Offshore oil and gas platform dynamics in the North Sea, Gulf of Mexico, and Persian Gulf: Exploiting the Sentinel-1 archive

This study utilizes Sentinel-1 satellite data and deep learning to generate a comprehensive, public dataset tracking the spatiotemporal dynamics, installation, and decommissioning of nearly 4,000 offshore oil and gas platforms across the North Sea, Gulf of Mexico, and Persian Gulf from 2017 to 2025, revealing a structural shift toward mobile units and demonstrating the viability of automated Earth observation for long-term maritime infrastructure monitoring.

The Gist

Imagine the world's oceans as a giant, busy construction site. For decades, we've been building massive steel cities underwater to drill for oil and gas. But because the ocean is so vast, dark, and often covered in clouds, it's incredibly hard for humans to keep a close eye on these structures. We don't know exactly how many there are, where they are, or when they are being built or torn down.

This paper is like a super-powered, all-seeing drone that finally gives us a clear, up-to-date map of this underwater construction site.

Here is the story of how they did it and what they found, explained simply:

1. The "All-Weather Camera" (Sentinel-1)

Usually, if you try to take a photo of the ocean from space, clouds or fog will block your view. But this study used a special satellite called Sentinel-1. Think of this satellite as having a "night vision" or "X-ray" camera. It uses radar instead of light, so it can see through clouds, rain, and darkness. It takes a picture of the ocean every few days, creating a massive library of images from 2017 to 2025.

2. The "AI Detective" (Deep Learning)

Looking at thousands of radar images to find tiny oil rigs is like trying to find a needle in a haystack while wearing blindfolds. So, the researchers trained an Artificial Intelligence (AI) detective.

• The Training: They showed the AI thousands of pictures of oil rigs, teaching it what they look like in radar images (they appear as bright, shiny blobs against the dark water).
• The Hunt: Once trained, the AI scanned the entire library of satellite images for three major oil zones: the North Sea (near Europe), the Gulf of Mexico (near the US/Mexico), and the Persian Gulf (near the Middle East).

3. The Big Reveal: What They Found

The AI didn't just count the rigs; it built a time-lapse movie of the last eight years. Here is the plot:

• The Three Neighborhoods:

  • The Persian Gulf: This area is like a booming construction site. It kept getting bigger and busier until 2024, with Saudi Arabia and the UAE building the most rigs.
  • The Gulf of Mexico: This area peaked around 2019 and has been slowly shrinking since then. It's like a neighborhood where old houses are being demolished, but not many new ones are being built.
  • The North Sea: This area has been shrinking the most. It's undergoing a "renovation" where old oil rigs are being removed to make way for wind farms or just to be taken away, as Europe moves toward greener energy.

• The "Pop-Up" Trend:
  The most interesting discovery is about the type of rigs. In the past, oil rigs were like permanent skyscrapers—built to stay for 30 or 40 years.
  The study found a rise in "pop-up" rigs (like mobile drilling units or jack-ups). These are temporary structures that move around, work for a few years, and then leave. It's like the difference between building a permanent house versus setting up a temporary tent. This suggests the industry is becoming more flexible and mobile.

• The "Ghost" Rig Problem:
  The study also tracked when rigs were installed and when they were removed. They found that for every new rig built, almost one was taken down. It's a constant cycle of construction and demolition, with over 2,700 rigs being moved or removed during the study period.

4. Why Does This Matter?

Think of this dataset as a publicly available "Google Maps" for offshore oil rigs.

• For Governments: It helps them know exactly who owns what and where, which is crucial for safety and laws.
• For the Environment: It helps scientists track how these structures affect the ocean and plan for their removal (decommissioning) so they don't become underwater junk.
• For the Future: As the world shifts from oil to green energy, this map helps us see where old oil rigs can be repurposed (maybe as carbon storage or artificial reefs) and where new wind farms can be built.

The Bottom Line

The researchers created a free, open-source tool that uses satellite radar and AI to keep a constant watch on the world's oil rigs. They found that while the Persian Gulf is still expanding, the North Sea and Gulf of Mexico are changing, with more rigs being moved or removed than built. This "living map" helps us understand how the ocean's energy infrastructure is evolving in real-time.

Technical Summary

1. Problem Statement

The monitoring of offshore infrastructure (specifically oil and gas platforms) is critical for maritime awareness, environmental management, and energy transition planning. However, systematic monitoring faces significant challenges:

• Inaccessibility: Maritime areas are vast and difficult to access physically.
• Data Limitations: Existing reporting mechanisms (e.g., AIS, regulatory filings) are often regionally limited, inconsistent, incomplete due to confidentiality, or lack historical continuity.
• Dynamic Nature: The offshore sector involves complex lifecycles, including installation, relocation, and decommissioning, as well as a shift toward mobile units (jack-ups, drillships) which are harder to track than fixed structures.
• Need for Scalability: There is a lack of consistent, long-term, global-scale datasets that combine spatial attributes with temporal lifecycle information.

