Plastic tunnel or glasshouse: reading structures in satellite imagery

When you're pulling a satellite pass over a growing region for the first time, the protected-cropping footprint usually reads as one thing from a distance: pale rectangles clustered along field edges. Getting from "there's something under cover here" to a usable inventory of plastic tunnels versus glasshouses takes a bit more than a glance at the thumbnail. Here's what separates the two in overhead imagery, and where it gets genuinely hard to tell.

The roofline is the first tell

Glass sheds sunlight in flat, angular flashes. Each pane sits at a slightly different tilt, so a glasshouse roof across a large nursery or tomato range often shows a stippled, broken glint pattern, almost like static, because you're seeing hundreds of individual reflective surfaces at once. Plastic film does the opposite. A polytunnel or multi-span plastic house reflects as one continuous, smoother sheen, sometimes with a soft gradient running along the curve of the arch.

On a tunnel, you'll also often pick up the rib shadows of the hoops themselves, faint parallel lines running across the structure at regular intervals. Glasshouses don't do that. Their internal structure is usually too fine to resolve even at sub-2m ground sample distance, so what you're reading is glazing bar geometry, not framework.

Roof shape matters too. Venlo-style glasshouses are built in repeating gable units, so from above you get a grid of small peaked roofs with gutters running between them, almost a waffle texture at full resolution. Tunnel houses read as long, unbroken barrel shapes, sometimes gathered in tight parallel rows (gothic arch tunnels for soft fruit or nursery stock) or as single wide spans over a hectare or more of glasshouse-scale plastic.

Shadow and shape give away the rest

Footprint geometry does a lot of the classification work before you even get to material. Tunnels tend to be narrow and long, often 6 to 10 meters wide and running the length of a field boundary, because they're cheap enough per meter that growers just keep extending the bay. Glasshouse ranges are usually squarer and more monolithic. A single glass range sized for year-round tomato or pepper production can cover several hectares in one connected roof, with service roads and loading areas built into the layout rather than tacked onto the edge.

Shadow length and direction at the time of the pass tell you about roof pitch, which correlates loosely with structure type. Glasshouse roofs pitch sharper for snow and rain shed in colder growing regions; tunnels are shallower arcs almost everywhere. On a low sun-angle pass, a glasshouse range throws a sawtooth shadow pattern off the gutters. A tunnel block throws a smooth, rounded shadow edge. Once you've seen both a few dozen times across a region, the pattern recognition gets fast, even on structures you've never walked past on the ground.

Where NIR earns its keep

RGB alone gets you most of the way, but near-infrared earns its keep on the ambiguous cases: old, degraded plastic that's gone cloudy and reflective enough to read as glass at low resolution, or a glass range under heavy whitewash coating that mutes the stippled glint signature. Polyethylene film and glass behave differently in NIR, especially where condensation sits on the inside of the cladding, which shows up on both materials but reads differently in the signal. Running RGB and NIR together, standard on the high-resolution satellite passes used for this kind of structure mapping, resolves a share of the cases a true-color image alone can't call with confidence.

Why this matters for territory sizing

None of this is interesting as a classification exercise on its own. It matters because tunnel and glasshouse growers buy differently, by crop choice, by input volume per square meter, by season length, and a regional map that lumps all protected cropping into one undifferentiated category tells an input supplier or planner a lot less than one that separates structure type and footprint. That's the gap Greenhouse Mapping is built to close: an annual, region-wide inventory of protected-cropping structures, sized and classed by type, built from the same high-res RGB and NIR passes described above.

If you're trying to size a protected-cropping territory and field survey alone won't cover the ground, it's worth seeing what a structure inventory for your region actually looks like.