Management Zones

The Zone Editor's Configure Zones step lets you pick a clustering method. The portal exposes four, each with different math and different field-result behavior.

This page explains what each one does, in plain English, with rules of thumb for when to reach for which.

K-Means#

"Equal frequency method that groups data into k clusters based on similarity."

K-Means iteratively picks k cluster centers and reassigns points to the nearest center until the assignment stabilizes.

Behavior: tends to produce roughly similar-sized zones (in pixel count). Outliers can pull a cluster center if they're extreme.

Best for: most fields. K-Means is the safe default. If you're not sure, pick this.

Less ideal for: bimodal distributions where two real "modes" exist (e.g., a stressed strip plus a healthy main field). K-Means will smear across the gap.

Jenks Natural Breaks#

"Minimizes the variance within each class while maximizing the variance between classes."

Jenks finds the breaks in the NDVI distribution where adjacent values become most dissimilar. It's the canonical method for choropleth maps in cartography.

Behavior: places zone boundaries at the gaps in the histogram. Great for fields where there are obvious "natural" tiers.

Best for: bimodal or multi-modal NDVI distributions. Stressed strips, drainage patterns, soil-type transitions.

Less ideal for: very smooth NDVI distributions where there are no obvious gaps — Jenks will produce nearly the same result as K-Means but is slower.

Quantile#

"Divides the data into equal-sized groups."

Quantile assigns each block to a zone such that every zone has the same area (the same number of pixels).

Behavior: forces equal-sized zones regardless of NDVI distribution. Each zone covers exactly the same acreage.

Best for: scenarios where you want predictable per-zone product totals. Common for compliance reporting where each "zone" must be the same size.

Less ideal for: agronomy. The math can split similar-NDVI pixels into different zones (or merge dissimilar ones) just to maintain equal counts.

Equal Interval#

"Divides the range of the data into equal NDVI intervals."

Equal Interval splits the NDVI value range into N equal slices. If NDVI runs from 0.2 to 0.8 and you want 3 zones, you get 0.2–0.4, 0.4–0.6, 0.6–0.8.

Behavior: predictable, absolute thresholds. Two zones with NDVI in [0.4, 0.6] mean the same thing across fields and dates.

Best for: when you want comparability across fields or seasons. Research trials. Multi-field reports.

Less ideal for: skewed distributions. If a field's NDVI is mostly in [0.5, 0.7], Equal Interval can produce empty zones at the extremes.

Choosing by symptom#

Symptom	Method to try
First time, no strong opinion.	K-Means
Field has clear stressed pockets.	Jenks Natural Breaks
Need every zone the same acreage.	Quantile
Want absolute NDVI thresholds for comparison.	Equal Interval
Zones look noisy (speckle).	Increase block size
Zones look blocky and miss small features.	Decrease block size
Two adjacent zones with nearly identical rates.	Merge Zones

Block size — independent of method#

The block size parameter (3–50 m) is independent of the clustering method. Smaller blocks = finer detail but larger output. Larger blocks = smoother zones but lose small features.

A practical heuristic:

• For drone OTM imagery (2–10 cm/px), block size of 5 m works well.
• For satellite Sentinel-2 (10 m/px), block size of 10 m matches the sensor.
• For very large fields where the drone's swath is wide, 15–20 m blocks may make more sense.

After clustering — refinement#

Whichever method you pick, you'll usually refine after the auto-clustering:

• Merge Zones to collapse two zones that came out too similar.
• Redistribute Blocks to manually reassign odd blocks.
• Bulk Editing to box-select and set a single rate.
• Exclusion Zones for areas with no application at all.

See VRA and zones for the refinement workflow.

Related#

• NDVI.
• VRA Prescription.
• Map Prescription reference.
• VRA and zones guide.