2. Methodology

The study proposes a scalable, automated workflow to detect and characterize offshore platforms using Earth Observation (EO) data and Deep Learning.

• Data Source:
  • Sensor: Sentinel-1 C-band Synthetic Aperture Radar (SAR) data (Level-1 GRD, VH polarization, ~10m resolution).
  • Timeframe: Quarterly time series from 2017 Q1 to 2025 Q1.
  • Regions: North Sea (NS), Persian Gulf (PG), and Gulf of Mexico (GoM).
• Preprocessing:
  • Data was processed in Google Earth Engine (GEE) to create quarterly median composites. This technique suppresses mobile targets (like ships) while retaining stationary infrastructure signatures.
  • Backscatter values were normalized and converted to 8-bit integers for efficient processing.
  • Images were split into 640×640 pixel chips with 20% overlap.
• Deep Learning Model:
  • Architecture: YOLOv10 (You Only Look Once version 10) object detection model.
  • Training Strategy: The model was pre-trained on data from the South China Sea, Caspian Sea, Gulf of Guinea, and Brazilian coast. Crucially, the three study regions (NS, PG, GoM) were completely withheld as an independent test set to validate transferability to unseen marine environments.
  • Performance: Achieved a precision of 0.91, recall of 0.89, and F1 score of 0.90 on the independent test set.
• Post-Processing & Analysis:
  • Consolidation: Detections across quarters were merged using Intersection-over-Union (IoU) clustering to define unique physical platforms and derive lifespans (first to last detection).
  • Filtering: Offshore wind farm structures were removed using external layers; low-confidence detections and noise were filtered based on backscatter thresholds.
  • Spatial Enrichment: Platforms were enriched with attributes including Exclusive Economic Zone (EEZ) affiliation, distance to coast, water depth (via GEBCO bathymetry), and estimated size (bounding box area).

3. Key Contributions

1. Comprehensive Dataset (OPD v1.0.0): Creation of a publicly available, quarterly time series dataset (2017–2025) containing 3,728 unique platforms across three major production regions.
2. Lifecycle Dynamics: Derivation of installation, decommissioning, and relocation dates, enabling the analysis of platform persistence and turnover rates.
3. Scalable Workflow: A reproducible, automated pipeline using freely available data (Sentinel-1) and open-source tools, capable of being extended to global scales.
4. Open Access: The dataset is published in GeoParquet format via Zenodo, providing a baseline for independent research and policy analysis.

4. Key Results

• Platform Counts (2025 Q1):
  • Persian Gulf: 1,731 platforms (highest growth until 2024).
  • Gulf of Mexico: 1,641 platforms (dominated by the US).
  • North Sea: 356 platforms (lowest count, showing a decline).
• Regional Trends:
  • Persian Gulf: Continued expansion until 2024, driven largely by Saudi Arabia.
  • Gulf of Mexico & North Sea: Declining platform numbers since 2018–2020. The North Sea decline is linked to net-zero energy strategies; the GoM decline is linked to decommissioning of aging infrastructure.
• Structural Shifts:
  • Mobility: A significant number of platforms have short lifespans (<5 years), indicating a structural shift toward mobile offshore units (jack-ups, drillships) rather than permanent fixed platforms.
  • Turnover: Between 2017 and 2025, ~2,763 platforms were newly installed/relocated, while ~2,718 were decommissioned/relocated.
• Spatial Characteristics:
  • Depth: ~90% of platforms are in waters <100m deep. The deepest detected infrastructure (Turritella FPSO) was at >2,900m in the GoM.
  • Size: North Sea platforms are larger on average (5.33 ha) compared to the GoM (2.45 ha) and PG (2.04 ha), but the largest platform complexes are found in the Persian Gulf.
  • Distance: North Sea platforms are generally further from the coast (median 92 km) compared to the GoM (24 km) and PG (53 km).

5. Significance and Implications

• Policy and Regulation: The dataset provides an objective, independent baseline for governments and regulators to monitor compliance, plan decommissioning, and manage Exclusive Economic Zones (EEZs).
• Energy Transition: By quantifying the decline in fixed platforms and the rise of mobile units, the study highlights the sector's adaptation to economic and environmental pressures. It also identifies potential sites for repurposing (e.g., CO₂ storage, artificial reefs).
• Environmental Monitoring: The high-resolution spatiotemporal data allows for better assessment of the cumulative impact of offshore infrastructure on marine ecosystems.
• Methodological Advancement: The study proves that deep learning models trained on synthetic or specific regional data can successfully generalize to complex, unseen global marine environments using SAR time series, overcoming the limitations of optical sensors (cloud cover) and manual reporting.

Limitations: The study notes potential underestimation of very small platforms due to SAR resolution (10m) and acquisition gaps in deep central basins (e.g., parts of the GoM). However, the automated nature ensures reproducibility and scalability for future updates.