While it’s possible for hand-made data visualization to be distorted (hopefully with intention) it’s not inherent to the medium—precise, meticulously drawn maps, charts, and graphs are quite doable, and were the norm before the arrival of computer-aided-design and drafting and R.
Consider these USGS 1:250,000-scale topographic maps of Yosemite Valley. The maps accurately represent the iconic topography, within the precision of the measurements they were based on.
They were originally created (as seen in the 1948 and 1963 maps) by painstakingly drafting contours from individual survey measurements, eventually aided by aerial photography. A method that was supplanted by an equally painstaking—but more precise—technique of tracing contours from pairs of aerial photographs, offset just enough to give a stereo (3D) perspective. “Planimetric detail partially revised by photo-planimetric methods” in the 1948 & 1963 versions, vs. “compiled in 1961 by photogrammetric methods” in the 1966 and 1971 editions. (And no, I don’t know why the dates don’t match.)
Also note the shifting use of background shading—I’m sure the evolution of USGS topographic maps is probably worth a blog post (PhD thesis?) on its own.
Although maps may be the most prevalent form of precision pre-computer visualization (they have to be correct to get their users from A to B, after all) they’re by no means the only rigorous hand-crafted graphics.
Before email and comma-separated variable (CSV) files made sharing data almost trivial, a graph was used to accurately transfer quantitative data as often as it was used to show patterns and trends. Researchers often read values off a graph in the pages of a scientific journal, often with the aid of a ruler or grid lines (graph paper wasn’t invented for D & D, after all).
This is illustrated by a plot of spring ozone concentrations above Antarctica from 1957 to 1984, which demonstrated the existence of the ozone hole for the first time. Dense gridlines—hardly chart junk—allow exact values (and uncertainties, in this case) to be read directly off the graph. Compare the draft plot (left) with a later version that appeared in the journal Nature (right). Interestingly, the published version features a double-axis plot (with the second axis inverted!) to show correlation between ozone levels and chlorofluorocarbons (F11 and F12).
Precision isn’t just useful for sharing quantitative data—it’s also effective at communicating qualitative data. Like the wonderful variety of bugs endemic to New Zealand.
Naturalists have long used paintings and sketches to describe and catalog Earth’s flora and fauna. A necessity before the invention of photography, and still a vital tool today. These drawings of New Zealand’s insects are from a collection by scientific illustrator Des Helmore, recently released into the public domain by Manaaki Whenua — Landcare Research.
Rather than try to describe Helmore’s illustrations, I’ll quote the introduction to Drawings of New Zealand insects:
The main purpose of these drawings is to provide an aid for identifying insects, the prime requirements being to show shape, proportions, pattern, and the different parts of the insect so clearly and accurately that it can be identified from the drawing. For this reason it is necessary to look at the insect as objectively as possible. There is no room for personal expression of how one feels about it, although empathy or a feeling for its form is necessary within the limits of accuracy to make the drawing convincing and life-like, and raise it above the impersonality of the diagram. This “feeling” need not necessarily be cultivated, as some involvement is inevitable. It is as if one visually feels the object as one draws it, the hand following the eye, so that one has the sensation of recreating it.
While I’m sympathetic to the view that “There is no room for personal expression” in visualization (I don’t consider my satellite image processing work art) Helmore’s own illustrations, simultaneously highly stylized and exacting, bely that notion. The line work and stipling (click the links for high-res versions) are clearly aesthetic choices, informed by the printing technology of the 1970s and ’80s. But they’re not purely aesthetic choices—they’re also in service of Helmore’s goal—communication:
Although the drawings may look completely realistic they are not attempts to copy the appearance of the insect. They are really only abstractions, in that only selected features from the mass of detail which confronts the eye are depicted, this selection being based on what is needed for identification. Sometimes it is necessary to disregard surface features to show structures that may be only partially visible. When the selected features are converted into the absolute black and white of an ink drawing the drawing can look very different from the actual appearance of the insect. This is because drawing an insect is a form of communication like drawing a map, where various marks and symbols are used to represent the selected features. Line, for example, is a “symbol” used to represent the edge of a form. Edges exist in nature, but lines do not. It is impossible to show everything within the limitations of the drawing medium, just as it is impossible to make a complete verbal or written description.
An earlier example of natural science illustration is found in John Chappelsmith’s Account of a Tornado near New Harmony, Ind., April 30, 1852. The report’s maps and drawings make a crucial additional step—they bridge the gap from cataloging the natural world to testing a scientific hypothesis.
