
When you scale a set of site-level machine counts into a regional installed-base number, the raw count from imagery is never the final number. It's an input that needs a correction step, and that step is what people mean by coverage correction.
The problem it solves
A satellite pass over a region doesn't see every site with equal clarity. Some sites get a clean, cloud-free image on the tasking date. Others get partial cloud, haze, a shadow from a nearby structure, or a revisit gap because the orbit path didn't line up with that site that week. Sum the machines you can count across whatever images you happened to get, and you're not counting the fleet. You're counting the fleet that was visible, which skews toward whatever region, season, or site layout happened to produce clean imagery.
That skew is coverage bias. It shows up as an undercount in cloudy regions, in areas with dense tree cover near yards, in sites where machines sit under open-sided sheds, and anywhere the usable image fraction ran lower than average. Left uncorrected, a regional installed-base figure built straight from satellite counts will systematically favor the clear-sky, open-yard regions over everything else. An analyst comparing market share across regions on that basis is comparing apples sized differently depending on how cloudy it was in July.
How the adjustment works
Coverage correction treats "percent of each site actually visible and interpretable in the image" as a measurable quantity, not a constant. For a given site and pass, you estimate how much of the yard, lot, or storage area was obscured by cloud, shadow, roof cover, or image edge effects, and how much was clean enough to count machines on. The raw count from that image gets scaled up in proportion to the obscured fraction, on the assumption that machine density in the hidden part of the site resembles density in the visible part.
Occlusion correction is the sibling of this step. It handles cover from fixed structures, sheds, canopies, stacked units blocking other units, rather than weather, but the two usually run together because both are forms of the sensor failing to see part of what's on the ground.
Roll the corrected per-site figure up across every site in a region and you get a regional fleet count adjustment that isn't just a sum of raw pixels-to-machines tallies. It's a sum of what each site was estimated to hold, with a wider correction applied to the sites where the image gave you less to work with.
Why this matters more than resolution
An OEM strategy team building a market-share denominator usually assumes the hard part is resolution, whether you can tell a skid steer from a compact track loader at a given ground sample distance. Resolution matters, but coverage correction is what determines whether your regional number holds up when someone across the table asks why it differs from the dealer survey estimate. A raw satellite count with no coverage correction tends to run low, and it runs low unevenly: worse in some regions than others, which is exactly what breaks a cross-region market-share comparison.
Dealer survey estimates have their own coverage problem too, namely which dealers responded and how they defined an active unit, but cloud cover isn't one of their variables. Correction is how satellite-based counting closes its version of that gap, since sharper imagery on its own doesn't fix a site that was cloud-covered or occluded on the day of the pass.
If you're trying to build a denominator you can defend in a quarterly review, the correction methodology is worth asking about directly: how is occlusion estimated per site, how is the correction factor derived, and is it re-estimated per pass or held constant across a region.
Installed Fleet Estimation applies coverage and occlusion correction as part of scaling site-level counts into a regional figure, the method described on the Installed Fleet Estimation homepage. If your current regional count has never had this step run against it, request access and see what the corrected number looks like.