Ocean Geophysics’ Post

Our latest publication in Exploration Geophysics unveils the application of unsupervised machine learning in resolving the architecture of deep-crustal and near-surface geological structures of Euler deconvolution depth results from magnetic data. While this is a stepping stone to modern applications of unsupervised machine learning in geophysics that reduce bias in conventional interpretation techniques, we highlight ways to mitigate over-interpretation of results irrelevant to modern structural geology concepts. Thanks to our co-authors and sponsors for their support. #geophysics #structuralgeology #machineLearning #exploration

This week and next, Ocean Geophysics is presenting oceanSweets - a candid overview of our software suite that we use to enhance and automate our geophysical data processing flows. TODAY’S OCEANSWEET: oceanClipper, for magnetic line overlap cleaning. Combining magnetic field data collected from multiple passes over the same area often results in distorted and low-quality data. Achieving high-quality maps of the observed magnetic field requires careful selection of the data used for interpolating the magnetic field grid, especially where lines overlap. Imagine making the prettiest and most accurate magnetometer grids in one click. 😊 oceanClipper streamlines the process of combining magnetic line data by swiftly cleaning input magnetic field data (in CSV format) from entire surveys in a single step, eliminating the need for days of manual processing. Utilising custom CUDA code, oceanClipper employs spatial indexing coupled with massively parallel pointwise comparisons to inspect lines and remove overlapping points within a specified radius. A post-processing stage then refines this initial clipping process, dynamically adjusting the removal or reinstatement of points. The cleaned data is output in CSV format, ready to be utilised alongside other geophysical data processing tools to generate high-quality magnetic field grids. #geophysics #offshorewind

the latest Datarock Applied Science blog is out. In this installment, we look at why understanding your data is so important and why AI won't take away your job, but can add a powerful tool to your kit

In this months technical blog from the Applied Science team, we take a look at the importance of subject matter input into a #machinelearning project. Having a deep understanding of the subject matter is essential when preparing #geophysical images for analysis using computer vision methods. #Geoscientists should carefully consider how image preparation intends to enhance or suppress texture that could relate to #geology. Visit our newest blog post for further insights: https://lnkd.in/gBh94Vde

Excited to share that our latest research paper, "Smoothness: The Key Factor in Well-Log Information-Assisted PINNtomo," has been published in the Journal of Applied Geophysics! In this study, we explore how integrating smoothed well-log velocity data into physics-informed neural networks (PINNs) enhances seismic traveltime tomography. Our findings highlight the importance of smoothness in well-log data for improving the accuracy of inverted background velocity models, which is crucial for full waveform inversion (FWI). Read the full paper here: https://lnkd.in/dPuXMSF6 #Geophysics #Research #SeismicTomography #PINNs #WellLogData #FullWaveformInversion

Enhance your geological data analysis with ElasticDocs. 🌏 #ElasticDocs allows #engineers and #geoscientists to swiftly navigate thousands of report pages and identify #geological analogs using advanced search-and-cluster techniques. This #innovative solution is crafted by Iraya’s team of expert machine learning specialists and #geoscientists. #UnstructuredData #LargeLanguageModelling #ComputerVision #DataAtelier #earth #energy #environment #readtheearthbetter #IrayaEnergies #DataInsights #DataDrivenExploration #DataDrivenDiscovery #InnovateWithIraya #AIForEarthInsights #EarthInsightsAI

It is always a great delight to see your work accepted and published in a peer reviewed journal like Near Surface Geophysics. André Bredeck, Volkmar Schmidt, and I would like to share our findings with those interested in #UXO #detection, #geophysics, #borehole and #GPR. We hope it helps with open questions, in case of upcoming problems, and inspires more research in this field. The article is open access and can be found at https://lnkd.in/e4BDngdG

Welcome to check out our new paper in Geophysical Journal International. In this work, we introduce a deep-learning seismic phase picker for OBS data using automated labeling and transfer learning (OBSTransformer). Phase picking on OBS data is a challenging task in seismology as OBSs are exposed to harsh noise levels from various sources, such as water column reverberations, marine mammal sounds, anthropogenic activities, etc. Here, we develop an automated method of dataset labeling and compile a comprehensive training dataset using data from eleven OBS deployments in different tectonic settings worldwide. We train OBSTransformer leveraging the transfer learning approach and novel deep-learning advances in EQTransformer land picker. Evaluation metrics on three distinct test datasets demonstrate picking improvements up to ~76% compared to the base model. You can access the codes and dataset at https://lnkd.in/g2NXqAqE. https://lnkd.in/e4-8rA89

🌟 Excited to share my latest project on geophysical depth prediction using machine learning techniques! 🌍🔍 I've just wrapped up a fascinating study at the University of Texas at Dallas, where we delved into the integration of machine learning methods for analyzing spatial data and predicting lithospheric depth variations. In our research, we leveraged Random Forest regression and K-means clustering to develop a novel methodology for analyzing gravity data across North America and Australia. By preprocessing datasets, performing clustering for segmentation, and training Random Forest models, we were able to predict lithospheric depth with high accuracy. Key findings from our study demonstrate the effectiveness of Random Forest combined with K-means clustering in analyzing spatial data and predicting lithospheric depth variations. Our evaluation metrics, including Mean Squared Error and R-squared, validate the robustness of our models, showcasing their potential for geological studies. Moreover, we discovered that a pretrained model could be effectively utilized for a different geographical area, as demonstrated in our analysis of Australia. Our project underscores the importance of machine learning techniques in advancing our understanding of geological processes and spatial data analysis. The methodology we've developed holds promise for similar studies and applications in geological modeling for different regions. If you're interested in learning more about our project or discussing potential collaborations, feel free to reach out! 🚀 #MachineLearning #Geophysics #DataScience #Research #SpatialAnalysis

My start-up has a geological machine-learning map of the USA, Canada, Mexico, South America, Australia, and Africa. We base all our work "only" on hard rock geochemistry, then get a supervised decision tree AI to predictively figure out the spots to look at based on ground shapes and gravity maps. It is superior to a geological time fast-moving vector such as soil geochemistry (no AI models work on soil geochemistry, we have found). Our approaches let you see what the geophysical anomalies likely contain, such as in the attached image that shows how well AI compares to highly expensive IP work. We are cheaper than IP, we show what metals likely are within targets for metal, an at a higher definition suitable for drill rig placements. This costs 5570 US per square km, and it delivers, as we can see when we compare the work to even IP. #geostatistical_reservoir_modeling