Why Variable-Rate Doesn't Always Work: Lessons from 4 Field Trials
Reading time — 7 minutes
A pattern analysis across four corn and winter rapeseed trials in Ukraine's Forest-Steppe and Polissia regions, 2023–2025.
Olha Matsera
Agronomist-expert at OneSoil,PhD in Agriculture (Plant Growing)
We pulled data from four field-level trials run on the OneSoil platform between 2023 and 2025, across Ukraine's Forest-Steppe and Polissia regions.
In one case, variable-rate seeding produced a clean 3.5 t/ha yield response curve across productivity zones. In the other three, the prescription either failed to read or actively worked against the grower's goal.
Here's what separated them.
In one case, variable-rate seeding produced a clean 3.5 t/ha yield response curve across productivity zones. In the other three, the prescription either failed to read or actively worked against the grower's goal.
Here's what separated them.
We pulled data from four field-level trials run on the OneSoil platform between 2023 and 2025, across Ukraine's Forest-Steppe and Polissia regions.
In one case, variable-rate seeding produced a clean 3.5 t/ha yield response curve across productivity zones. In the other three, the prescription either failed to read or actively worked against the grower's goal.
Here's what separated them.
In one case, variable-rate seeding produced a clean 3.5 t/ha yield response curve across productivity zones. In the other three, the prescription either failed to read or actively worked against the grower's goal.
Here's what separated them.
Olha Matsera
Agronomist-expert at OneSoil,PhD in Agriculture (Plant Growing)
A note on what we're measuring
Two map types come up repeatedly in the analyses below:
❗Treating them interchangeably is one of the more common ways trials get muddled.
- productivity zones describe a field's long-term yield potential, derived from multi-year satellite and soil data.
- current contrast NDVI describes how plants are actually performing right now, this season.
- productivity zones answer "where is this field's ceiling?" - useful for decisions made before the crop is in the ground.
- current NDVI answers "how are plants doing today?" - useful for in-season interventions.
❗Treating them interchangeably is one of the more common ways trials get muddled.
🌽 Case 1 - Field A: When rate contrast is too small to read
Setup: 105 ha corn field, Forest-Steppe region, 2025 season.
A drought-affected year - accumulated NDVI of 0.11 against a potential of 0.44, with regional average corn yields running 6.7–7.13 t/ha against a field average here of 3.68 t/ha.
What was applied: Seeding rates of 60 / 64 / 68 thousand seeds/ha across low, mid, and high productivity zones - a deviation of roughly ±6% from the mid-zone rate.
What the data shows: The expected pattern - higher rates lifting yield in high-potential zones - didn't appear.
The lowest rate (60k) produced the highest yield in both the low zone (3.75 t/ha) and the high zone (3.85 t/ha). The highest rate (68k) produced the lowest yields in those same zones (3.32 and 3.70 t/ha respectively).
What the data suggests:
A ±6% rate contrast sits inside natural field variability. The trial recorded noise more than rate effect.
The differentiation needed to produce a readable signal is ±10–15% from the mid-rate - for this field, that would have meant roughly 57.6 / 64 / 70.4 thousand seeds/ha.
A drought-affected year - accumulated NDVI of 0.11 against a potential of 0.44, with regional average corn yields running 6.7–7.13 t/ha against a field average here of 3.68 t/ha.
What was applied: Seeding rates of 60 / 64 / 68 thousand seeds/ha across low, mid, and high productivity zones - a deviation of roughly ±6% from the mid-zone rate.
What the data shows: The expected pattern - higher rates lifting yield in high-potential zones - didn't appear.
The lowest rate (60k) produced the highest yield in both the low zone (3.75 t/ha) and the high zone (3.85 t/ha). The highest rate (68k) produced the lowest yields in those same zones (3.32 and 3.70 t/ha respectively).
What the data suggests:
A ±6% rate contrast sits inside natural field variability. The trial recorded noise more than rate effect.
The differentiation needed to produce a readable signal is ±10–15% from the mid-rate - for this field, that would have meant roughly 57.6 / 64 / 70.4 thousand seeds/ha.
The principle: If the rate difference is smaller than the field's normal variation, the trial cannot be read — regardless of how cleanly the prescription map is drawn.
