bayesian Evidential Learning 1D imaging (BEL1D). BEL1D is combined with SimPEG to solve the forward problem. BEL1D bypasses the inversion step by generating random samples from the prior model distribution (with predefined ranges for thickness, electrical conductivity and salinity for the different layers). It then directly generates the corresponding data in order to learn a direct statistical relationship between data and model parameters. From this relationship, BEL1D can generate posterior models fitting the field observed data, without additional forward model computations, making it a very efficient way to stochastically solve the inverse problem. The output of BEL1D shows the range of uncertainty for subsurface models. It enables to identify which model parameters are sensitive and can thus be accurately estimated from TDEM data. In our case, it reveals the uncertainty on the depth of fresh saline interface as well as the total dissolved solid content of groundwater. The application of BELID together with SimPEG for stochastic TDEM inversion is a very efficient approach as it allows to estimate the uncertainty at a limited cost. We thus expect our approach to be also valuable for the inversion of airborne data sets