Spectral Schemes: Controversial Color Use on Maps.

ABSTRACT. Cartographers have long discouraged the use of spectral, or rainbow, color schemes on thematic maps of quantitative geographic data, though such color use is common in GIS and scientific visualization. Recent research, however, has shown that spectral schemes are preferred and are interpreted accurately when used as multi-hue renditions of diverging schemes. Both spectral and diverging schemes can emphasize a critical point within a data range with light colors and emphasize both high and low extremes of the data with dark colors. Although spectral schemes include multiple saturated hues, they can be designed to accommodate map reading by people with red-green impaired color vision by skipping over the yellow-greens in the spectral sequence. Cartographers should encourage use of spectral color schemes for depicting diverging quantitative data, rather than insisting that these schemes should not be used.

KEYWORDS: cartography, map design, spectral color schemes, color vision

Introduction

Spectral color schemes are common on rainbow weather maps in the daily news, in scientific visualizations, and in GIS-based mapping. Yet, cartographers continue to discourage their use for representing quantitative geographic data on thematic maps. Spectral schemes are preferred and interpreted accurately when used as multi-hue renditions of diverging schemes (Brewer et al. 1997), and careful variation in both hue and lightness within spectral schemes aids map reading. Like a diverging scheme, spectral schemes can be designed with light colors emphasizing a critical point within a data range and dark colors emphasizing both high and low extremes of the data. The multiple saturated hues of spectral schemes may confuse people with red-green color-vision impairments but, again, careful design permits spectral schemes to accommodate these map readers. Cartographers should encourage use of well designed spectral color schemes for thematic maps of diverging data, rather than generally discouraging use because of inappropriate applications of spectral schemes to sequential data.

This paper begins with a critique of the assumption that spectral schemes are illogical. Examples of diverging data and spectral color use are presented from mapping in cartography, geography, and scientific journals. Results from two research projects in which spectral schemes led to good map reading performance by people with both normal and impaired color vision are described. Finally, the suggestion that spectral schemes do not accommodate people with impaired color vision is addressed with a recommended adjustment to the spectral sequence.

Literature Review

Bertin's writings (1981; 1983) are influential in shaping our discipline's approach to data representation. His opinions about the use of color hue are cast in a strident tone: "I am indeed against color when it masks incompetence; ... when people believe it capable of representing ordered data" (Bertin 1981, p. 222). The following sample of recent quotes about spectral schemes by cartographers and other authors echo and elaborate Bertin's opinions. Rarely do we find this degree of consensus or emotion expressed about a subjective challenge of symbolization. We have been recommending that spectral schemes should not be used to represent ordered data. These claims have established a conventional wisdom that spectral schemes are illogical and inappropriate for the representation of quantitative data. My own earlier writings on the issue are no exception, as can be seen in the following list of statements:

   Do not use the saturated spectrum as a sequential scheme. [Brewer 1994a, p.
   138] There appears to be general agreement among cartographers that the
   color dimension of value [lightness] be used to symbolize an ordered array
   of data magnitudes ... Hue differences alone should not be used, but if
   they are, part-spectral schemes are preferred over the full-spectral ones.
   [Dent 1996, p. 306; cites Cuff 1973]

   A common objection by cartographers to maps of quantities produced by
   noncartographers is that these maps often ignore the importance of the
   linear order schema and employ a set of eye-catching (but randomly ordered)
   hues. Sometimes the hues are ordered, but according to wavelength of hue.
   Wavelength ordering is not immediately recognized by our visual system, and
   therefore is unlikely to prompt the appropriate linear order schema on the
   part of the viewer. [MacEachren 1995, p. 188]

   Hue differences usually fail at portraying differences in percentages,
   rates, median values, and other intensity measures because spectral hues
   have no logical ordering in the mind's eye ... there is no simple, readily
   remembered and easily used sequence of hues that would obviate a map
   reader's needing to refer back and forth repeatedly between map and key ...
   The use of spectral hues to portray intensity differences is a strong clue
   that the mapmaker either knows little about map design or cares little
   about the map user. ... most users will find [a] full spectral scale of
   primary hues confusing, complex, and comparatively difficult to decode.
   [Monmonier 1991, pp. 150 and 152]

