Using Machine Learning in the Search for Petroleum

Demonstrating how ArcGIS Pro forest and boosted regression models predict petroleum locations using historic wells, radiometric, lidar, and Landsat data, validated by 2D seismic AVO.

ArcGIS Pro Landsat-9 LiDAR 2D seismic AVO interpretation

Overview

This project attempts to predict the location of potential petroleum deposits using ESRI’s ArcGIS Pro Spatial Statistics Forest-based and Boosted Classification and Regression deep learning algorithm. For the variable to predict, known historic oil wells in the 2nd Trenton pay of Grant and Madison counties of Indiana were used. The explanatory training rasters were multiple raster layers derived from airborne radiometric surveys, airborne lidar surveys and Landsat9 data. Results were compared to existing 2D seismic using prestack processing and AVO (Amplitude Velocity Offset) interpretation in an attempt at ground truthing the results.

Tech stack

ArcGIS Pro

ArcGIS Pro is Esri's premier desktop GIS application: advanced 2D and 3D mapping, data management, and spatial analysis with over 2,000 geoprocessing tools.

This is the world's leading desktop GIS software, relied on by hundreds of thousands of organizations for professional-grade geospatial work. ArcGIS Pro delivers a powerful, project-based workflow, enabling simultaneous operation on multiple maps, layouts, and integrated 2D/3D scenes. Users leverage over 2,000 geoprocessing tools to perform complex spatial analysis, manage authoritative data, and automate tasks via Python or ModelBuilder. It fully supports AI/machine learning integration and ensures seamless data sharing and collaboration across ArcGIS Online and ArcGIS Enterprise portals.

https://www.esri.com/en-us/arcgis/products/arcgis-pro/overview

View projects
Landsat-9

Landsat 9: The latest satellite in the joint NASA/USGS program, extending the 50-year global land imaging record and doubling data collection with Landsat 8 for an 8-day revisit time.

Landsat 9 is the newest satellite in the NASA/USGS Sustainable Land Imaging Program, successfully launched on September 27, 2021. The observatory carries two key instruments: the Operational Land Imager 2 (OLI-2) and the Thermal Infrared Sensor 2 (TIRS-2). These sensors provide high-quality, science-grade data with 14-bit radiometric resolution and 30-meter spatial resolution (multispectral). Operating in tandem with Landsat 8, the combined constellation achieves an 8-day revisit cycle for global landmasses, significantly increasing the frequency of critical Earth observations for resource management and change detection.

https://www.usgs.gov/landsat-9

View projects
LiDAR

LiDAR (Light Detection and Ranging) is an active remote sensing method: it uses pulsed laser light to measure distance and generate precise, real-time 3D models (point clouds) of an environment.

LiDAR systems operate by emitting millions of laser pulses per second and measuring the 'Time of Flight' (ToF) for the light to return to the sensor. The core instrument package—laser, scanner, and a specialized GPS/IMU—calculates distance using the constant speed of light (approx. 299,792,458 m/s), compiling the results into a dense 3D point cloud. This high-resolution data is critical for multiple applications: autonomous vehicles (for real-time object detection), high-accuracy topographic mapping (e.g., the USGS 3DEP program), and infrastructure inspection (power lines, bridges).

https://oceanservice.noaa.gov/facts/lidar.html

View projects
2D seismic

2D seismic is the cost-effective, single-line method for initial subsurface imaging, delivering a vertical cross-section for regional reconnaissance.

This is your foundational exploration tool: fast, focused, and budget-conscious. 2D seismic uses a single-line array (geophones or hydrophones) to record reflected energy from a source (e.g., an airgun or Vibroseis truck). The output is a clear, vertical, two-dimensional cross-section of the geology beneath the line. We deploy this for regional reconnaissance or preliminary basin-scale mapping, efficiently identifying large structures like major faults or folds before committing to a higher-cost 3D survey. It’s a proven method for initial resource assessment (oil, gas, or CO2 geosequestration).

https://www.silverthorneseismic.com/understanding-2d-and-3d-seismic-acquisition-whats-the-difference/

View projects
AVO interpretation

AVO interpretation analyzes seismic reflection amplitude changes across increasing offset (angle) to quantitatively predict subsurface rock properties (e.g., lithology, porosity) and identify fluid content, specifically hydrocarbons.

AVO (Amplitude Variation with Offset) interpretation is a proven, high-ranking technology for direct hydrocarbon detection (DHI) and reservoir characterization. The process analyzes how seismic reflection amplitudes change with increasing offset (angle), a variation governed by rock property contrasts (Vp/Vs, density) across an interface. Interpreters commonly use cross-plots of the zero-offset amplitude (Intercept, R0) versus the amplitude gradient (A1) to define AVO classes (e.g., Class 2, Class 3) and isolate fluid anomalies from background lithology. Initially applied successfully in the 1980s to detect gas in California sandstones, the technique now provides critical, quantitative information on lithology, porosity, and seal properties, significantly reducing exploration risk and reserves determination uncertainty.

https://wiki.aapg.org/Amplitude_versus_offset_(AVO)_analysis

View projects

Using Machine Learning in the Search for Petroleum

Related talks