The 7.5% Visual Distortion Penalty of the Rainbow Colormap

On October 28, 2020, the journal Nature Communications published a systematic review quantifying a specific failure in scientific communication: the rainbow colormap. Geoscientist Fabio Crameri and his colleagues demonstrated that the ubiquitous red-to-blue gradient, often called 'Jet', was visually distorting data by up to 7.5% through uneven color gradients.

The paper marked a formal turning point in a quiet war against a visualization default that had dominated scientific software since the 1990s. By measuring the incremental lightness difference along the color map, the researchers proved that the rainbow's lack of perceptual uniformity was not merely an aesthetic flaw, but a mathematical one that manufactured false boundaries where none existed in the underlying data.[2]

The mechanism behind this distortion lies in how the human eye processes luminance, or brightness. A perceptually uniform colormap ensures that equal steps in data correspond to equal perceptual steps in color. In a rainbow colormap, the brightness climbs and falls erratically.

The transition from blue to cyan is bright, the green section is dark, the yellow is intensely bright, and the red drops back to dark. When a viewer looks at a heatmap rendered in Jet, the bright yellow and cyan bands draw the eye immediately, creating the illusion of sharp gradients or anomalies in regions where the actual data is flat.[1]

This perceptual failure has concrete consequences in high-stakes environments. When researchers tested how physicians read artery visualizations, they found that doctors using a rainbow color scheme took longer to reach a diagnosis and made significantly more errors than those using a perceptually designed scale.

As one analysis noted, "The color map was not a neutral container for the data. It was an active participant in the diagnosis — and a misleading one." The bright red patches in the Jet colormap artificially amplified the appearance of blood-flow blockages, leading cardiologists to see problems that the raw numbers did not support.[1]

The exclusion extends beyond diagnostic errors to basic accessibility. Approximately 300 million people worldwide—roughly one in 12 men and one in 200 women—live with some form of color vision deficiency. For a large share of this population, the red-to-green journey at the heart of most heatmaps is partly or wholly invisible. When a rainbow colormap is desaturated to grayscale, its non-monotonic luminance causes different data values to map to the exact same shade of gray, rendering the chart entirely illegible to anyone relying on brightness to interpret the scale.[1][3]

The software industry's pivot away from Jet began in earnest on July 20, 2015, at the SciPy conference in Austin, Texas. Developers Stéfan van der Walt and Nathaniel Smith presented a new default colormap for Matplotlib, the foundational plotting library for the Python programming language. They named it 'Viridis'.

Engineered to be perceptually uniform, monotonic in brightness, and legible to viewers with common forms of color blindness, Viridis transitions smoothly from dark purple to blue, green, and finally bright yellow. Equal steps in the data look like equal steps in color, and the order survives even when printed in black and white.[3]

The adoption of Viridis triggered a cascade of changes across the data science ecosystem. The R programming language integrated the colormap through the 'viridisLite' package, which has since become a standard dependency for spatial and statistical visualization. In 2019, a comparative study published in MDPI evaluated the task completion time and correctness rate of participants using different colormaps in node-link diagrams. The researchers found that Viridis was the least error-prone among the colormaps studied, while single-hue blue and the Jet rainbow produced significantly higher error rates in data comprehension.[4]

Despite these empirical findings, the rainbow colormap remains stubbornly entrenched in certain disciplines. A 2026 study published by the American Meteorological Society analyzed the visualization of lidar data—laser-based remote sensing used in atmospheric science.

The authors noted that despite three decades of publications showing the artifacts produced by the rainbow colormap, it remains widely used in remote sensing, where it appears 14 times more frequently than in information visualization literature. The high-frequency noise inherent in lidar signals is particularly vulnerable to the hue banding effect of the Jet colormap, which exacerbates the visual clutter.

The persistence of Jet is largely a product of institutional inertia and the default settings of legacy software. Older versions of MATLAB cemented the rainbow as the standard for engineering and physics, and it continues to lurk in spreadsheet conditional formatting and commercial dashboards.

Many domain experts accept it as the standard simply because it is familiar, equating its vibrant use of the full visual spectrum with scientific precision. However, as Crameri noted in his 2020 review, "Colour maps that visually distort data through uneven colour gradients or are unreadable to those with colour-vision deficiency remain prevalent in science."[1][2]

To correct this, visualization researchers advocate for a strict taxonomy of colormap applications. Sequential data, which progresses from low to high, requires a sequential colormap with monotonically increasing luminance, such as Viridis, Magma, or a simple light-to-dark single-hue ramp. Diverging data, which has a meaningful midpoint like zero or an average, requires a diverging colormap that uses two different hues that meet at a neutral, light color in the center. Categorical data, where there is no inherent order, is the only appropriate use case for distinct, contrasting hues.[2]

The transition to perceptually uniform colormaps represents a broader shift in how the scientific community treats data visualization. It is no longer viewed as an aesthetic afterthought, but as a core component of the analytical pipeline that is subject to the same rigorous standards as the underlying mathematics. A chart that misleads the human visual system is fundamentally no different from a statistical model that introduces bias. By replacing the rainbow with scales engineered for human perception, researchers are closing a critical vulnerability in the communication of quantitative evidence.[1][2]

The mathematical foundation of perceptually uniform colormaps relies on advanced color spaces, specifically the CIELAB or CAM02-UCS models. Unlike the standard Red-Green-Blue (RGB) color space, which is tailored to the hardware of computer monitors, these models are designed around human visual perception.

In the CAM02-UCS space, the Euclidean distance between two color coordinates directly corresponds to the perceived difference between those colors. By plotting a straight line through this space and mapping it to data values, designers ensure that a 10 percent change in the data always produces exactly a 10 percent change in perceived color.[3]

This mathematical rigor exposes exactly why the RGB-based rainbow fails. When the Jet colormap is plotted in a perceptual color space, its path is highly erratic, featuring sharp corners and varying speeds. The distance between the green and yellow segments is vastly larger than the distance between the blue and cyan segments. This uneven velocity means that the colormap artificially compresses some data ranges while expanding others, effectively applying a hidden, non-linear transformation to the data before it reaches the viewer's brain.[2][3]

The push for better defaults has also spawned specialized colormaps tailored to specific physiological needs. The 'Cividis' colormap, for example, was developed specifically to optimize viewing for individuals with color vision deficiency. It transitions from dark blue to yellow, avoiding the red and green channels entirely. Because it was generated by simulating how a colorblind viewer would perceive the Viridis scale and then optimizing the result, Cividis ensures that the visual experience is nearly identical for viewers with and without normal color vision.

The evidence against the rainbow colormap is now overwhelming, spanning cognitive psychology, medical diagnostics, and geosciences. Yet the debate highlights a persistent gap between visualization research and applied science. While the tools to create accurate, accessible charts are now freely available and set as defaults in modern libraries, the responsibility falls on the individual researcher or analyst to understand why these choices matter. The data does not speak for itself; it relies entirely on the visual encoding chosen to represent it.[1]