Bayesian evidential learning: validation à partir de tests expérimentaux d’injection-pompage

Abstract

Recent developments in uncertainty quantification show that a full inversion of model parameters is not always necessary to forecast the range of uncertainty of a specific prediction in Earth Sciences. Instead, Bayesian evidential learning (BEL) uses a set of prior models to derive a direct relationship between data and prediction. This recent technique has been mostly demonstrated for synthetic cases. This paper demonstrates the ability of BEL to predict the posterior distribution of temperature in an alluvial aquifer during a cyclic heat tracer push-pull test. The data set corresponds to another push-pull experiment with different characteristics (amplitude, duration, number of cycles). This experiment constitutes the first demonstration of BEL on real data in a hydrogeological context. It should open the range of future applications of the framework for both scientists and practitioners.

Investigations into the First Operational Aquifer Thermal Energy Storage System in Wallonia (Belgium): What Can Potentially Be Expected?

“…It is believed that the hydraulic conductivity of the aquifer and its variance are the most sensitive parameters of the model, as the sensitivity analysis results of a study undertaken in the same aquifer (but at another location) have shown. This study was carried out with similar parameters and boundary conditions, by performing and simulating a heat push-pull test [34]. Inverse modeling has been widely applied in previous research projects [31,73] and recent developments of Bayesian evidential learning techniques push forward the allowance for direct predictions to be performed based on prior models [34].…”

“…In Flanders, a decree on deep geothermal use was approved in 2016 and new regulations are expected to be implemented soon [27]. In Wallonia, efforts to develop ATES systems have been made [28,29], with several studies performed on shallow aquifer thermal injection and recovery highlighting some potential ATES target aquifers [30][31][32][33][34][35][36].Reproducing heat transport originating from ATES systems with numerical models has been successfully achieved by various authors [37]. By using the finite element FEFLOW modeling code [38,39], Bridger and Allen [40] simulated 3D seasonal ATES in a heterogeneous unconfined aquifer, highlighting thermal short-circuit issues, which were reported in several other studies [41,42].…”

“…In the latter case, a 3D model was recently developed at high temperatures considering the geothermal gradient [45]. Other thermal transport models were developed with other modeling codes, such as HydroGeoSphere [30,34], SEAWAT [46,47], HST3D/HSTWin [48], or METRA [49]. To the authors' knowledge, most of the ATES numerical applications span over an annual or seasonal timescale with very few available data.In the present study, extensive datasets were acquired to conceptualize and calibrate a coupled groundwater flow and heat transport model.…”

“…In Flanders, a decree on deep geothermal use was approved in 2016 and new regulations are expected to be implemented soon [27]. In Wallonia, efforts to develop ATES systems have been made [28,29], with several studies performed on shallow aquifer thermal injection and recovery highlighting some potential ATES target aquifers [30][31][32][33][34][35][36].…”

Automated Monte Carlo-based quantification and updating of geological uncertainty with borehole data (AutoBEL v1.0)

“…Global sensitivity analysis is therefore applied to select a subset of the PCA orthogonalized m that is most informed by the data variables. The subset m may retain only a few principal components (PCs) (Hoffmann et al, 2019), depending on how informative the boreholes are. For unselected (non-sensitive) model variables, they remain random according to their prior empirical distribution.…”