Automating Offshore Infrastructure & Vessel Identifications Using Synthetic Aperture Radar & Distributive Geoprocessing

Abstract

Global Fishing Watch (GFW) recently published the first worldwide industrial fishing effort data set learned from processing 22 billion Automatic Identification System (AIS) observations. Despite quantifying 40 million hours of fishing activity that extended to over 55% of the ocean’s surface area in 2016, GFW now aims to quantify fishing effort not captured by current analyses through multimodal remotely-sensed imagery. Such imagery-based vessel identifications are commonly confounded with offshore infrastructure, though, so a global offshore infrastructure data set is first required to disentangle the two. This study first establishes robust and scalable methods for automating offshore infrastructure identifications using synthetic aperture radar in the Gulf of Mexico, and then evaluates the feasibility to adopt these methods for vessel identifications. Results indicate our model identifies offshore infrastructure with a probability of detection of 96.3%, an overall accuracy of 91.9%, a commission error rate of 4.7%, and an omission error rate 3.7%. Additionally, a cloud-native geoprocessing framework using the Google Earth Engine Python API was implemented to automate vessel identifications globally. Over 45,000 SAR images or approximately 100TB of data were processed to build a new database overlaying both SAR-derived and AIS-derived vessel locations.