While radar analysis can expose “dark vessels”, know its limits: False positives from waves and islands are common. Ships under 30m are hard to detect. Coastal areas create noise. Best results come from combining radar data with OSINT methods.

#OSINT #InvestigativeJournalism #SAR #ShipTracking

Tech explainer: Synthetic Aperture Radar (SAR) satellites beam radar waves to Earth. Unlike optical satellites, they don’t need daylight or clear skies to spot things.

ESA's SNAP software comes with a powerful Ocean Object Detection feature for analyzing satellite radar images. This tool allows journalists to spot vessels even when they've turned off their AIS transponders. The interface is designed for non-programmers.

step.esa.int/main/download/snap-download/

The NYT investigation used Truthbrush, a powerful tool developed by the Stanford Internet Observatory. From my experience in social media research and data analysis, I can recommend Apify’s scrapers as an effective solution for this type of investigation.

apify.com

Before the elections, the NYT team analyzed 5,641 Trump posts on Truth Social – a platform that presents itself as a “free speech haven”. The reporters identified recurring themes and revealed a universe where Trump’s opponents became conspiracy plotters by default.

www.nytimes.com/interactive/...

It proved a crucial point: While data visualization helps grasp complex patterns, it’s the blend of satellite insights and on-ground reporting that tells the complete story.

3. Data needs human stories. In 2024 Bloomberg used satellite imagery to locate tents around Rafah to track displacement patterns.

www.bloomberg.com/graphics/202...

2. In journalism, results must be manually verified.

When the New York Times used AI to analyze satellite imagery revealing around 2,000 potential bomb craters in Gaza, every single impact site required careful human verification.

www.nytimes.com/2023/12/21/w...

The goal was to automatically analyze what the documents were about – without having to read each one, and to set up email alerts for specific matches. The payoff was worth every hour, even though untangling messy data turned what seemed straightforward into a time-intensive challenge.

1. Before building complex machine learning models, question whether manual analysis might be faster. In summer 2024, I worked on automating a database that regularly receives new documents.

In the latest episode, data scientist Krishna Karra from Bloomberg explains how he uses machine learning and satellite imagery for stories.

3 key takeaways :

Intrigued by Gaub’s insights on Russian military data in Syria, I discovered an article from the BBC that leveraged OSINT techniques to identify more than 500 Russian military personnel and Wagner members who died during Syrian operations between 2015 and 2024.

In Anne Will’s podcast, @florencegaub.bsky.social shared her analysis of assassination figures in Syria from two years ago, noting an uptick in terrorist attacks on Russian and Syrian officers. I highly recommend listening to her conclusions.

@trbrtc.bsky.social and Riley Mellen of The New York Times crafted a compelling analysis of the Tartus port using satellite imagery and marine tracking data. Their work exemplifies how precise insights can be gleaned from quality visuals and data without the need for extensive scrollytelling.

The Washington Post’s swift release of a visual timeline chronicling the fall of Assad’s regime, utilizing videos, photos, and maps. The chronological arrangement of videos provided a much clearer sense of how events unfolded.