The aim of this study was to confirm if predictive regression algorithms can provide reliable results in missing geophysical logging data in the western and eastern parts of the Drava Super Basin, especially Gola Field, and to apply unsupervised machine learning methods for a better understanding of lithological subsurface relations. Numerous regression models have been used for the estimation of prediction accuracy, along with some clustering algorithms to support the estimation of lithology distribution estimations in well log datasets, consisting of 20 wells in total. Tree-based algorithms and the boosting algorithm have been optimized and proven valuable in predicting well log data when they are not measured or are unavailable at all depth intervals. For blind datasets, predictions become much less reliable. For this purpose, neural networks with at least one Long Short-Term Memory (LSTM) layer have significantly improved the accuracy and reliability of predictions, not in terms of absolute values but in the aspect of the trends in values that change with the depth and other well features, as well as in terms of the magnitudes. Trendlines can further be used for pattern recognition or as a newly engineered feature. Unsupervised learning has confirmed reliability in lithology recognition on validation sets and has proven to be a great asset in distinguishing variabilities in the petrophysical properties of sediments.

In subsurface geological mapping, it is more than advisable to compare different solutions obtained with neural and other algorithms. Here, for such comparison, we used the previously published and well-prepared dataset of subsurface data collected from the Bjelovar Subdepression, a 2900 km² large regional macrounit in the Croatian part of the Pannonian Basin System. Data on depth were obtained for the youngest (the shallowest) Lonja Formation (Pliocene, Quaternary) and mapped using neural network (NN), inverse distance weighting (IDW), and ordinary kriging (OK) algorithms. The obtained maps were compared based on square error (using k-fold cross-validation) and the visual interpretation of isopaches. Two other algorithms were also tested, namely, random forest (RF) and extreme gradient boosting (XGB) algorithms, but they were rejected as inappropriate for this purpose solely based on the visuals of the obtained maps, which did not follow any interpretable geological structures. The results showed that NN is a highly adjustable method for interpolation, with adjustment for numerous hyperparameters. IDW showed its strength as one of the classical interpolators, and its results are always located close to the top if several methods are compared. OK is the relative winner, showing the flexibility of variogram analysis regarding the number of data points and possible clustering. The presented variogram model, even with a relatively high sill and occasional nugget effect, can be well fitted into OK, giving better results than other methods when applied to the presented area and datasets. This was not surprising because kriging is a well-established method used exclusively for interpolation. In contrast, NN and machine learning algorithms are used in many fields, and these algorithms, particularly the fitting of hyperparameters in NN, simply cannot be the best solution for all.

The Gola Gas field in the northwestern Drava Basin lies within the structurally complex petroleum system. Its sandstone reservoirs exhibit spatial heterogeneity due to deltaic fan sedimentation, necessitating robust interpretation tools. We present a fully Python-based seismic horizon extraction and 3D geological modelling workflow tailored to this challenge. The method leverages the greedy propagation method to track horizons with optimal reflector continuity, overcoming limitations of manual interpretation and traditional auto-pickers in faulted or noisy zones. An initial approximation is provided via seed-guided interpolation. Around this surface, a seismic sub-volume is extracted and flattened to align reflections. Within this volume, the greedy algorithm identifies a locally optimal path, maximizing continuity and amplitude strength. Our implementation extends the cost-based framework with slope guidance, local correlation, and robust outlier filtering to ensure geologic realism and repeatability. Surfaces are converted to 3D meshes using interpolation and smoothing, and the results are integrated into an interactive 3D geological model of the Gola field. For the first time in the research area, the model was built entirely in Python, enabling automated horizon extraction, geobody visualization, and seismic interpretation within a flexible, open-source environment.

Gamma-ray parameters are widely used to screen organic-rich intervals, yet the influence of lithofacies heterogeneity on the reliability of gamma-ray-based TOC predictions remains incompletely evaluated. This study aims to (1) assess the relationships between gamma-ray parameters and TOC, (2) evaluate how lithofacies heterogeneity influences predictive reliability, and (3) establish practical screening criteria for identifying organic-rich intervals in two Middle Miocene marine successions from the Croatian part of the Pannonian Basin System. Outcrop gamma-ray spectrometry measurements (K, U, Th) and dose rate (DR) were paired with co-located laboratory TOC analyses (Voćin n = 45; Podsused n = 96). Pearson correlation, multiple and simple linear regression, logistic regression for TOC ≥ 1 wt.%, and threshold analysis were applied to evaluate relationships between gamma-ray parameters and organic matter enrichment. U and DR show moderate positive relationships with TOC in both sections, whereas K and Th are not statistically significant predictors. Regression and threshold analyses identify U as the primary predictor of TOC and show more stable and systematic predictive behaviour in the homogeneous Podsused succession than in the heterogeneous Voćin section. The results indicate that lithofacies heterogeneity primarily influences predictive stability rather than the existence of TOC–gamma-ray relationships. The study demonstrates that lithofacies heterogeneity is an important factor influencing the reliability of gamma-ray-based TOC screening and provides a framework for assessing the suitability of uranium and dose-rate proxies in different sedimentary settings.