Rainbow palettes create false boundaries

What it is

The rainbow palette (jet and kin) spreads the visible spectrum from blue to red across the value range and served as scientific computing's default ramp for decades. Its structural defects stack three layers: perceptually non-monotone — lightness descends then ascends along the ramp (dark blue, bright cyan-green, darker, bright red again), making two distant values equally strong; perceptually non-uniform — the green-yellow stretch commandeers enormous perceptual bandwidth while blue-violet crams together, so equal value increments earn wildly unequal visual differences; false boundaries — hue transitions (yellow-to-green, cyan-to-blue) read as data "steps," and readers see dividing layers in smooth gradients that do not exist. Together, half the "structure" on a rainbow chart is the ramp's product.

Why it happens

False boundaries come from hue discontinuity: the visual system is acutely sensitive to categorical hue switches (green-to-yellow is a "boundary"; gradients within yellow are not), and the ramp's hue segmentation carves a continuous range into strips that "look like different kinds." Add the non-monotone lightness pulling distant values together, and rainbow charts spontaneously generate pseudo-stratification: a linearly varying field renders as sandwiched bands. For scientific data this is directional damage — fronts, phase transitions, and anomaly zones identified from the chart are partly ramp artifacts; change the ramp and both the count and the placement of "structures" move.

Studying it

The systematic critique has a clear anchor: Borland and Taylor's "Rainbow Color Map (Still) Considered Harmful" (IEEE CG&A 2007) itemized rainbow mapping's perceptual defects (non-monotone lightness, non-uniform perception, hue segmentation manufacturing false boundaries) with replacement cases from medical imaging — the standard citation for this issue; the subsequent spread of perceptually uniform ramps (the viridis family) demonstrated the alternatives' viability. Replication is straightforward: render one continuous field under rainbow and perceptually uniform ramps, have domain readers mark "boundaries/layers," and the divergence between the two annotation sets is the rainbow's pseudo-structure contribution.

Where it stops holding

The rainbow is not without a seat: in categorical use (discretizing a continuous range into a few named intervals) hue segmentation is the function — but that is in substance "discretize plus categorical encoding," with intervals declared; in teaching demonstrations the rainbow's familiarity carries communicative value. Continuous-gradient ordinal encoding (heatmaps, terrain, medical imaging, depth maps) is the disaster zone — replace with perceptually uniform ramps without exception. The jet default in legacy systems is historical inertia; replacement is cheap and the payoff direct.

Applying it

• Ban jet/rainbow as the continuous-encoding default in spec; replace with viridis/magma or ColorBrewer sequential, keeping the old rendering alongside for a week to ease migration.
• Where segmented emphasis is genuinely wanted, switch to explicit binning with stepped or diverging palettes, turning "boundaries" from ramp artifacts into declared design.
• Verification: two ramps, one dataset, ask readers to count "visible layers/boundaries"; the rainbow's excess is the false-boundary count — keep it as replacement-review evidence.

Related

• Same group: U4.02.1 Sequential palettes must increase monotonically in perception · U4.02.3 Luminance carries the sense of order
• Nearby: U2.10.3 Colour scale choice directly shapes the conclusion · U4.04.3 Palettes must remain readable in greyscale
• Search terms: rainbow colormap · jet colormap · false boundaries · perceptually uniform