Combination of Active and Passive Seismic Methods with Artificial Intelligence for Hydrogeophysical Monitoring of the Vadose Zone

Despite significant advancements and numerous applications in the study of the critical zone, seismic methods remain underutilized for exploring the vadose zone compared to hydrogeophysical approaches dominated by electrical and electromagnetic methods. These latter methods are often preferred due to their sensitivity to water content and salinity. However, seismic techniques offer a valuable complement through their sensitivity to mechanical properties essential for characterizing subsurface heterogeneity, as well as key hydraulic parameters such as porosity, permeability, and saturation. This is particularly relevant in clay-rich environments, where clays tend to obscure saturation contrasts for electrical and electromagnetic methods. The aim here is not to pit these approaches against each other but to highlight their complementarity, as recently demonstrated in studies that also underscored the efficiency of electrical methods in terms of implementation and interpretation. While the combination of seismic refraction tomography (SRT) and surface-wave dispersion analysis (MASW) produces useful images for identifying significant saturation contrasts, it remains limited in detecting subtle spatial or temporal variations. These limitations are especially pronounced in time-lapse experiments, which in addition are often complex and resource-intensive to implement. Current inversion techniques struggle to represent continuous saturation variations between the surface and the water table, and interpretations frequently adopt a binary perspective, distinguishing only between partially and fully saturated zones. A promising alternative lies in the use of recently developed rock physics models capable of simulating the impact of saturation variations on seismic-wave velocities. Recent studies have shown that surface-wave dispersion is particularly sensitive to these variations, providing insights into the influence of hydrological and hydrogeological dynamics on passive seismic results. Although this approach is rapidly advancing in environmental seismology, it remains relatively underexplored in hydrogeophysics and agrogeophysics. Through examples obtained at a study site, we illustrate how passive seismic methods could enable precise monitoring of hydrological and mechanical properties. We also show that simple neural networks can effectively extrapolate water table maps from 2D seismic velocity data obtained through hybrid approaches, using a limited number of spatial piezometric observations. Finally, we test how high-density surface-wave dispersion monitoring data, combined with artificial intelligence algorithms (inspired by neural machine translation and speech recognition architectures), can deliver precise petrophysical and hydrogeological descriptions.