🌽 Case 2 - Field B: When methodology aligns
Setup: 31.6 ha corn field, Polissia region, 2023 season.
Favorable weather conditions: ~17.3°C average growing-season temperature, 456 mm accumulated precipitation, vegetation NDVI of 0.44.
What was applied: Seeding rates of 65 / 74 / 80 thousand seeds/ha - contrast of approximately ±10–13% from the mid rate, within the recommended differentiation range.
What the data shows:
In the high productivity zone, yields rose cleanly with seeding rate — 7.12 → 8.77 → 10.6 t/ha.
In the low productivity zone, all three rates produced equivalent yields around 8.28 t/ha, meaning the lower rate captured the same yield with less seed cost.
The mid zone produced the field's highest yields, up to 12.8 t/ha at the elevated rate. Field average: 9.11 t/ha.
What the data confirms: When the rate contrast is sufficient, methodology delivers what the model predicts. In the low zone, the grower can confidently apply the lower rate going forward, saving on seed without yield penalty. In the high zone, higher rates are justified by the yield-response curve.
Favorable weather conditions: ~17.3°C average growing-season temperature, 456 mm accumulated precipitation, vegetation NDVI of 0.44.
What was applied: Seeding rates of 65 / 74 / 80 thousand seeds/ha - contrast of approximately ±10–13% from the mid rate, within the recommended differentiation range.
What the data shows:
In the high productivity zone, yields rose cleanly with seeding rate — 7.12 → 8.77 → 10.6 t/ha.
In the low productivity zone, all three rates produced equivalent yields around 8.28 t/ha, meaning the lower rate captured the same yield with less seed cost.
The mid zone produced the field's highest yields, up to 12.8 t/ha at the elevated rate. Field average: 9.11 t/ha.
What the data confirms: When the rate contrast is sufficient, methodology delivers what the model predicts. In the low zone, the grower can confidently apply the lower rate going forward, saving on seed without yield penalty. In the high zone, higher rates are justified by the yield-response curve.
The principle: When the math is right, agronomy follows.
🌽Case 3 - Field C: When the trial doesn't trial
Setup: 59.4 ha corn field, Polissia region, 2023 season - favorable conditions, field average 10.1 t/ha.
What was applied: Rates of 68 / 78 / 80 thousand seeds/ha. In the high productivity zone, that's a difference of only 2,000 seeds/ha (~2.5%) from the mid-zone rate.
What the data shows:
In the high zone, all three rate variants produced identical yields - 9.7 t/ha across the board.
The mid zone, with broader contrast, showed a meaningful response (9.27 vs 11.5 vs 12.1 t/ha).
The low zone behaved similarly to Field B: the lower differentiated rate captured equivalent yields (~10.3 t/ha), supporting a seed-cost-saving strategy.
What the data suggests: A 2k-seed difference falls within seed-meter precision tolerance and natural in-zone variability. In the high zone, the trial measured planter accuracy, not seeding-rate effect. To actually test the hypothesis, the high-zone rate would have needed to push ±10–15% beyond the mid rate — roughly 84–88k seeds/ha.
What was applied: Rates of 68 / 78 / 80 thousand seeds/ha. In the high productivity zone, that's a difference of only 2,000 seeds/ha (~2.5%) from the mid-zone rate.
What the data shows:
In the high zone, all three rate variants produced identical yields - 9.7 t/ha across the board.
The mid zone, with broader contrast, showed a meaningful response (9.27 vs 11.5 vs 12.1 t/ha).
The low zone behaved similarly to Field B: the lower differentiated rate captured equivalent yields (~10.3 t/ha), supporting a seed-cost-saving strategy.
What the data suggests: A 2k-seed difference falls within seed-meter precision tolerance and natural in-zone variability. In the high zone, the trial measured planter accuracy, not seeding-rate effect. To actually test the hypothesis, the high-zone rate would have needed to push ±10–15% beyond the mid rate — roughly 84–88k seeds/ha.
The principle: A trial without sufficient contrast is not a trial - it's a confirmation that the equipment works as designed.
