Dashed and patterned lines for visualization (aka 1D texture)

Bertin and Texture

Jacques Bertin discusses texture in visualization, in his landmark book Semiology of Graphics (1967 original French edition, 1983 English translation). Much of modern visualization theory and implementation hinges on Bertin’s framework, including the notion of visual attributes, marks and layouts (for example, Wilkinson’s Grammar of Graphics, or Bostock’s D3).

Bertin talks about texture as a visual attribute and shows quite a few examples of texture applied to points, lines and areas. Bertin’s work exists at the heyday of Letraset – transfer film used by graphic designers to create professional graphics. Letraset provided professional fonts, graphics and textures that could be applied by any designer to create professional graphics for use in glossy magazine ads, to black-and-white screen graphics in newspapers, down to small scale zines – it was the democratization of desktop publishing before computation desktop publishing appeared in the mid 1980’s. Letraset and their competitors offered a wide variety of ready-made textures:

Samples from 300+ ready-made textures from Letraset (1,2,3,4).

And these kinds of textures appear in Bertin’s lines and maps and substantially influenced his work. Note how textures can be combined together to represent multiple variables:

Bertin uses textures in many visualizations, to represent quantitative values (left), or layer multiple categories (center and right).

But what is texture? What are the variables that make up texture? What are the parameters for use? Did Bertin articulate all the possibilities, or are there more?

This is too big of a question for a single blog post, so instead, consider a very narrow use of texture as applied in only one dimension – that is – a texture applied to a line. Texture constrained to a line is very limited, particularly if the line is understood to have no width (at least for this blog post).

Dash Patterns: Length, Gaps and Rhythm

Like Bertin, we can think of texture as a sequence of on/off values applied to a line to create a sequence of dashes. The regularity of the sequence allows the viewer to distinguish between a line with a short dash pattern with big gaps and or long dash pattern with small gaps, such as A50 and F90 in the image below. The wide range in variation indicates that these lines could be used to show quite a few different categories within one visualization. Also, both the dash length and the the proportion of ink could be utilized to represent quantitative values associated with the line, e.g. less ink for lower contour lines, more ink for higher contour lines; short dashes for low certainty, long dashes for high certainty, etc.

Dashes varying in length, gap proportion, rhythm, and randomness. Link to code using D3 and dasharray.

In the bottom half of the image, other variants are considered. Rhythm can vary: on the left, the lines have a consistent dash-gap rhythm ABABAB, as seen in line G0. In the center column, a dot is added into the gap: the line now has the rhythm ABCBABCB, such as G1. It has the same ink as its counterpart in the 0 column, and is clearly differentiated. The right column has two dots (rhythm ABCDCBABCDCBA – which is clearly easier to describe as a visual line shown in G2 rather than as an alphabetic sequence). Lines with none, one, or two dots are clearly differentiated, allowing for many more possible combinations dash sequences to be used.

Randomizing Lengths and Gaps

In the bottom half of the image, within each column the regularity of the line pattern is perturbed. That is, with each successive line a bit more randomness is added. In G0 the dash lengths and gaps are consistent, in H0 the difference is almost unnoticeable, and by K0 there is some noticeable differences in dash lengths and gaps. G0 is the same line as E70. The K0 line retains the highest similarity to line E70, even though dash lengths and gaps have been modified more heavily. The ability to make minor adjustments to dash lengths and gaps while retaining the same line identity is a property exploited by old mapmakers – essentially it is preferable for the dashed line to be solid at a corner, not a gap:

Dashed lines are solid at corners and gap over other lines.

As seen on this Ordnance Survey map from 1900, the dashes are solid at corners. The draftsman (drafts-person) has made minor adjustments to the intervals to retain the dash style while making the point of change in line direction visually explicit. Another interesting adjustment is that the gaps in the dashed lines occur when a dashed line crosses another line, as can be seen on the dotted line crossing lines in the far left and far right in the above thumbnails. This facilitates visual separation of the lines, not allowing them to be become an intersecting blob.

These kinds of adjustments are critical to making dashed lines effective in visualization. Consider a timeseries chart using a dashed line. If the gap occurs at a high point or low point, then it is impossible to determine the value from visually reading the chart, significantly reducing the effectiveness of the chart:

Left chart from 1912 shows dashed line with every high point and every low point solid. Right chart from Excel some high points and low points are gaps making it impossible to visually determine the value.

