Let’s make more dot grid maps
I am excited about dot grid maps. And I’m surprised that we don’t see more of them. Dot grids are a clear, informative, multidimensional and flexible cartographic technique. They effectively leverage patterns of human perception to present information dense but readily comprehensible maps. Compared to choropleth maps, dots retain the base map context, and invite us to fill in the gaps. They emphasize the limits of data sampling. Dot grids can be joined together across different boundaries flexibly. The density of a dot grid can be varied depending on the scale. And that visual regularity .. it just looks so cool.
Here’s a quick survey of how we have been using dot grids at Earth Genome, and usage of dot grid maps out in the wilds, along with notes on why I think they’re so effective. Also I pulled on the threads to learn about the origins of dot grid maps over 50 years ago with the foundational work of Jacques Bertin, and further back to Gestalt psychology.
Dot grids at Earth Genome
The cartographic challenge to launch the map of individual emissions sources on Climate TRACE was high, as emissions come from so many different kinds of sources with very different geographic representation. Vehicles and ships move, agriculture covers large areas, and oil processing facilities are at a specific location. The visual interaction of tens of thousands of lines, polygons and points together on a global map can get messy.
To represent 70,000 sources, we chose to rely on points at the global level, and reveal more detail of a fewer number of features at higher zooms. Several land use emissions, like cropland fires, are calculated using gridded data. Gridded data is typically shown with a square grid completely covering the map area with a color representing a category and/or value. For Climate TRACE, at low zooms, we use the centroid of each grid square to plot a point, and scale the point by the volume of emissions. The result is a dot grid that works well with other kinds of sources, and keeps the base map context fully in view.
The tracks of emitting ships presented a similar problem. We first tried connecting the tracks of individual ships by lines, which ended up visually dominating the map, filling in large parts of the oceans. Taking ship tracks aligned to a global grid, we plotted centroids of the grid squares and scaled opacity by the total emissions that could be attributed to ships passing through that point. The result gives an impression of major shipping routes and their relative contribution to emissions, in a visually complementary way to other kinds of emissions.
Our work on food systems and climate has led us to put particular focus on food security. Food security predictions and actions operate on relatively short term cycles, but involve similar analysis of data, identification of trends and modeling of disruptions that can occur on the longer timeline of climate change. We are working with data from the Famine Early Warning Systems Network, and looked to dot grids rather than choropleth maps to bring multidimensionality to the data.
In the demo, we took regional forecasts of food insecurity, and derived a dot grid. The colors are assigned according to food security risk, with red being highest. The size of the dots are scaled by the size of population estimates at that point. This multidimensionality gives some sense of where potential crises are concentrated.
Earth Index is our platform to organize satellite imagery for search. The fundamental unit of analysis in Earth Index is a grid of satellite image tiles. In the process of search, you select tiles that match, or don’t match, the kind of features you’re trying to locate. In the results above for a search on concentrated animal feeding operations (or CAFOs), the dot grid indicates degree of confidence in a positive match, with red the strongest match, yellow some uncertainty, and white no match. In this application of dot grids, the satellite base map is critical to see, and the dot grid gives strong but unobtrusive signal on where to focus attention on the search results.
Dot grids in the wild
Our efforts have been inspired by amazing dot grid maps out in the wild.
eBird recently refreshed Trend Maps tracking where and how bird populations change through time. They used dot grids to illustrate trends in abundance of observations. The size of each observation point in the grid scales with abundance of bird sightings, while the range of color indicates the change in abundance from 2007, with red indicating a decrease and blue an increase. For the red-tailed hawk, immediately areas of larger populations are visible in denser regions of the Northwest and Canada, which also show a strong increasing trend. The eastern half of the country shows a broad decrease in abundance, with only a few areas of impacted denser populations. Areas of concern jump out as dense, redder “edges” of high population regions.
In contrast, eBird uses square grids to map only the abundance of birds. Square grids are the most common way to visualize gridded data on maps. I still think this is an effective map, but comes at the price of multi-dimensionality. In order to uncover abundance in different seasons, layers are toggled off and on. Visually, the grid reduces the resolution of underlying natural features like coastlines and lakes. Ultimately, my subjective impression is that the dot grids feel more “live”, representing active data, while the square grids feel heavy and static.
Perhaps the most widely seen example of a “dot grid”-like map comes from alternate U.S. presidential election maps. Visualizing each county as a solid polygon colored according to proportion of parties votes visually overweights proportions going to Republicans, which tend to hold lower population areas. The map above takes the center point of counties, and colors by vote and scales by population, giving a more accurate picture of two dimensions.
This map on climate and conflict takes a slightly different approach, with three dot grid layers each representing a different variable slightly offset from each. Without toggling anything, the eye can look at one variable separately, or narrow in on when there’s a convergence of factors.
Direct Relief just blogged about ReadyMapper, where population change source from mobility data can be examined alongside emergency events like fires. In the map above, the colors of the dot grid represents change in population due to the Caldor Fire, with red indicating a decrease in population. Overlaid is a county boundary (El Dorado) and the boundary of the fire. The dot grid presents high fidelity of information, easily visually separated and comparable to various polygons.
The origins of dot grids
Why are dot grid maps so effective, yet not widespread? Going down this rabbit hole unearthed a whole system of thought about what makes effective maps. Turns out dot grids are a relatively recent cartographic invention, originating with the work of Jacques Gertin in 1967 with the Semiology of Graphics. Gertin’s approach was to synthesize fundamental principles and apply logical rules, to systematically explain and develop graphical techniques.
One result of this is Bertin’s Visual Variables, described by Axis Maps as “important as a step toward knowing how those different types of maps work, and knowing how your choice of symbols implies or doesn’t imply patterns, groups, order, and quantity.”
Visionscarto frames this approach as a “change of paradigm .. until now, “good representation” essentially had to be precise and complete. With Bertin, the priority becomes effective communication”. The key criteria of visual representation become “efficiency”: “to obtain a correct and complete answer to a given question, and, all things being equal, a construction requires a shorter observation time than another construction, we will say that it is more efficient for this question”. This metric of mental efficiency resonates strongly.
And while not a direct influence on Bertin, this article by Elena Kazakova draws linkage from Bertin to Gestalt principles, a system developed by psychologists in the 1920s to describe elements of perception. This approach of deriving fundamental “laws” of perception like proximity and similarity again suggests a way to assess efficiency of visual processing of information.
And to bring it back to the start, I love how Wikipedia illustrates the Law of Similarity with a dot grid.
So I have more to learn and maps to make. I hope dot grids inspire you too, and I’d love to hear about anything built using them!
