We're actively working to expand imagery coverage at LGND with open source collections and commercial imagery partnerships. We’ve recently expanded coverage across Europe and want to highlight Spain’s PNOA survey - a national dataset that captures high-resolution aerial photographs every 2-3 years - to demonstrate LGND’s capabilities, in service of the world's greatest sport. 

Football Pitch Expansion

Spain currently holds both the men's and women's FIFA World Cup trophies — the first country ever to hold both at once. That success rests on decades of investment in grassroots facilities.

Using Spain's national aerial survey and LGND's technology, we built La cantera, from above to trace the ground game behind it.

The application steps through Spain's PNOA survey epoch by epoch — 2006, 2009, 2011, 2014, 2017, 2020, 2023 — plotting every football pitch across greater Madrid that changed state between flights. Grey dots are existing natural or dirt pitches; yellow means a pitch was resurfaced, almost always to artificial turf; red means a pitch appeared where there was nothing before. Scrubbing the timeline plays the city's build-out in real time, with a running count of resurfaced and new-build pitches ticking up at each epoch — 86 changed pitches detected between 2006 and 2009 alone, tapering to 34 in the most recent survey window. Click any dot and the panel shows the actual before/after image chips LGND classified to make that call, so the detection isn't a black box — you can see the dirt lot and the astroturf pitch that replaced it, dated to the exact pair of flights that caught the change.

Under the hood, that's the Embeddings API and Studio workflow product doing the work: each PNOA epoch is embedded as a set of image chips, and LGND compares chip embeddings across consecutive epochs at the same location to classify what changed — nothing to pitch, dirt to turf, or no change at all. Run across all seven epochs, that comparison surfaced at least 257 pitches built or resurfaced, from full 11-a-side grounds down to 5-a-side courts, individually tracked over seventeen years. 

Why resolution makes a difference

A Sentinel-2 pixel covers about 100 square meters (10 meter x 10 meter) — roughly the footprint of a small house. That's genuinely useful for things such as tracking land cover or spotting large-scale construction. But at that resolution, a football pitch is a handful of pixels at best, indistinguishable from the park or lot around it. 

Spain's PNOA survey runs far finer than that with most recent surveys at a resolution of 0.25 meters. At that resolution a pitch stops being a blur and becomes a shape — one an embedding can recognize, compare across years, and tell apart from a car park or a rooftop of the same size. That's the general capability the La Cantera project points at: once imagery is sharp enough to represent individual objects rather than blended land cover, the LGND stack can find, count, and track those objects at scale. 

Try it 

The Madrid pitches project is one example of what high-resolution imagery makes possible in our stack. Elsewhere, the same approach applies to:

• Wildfire risk assessment. Find every rooftop with wood shake or shingle roofing versus tile or metal across a wildland-urban interface — a distinction that's invisible at 10m but a first-order signal for insurers and fire agencies.
• Insurance and property underwriting. Search for pools, trampolines, solar installations, or detached structures that change a property's risk profile, at the scale of an entire portfolio rather than a handful of site visits.
• Code and permit compliance. Run change detection on individual parcels to flag new construction, a converted outbuilding, or a structure that was never permitted — object-level events, not a blurred regional trend.

We're always interested in what problems people are trying to solve with it. If you have a use case in mind connect with us to talk through the possibilities with our team. Or skip the conversation and try it yourself:

• Building your own application? Spin up a new collection against any of these imagery sources through the Embeddings API and start running similarity search, change detection, or natural language queries over the new coverage.
• Prefer a visual workflow? Contact us to get access to LGND Studio and start exploring what's newly discoverable in your region.