Something Spatial: Visualizing Density

No doubt, the technology needed to generate maps is becoming increasingly more user-friendly through an assortment of web-based map-making technologies like Tile Mill, Leaflet, Google Maps, Google Fusion Tables, and CartoDB. With these technologies, anyone can upload geo-referenced data and share interactive maps with friends, co-workers and the public.

For some, especially cartographers, this new trend is stepping on some toes within traditional academia. Eric Steiner – of Penn State fame – makes a salient observation about this phenomenon:

“Traditional cartographers today might say some form of, ‘Kids these days, they don’t know the rules,'” says Eric Steiner, a former president of the North American Cartographic Information Society. “I hear that sometimes at conferences. People lament that there’s this huge influx of people doing cartography who aren’t cartographers.” By “cartographer,” they mean someone who is skilled in trade techniques like projection (transforming a globe into a flat map) or who knows how to interpret line weights. Instead, new cartographers are increasingly software engineers or developers using programming languages like JavaScript and Python. Steiner, himself a graduate of Penn State’s prestigious cartography program, sees the plurality of technique as beneficial. Whether a map is good or bad shouldn’t be based on the narrative of the individual making the map, he says, but rather on the map’s ability to evoke, inspire and question.

Despite these new possibilities, there are still some limitations to how far one can get with these services. Although they provide the most basic answers in spatial analysis – representation of location – there is still some difficulty in answering more interesting spatial questions, like determining the extent to which certain spatial phenomena cluster. Being able to represent clustering – and the extent to which it is occurring – can enhance the power of the “citizen geographer.”

Map Box offers a tutorial that suggests a method for visualizing clusters. With the open-source (read: free) QGIS software, users can generate “heat maps,” or kernel density maps. In this case, QGIS can generate a special raster file (called “.GeoTIFF”) optimized for web, which can then be used in beginner-friendly software like Tile Mill. However, the author suggests an even simpler method, which uses opaque circles around data points:

The effect is less dramatic, although it suggests a simple way to conduct cluster analysis.

The geniuses at the Politecnico di Milano use a similar method for visualizing clustering of restaurants:


Here, clustering is captured merely by generating buffers around points in space, and setting opacities to those buffers. This is not only visually pleasing, it allows for some sense of clustering.

Using data from my own research, I used ArcGIS’s kernel density spatial analyst tool to generate rasters of how banks in Cambridge cluster:

Screen Shot 2013-08-10 at 6.36.59 PM

Clearly, this approach to cluster analysis is not as accessible as Tile Mill, but it is interesting when comparing this method to the method suggested by Tile Mill and the Politecnico di Milano:

Screen Shot 2013-08-10 at 6.39.12 PM Screen Shot 2013-08-10 at 6.43.38 PM

These two images show the workflow for this approach to cluster analysis. In the first image, I used Arc to generate buffers. Then I exported those buffers (with their associated points) into Adobe Illustrator, where I was able to manipulate opacity and color.

The way Arc exports into Illustrator is not always intuitive. Be aware that Arc tends to group layers oddly, and in order to manipulate the vectors, you must select the layer group, expand the group tree completely, and uncheck the eyeball icon (visibility). This will select the vectors and allow you change formatting.

Compared to the kernel density analysis, the second method is obviously weaker in its accuracy. For serious insight into clustering, it is necessary to use algorithmic techniques like kernel density. However, for the purpose of “citizen geography,” the second method more easily captures a sense of how things cluster visually. They both seem to suggest three high-density clusters, with surrounding clusters of weaker density.