This research evaluates the applicability of Multivariate Imputation by Chained Equations (MICE) for estimating missing well-log data across different sedimentary basis. Utilizing various machine learning techniques including XGBoost (XGB), Random Forest (RF), K-Nearest Neighbors (KNR), and Bayesian Ridge (BR), the performance of MICE was tested on three different data sets from distinct geological contexts and preprocessing conditions with minimal user input.
The main results indicate that the performance of MICE varied across different data sets and well-logs, highlighting the complexity of imputing missing data in heterogeneous sedimentary basins. The number of iterations in MICE did not significantly impact the performance of the models, while data quality, pre-processing, and geological complexities played crucial roles. The Force-200 data set, which underwent extensive preprocessing, demonstrated better imputation performance compared to the Montney and Beetaloo data sets. Additionally, XGB often outperformed other algorithms, predicting missing values with different number of iterations.
The main conclusions drawn from this study emphasize the need for more research to minimize user input and to develop more robust and flexible approaches to imputing missing data in well-logs. The study highlights the challenge of determining a single set of hyperparameters optimal for all the well-logs, suggesting the need for more adaptable models or even advanced techniques like deep learning techniques. The research also suggests the importance of refining pre-processing techniques, exploring further combinations of well-logs, and developing cross-validation approaches that effectively replicates real-world scenarios to advance the application and reliability of MICE in data imputation of subsurface data with missing values.

Numerical modelling in Geo-Engineering is used to solve complex problems by simulating, analysing, or predicting soil behaviour under certain loading and boundary conditions. The soil behaviour is simulated by constitutive models that describe the relationship between stresses and strains through a mathematical formulation. Model parameters are used to calibrate model behaviour to physical soil behaviour measured during in-situ testing (e.g. CPT) or laboratory testing (e.g. triaxial testing). The selection of model parameters is challenging as it needs to cope with aspects as, constitutive model limitations, laboratory test limitations, sample disturbance, soil heterogeneity and many other. This study shows how these model parameters can be determined, optimised and selected by using over 3000 triaxial test results performed on dutch soils (stored in text files) and machine learning tools.

Permeability, a key reservoir characteristic, governs the rate of fluid flow through reservoir rocks. Accurate permeability estimates are paramount for robust reservoir simulation, history matching, and production forecasting. Due to limited core data availability and intrinsic heterogeneity of permeability at different scales, establishing reliable permeability models can be challenging. This study aims to overcome these hurdles by predicting lab-measured core permeability from commonly acquired well logs, using various machine learning algorithms such as Support Vector Regression (SVR), Random Forest (RF), XGBoost, and LightGBM.
We examined two diverse datasets, representing a carbonate platform (Costa Field) and clastic formations (Volve Field). The Costa dataset, including 17 wells across a single reservoir, and the Volve dataset, comprising three wells across three different reservoirs, allowed for evaluating the robustness of our approach under different geological conditions. A critical part of our methodology is feature engineering, particularly incorporating vertical variability. We integrated measurements from adjacent well log readings into our models, recognizing the importance of spatial context and the smoothing effect of well logs over small-scale heterogeneities. This improved prediction accuracy by accounting for shared geological history and depositional environments in proximity.
In Costa Field, blind tests showed R2 scores up to 0.64, and validation R2 scores reached up to 0.8 using a leave-one-well-out cross-validation method. For the Volve Field, blind test R2 scores were up to 0.84, 0.76, and 0.78 for Hugin, Sleipner, and Skagerrak formations, respectively. These results, while satisfactory, underscore the potential of machine learning methods in accurately predicting permeability and highlight the need for effective feature engineering.
This work advocates that while machine learning holds promise for automated feature engineering, human intervention, specifically to incorporate spatial context, can still significantly enhance predictions. Future advancements may seek to internalize this spatial awareness within the machine learning algorithms themselves

