Agritech/United States
Only 17 percent of the satellite curves were usable for setting a nitrogen rate
Three on-farm corn trials tested whether satellite vegetation indices can set an in-season nitrogen rate. Only 17 percent of the response curves were usable.
Three rainfed on-farm corn trials in Indiana with contrasting tillage systems and four to five nitrogen rates used 3-metre PlanetScope imagery to compute sixteen vegetation indices. Vegetation index to yield relationships during the vegetative period were weak, R-squared at or below 0.31, and only 17 percent of the index-to-nitrogen response curves met the criteria for estimating an agronomic optimum nitrogen rate.
What happened
Variable-rate nitrogen application guided by satellite imagery is one of precision agriculture’s established products. The logic is clean: a vegetation index computed from satellite data acts as a proxy for how the crop is doing, which lets you estimate the economically optimum nitrogen rate during the season, in time to apply it.
The traditional method needs grain yield data, which arrives after harvest and is therefore useless for this year’s decision.
A team at the Department of Agronomy at Purdue University, the Department of Plant and Soil Sciences at the University of Kentucky and the Precision Agriculture Center at the University of Minnesota tested whether the satellite route works on real farm fields.
Seventeen percent of the response curves were good enough to use.
The test
Three rainfed on-farm field trials in Indiana in 2021, deliberately chosen with contrasting tillage systems and therefore contrasting amounts of crop residue on the soil surface. Each trial had four to five nitrogen rates.
PlanetScope imagery at 3-metre resolution was used to calculate sixteen vegetation indices across multiple growth stages – eight based on near-infrared and eight on red, green and blue bands alone.
The analysis had three parts. First, how strongly do vegetation indices relate to yield during the vegetative period, and does that vary by field. Second, can vegetation index to nitrogen response curves be used to estimate the optimum rate. Third, how close is that estimate to the optimum derived from actual yield.
Using on-farm trials with commercial tillage rather than research plots is what makes this a field-scale evaluation rather than a proof of concept.
The numbers
| Measure | Result |
|---|---|
| Fields | 3 rainfed on-farm trials, Indiana, 2021 |
| Nitrogen rates per trial | 4 to 5 |
| Vegetation indices computed | 16 (8 near-infrared, 8 RGB) |
| Imagery | PlanetScope, 3 m resolution |
| Vegetation index to yield relationship | generally weak, R-squared at or below 0.31 |
| Response curves meeting selection criteria | 17% |
| Effect of higher crop residue | fewer significant relationships, lower proportion usable |
| Best imagery timing | V10-V11 for near-infrared indices |
| Best single index | none consistently across fields and timings |
The result
During the vegetative period, vegetation index to yield relationships were generally weak, with R-squared at or below 0.31, and they varied across fields. Fewer significant relationships were found under higher crop residue conditions.
Of the vegetation index to nitrogen response curves evaluated, 17 percent met the selection criteria for estimating an index-based optimum nitrogen rate, with lower proportions in higher-residue systems.
Mid-vegetative period imagery at growth stages V10 to V11 produced the smallest deviations from the yield-based optimum for near-infrared indices. No single vegetation index consistently performed best across fields and timings.
The authors conclude that 3-metre satellite imagery has potential to detect crop nitrogen response under commercial field conditions, but its reliability for estimating in-season optimum nitrogen rate depends on management context, image timing and spectral domain.
How this compares to what came before
The research base for vegetation-index-guided nitrogen is substantial and most of it comes from small research plots with bare soil between rows, uniform management and intensive ground measurement. Under those conditions the relationships are much stronger.
The move to commercial fields with real tillage systems is where it falls apart, and the residue effect explains why. A satellite looking at a field under conservation tillage sees crop residue mixed with the crop canopy, and that contaminates the spectral signal the index depends on.
This matters because conservation tillage is increasing, for good soil and carbon reasons. The direction agriculture is moving is the direction that makes this technology work less well.
What it means
For farmers being sold satellite-guided nitrogen, the honest summary is that it may work on some fields in some years at some growth stages, and nobody can tell you in advance which.
