Parking Ticket Map from SFGate.com

The San Francisco Chronicle’s website, SFGate.com, has a nice map showing the top locations in San Francisco where parking citations are issued. The dataset includes individual locations where 100 or more citations were issued, so it’s a map of the single places you’re most likely to get ticketed (but note that it doesn’t include the full dataset, only the top 576 locations). They’ve created a map that uses the Google Maps API and they overlay their own custom markers that use graduated circles to represent the number of tickets issued at any given location. The size of the circle indicates how many tickets were issued, and each unique location has one circle centered on the location.

But there are a few problems with this visualization. The two most obvious things that stand out are the difficulty in understanding the density of many map markers all overlapping one another, which is seen in the north-east area of the city (downtown), and the second issue is the fact that the one huge marker at the southern edge of the city makes every other marker look tiny and unimportant. A single glance at this map would lead me to conclude that there must be more parking citations issued in the southern area of the city than in the other areas. But that’s the wrong conclusion to draw.

The problem with overlapping markers

One of the big issues that we see with maps that contain a lot of data points is that the kind of markers that are typically used in online maps start to become unreadable when you get dense areas of data. In this dataset there aren’t even that many data points (576 total), but the concentration downtown makes that area a jumble of markers.

You can tell that there are a lot of points in the area, but you can’t tell how many there are. And in this case, each marker doesn’t just represent a single point, the size of the circle also represents how many tickets were issued at that location, so ideally I would be able to look at this map and tell where the most citations are issued, but I simply have no way of knowing that.

The problem with relative sizing of individual markers

The second problem has to do with that large marker near the bottom of the map. This map leads to a confusing conclusion because every map marker, regardless of how close it is to other markers (or even if it overlaps others) is showing the value of a single location. This means that if there are 10 locations all within a single city block that each have 100 citations, and then there is a separate location elsewhere in the city that has 500 citations, that location with 500 citations will appear 5 times as large as any of the other locations, and you will have no way of knowing that within a single block there were actually 1,000 citations issued (making that block a far more likely area for receiving a parking ticket). What we really want to see is the total citations issued within a certain geographic radius, so we can view which areas have the most total citations, not just the single locations.

SpatialKey to the rescue with aggregated heatmaps

To try to better understand the underlying data, I decided to bring the same dataset into SpatialKey. The data on the SFGate website was loaded into their Google Maps application in JSON format, and to get it into SpatialKey I simply grabbed the JSON feed, opened it up in a text editor, and did a bit of find and replace to convert the data to CSV (the whole process took a few minutes to get the CSV ready for import). Then I imported the data using the SpatialKey CSV import feature (for more on this and other features, check out the feature videos).

Once I had the data imported I loaded up a new report with a map and here’s what I got:

San Francisco parking citations heatmap in SpatialKey (click to enlarge)

This map shows a heatmap that visualizes the total citations issued. But the important difference is that the clusters of data points downtown are aggregated by geography so items that are very close together are all factored into the hotspots. Now we’re able to see the real relationship between areas of the city. That point down in the southern part of the city is still visible, but it becomes clear that there are far more citations issued downtown. This deeper understanding is possible because we aren’t simply throwing a marker for each point up on the map, we’re aggregating the total value for all markers within a certain geographic area.

If we zoom in downtown we can see another view that shows the more specific hotspots:

Heatmap of parking citations in downtown San Francisco (click to enlarge)

Looks like they get people along Market street. Right at the Westfield Shopping Center is a prime spot, as well as the intersections of  Market and O’farrell and near Market and Sutter (if you’re parking there, look out!). You can see that out of the total parking citations in this dataset (82,911) about 42% (34,695) are issued just within the downtown area shown in the above screenshot.

I hope this example shows how important it is to be able to tell the right story with your data. SpatialKey gives you the flexibility to visualize your data in complex ways that go beyond simply throwing markers on a map. Have you run into similar problems with the current tools for web-based mapping? If so let us know in the comments and then sign up for the SpatialKey beta program!

Droughts, floods, storm events — all are visible in our visualization templates utilizing this free government data.  See, for example, the Great Flood of the Missouri and Mississippi Rivers in 1993 by visualizing streamflow (in cubic feet per second) measured at dozens of stations along these rivers.

Launch this visualization

The symbols on the map are sized proportional to the streamflow readings, allowing us to see the peak of the flooding in July and August.  This dataset shows off many unique aspects of the SpatialKey visualization system.  First, the templates excel at aggregating both spatially and temporally.  In the screenshot above, the St. Louis and Grafton stations are aggregated together, because they are so close and would overlap, producing illegibility and reducing clarity, if they were displayed separately.  The average of the two stations’ readings is symbolized; zoom in to reveal more detail, separating the St. Louis and Grafton stations’ symbols:

In addition to the spatial aggregation, this example also demonstrates temporal averaging.  In the above screenshot, each of the symbols represents and average of one month of daily readings.  Selecting different time ranges on the time chart instantly reveals the new average, and any range can be animated in the Playback template.

