Exploring electrofacies for property mapping of the Triassic: an improved approach for geothermal development in Brabant.

Abstract

Geothermal energy in the Netherlands has been receiving increased attention recently since it may be a promising contributor to the energy transition. Public well data that has been acquired for hydrocarbon exploration can now be used to assess the potential of geothermal plays. Similarly to hydrocarbon reservoirs, the reservoir quality of a geothermal play is largely controlled by three factors: facies, maximum burial depth and diagenesis. The direct onset for this project has been the release of the regional property maps by TNO, 2017. The aim of this study was to improve the methodology for property maps of the Triassic formations from the Roer Valley Graben in Noord-Brabant, SE Netherlands. The approach has been twofold by (1) a detailed analysis of the current mapping methodologies and their limitations, followed by (2) a lithology prediction analysis as an addition to the existing workflow. In previous studies, regional porosity maps have been made for the West Netherlands Basin (Vis et al., 2010) and for the entire Netherlands (including offshore) in 2017. Both mapping projects were based on porosity data from wells, although different driving maps have been used; respectively facies and maximum burial depth. Given the strong relation between maximum burial depth and provenance, the maximum-burial driven maps provide a good indication for regional property trends. However, this relation is not valid for individual wells, hence neither in local geothermal enterprises. Given the heterogeneous nature of e.g. a braided channel complex, a local geological lithofacies interpretation is required. The facies-driven property maps as constructed for the West Netherlands Basin use lateral extrapolation of facies modelling and assign porosity values to each facies. Due to the sparse amount of data available and related loss of detail and oversimplification this method is not preferred for the Roer Valley Graben. To improve the current mapping methodology, our knowledge of the wells should be enhanced first. Therefore lithology characterization and prediction have been explored. The workflow starts with the use of cluster analysis of interpreted facies intervals. Cluster analysis has been performed by Wkmeans on intervals of three wells within the area: KDK-01, WWK-01 and WWS-01-S1. Since lithology prediction is not possible with this tool, this method requires further development. Lithology prediction has been performed in Petrel on intervals of three wells within the area: BRAK-01, KDK-01 and WWN-01-S2. The input consisted of lithofacies interpretations, followed by either of the two classification techniques: a Bayes classifier and a neural network. The classification based on Bayes algorithm generally yielded poor, discontinuous lithofacies logs. The poor results were mostly due to insufficient input data, e.g. only 1 interval with a distinct lithofacies. Lithology predictions with a neural net created good lithofacies logs. Cross-checks within the wells validated the fairly good results. Wireline logs (e.g. GR, RHOB, RT) resulted in continuous lithofacies logs, whereas core plug logs (porosity, horizontal permeability, grain density) resulted in discontinuous lithofacies logs. Furthermore sufficiently long intervals with separate facies intervals drastically improve the lithology prediction. The neural net classifier has proven to be a promising technique to validate and predict lithofacies within a well. The lateral extension to other wells should be approached with caution, at least for this particular dataset. Since a large number of logs and interpretations will yield the best predictions, the need for extensive, full-range data acquisition is emphasized. Finally, this study also stresses that the intended users of regional property maps should become well-informed on the uncertainties related to the making of these maps.