The definition of the search neighbourhood in kriging can have a significant impact on the resulting estimates. Stationary domains are usually estimated using a unique search strategy for the entire domain. However, the use of a global search neighbourhood ignores the local variations within each domain, i.e. all blocks are interpolated using a unique search strategy. In this paper, localised kriging parameter optimisation (LKPO) is proposed as an alternative methodology that considers the best ‘local estimation parameter settings’ block by block. The optimisation process is based on absolute error minimisation obtained in cross-validation. Two datasets are presented, the first is a synthetic mineral deposit (2D) and the second is a gold deposit (3D). A wide variety of validation checks show that the use of local kriging parameters significantly improves the grade estimation, obtaining more precise and accurate results than the methodologies currently available in the geostatistical literature.

The geological modeling of stratiform deposits can become very complex, often making use of geological envelopes of small thickness and requiring the use of subblocks (based on Cartesian coordinates) to produce a coherent block model. However, geological events after the formation of the deposit (folds, faults, etc.) can change the direction of spatial continuity of certain attributes, with the mixing of samples belonging to different formation eras (in the case of stratiform deposits) in the same elevation. This study presents a solution for deposits with stratigraphic grades combined with samples of different origins. The solution is a two-dimensional estimate obtained by accumulating the thicknesses of P2O5 in a phosphate deposit (as compared to traditional statistical analysis in three dimensions).

Kriging methods require parameters to define search strategy (kriging neighbourhood). These parameters affect the precision and accuracy of its estimates. Frequently, the choice of these parameters is merely subjective. Some practitioners prioritise estimates that lead to models with a reduced smoothing effect or a regression slope as close as possible to one. However, it is prevalent to use the same kriging neighbourhood or search strategy for all blocks estimated within a stationary domain. This study presents a contribution that challenges this concept by using a block-by-block optimisation approach focused on the localised kriging parameter optimisation (LKPO) methodology. A comparative study is carried out, and some of the metrics analysed include the kriging efficiency and the slope of regression (typical in optimising methodologies in the mining industry). The results indicate that the LKPO methodology provides more accurate and precise estimates than those based on a global kriging neighbourhood.

Estimates of natural phenomena with spatial correlation, i.e. stationary domains, are more precise and accurate when performed using geostatistical techniques (e.g. kriging). The kriging estimates require the definition of the spatial continuity model and a search strategy. Many techniques, such as unfolding and dynamic anisotropy, try to give some improvement in the estimates, considering the variations of the distributions in the geological bodies, however, the definition of the search strategy in the other parameters is unique. This study presents an alternative to this, called Localized Kriging Parameters optimization (LKPO). LKPO considers the best local kriging parameters settings (block by block) through the local uncertainly (simulations). To illustrate this methodology, a synthetic dataset is presented, and the results are compared with the methodologies currently available in the geostatistical literature. Validation checks show a significant improvement in precision and accuracy on the estimates when using local kriging parameters.

Some mineral deposits show mineralization along layers. These layers may pass through several subsequent geological events such as folding and/or severe erosional processes. Grades within these deposits tend to be correlated along orientations where the mineralization was originally deposited or along the same geological period (stratigraphic level). Consequently, some locations close to each other in terms of geographical coordinates can show uncorrelated grades. Spatial continuity analysis can also be affected by error inflicted by combining samples from different stratigraphic levels. This article uses the coordinate transformation (unfolding) to align the grades measured along the same stratigraphic level. The modification in coordinates improved the spatial continuity modeling and the grade estimates at non-sampled locations. The results showed that the mean of the relative error between the estimated value and the real value of the samples using unfolding is -0.10%. However, when using the original coordinates, the mean of the relative error is -0.65%. Furthermore, the correlation between the real and estimated value using crossvalidation is greater using stratigraphic coordinates. A complete case study in a manganese deposit illustrates the methodology.

This study compares the performance of three YOLO-based object detection models—YOLOv3, YOLOv5, and YOLOv8—for vehicle detection and classification at an urban intersection in Montería, Colombia. Recordings from five consecutive days, spanning three time slots, were used, totaling approximately 135,000 frames with variability in lighting and weather conditions. Frames were preprocessed by maintaining the aspect ratio and were normalized according to each model. The evaluation employed models pre-trained on COCO, without fine-tuning, enabling an objective assessment of their generalization capacity. Precision, recall, F1-score, and mAP@0.5 were computed globally and by vehicle class. YOLOv5 achieved the best balance between precision and recall (F1-score = 0.78) and the highest mAP (0.63), while YOLOv3 showed lower recall and mAP, and YOLOv8 performed competitively but slightly below YOLOv5. Cars and motorcycles were the most robust classes, whereas bicycles and trucks showed greater detection challenges. Visual evaluation confirmed stable performance on cloudy days and in light rain, with reduced accuracy under sunny conditions with high contrast. These findings highlight the potential of modern YOLO architectures for intelligent urban traffic monitoring and management. The generated dataset constitutes a replicable resource for future mobility research in similar contexts.