The Dutch Ministry of Defence is interested in developing a tool or workflow that can be used to remotely predict Unified Soil Classification System (USCS) classes of any area to aid mobility-related decisions. Therefore, this study was initiated by Cohere Consultants in collaboration with NEO, a Dutch remote sensing company and Utrecht University to create such tool or workflow. This MSc thesis provides a study in how the total geological history (TGH) approach as proposed by \cite{TGHA} can be used for predicting soil USCS classes. This is done by implementing the core concept of the TGH into a decision tree model that makes use of many modules, sub-modules and supporting modules to characterise the tectonic, geologic and geomorphological setting of a soil unit. The modules are divided depending if the unit dealt with is a soil, rock or part of a mountain, whereby the soil module incorporates 7 sub-modules that characterize alluvial, lacustrine, coastal, marsh, aeolian, evaporitic and glacial/periglacial environments. The rock and mountain modules, on the other hand, attempt to determine the presence and characteristics of residual soils using a weathering grade system and a table with the weathering products of 23 common rock types. The performance of the decision tree model was tested using two pilot studies in Konna, Mali and Zamora, Spain and one validation study in 's Hertogenbosch. For each study, a map with the predicted USCS soil classes was generated for the study area. The pilot studies explored the possibility of combining the predicted USCS soil maps with topographic wetness index (TWI) and slope angle maps to make a qualitative prediction on the trafficability of the area. The pilot studies showed that the TGH-based decision tree model has potential for being expanded into a tool for aiding military mobility predictions. Next, the validation study compared the predicted USCS map for 's Hertogenbosch to 5 ground truth data points collected by the Dutch Ministry of Defence. The validation study concluded that the decision tree was in general able to distinguish between coarse grained soils and fine grained soils, however struggled with correctly predicting if a soil has high or low plasticity. Finally, the Mali pilot study was able to compare USCS soil predictions made using the decision tree model to those made by a classification of hyperspectral data (made by \cite{flipsen_2022}). Based on the comparison, there seems to be promise for future works to integrate the two methods to benefit from the detail achievable by the hyperspectral method and the qualitative soil descriptions using the TGH-based decision tree. Future recommendations include fine-tuning the decision tree model so that is is able to incorporate more detailed geologic or soil maps (currently has been trained with maps of about 1:500,000), incorporating remote sensing data to create soil units, quantifying uncertainty and possibly automatising the workflow.

Mineral Resource Modeling (MRM) is used to predict the properties of an orebody, however it does not come without uncertainties. Multiple approaches can be used to reduce the latter. Due to limited knowledge about the subsurface, predictions are difficult to be made. In this thesis the application of Bayesian Evidential Learning (BEL), in order to reduce uncertainties on MRM, will be researched. The uncertainties present in MRM will be linked to the knowledge obtained from case studies in different geological domains where BEL has been applied successfully and reduced certain parameter uncertainties. The gap in todays industry is the knowledge and proof that using BEL for MRM will reduce uncertainties and risks, works effectively and will consume less money. The aim of this study is to show that BEL is a useful approach to reduce uncertainty of the predictions in MRM. The success of a project is supported by the accuracy of the model utilized and the geological interpretation. BEL is a framework based on statistical relationships between data and prediction variables. It will predict the posterior distribution of the prediction variable. The various case studies that have been discussed are (Hermans et. al, 2019), (Hermans et. al, 2018), (Thibaut et. al, 2021) and (Tadjer and Bratvold, 2021). BEL has successfully been able to reduce uncertainties related to geological problems of the following:
• The temperature in an alluvial aquifer
• The efficiency of the thermal energy storage capacity in an alluvial aquifer
• The wellhead protection areas surrounding the pumping well using tracing experiments as predictors
• The prediction of leakages of CO2 and the storage of CO2
Thus, BEL can be seen as a potential approach to also reduce uncertainties in a mineral resource domain; the research of this thesis. In attempt to prove this, a descriptive case study on Tropicana Gold Mine has been executed. The aim is to show that the use of BEL will, based on the geological and geochemical properties such as lithology, grade and mineral type, reduce the uncertainty in the prediction of a geometallurgy property, namely the hardness of a rock. The six steps of BEL’s framework will be followed consisting of Monte Carlo simulations, Principal Component Analysis (PCA) and Canonical Correlation Analysis (CCA) in order to obtain relations between the data and predictor variables. Where the data variables are from exploration drillhole data and prediction variable the hardness of the rock. It shows that BEL is able to reduce the uncertainty of a geometallurgy property by using geological and geochemical properties. Meaning BEL can be applied to MRM to reduce uncertainties.