Drone and satellite agro-analytics

Remote sensing of a field starts from a physical fact rather than an algorithm. Chlorophyll absorbs strongly in the blue and the red, leaving a reflectance trough near 670 nm; the mesophyll cell walls and air spaces inside the leaf, by contrast, absorb almost nothing beyond 750 nm and scatter near-infrared, reflecting it at 40–50%. Between the two regimes reflectance rises almost vertically across 700–730 nm — the red edge, the most informative stretch of the vegetation spectrum. A healthy dense canopy gives a deep red trough and a high infrared shoulder; a sparse, wilted or chlorotic one flattens both.

The normalised difference vegetation index is simply the shortest way of writing that contrast: infrared minus red over their sum. The normalisation cancels part of the effect of illumination and view angle, which is why the index compares between images better than the raw bands do.

Where it stops working

The limitation follows from the same physics. Red absorption saturates: by a leaf area index of about three the upper leaves intercept nearly all the red light, and another layer changes nothing in that band. NDVI flattens and stops distinguishing a good canopy from a very good one — precisely at the growth stage when nitrogen decisions in cereals are made. Red-edge indices such as NDRE saturate later, because the edge shifts in wavelength as chlorophyll content rises rather than merely changing in amplitude.

The second limitation is the mixed pixel. A 10-metre satellite pixel contains plant, soil and shadow together; soil reflectance varies with moisture and colour, and early in the season the soil contribution can exceed the crop’s.

Resolution, revisit, cloud

Choosing a platform is a trade among three quantities. Open satellite imagery from Sentinel-2 gives 10 metres every few days; Landsat gives 30 metres every 16; commercial constellations give metres at daily coverage. All of it is optical, so cloud deletes dates — and deletes them in the wet spells when disease develops fastest. Radar (Sentinel-1) sees through cloud but measures structure and moisture rather than chlorophyll; that is a different quantity, not a fallback.

A drone flies on demand at centimetre resolution, resolving individual plants and so supporting stand counts and plant-level treatment. It pays in coverage and radiometry: tens of hectares per flight, and comparability between flights requires a calibration panel and some handling of the angular dependence of reflectance.

An index is not a decision

The point that ties all of this back to the precision agriculture page is that an index is a proxy. It measures neither nitrogen nor yield. Turning it into an application rate requires local calibration — most commonly a reference strip given deliberately excess nitrogen, against which the rest of the field’s relative deficit can be read. Without that anchor, an index map is attractive and agronomically empty.