What is a mixed pixel in satellite imagery?

A mixed pixel is what you get when a satellite's ground sample distance is bigger than the thing you're trying to measure. Say your sensor resolves 10 meters per pixel. A plot that's 20 by 15 meters doesn't get its own reading. It gets folded into two or three pixels, each of which also covers part of the neighbor's field, a strip of bare track, maybe a line of trees at the boundary. The reflectance value that comes back for that pixel isn't the crop. It's an average of everything the ground sample touched.

This is just arithmetic, not a sensor flaw. The instrument is doing exactly what it's built to do: integrating light over its footprint and reporting one number for that footprint. The problem shows up when the footprint is larger than the management unit you care about.

Where the averaging happens

On a half-hectare block with a single crop and a clean boundary, a wide-swath pass at 10 to 30 meters can still land a handful of pixels that are mostly crop. The averaging barely matters. On smallholder parcels, it matters a lot more, because the pixel footprint routinely straddles two or three different plots, each at a different crop stage, different variety, different water status. One pixel might be half a thriving maize stand and half a fallow strip. The index value for that pixel sits somewhere in between, and it belongs to neither field.

This is sub-pixel heterogeneity: real variation happening at a scale finer than the sensor can separate. The satellite doesn't know there are two fields down there. It reports one blended signal and calls it done.

Why it matters for NDVI on smallholder plots

NDVI is a ratio built from reflectance in two bands, so it inherits whatever averaging already happened at the pixel level. If a pixel is 60% healthy crop and 40% bare soil or stressed crop, the NDVI it returns isn't the healthy crop's number pulled down a bit. It's a blend that can mask a real stress signal, or invent one where there isn't any, depending on what's sitting next to the field in that footprint.

For an agronomist working a regional average or a district-level map, this washes out as noise across hundreds of fields and mostly isn't a problem. For someone trying to flag which individual plots need a visit this week, it's the whole problem. A field reading that's actually the field next door's irrigation canal plus half your client's sorghum doesn't tell you anything you can act on. You end up either ignoring the index for small plots entirely, or trusting a number that's wrong without any flag on it, and you usually don't find out until the agronomist on the ground tells you the map was off.

Finer-resolution imagery exists but isn't always available on a weekly cadence, and buying a tasking pass for every small field in a portfolio isn't how most advisory budgets work. Combining what the wide-swath pass sees every week with sharper looks at field boundaries resamples the index down to something closer to the actual field shape, instead of leaving it locked to the sensor's native grid.

That's the gap Field Scale Index is built around: fusing wide-swath multispectral passes with sharper imagery to produce a weekly index reading sized to the field, not the pixel. It's early access and being built out field type by field type rather than as a finished national layer, but if sub-pixel blending has been dragging down your plot-level NDVI without you noticing, it's worth a look.

If smeared pixels are costing you confidence in your smallholder index readings, get in touch about early access.