In the left chart, the drafts-person has a solid dash at every corner, whereas the Excel chart on the right has arbitrary dash lengths and gaps leading to corners not being explicitly visible. Since the corners on a line chart represent the actual data points, the arbitrary rendering results in data points not being visible! Yes, it is feasible in Excel to also plot the data points (as boxes, circles, stars, etc), but that only clutters the chart and breaks the rhythm of the line style — the drafts-people didn’t need to add extra marks for each point.

Given that a data visualization is supposed to show data, it is strange that the line style pattern overrides the visibility of the datapoints. Therefore, implementing effective dashed lines for visualizations isn’t as simple as defining a dasharray in SVG or D3. A lower-level dashed line module would be needed to make sure solid dashes occur at corners. This is non-trivial: what if there are multiple sharp angles in a short sequence (e.g. on the 18th day in the 1912 chart above), or multiple line crossings?

Showing Rhythm with Transparency, Saturation, Brightness or Hue

In Bertin’s example, texture is binary: it’s on or off. Much more is possible using computer graphics as opposed to transfer film. Transparency, saturation, brightness or hue can be modified with the regularity of the dash sequence:

Texture patterns represented with varying gradients and color. Link.

In the top block, L50 – Q90, gradient transparency varies from solid to transparent to create the same pattern as previously seen in the dashes on lines A50 – F90. The lines in the 50 column use a linear gradient that goes smoothly from min to max and back to min, whereas in column 70 the gradient drops off more quickly than it rises, and in column 90, the gradients have a sharp drop off. Interestingly, the non-equal application of the gradient appears to give the lines the illusion of motion, like motion blur in photographs.

In the bottom block, Rs – Wh, the same patterns are used with attributes of color. In the first column, only saturation is changed (from orange to grey) which is difficult to perceive as there is no contrast between the two colors. In the second column, only brightness is changed, with the alternating pattern of light orange, dark orange clearly visible. And the third column changes hue, resulting in a repeating rainbows along the lines. Again, attempting to use SVG and D3 to implement these is difficult and the approach used in my quick code won’t work with curving lines.

More importantly, these examples suggest that texture is operating at a different level than visual attributes such as length, transparency, hue and so forth; as the texture can be represented in any these other variables. So Bertin’s notion of texture and how texture fits into the formal definitions of visual attributes of data visualization may be more nuanced than the current models used in visualization research.

Conjunction of Textures

Bertin uses multiple overlapping textures to convey multiple variables in his maps: textures of small dense dots, with large coarse dots, with lines oriented in one direction, and lines oriented in a different direction. Can multiple dash patterns be combined on one-dimensional lines? Yes:


In the top row, the line X1 has a very short pattern with a lot of white space is shown. Then, on the left, a line Y2 also has a short pattern and an interval that is a multiple of the interval for X1. The two patterns can be interleaved form the line XY12. The right image from Bertin back at the top of this article uses this interleaving appproach: it uses carefully designed textures such that dots fit neatly between the diagonal lines.

On the right, the first line X1 is the same short dash pattern, while the second line Y7 has a long dash pattern: there is no way to combine the two patterns without intersection. Here the patterns are combined to alternate on/off to create a reverse video effect (i.e. the patterns are XORed together).

This kind of approach might be desirable for charts with many different dimensions, for example, a census chart plotting unemployment overall/minorities, and for all ages/and those under age 25. Thus lines can be overall solid, minorities short dash; all ages solid (same data as overall), under 25 long dash, and the conjunction of minorities under 25 in the conjunction of the two dash patterns.

So What?

If dash patterns are problematic, then why use them at all? Sometimes there may be a need to use many lines, more than can be comfortably differentiated using color. Dashed lines are also common on line charts to show predicted data, or on maps to show unpaved paths: semantically a dashed line can effectively convey the data has uncertainty. Dashed lines have real uses.

Texture is an under-explored area of data visualization. Historical charts and maps do show that these can be effectively used. Visualization tools and libraries, however, use dashes arbitrarily and don’t take into account how to draw dashes to suit perceptual needs. Furthermore, our definition of the design space around texture may be somewhat lacking – perhaps some future grad student will want to take on some of these issues.

About richardbrath

Richard is a long time visualization designer and researcher. Professionally, I am one of the partners of Uncharted Software Inc. I have recently completed a PhD in data visualization at LSBU. The opinions on this blog are related to my personal interests in data visualization, particularly around research interests related to my PhD work- this blog is about exploratory aspects of data visualization not proven principles.
This entry was posted in Data Visualization, texture. Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s