This study explores dimensionality reduction techniques, namely, PCA (Principal Component Analysis) and ICA (Independent Component Analysis), to condense Earth Observation (EO) data obtained from Landsat 9 and PRISMA satellites to detect alteration zones related to Cobalt (Co) mineralization in the Áramo mine, situated in Asturias, Spain, by employing Support Vector Machine (SVM) Machine Learning (ML) algorithm. The ICA-based models exhibit slightly better performance than PCA-based ones, particularly in delineating alteration zones in the Landsat 9 image, showing promising results in distinguishing alteration zones from host rocks, demonstrating the viability of these techniques applied to mineral exploration. However, the results show the need for refined field data collection methodologies to enhance prediction accuracy for more robust results, in the scope of the HORIZON Europe S34I project (https://s34i.eu/).

The Horizon Europe S34I project aims to enhance exploration methods to secure critical raw materials (CRM) and location management through innovative methods to process Earth Observation data. Placer detection using Copernicus optical data was recently assessed on the Iberian Peninsula Atlantic coast, but radar data potential is still unknown. This search evaluates the contributions from Sentinel-1 data for placer exploration through textural analysis and unsupervised learning with K-means. Different numbers of clusters and iterations were tested, and different attempts were created using several input features for unsupervised classification. RGB compositions were tested to further explore the contribution of the textural indices. The results show the potential of Sentinel-1 data and unsupervised learning in identifying distinct classes in the foreshore and intertidal zones. The Homogeneity textural index allowed for the discrimination of different classes within the Spanish rias while onshore it highlighted geological contacts and fault zones. This study showcases radar data's potential for placer exploration, paving the path for new future applications.

This study addresses the challenge of subpixel occurrence in identifying Lithium Cesium Tantalum (LCT) pegmatites, crucial sources of lithium for electric batteries. Previous spectral unmixing methods have been applied in brownfield sites with pre-existing large mines, facilitating pegmatite endmember acquisition. However, this work focuses on a method for greenfield exploration, presenting a knowledge transfer approach to transfer pre-extracted pegmatite endmembers to new areas using a spectral unmixing based method. Two study areas in Minas Gerais, Brazil, were chosen: Area 1 (A1) as a brownfield site for deriving a pegmatite endmember, and Area 2 (A2) simulates a greenfield area to test the efficacy of the pre-extracted endmember. The Mixture Tuned Matched Filtering (MTMF) classification method was employed, showcasing potential value in scenarios where no pegmatite occurrences are known.

Several methodologies can be employed in the prospection of Lithium (Li) in hard-rock (pegmatites). Spectrometry analysis, a Remote Sensing (RS) technique, can be applied to understand the surface spectral response of a sample, both to identify the rock-forming or alteration minerals in its composition and to validate the data collected in situ (by sensors onboard satellites/drones, for example). This paper aims to make available for public use the information acquired on the spectral composition of rock samples from the Barroso pegmatite field in Portugal, within the scope of the INOVMINERAL4.0 project. As a result, a spectral library was created with 47 spectra, collected from 11 different samples. All data is made available in a universal format, thus contributing to open science, corroborating the validation of local spectral data, and stimulating the creation of other databases, in other locations.

Remote Sensing has been successfully applied in the identification of pegmatitic targets. However, the spatial resolution of open data satellites, which is often much larger than the outcrop size of the target mineral or rock, has been a recurrent challenge in this scientific field. This restricts remote sensing methods that are dependent on large outcrop sizes for successful identification. This work applied spectral unmixing approach on WorldView-3 satellite imagery, to evaluate the potential for high spatial resolution imagery on the Tysfjord Niobium-Yttrium-Fluorine (NYF) pegmatite field (Norway). The preliminary results of this research are encouraging and make a strong contribution to the scientific field of mineral exploration.

PRISMA data is still underexploited in what concerns mineral exploration, while the high demand for battery components such as lithium (Li) instigates exploration of other low-grade deposits such as St. Austell (Cornwall, UK). This study assesses the potential of PRISMA data to target such Li deposits through the detection of topaz as a proxy to the mineralization using band math and partial unmixing techniques. The topaz distribution maps obtained are coherent between each other and with the known geology of the area, highlighting the PRISMA potential, although there could be some shortcomings related to its spatial resolution. Comparison with Sentinel-2 or Worldview-3 data shows the limitations of multispectral products, despite some potential to use Worldview-3 that need to be further investigated. In the future, new approaches to directly detect Li-micas and field validation of the results must be conducted.

