In the Netherlands, geothermal energy is considered a major contributor to achieving climate and energy goals. The success of geothermal projects depends strongly on understanding the reservoir. Knowing the thermo physical and mechanical properties of reservoir rocks, which govern heat transfer and mechanical stability, is therefore essential. However, these properties are often poorly constrained due to high measurement costs, the substantial time required for laboratory testing, and limited availability of suitable rock material. In addition, reliable rock property prediction is challenging because of strong heterogeneity in lithology, mineralogical composition, and diagenetic history.
This thesis generates a comprehensive database of thermal, acoustic, and mechanical properties for key Dutch geothermal formations. Based on measured data and their integration with downhole petrophysical logs, several predictive equations and models were developed, including machine learning approaches. These models improve property prediction tailored to the Dutch subsurface and enhance geothermal reservoir characterisation in general.
The research begins with a comprehensive study of Permian Rotliegend sandstones, a key geothermal reservoir in the Netherlands. More than 1100 core plugs were analysed to determine porosity, density, acoustic velocities, thermal properties, and mineralogy. The results confirm that porosity is the primary control on most rock properties. Higher porosity corresponds to lower density, acoustic velocity, thermal conductivity, and diffusivity. Systematic deviations from porosity trends were linked to mineralogical and diagenetic factors. For example, nacrite and other kaolinite group minerals enhanced thermal conductivity beyond porosity based predictions, whereas other clay types reduced it. Porosity dominates, but mineralogy and texture impose measurable secondary effects.
The analysis was extended to the Triassic Main Buntsandstein Subgroup, with more than 700 core plugs studied and compared directly to the Rotliegend dataset. Similar porosity dependent trends were observed, but systematic inter formation differences emerged. At equal porosity, Buntsandstein samples show lower thermal conductivity than Rotliegend samples. This difference is attributed to variations in clay type and distribution, as well as mineralogical features such as dolomite cementation and replacive clays. The lower Cretaceous Delft Sandstone Member was investigated to assess coupled mechanical and thermal behaviour. Laboratory tests included ultrasonic velocity measurements, thermal properties, and mechanical loading. Dynamic elastic moduli derived from ultrasonic data were systematically higher than static moduli measured during loading. A lithology specific workflow was developed to convert dynamic to static Young modulus, enabling continuous static modulus logs. Sandstones follow trends comparable to Permian samples, while clay rich intervals exhibit distinct but explainable behaviour due to low porosity.
The final part focuses on machine learning based prediction of thermal properties using laboratory and well log data. Ensemble models and regularised regression achieved promising results for thermal conductivity prediction, even in unseen wells. Thermal diffusivity remained poorly predictable, reflecting its sensitivity to mineralogical and microstructural factors. Density and acoustic features dominate conductivity prediction, whereas no single parameter controls diffusivity.
Overall, this thesis establishes a coherent framework for predicting thermo physical and mechanical properties of Dutch geothermal sandstones. It combines laboratory measurements, petrophysical analysis, and machine learning to improve reservoir characterisation and support reliable geothermal resource assessment.

The ambition is to significantly enhance the knowledge, mapping, and prediction of the geological character of Dutch geothermal plays, focussing on thermal and geomechanical properties. The aim is to improve fundamental understanding of the geological causes behind rock properties, the understanding and quantification of the conversion of downhole log responses into rock properties, the statistical approaches to study these properties, and the implementation of rock properties into reservoir models. The study aims at leading to lower research costs for geothermal operators, lower uncertainties concerning production prediction and risk assessment, and improved productivity due to more optimal well placement and production strategies. With that, our fully open-access results will be applicable to all geothermal targets in the Netherlands and thereby be able to calibrate reservoirs, geomechanical and thermal reservoir models with the ultimate goal to optimise the exploitation of geothermal heat in the Netherlands in a sustainable and safe way. The project has a measuring component producing data and geological understanding, a correlating component, linking rock properties to petrophysical log data by various innovative means, and an implementation component setting the findings into geomodel application. Novel microstructural scanning data enables to link the nano-scale rock composition to decimetre scale bedding and logs. The focus will be on pore throats, detrital, authigenic, and new-grown cement and their types and how these relate to flow, and in particular thermal, and mechanical properties of the rocks. Next, petrophysical downhole logging data will be analysed using statistical and machine learning techniques to produce a much enhanced methodology to relate petrophysical log responses to different rock properties. Improved correlations will be produced per play investigated and can be applied to clastic geothermal reservoirs in general. This will allow for quick and improved screening of rock properties through different wells and finally beyond those into white spots.