All of the SpatialKey visualization templates include extent-filtering, which updates the time chart to only display the temporal trend for points in the current map extent.  Seeing the streamflow trend for an individual station is therefore as easy as zooming into that station:

The dataset shown off in this example is different from the others in our gallery.  While the others demonstrated an ability to aggregate and visualize thousands of independent points across space and time (home sales, Wal-Mart store openings, arrests), the dataset presented here has only dozens of points, but includes thousands of streamflow readings (the dataset is over 12,000 rows long).  This “remote sensor” style of dataset is easily accomodated by our templates.  A third type of point dataset tracks assets (shipping containers, police cruisers) whose locations and attributes (speed, weight, etc.) may change over time.  We have developed solutions in the past for this latter type of dataset, and will show off asset tracking in future visualization templates.

Please comment here or contact us if you have any comments or questions.

This technology preview is not a formal release, and is designed to show what’s possible with Flex mapping and browser-based visualization, and to generate feedback on what we’ve done so far. For more information on this preview, please see spatialkey.com. If you’re anxious to see your own data, stay tuned for SpatialKey Personal, a soon-to-be-released version of our system that will allow you to map thousands of points of your own geotemporal data. If you’re interested in mapping larger datasets, say millions of points, or would like to speak to us about a custom visualization solution, please contact us.

This technology preview is all about examples, specifically dataset + visualization template pairings. These are meant to show off our advanced visualization and mapping tools, and to present potential use cases for this technology. Each template (there are currently four in this preview) allows you to look at a dataset in a particular way, while each remains general enough to accommodate most geotemporal datasets. Below are screenshots and links to the examples in the current preview; we’ll be posting more over the coming weeks. For more information on these examples, please see the Gallery page. To stay informed about SpatialKey, join our beta, subscribe to this blog, and/or check back often!

the radial spread of Wal-Mart

where prostitutes are arrested in San Antonio

construction slowdown in D.C.

Sacramento residential burglaries

the spread of a geo-social network: BrightKite

Wal-Mart stores, supercenters, and distribution centers

increasingly violent terrorist attacks on infrastructure

Sacramento sex crime arrests

a month of crime in Sacramento

housing slump in Sacramento

The growth of Wal-Mart provides a particularly compelling use case for the SpatialKey Animation Template.  Though most know of Wal-Mart’s southern provenance (Sam Walton opened the first Wal-Mart in Rogers, Arkansas, in 1962), the chain’s subsequent spread and resulting dominance over American retail is a more complex phenomenon.  Utilizing a dataset of over 3000 Wal-Mart store openings, from 1962 to 2005, our animation template shows the radial spread, increasing density, and overall coverage of the retail giant.

Launch this dataset in the animation template

The dataset used in our animation template comes from an economics paper, “Diffusion of Wal-Mart and Economies of Density” by Thomas J. Holmes, and is freely available here.  As Holmes notes,

Wal-Mart started in a relatively central spot in the country (near Bentonville, Arkansas) and store openings radiated from the inside out. Wal-Mart never jumped to some far off location to later fill in the area in  between. With the exception of store number one at the very beginning, Wal-Mart always placed new stores close to where they already had store density.

This pattern is clearly visible in our cumulative animation – a playback mode in which points accumulate over time on the map.  Switching the animation style to noncumulative allows you to select a decade at a time (or any time range), which can then be played over the timeline.  Highlighting only the store openings in the 1990s reveals a strategy aimed at the Northeast and California, with relatively few openings in the rest of the country.  Though this differs markedly from the early history of Wal-Mart, it is much more in line with the population centers of the United States, and reveals a company no longer rooted in the South.

To demonstrate some of the additional filtering and aggregating capabilities of our geovisualization toolset, the same Wal-Mart dataset can be visualized in our Drill Down template, which allows filtering by map extent, time, type, or list. Filtering by type allows you to show only the stores that have been converted to, or have always been, Supercenters. As noted by Holmes,

With this [Supercenter] format, Wal-Mart added a full-line grocery store alongside the general merchandise of a traditional Wal-Mart. Again, the diffusion of the Supercenter format began at the center and radiated from the inside out.

Thus, the spatial trend of Supercenter distributions appears about a decade behind the trend when all stores are included. Is this because Supercenters are largely made up of converted regular Wal-Marts? Or because shoppers in the South are more amenable to grocery shopping at Wal-Marts than shoppers in the rest of the country, where the presence of Wal-Mart is still somewhat novel? We’re not sure. But visualization is often as much about spurring questions as answering them. This dataset has been mapped before (see here and here), but doing so in a flexible geovisualization environment reveals interesting geographic and temporal patterns, only a few of which have been explored in this post.