The dissolution features in carbonate rocks are common and highly variable; ranging from millimetric pores to enlarged conduits, open fractures, and even caves. Identifying these features is essential in multiple contexts, particularly in the oil industry, as they significantly impact fluid flow capacity and reservoir performance. Furthermore, their identification helps drilling campaigns predict and mitigate zones with high losses of drilling mud. In this context, we propose a novel methodology for identifying dissolution features in carbonate rocks using BHI and CNN models. For this task, seven wells with acoustic BHI from carbonates from Barra Velha Formation located at Santos Basin, Brazil, were used. The proposed workflow includes: (i) defining the display image scale and cut height size (CHS); (ii) class definition and dataset labeling; (iii) CNN model training; and finally (iv) applying the best CNN model to a blind well to check the performance in an untrained dataset. The display resolution of 1:10 was defined and four CHS scenarios were tested: 10 cm (S1), 40 cm (S2), 70 cm (S3) and 100 cm (S4). Six classes were defined based on the type and intensity of dissolution, encompassing a range from non-vuggy rock to facies dominated by vuggy matrix porosity, fracture related dissolution, and large scale conduits or caves. Four CNN models were tested to predict dissolution features: ResNet, RegNet, ShuffleNet, and MobileNet. For scenarios S1 to S3, the ResNet model achieved the best performance, with accuracy values around 0.90 and F1-macro of 0.80. In S4, MobileNet was the best model, with accuracy values close to 0.70 and F1-macro of 0.58. Finally, a qualitative analysis was made using the blind well to evaluate the performance of each model in an untrained/unseen dataset. Although CNN performance decreases with increasing CHS, the results indicate that a multi-scale approach combining 10 cm and 70 cm sample heights provides a robust workflow for capturing both fine-scale dissolution textures and larger-scale geological trends along the wellbore.

Abstract Earthquake early warning (EEW) systems issue alerts in the brief interval between the first P-waves and the arrival of strong S-waves or surface waves. Traditional algorithms degrade under heterogeneous, noisy conditions, motivating complementary artificial intelligence (AI) approaches. However, most previous studies emphasized model accuracy alone; we provide a benchmark of modern deep-vision backbones within a standardized time–frequency pipeline to characterize the accuracy–latency trade-off and assess real-time feasibility for EEW event/noise discrimination. We fine-tune and evaluate six pretrained vision models for event/noise discrimination on 10 s vertical-component windows aligned so that the P onset occurs $$\sim$$ 7–8 s into the window, i.e., only the first 2–3 s of post-onset P-wave signal are included. Each window is transformed to a continuous wavelet transform power scalogram and fed to pre-trained convolutional neural networks (CNNs; EfficientNet-B0, ResNet-18/50, ShuffleNetV2) and vision transformers (ViT-B/16, Swin-Tiny), fine-tuned for binary classification. Models are trained on the STanford EArthquake Dataset (STEAD, global) and externally evaluated on the Italian Seismic Dataset for Machine Learning (INSTANCE, Italy) and the Red Sísmica del Noroeste de México (RESNOM; the CICESE broadband network in northwestern Mexico), with decisions derived from either a fixed threshold or dataset-specific calibration. The models achieve near-ceiling discrimination on STEAD and strong but dataset-dependent transfer to independent networks, with external performance in the range MCC $$\approx$$ 0.87–0.97 and F1 $$\approx$$ 0.93–0.98, depending on model, training regime, and thresholding strategy. The lightweight CNNs evaluate a single trace in only 0.6 ms, a latency compatible with real-time EEW requirements. This study introduces a unified framework for benchmarking early event/noise discrimination based on short P-onset windows, explicitly incorporating end-to-end computational latency. By demonstrating cross-network generalization, millisecond-scale inference, and physically interpretable focus on the P onset, the results provide a realistic foundation for next-generation AI modules within operational early warning systems.

Abstract With the growing complexity and frequency of oceanic sampling, more efficient systems are needed to manage the information. Developing robust infrastructures that facilitate the search and access to oceanographic data becomes essential to ensure effective use of information — meaning its integration, interoperability, and reuse across research and operational contexts — as well as to promote collaboration and interoperability among organizations. In this context, the concept of geoportals as human-machine interfaces emerges as one of the key solutions for accessing and disseminating spatial data. The Spanish Institute of Oceanography (IEO-CSIC), as part of the National Oceanographic Data Centres (NODC) network, has adopted open-source solutions such as GeoNetwork to develop a standardized metadata repository that adheres to FAIR principles (Findable, Accessible, Interoperable, and Reusable). This approach has enabled IEO to organize and present an extensive catalogue of multidisciplinary oceanographic data, supporting data discovery, integration, and global exchange. The implementation of proper hierarchization mechanisms has improved search capabilities, allowing users (individuals or data hubs) to efficiently access marine datasets for research and applications. Furthermore, the cataloguing of ocean data at IEO plays a crucial role in supporting the implementation of the EU Marine Strategy Framework Directive (MSFD), demonstrating the impact of structured data management on environmental policies and ocean governance.