For the technology, the finding is not fatal. 17 percent of curves being usable means the signal is sometimes there. The problem is the lack of a rule: no single index worked best, and usability depended on residue, field and timing. A product needs a rule.
The timing finding is the most immediately useful. V10 to V11 with a near-infrared index produced the smallest deviations, which is at least a starting point for anyone trying to make this work.
The broader lesson is about where agricultural technology is validated. Research plots are designed to produce clean signals. Commercial fields are not, and the gap between them is where precision agriculture products keep disappointing buyers.
Business ideas from this paper
Residue-corrected vegetation indices. The residue problem is physical and tractable. An index or correction specifically designed for high-residue systems would address the case where the current approach fails worst.
Field suitability screening. If only some fields give usable response curves, there is value in telling a farmer in advance whether theirs will. That screening service is cheaper to deliver than the nitrogen recommendation itself.
Higher-resolution or multi-source imagery. 3-metre PlanetScope was the input here. Drone imagery at centimetre resolution, or fusion with radar, might separate canopy from residue where satellite alone cannot.
On-farm trial networks. The evidence gap is between research plots and commercial fields. An organisation that runs replicated strip trials across many commercial farms, and sells the resulting evidence, serves both growers and vendors.
What could be wrong with it
Three fields in Indiana in a single season, 2021, all rainfed. That is a small base for a general conclusion, and season-to-season variation in corn is large.
The 17 percent figure applies to curves meeting the authors’ selection criteria, and different criteria would give a different number.
Only PlanetScope imagery at 3-metre resolution was tested. Finer imagery might perform differently.
The comparison standard is the yield-based optimum nitrogen rate, which is itself estimated from a limited number of rates per field.
No economic analysis is reported, so the paper does not say what following a wrong recommendation would cost.
How sure can you be?
We score this 7 out of 10.
It is a peer-reviewed paper in Frontiers in Plant Science, based on on-farm field trials under commercial tillage with replicated nitrogen rates, from three named institutions – Purdue University, the University of Kentucky and the University of Minnesota.
It loses points because the data are not openly deposited beyond the article and supplementary material, there is no independent replication, and the field and season base is small.
The finding is a careful negative result from a multi-institution team with no product to sell, which is the kind of evidence that is hardest to fund and most worth reading.
Do this today
If you are using satellite-guided nitrogen, check what your tillage system does to the signal. Under high residue, this trial found fewer usable relationships, and that is the condition most conservation-minded farms are in.
If you are evaluating a precision agriculture product, ask whether the validation was done on research plots or commercial fields. The difference is where the performance goes.
If you are a researcher in this area, the residue effect is an open and tractable problem. Correcting for it would make a technology that half works into one that works.
Source: Field-scale evaluation of satellite-derived vegetation indices and image timing for in-season nitrogen management in corn, Frontiers in Plant Science, April 2026. PMID 41937772 · doi.org ().
Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.
- Vegetation index to yield relationships during the vegetative period were weak, R-squared at or below 0.31.
- Only 17 percent of index-to-nitrogen response curves met the selection criteria, and fewer under higher crop residue.
- Mid-vegetative imagery at V10-V11 produced the smallest deviations for near-infrared indices.
- No single vegetation index performed best across fields and timings, so there is no rule to hand a farmer.
The evaluation used three rainfed on-farm trials under contrasting commercial tillage systems with replicated nitrogen rates and commercial satellite imagery.
Questions people ask
What is the agronomic optimum nitrogen rate?
The nitrogen application rate that maximises economic return, traditionally determined from grain yield data after harvest - which is too late to inform this season's sidedress application.
Why use vegetation indices?
They can be computed from satellite imagery during the growing season, so they could in principle estimate the optimum rate in time to act on it.
Why does crop residue matter?
Residue from the previous crop covers the soil and alters the spectral signal the satellite sees. Fewer significant relationships were found in higher-residue systems, which is exactly where conservation tillage is heading.
What is V10-V11?
Corn growth stages, roughly the mid-vegetative period when the plant has ten to eleven visible leaf collars. Imagery from this window produced the smallest deviations from the yield-based optimum.