   The spectral progression on relief maps is a convention of long standing
   and thus is well known through regular appearance on school maps and atlas
   maps for the general public. Except for its familiarity, it is graphically
   illogical with little to recommend it ... [Robinson et al. 1984, p. 186]

   In a later edition of the text, an added advantage of spectral schemes is
   recognized but they are still dismissed, albeit with softened language:

   Except for its familiarity and map-legend matching advantage, there is
   little reason to use spectral progressions for quantitative data.
   Communicating the geographic pattern of magnitude variation generally is a
   more important objective, and a spectral hue progression does not
   inherently carry a magnitude message. [Robinson et al. 1995, p. 389]

   Others have commented as follows:

   ... there is no logical sequence for ordering colors that differ by very
   large steps in color space ... It is erroneous to assume that we have some
   hard-wired intuitions for a spectral sequence ... If this were true, school
   children would not find it necessary to learn mneumonics such as "Richard
   Of York Gains Battles In Vain." Such mneumonics are not required to rank
   colors in saturation or brightness, or when small steps in hue are
   considered ... [Travis 1991, p. 126 (Travis's bolds and italics)]

   If the categories are ranked or naturally ordered, assign colors in order
   of lightness or value, not by hue or wavelength. A color spectrum of pure
   hues contains two scales of value on either side of yellow ... For ordered
   concepts, never mix the two scales. Value, even within colors, is a
   stronger ordering force than hue. [Horton 1991, p. 232 (cites Bertin 1983)]

   A widely used alternative is a scale of rainbow colors, replacing the clear
   visual sequence of light to dark with the disorderly red, orange, yellow,
   green, blue, indigo, and violet--an encoding that now and then reduces
   perplexed viewers to mumbling color names and the numbers they represent
   ... Despite our experiences with the spectrum in science textbooks and
   rainbows, the mind's eye does not readily give order to ROYGBIV. [Tufie
   1990, p. 92]

Kumler and Groop (1990) tested spectral and part-spectral schemes in the context of evaluating continuous-tone representations of smooth and continuous surfaces. In contrast to the above quotes, they found that subjects scored better with spectral schemes than part-spectral (both with isarithms) and that 73 percent of subjects preferred spectral schemes to the other representations used in their test.

Graphic Critiques of Spectral Schemes

I have seen three similar and compelling graphic demonstrations by Bertin (1981), MacEachren (1994), and Livingstone (interviewed by Grady 1993, p. 63) of how spectral schemes misrepresent distributions. Livingstone presents a photo of Eisenhower's face. With a sequential grayscale representation, Eisenhower is immediately recognizable. The same photo is shown with hues that are ordered sequentially from light to dark: yellow, orange, green, blue, and dark purple. The face is still recognized as familiar, though strangely colorful. With a full spectral scheme applied to the `data' (dark red, orange, yellow, green, blue, dark purple), the image becomes almost unrecognizable as a face, and-certainly not as a familiar one.

Figure 1 presents an example similar to Livingstone's with a shaded relief representation of a readily recognized region of faulted terrain in the southwest U.S. Figure 1a is a standard grayscale shaded relief representation of the area. Figure 1b presents spectral colors in lightness order that produces a strangely colored but recognizable relief. In contrast, Figure 1c presents a diverging spectral scheme that makes the terrain difficult to interpret; larger structures can be seen but the river valleys of the Sierra Nevada, for example, are completely masked. [See Moellering and Kimerling (1990), Brewer and Marlow (1993), and the cover of MacEachren and Taylor (1994) for examples of effective spectral colors in terrain mapping for slope-aspect representation.

[Figure 1 ILLUSTRATION OMITTED]

The Livingstone and Figure 1 demonstrations are effective because it is only appropriate to represent the original data (high to low reflectance from a face or landforms) with an ordered lightness sequence to understand the `distribution.' Emphasis on mid-range reflectance with light yellows is not suitable because this is not an important range in the data (unlike a median mortality rate, Figure 2). These reflectance data are suited only to a sequential scheme, and an application of a diverging spectral scheme makes them difficult to interpret.

…