The drawing above shows trees downed by the passage of a strong tornado near the border between Indiana and Illinois (a not infrequent occurrence in the Midwestern U.S.), along with a detailed diagram showing the orientation of each fallen tree he had depicted. The drawing accompanies a regional map of the storm’s damage — an attempt to solve a major dispute in 19th-Century meteorology. Do the winds of tornadoes and hurricanes spiral around a central axis, or rush in towards a column of rising air?
What is well known today—with the structure of tornadoes monitored by video, radar, satellites, and even in-situ instrumentation—was a gigantic mystery in the 19th Century. A tornado could only be safely studied from afar, or after it had passed. A handful of gentleman scientists used scarce data collected from widely scattered storms to lend support to their conflicting theories.
To address this data deficit, Chapplesmith plotted the location and orientation of thousands of downed trees.
Knowing the tendency which exists in most minds to see chiefly those facts which favor a preconceived hypothesis, it seems to me that to select a few groups of trees out of thousands, would not afford sufficient evidence to others, however satisfied one might be with the truthfulness of his illustration. I therefore present the plot of a square mile of track on which some 7 or 8,000 prostrated trees are represented in their relative positions, with the hope that the means will thus be furnished for more satisfactorily determining whether the “immediate mechanical cause of devastation in tornadoes” be a spirally involuted rotating moving column of air, or a vertical current at the centre of the tornado with a horizontal conflux from surrounding space to the moving access.
The diagram at the bottom of the map illustrates the patterns that would be expected by each type of storm “to more readily compare the phenomena with the hypotheses”.
For a “spirally involuted rotating moving column”, trees downed in each quadrant would be facing a different direction, with the most damage occuring where the spiraling winds were aligned with the motion of the storm. In the case of a “horizontal conflux from surrounding space to the moving center” downed trees on the margins of the storm’s path would have been blown towards the storm, while trees near the center would be aligned with the storm’s motion.
Despite his considerable effort, Chappelsmith’s interpretation of his own data turned out to be a bit off base. He believed the evidence showed that “tornadoes are an inward, upward, and onward moving column of air”, rather than a spiral. It turns out that the real world is messy, and tornadoes consist of both spiraling winds (driven by the Coriolis effect) and a central column of rising air (driven by energy released as water vapor condenses at altitude).
100 years later, another American scientist and keen observer of natural phenomena would help unravel the role of water vapor in weather and climate.
Joanne Simpson—the first woman to receive a Ph.D. in meteorology—had a long and distinguished career with the University of Chicago, the Woods Hole Oceanographic Institution, UCLA, NOAA, the University of Virginia, and NASA. She’s known for unraveling the driving forces behind tropical cloud formation and studying the intensification of hurricanes.
In the pre-satellite era of the 1950s, Simpson needed to go into the field to gather data to validate the models. In this case the “field” was the pristine atmosphere above the vast stretches of the Pacific Ocean. Then named Joanne Malkus, she spent the summer of 1957 filming cloud patterns during a series of flights in a Navy flying boat.
With the data collected during the expedition, she carefully sketched a series of maps and diagrams showing the type, distribution, and structure of clouds. These drawings turned raw data (hours of film and extensive meteorological measurements) into evidence to show how the energy released by the condensation of water vapor during cloud formation drives larger circulation patterns in the atmosphere. It was an essential act of interpretation, relying on Simpson’s domain expertise and years of work.
My final example is a plot of the path of Apollo 16 from Earth to the Moon and back—one of a series of “Trajectory Plotting Charts” made for each of the Apollo lunar missions. These charts embody precision; from the arcing trajectories to the daily lunar phases to the generous use of the razor-sharp Futura typeface. I love details like the silhouette of the Moon’s position at key moments of the flight, and the sun’s position at the beginning of the lunar month.
Unfortunately, I know nothing else about them, other than that they’re a popular item in auctions of space memorabilia. If you do know something—please leave a comment?
A common thread between each of these visualizations is the sheer amount of work that went into each of them. The painstaking effort of transforming a dataset into a graphic by hand grants a perspective on the data that may be hindered by a computer intermediary. It’s not a guarantee of accurate interpretation (see Chapplesmith’s flawed conclusions), but it forces an intimate examination of the evidence. Something that’s worth remembering in this age of machine learning and button-press visualization.
Enough abstraction. My next post in this series will focus on physical data visualization.