The doctoral dissertation in the field of Applied Physics will be examined at the Faculty of Science, Forestry and Technology, Kuopio campus and online.
What is the topic of your doctoral research? Why is it important to study the topic?
My doctoral research investigates how seismic measurements and deep learning can be combined to monitor groundwater stored in porous geological formations. Groundwater is an essential freshwater resource, but estimating changes in underground water reserves can be difficult using conventional methods alone.
I used a computational framework in which seismic wave propagation in water-saturated porous materials is modelled using physical theory, and the resulting seismic signals are analysed with neural networks. The aim is to estimate important aquifer properties, particularly the water table level and porosity, and use them to determine groundwater volume. The research contributes to the development of faster and more efficient methods for groundwater monitoring.
What are the key findings or observations of your doctoral research?
The research demonstrates that neural networks trained with physics-based synthetic seismic data can estimate properties of a porous water reservoir and its water volume. The work progressed from real-valued neural networks to complex-valued neural networks designed to process complex-valued seismic data directly. A physics-based constraint was also introduced to improve consistency between predicted water table level, porosity, and water volume. Field measurements from a controlled aquifer test site in Laukaa, Finland, were used to evaluate the approach. Transfer learning enabled models trained mainly on synthetic simulations to be adapted using a small amount of field data.
The research also used explainable artificial intelligence to investigate which seismic receivers contributed most strongly to predictions, showing potential for designing more efficient measurement configurations.
How can the results of your doctoral research be utilised in practice?
The developed approach could contribute to more efficient monitoring of groundwater resources. Once a neural network has been trained, it can estimate aquifer properties from seismic measurements much faster than repeatedly solving computationally expensive inverse problems. In the future, such methods could support repeated or near-real-time monitoring of changes in groundwater storage. The receiver-importance analysis may also help optimise seismic surveys by identifying informative measurement locations and potentially reducing the number of sensors required. Further validation on different and more heterogeneous natural aquifers is needed before the approach can be used in groundwater management.
What are the key research methods and materials used in your doctoral research?
The research combined numerical modelling, seismic measurements, and deep learning. Three-dimensional seismic wave propagation in porous materials was simulated using Biot’s poroviscoelastic theory and a high-order discontinuous Galerkin numerical method. These simulations generated synthetic seismic datasets for training neural networks to estimate aquifer properties. Both real-valued and complex-valued neural networks were investigated. Seismic field measurements were collected at a controlled sand aquifer in Laukaa, Finland, at different water table levels. Transfer learning was used to adapt models trained on synthetic data to field measurements, while SHAP analysis was used to examine the importance of different seismic receivers. The research consisted of three interconnected studies progressing from synthetic experiments to field validation.
Is there something else about your doctoral dissertation you would like to share in the press release?
One important aspect of the dissertation is the combination of physical modelling and artificial intelligence. Rather than training neural networks only on experimental data, which are difficult and expensive to collect in large quantities, the research uses numerical simulations based on the physics of seismic wave propagation to generate training data. Field measurements are then used to evaluate and adapt the models. This combination offers a promising direction for applying artificial intelligence to geophysical monitoring when real labelled data are limited. At the same time, the dissertation identifies important challenges for future research, particularly uncertainty quantification and generalisation to more heterogeneous natural aquifers.
The doctoral dissertation of Mahnaz Khalili, MSc, entitled Deep Learning for Seismic Monitoring of Porous Water Reservoirs will be examined at the Faculty of Science, Forestry and Technology, Kuopio campus and online. The opponent will be Professor Lassi Roininen, LUT University, and the custos will be Associate Professor Timo Lähivaara, University of Eastern Finland. Language of the public defence is English.
For further information, please contact:
Mahnaz Khalili, [email protected]
