How small can a field be before satellite NDVI stops working?

If you've ever pulled an NDVI layer over a smallholder plot and watched the whole field disappear into one blurry square shared with the neighbor's maize and a strip of bare ground, you've already found the limit. The question isn't really "how small is too small." It's "how many pixels does your field actually fill," and for a lot of smallholder ground, the honest answer is less than one.

The pixel math nobody puts on the label

Most of the free, wide-swath multispectral satellites agronomists reach for first sit somewhere in the 10 to 30 meter ground sample distance range. That sounds fine until you do the area math. A 10 m pixel covers about 100 square meters, call it 0.01 hectare. A 30 m pixel covers roughly 900 square meters, about 0.09 hectare. A single pixel from the coarser end of that range can be bigger than an entire quarter-acre plot.

Now put an actual field inside that grid. A half-hectare plot at 10 m resolution might span 40 to 50 pixels, which sounds workable until you remember that the pixels along every edge are mixing your crop's reflectance with the access track, the irrigation channel, the neighbor's fallow strip, or the farmhouse roof. At 30 m resolution, that same half-hectare field might be smeared across three or four pixels total, most of them edge-contaminated. What comes back is an average of the crop and everything touching it along its edges.

This is the mixed-pixel problem, and it's the real reason "minimum field size for satellite monitoring" doesn't have one clean number. It depends on field shape (a long thin plot loses more of its interior to edge mixing than a square one of the same area), on how much contrast there is between the crop and its surroundings, and on which sensor's swath you're pulling from. But as a rule of thumb: once a field's shortest dimension gets close to the pixel size itself, you're monitoring a blend of several ground covers, not a single crop.

What this looks like in the field book

For an advisory agronomist, this shows up as a specific, recurring annoyance. You pull up the index for a farmer's parcel ahead of a visit, and the reading looks suspiciously similar to the three fields around it, because the regional pixel grid doesn't know where one farmer's boundary ends and the next one starts. Two adjacent plots with genuinely different stress levels, one short on water, one not, come back reading nearly identical, because they're sharing pixels. You end up driving out to check anyway, which defeats the point of pulling a remote reading in the first place.

The satellites with the shortest revisit times, the ones giving you a weekly look, are exactly the wide-swath, coarser-resolution instruments, so a sharper single sensor alone doesn't fix the revisit problem. The practical answer combines passes: a wide-swath multispectral pass for frequent coverage, fused with sharper looks, so the result lands closer to field scale instead of regional-average scale. That's a fusion problem, sitting in the gap between satellite imagery existing for your district and being able to tell field A from field B.

Where field size vs. pixel resolution leaves you

If your fields run several hectares and square-ish, a straight wide-swath NDVI pull will usually behave itself, because the field fills enough pixels that edge mixing is a small fraction of the total. If your advisory book is mostly smallholder plots under a hectare, narrow or irregular in shape, or tucked between other crops, treat any single coarse-resolution reading with suspicion until you've checked how many clean, non-edge pixels fall inside the boundary.

Field Scale Index is built for that second case: it fuses wide-swath passes with sharper looks to get a weekly index reading near field scale instead of a regional smear, for the field types the fusion pipeline covers today. It's early access, rolling out field type by field type rather than arriving as a finished national layer, so the honest move is to see what's covered so far before you build it into a visit schedule. If your parcels keep getting lost in somebody else's pixel, that's worth five minutes to check.