Lithium-Cesium-Tantalum (LCT) pegmatites are enriched in several raw materials. However, their small size and the limited penetration depth of the sensors, limits remote sensing approaches. This study evaluates the usefulness of hyperspectral data to identify geochemical halos related to LCT pegmatites by exploiting the information acquired in European projects. It was possible to identify key minerals and related mineralogical changes that can be due to hydrothermal alteration. Partial Least Squares Regression (PLSR) was used to model the abundance of Li, Rb and Cs. The most reliable results were obtained for Cs, with results being influenced by lithological and weathering factors. New outcomes are expected, namely mineral chemistry studies that will complement the hyperspectral results.

Several algorithms were developed to map lithium (Li)-pegmatites in recent years. This preliminary study attempts to validate the same algorithms in the Trancoso region (Portugal), an area with unknown economic potential. The application of RGB combinations, Band Ratios (BR), and Selective Principal Component Analysis (PCA) to Sentinel-2 and Landsat-8 images allowed to identify four target areas for Li-exploration. Some targets correspond to an inert exploitation of the former mine's waste, confirming the potential of this methodology. Moreover, outcropping metric Li-dykes were missed with this approach, highlighting the need to refine the existing algorithms. Reflectance spectroscopy studies of Li-dykes and host rocks support these conclusions and will be crucial to improve the methodology. The collected spectra increased the knowledge on this greenfield area and its economic potential and provided information on new mineralization proxies to be used in future satellite detection studies.

The clustering problem, or grouping, has two variants. If the number of clusters is predefined, this problem is known as the Clustering Problem (CP) or k-Clustering Problem, but when the number of clusters is not defined, the problem is known as the Automatic Clustering Problem (ACP). This paper proposes a new multi-population Biased Random-Key Genetic Algorithm (BRKGA) for the ACP, considering the silhouette index as similarity measure. In algorithm, several BRKGA populations evolve independently, such that each population is responsible for searching the best clustering for a given cluster number, i.e., each population solves one k-Clustering Problem for a particular k. Extensive experiments in 53 benchmark instances commonly used in the literature show that the algorithm obtained very competitive results compared to the state-of-the-art algorithms.

In this work, we conduct a comparative study of GPU accelerated metaheuristics for data clustering. Three population-based metaheuristics were implemented in GPU: Particle Swarm Optimization (PSO), Differential Evolution (DE), Scatter Search (SS). These metaheuristics were compared with the state-of-the-art methods for data clustering considering both runtime efficiency and solution quality. GPU-PSO and GPU-DE algorithms demonstrated competitive performance in the data sets proposed by the literature, as well as real-world problems. Moreover, experimental results show that our GPU proposal obtained an average speedup of 175x over the CPU-only implementation.

Satellite-based mineral exploration will be increasingly relevant in the future to achieve more conscious exploration models. Therefore, new applications to high-demand mineral commodities such as lithium (Li) are emerging. Previous applications to the study area of Fregeneda (Spain) - Almendra (Portugal) showed a high number of false positives despite their ability to identify Li pegmatites. So, the objective of the present work is to improve the classification results using Artificial Neural Networks (ANNs). For that, the same Sentinel-2A images were used and a three-layer feedforward network was computed using the backpropagation method. The ANN created was able to identify all the open-pit mines exploiting Li pegmatites in the area. However, the high number of Li false positives persisted. Future applications may include Convolutional Neural Networks (CNNs) to improve these results.

Raw-materials like lithium (Li) are crucial to the current global decarbonization, but Li-exploration presents some technical challenges. Therefore, new solutions for Li-exploration are needed. Consequently, the aim of this study is to present a unique multi-scale remote sensing approach for Li-pegmatite exploration integrated within the LIGHTS project, considering as study area the Bajoca mine (Portugal). Satellite data allowed the identification of the spectral signatures of Li-pegmatites at a district scale, while drone-borne hyperspectral measurements provided data at the target scale. Handheld spectroscopy and in situ hyperspectral scans of the mine walls were carried out to validate the satellite and drone data. Hyperspectral field and laboratory scans also aim to collect information at the mineral scale, to distinguish different lithological materials, and to identify the Li-rich areas. In the future, machine learning algorithms will deliver an automated integration of all acquired data.

Researchers are constantly looking for ways to improve the execution time of parallel applications on distributed systems. Although compile-time static scheduling heuristics employ complex mechanisms, the quality of their schedules are handicapped by estimated run-time costs. On the other hand, while dynamic schedulers use actual run-time costs, they have to be of low complexity in order to reduce the scheduling overhead. We investigate the viability of integrating these two approaches into a hybrid scheduling framework. The relationship between static schedulers, dynamic heuristics and scheduling events are examined. The results show that a hybrid scheduler can indeed improve the schedules produced by good traditional static list scheduling algorithms.