Two-stage random forest framework for offshore platform identification to analyze spatial patterns and activity levels

Offshore oil and gas platforms constitute critical infrastructure for national energy resource development, with their spatial density and operational activity exerting direct influence on national energy production capacity and the stability of maritime energy corridors. This study develops a multi-source remote sensing framework for identifying and analysing global offshore oil and gas platforms, integrating Sentinel-1 radar, Sentinel-2 multispectral, and VIIRS nighttime light imagery. A two-stage random forest approach is used for platform identification, supporting the construction of a global long-term platform dataset and subsequent spatial and temporal analyses. Utilizing monthly VIIRS nighttime light time series from 2017 to 2024, platform activity patterns are characterized through a radiometric contrast approach, revealing temporal variations and distinct seasonal characteristics across major producing countries. Furthermore, a framework for analysing the global spatial patterns of offshore oil and gas platforms was developed. Using the Exclusive Economic Zone (EEZ) as the basic unit of analysis, the framework employs indicators including platform density, average nearest-neighbor distance, nearest-neighbor index, and local clustering intensity to quantitatively characterize their spatial distribution and clustering characteristics, revealing regional differences and spatial heterogeneity in offshore oil and gas development. The multi-source remote sensing framework and quantitative analyses developed in this study provide a consistent basis for characterising the global spatial and temporal patterns of offshore oil and gas platforms, and may support further research on offshore energy development, marine spatial planning, and related resource management.