
Before anyone pulls a scene or counts a machine, somebody has to draw a line on a map and decide that everything inside it counts and everything outside it doesn't. That line is the whole estimate. Get it wrong and the denominator you hand to the VP of strategy is wrong no matter how clean the counting was.
Most analysts don't think about the boundary as a methodology choice. They think of it as a formality, something the GIS person sorts out after the real work of counting starts. It isn't. The boundary sets what gets sampled, what the coverage ratio is measured against, and what the final scaled number means when someone asks "installed base of what, exactly, and where."
Why the boundary decides the math before you've counted a single machine
A satellite-based estimate works by counting machines on a set of sites you can image cleanly, then scaling that count up to the full region using a coverage ratio. The coverage ratio is just: how much of the region did we get clean, high-res, unoccluded imagery for, versus how much did we have to infer. If your sample region is drawn too tight around the easy, cloud-free sites, your coverage ratio looks great and your scaled number ends up quietly wrong, because you've sampled the parts of the territory that aren't representative of the whole.
If you draw the boundary too loose, pulling in adjacent counties or a neighboring state's agricultural belt just because the imagery happened to cover it, you end up scaling a number that doesn't match any denominator anyone downstream can use. Sales reports by dealer territory. Finance reports by state or by census region. Nobody reports by "wherever the satellite pass happened to be clean."
So the boundary has to do two jobs at once. It has to be a real, defensible geography that matches how the number will get used, and it has to be a geography you can get coverage for without occlusion correction doing most of the work.
Matching the boundary to how you'll report the number
Start from the denominator you need, not from the imagery you have. If the share number goes into a board deck as "installed base, upper Midwest," then the boundary is the upper Midwest as your own planning team already defines it, probably a set of states or a set of ag census regions, not a shape that happens to minimize cloud cover for Q3.
A few boundary choices come up often enough to name directly:
Dealer territory lines. These match sales reporting cleanly but rarely align with crop-type or terrain boundaries, so coverage and occlusion can vary a lot inside one territory. A hilly eastern edge of a territory might need more correction than the flat western two-thirds.
State or provincial lines. Easy to defend, easy to cross-check against public shipment and registration data, but they cut through machine populations that don't respect the line. A fleet that works both sides of a state border gets split in a way that doesn't reflect how the equipment moves.
Agro-ecological or census regions. These line up with how machines get used, since terrain and crop mix drive fleet composition, but they're harder to explain to a sales VP who thinks in territories and quarters.
None of these is automatically right. Pick the geography the downstream report needs, then treat gaps in that region, cloud cover, resolution gaps, terrain shadow, as something to correct for. Don't route around them by redrawing the boundary until the coverage looks clean.
Before you lock a boundary, check one more thing: the sample region you can actually photograph this quarter and the region you scale the estimate to don't have to match in size. You might only get clean imagery over 60% of a state this quarter. That's fine. The boundary you scale to should still be the full state your report needs, with the coverage gap corrected for rather than left out of the number.
Installed Fleet Estimation builds the scaling and coverage correction around exactly that distinction, which is worth understanding before you fix your own region boundary for a quarter.
If you're setting a region boundary for the first time and want a second set of eyes on whether it'll hold up as a market-share denominator, it's worth getting in touch before you lock it in.