Header image

Huge areas of PHL are still missing from OpenStreetMap, satellite data reveals

March 12, 2019 blog-post geospatial maps openstreetmap open-data remote-sensing satellite-imagery computer-vision

When Typhoon Haiyan (Yolanda) ravaged the Philippines in 2013, humanitarian aid workers on the ground urgently needed up-to-date maps of the affected areas.

For this, they turned to OpenStreetMap, the world’s Wikipedia of Maps. In the aftermath of Haiyan, hundreds of volunteer cartographers from around the world made 1.5 million edits to OpenStreetMap, painstakingly tracing thousands of roads and buildings over post-typhoon satellite imagery. Their work helped disaster response agencies locate damaged buildings, route relief efforts, and plan rescues.

Today, OpenStreetMap is still the world’s best free and open alternative to proprietary maps, such as Google Maps, which do not make their underlying data available to the public. Open maps are hugely useful, not only for crisis response, but also for transportation planning, forest management, and more. At Thinking Machines, we’ve even used OSM data to estimate poverty in remote areas of the Philippines.

Yet while citizen volunteers around the world have worked hard to make OpenStreetMap as complete as possible, there’s still a lot of work to be done!

Our team wanted to identify areas of the Philippines where OSM data is still missing. To do this, we compared a dataset of built-up areas as derived from satellite imagery with building locations on OSM (as of March 1). The result is this interactive map below, which visualizes levels of OSM completeness across the country.

This map also shows the power of satellite imagery and computer vision to augment mapping done by people. While it would take an army of volunteers months or years to map every single building and road in the Philippines, a machine learning algorithm can potentially do the same task much more quickly.

At Thinking Machines, we're working on developing building and road detection models trained specifically for Philippine infrastructure. Government teams like the Department of Science and Technology's Remote Sensing and Help Desk or DATOS are also doing similar work, using machine learning to map crops, roads, and flood-prone areas.

Yet for their speed and scalability, machines can never replace local knowledge. An algorithm may be able to identify a building, but only local communities can provide ground-truth insight on how these buildings or roads are used, what they're called, and the people who built and use them. For this, volunteers are irreplaceable.

Explore the map to find OSM’s uncharted territories in the Philippines. Or, better yet, create an account on OSM so you can help #MaptheGap too!


Harnessing the Potential of Claims Data To Detect Insurance Fraud

Developed a fraud detection model to help identify insurance claims outliers for further investigation.

When did everything ML get 'deep'? Using fasttext to examine semantic trends in 30 years of NIPS papers

We use word embeddings to look into the evolving landscape of research topics in NIPS, AI’s most prestigious conference, through the years.

Kailan ka ikakasal? Four data-driven comebacks to use at your next family reunion

Actually, Tita, the number of Filipinos getting married has declined 15% in the past 10 years. It's part of a nationwide trend.