🌾Case 4 - Field D: When input type inverts the logic
Setup: 23.2 ha winter rapeseed field, Western Ukraine, 2024–2025 season. Operation: early-spring UAN application with a nitrogen stabilizer (urease and nitrification inhibitors), applied on frozen-thawed soil to support vegetation restart.
What was applied: Based on a current contrast NDVI map, doses of 169 / 245 / 321 l/ha — more nitrogen to high-NDVI zones (already-strong plants), less to low-NDVI zones (struggling plants).
What the data shows: The yield gap widened. Low-NDVI zone: 2.29 t/ha. Mid: 2.32 t/ha. High: 2.55 t/ha. Field average: 2.03 t/ha. Plants that were already strong got stronger. Plants that were already weak stayed weak.
What the data suggests:
For early-season nitrogen applied to existing plant tissue, the logic is compensatory, not amplifying. Plants in the low-NDVI zone - visibly slower to restart vegetation - needed more N to catch up.
Plants in the high-NDVI zone - already drawing adequate N from soil reserves and reflecting that in their biomass - needed less. An inverted gradient (something closer to 321 / 245 / 169 l/ha) would likely have narrowed the yield gap and raised the field average.
What was applied: Based on a current contrast NDVI map, doses of 169 / 245 / 321 l/ha — more nitrogen to high-NDVI zones (already-strong plants), less to low-NDVI zones (struggling plants).
What the data shows: The yield gap widened. Low-NDVI zone: 2.29 t/ha. Mid: 2.32 t/ha. High: 2.55 t/ha. Field average: 2.03 t/ha. Plants that were already strong got stronger. Plants that were already weak stayed weak.
What the data suggests:
For early-season nitrogen applied to existing plant tissue, the logic is compensatory, not amplifying. Plants in the low-NDVI zone - visibly slower to restart vegetation - needed more N to catch up.
Plants in the high-NDVI zone - already drawing adequate N from soil reserves and reflecting that in their biomass - needed less. An inverted gradient (something closer to 321 / 245 / 169 l/ha) would likely have narrowed the yield gap and raised the field average.
The principle: Input type determines logic direction.
Base-rate fertilization, applied pre-season into soil, tracks long-term zone potential - more input where potential is higher, positive correlation.
In-season foliar or quasi-foliar applications, applied to standing biomass, follow inverted logic - more input where plants are currently weakest, negative correlation.
Base-rate fertilization, applied pre-season into soil, tracks long-term zone potential - more input where potential is higher, positive correlation.
In-season foliar or quasi-foliar applications, applied to standing biomass, follow inverted logic - more input where plants are currently weakest, negative correlation.
The pattern across four fields
What this means in practice
The technology side of variable-rate is the mature part.
Prescription maps generate cleanly, machinery executes them accurately, satellite data refreshes on schedule. The friction sits one layer up - in the methodology decisions that determine whether the prescription was well-designed in the first place.
Most growers adopting VRA expect the technology to make those decisions for them. Field A through D suggest the opposite is true.
For growers running their first variable-rate trials, three checks are worth doing before drawing conclusions from the data:
Prescription maps generate cleanly, machinery executes them accurately, satellite data refreshes on schedule. The friction sits one layer up - in the methodology decisions that determine whether the prescription was well-designed in the first place.
Most growers adopting VRA expect the technology to make those decisions for them. Field A through D suggest the opposite is true.
High-resolution data about a field is necessary but not sufficient - the difference between Field B's clean 3.5 t/ha rate-response curve and Field D's widening yield gap sits in how the decisions were framed, not in which platform generated the maps.
For growers running their first variable-rate trials, three checks are worth doing before drawing conclusions from the data:
- Did the rate or dose contrast meaningfully exceed natural variability? If not, the result is unreadable regardless of what the numbers say.
- Did the logic direction match the input type? Base-rate and in-season treatments follow opposite rules.
- Was the result interpreted in the context of season weather, soil agrochemistry, and within-zone heterogeneity — or treated as a clean A/B/C test?
Variable-rate isn't a single technology decision. It's a chain of four - and any weak link sets the whole prescription back toward field-average performance.
More Articles
People, fields and technology
Get all new articles straight to your inbox! You can read them even in the field.