Improved Modelling Approach for LT-ATES Systems under Uncertainty in Fractured Reservoirs: A Case Study in Brussels, Belgium

Most operational low-temperature aquifer thermal energy storage (LT-ATES) systems are implemented in productive sandy aquifers, the traditional storage target. In many regions, such resources are scarce or absent, and we have to resort to more complex subsurface settings. In Belgium, 14 LT-ATES systems already operate in the fractured basement rock. The heterogeneous distribution of fractures leads to uncertainty on preferential flow paths and thereby complicates predictions of thermal recovery efficiency. Although many studies focused on characterizing the fractured basement aquifer, their practical use to build models remains limited because spatial correlation scales are smaller than average ATES well spacing. To enable larger-scale adoption, we propose an improved modelling approach that captures the well connectivity to the surrounding rock and the connectivity between both wells. The IC-FERST simulator is used to model groundwater flow and (heat) transport, as it efficiently handles highly complex models with a (simplified) explicit fracture representation. Fracture zones were conceptualized to only represent distinct connectivity features. Several alternative conceptual models were semi-automatically generated, including uncertainty on the number and size of fracture zones and fracture orientation. As such all concepts have a different degree of connectivity. A short-term fluorescence tracer test was performed and the simulated breakthrough curve for each scenario was compared to field data. Several concepts led to inconsistent tracer behaviour and were quickly falsified, while others captured the first arrival, peak or tail of the breakthrough curve. Recirculation of the tracer showed to be essential to explain the trailing breakthrough behaviour. The concepts that best explained field data were kept for further uncertainty analysis using Monte Carlo simulations. Variability was included on fracture and matrix porous properties, natural groundwater flow, diffusivity and dispersivity. The resulting prior distribution was not falsified by the observed field data. Finally, long-term ATES simulations were performed for the selected concepts. The (simplified) explicit fracture representation produced highly complex storage volume geometries. Concepts that fit field data well, still showed large variability in the prediction of the outflow temperature of the ATES system, highlighting that uncertainty quantification is indispensable for ATES feasibility studies in fractured reservoirs. In conclusion, the well connectivity modelling strategy to simulate an ATES doublet successfully predicts field data, outperforming previous equivalent porous media modelling efforts in the fractured reservoir. This study also highlights that feasibility studies and design standards for ATES in complex subsurface settings should differ from those developed for the traditional setting. In the future, we aim to apply the Bayesian Evidential Learning (BEL) framework to predict the probability of reaching certain ATES recovery efficiency and to optimize ATES design. Since BEL maps a direct relationship between data and prediction without any model calibration, it is well suited to systems with complex heterogeneity.