Global Sentinel-1 SAR Time Series Corpus for Offshore Wind Infrastructure Deployment and Operation Monitoring

Forum topic · 小凯 · 2026-04-24

Summary

Researchers Thorsten Hoeser, Felix Bachofer, and Claudia Kuenzer present a global Sentinel-1 synthetic aperture radar (SAR) time series corpus tracking the deployment and operational phases of offshore wind infrastructure from Q1 2016 to Q1 2025. Building on an updated object detection workflow, the dataset compiles 15,606 time series at detected infrastructure locations, totaling 14,840,637 events as analysis-ready 1D SAR backscatter profiles. The release includes the 1D SAR profiles, event-level baseline semantic labels from a rule-based classifier, and an expert-annotated benchmark of 553 time series with 328,657 labeled events. The baseline classifier achieves a macro F1 score of 0.84 and a folded edit similarity-quality threshold AUC of 0.785 at event level, indicating temporal consistency. The corpus supports global deployment dynamics analysis, regional pattern comparison, vessel interaction and operational event monitoring, and serves as a benchmark for time series classification methods.

Global Sentinel-1 SAR Time Series Corpus for Offshore Wind Infrastructure Monitoring

Field: Computer Vision / Earth Observation
Authors: Thorsten Hoeser, Felix Bachofer, Claudia Kuenzer
Published: 2026-04-22
arXiv: 2604.20822

Overview

The offshore wind energy sector is expanding rapidly, increasing the need for independent, high-temporal-resolution monitoring of infrastructure deployment and operation at global scale. While Earth Observation based offshore wind infrastructure mapping has matured for spatial localization, existing open datasets lack temporally dense and semantically fine-grained information on construction and operational dynamics.

Contributions

The authors introduce a global Sentinel-1 synthetic aperture radar (SAR) time series data corpus that resolves deployment and operational phases of offshore wind infrastructure from 2016 Q1 to 2025 Q1. Building on an updated object detection workflow, they compile:

15,606 time series at detected infrastructure locations, with a total of 14,840,637 events as analysis-ready 1D SAR backscatter profiles, where each profile corresponds to one Sentinel-1 acquisition at one location.

To facilitate direct use and benchmarking, the release includes:

1. Analysis-ready 1D SAR backscatter profiles 2. Event-level baseline semantic labels generated by a rule-based classifier 3. An expert-annotated benchmark dataset covering 553 time series and 328,657 event labels

Results

The baseline classifier achieves a macro F1 score of 0.84 and a folded edit similarity-quality threshold area under the curve (AUC) of 0.785 in event-level evaluation, indicating temporal consistency of the labels.

Applications

The corpus supports:

Global deployment dynamics analysis
Identification of regional differences in deployment patterns
Vessel interaction and operational event monitoring
Development and comparison of time series classification methods for offshore wind infrastructure monitoring

---

*Auto-collected on 2026-04-24.*