I’m a co-author on a new article out in Environment and Planning B: Urban Analytics and City Science titled “A Roundtable Discussion: Defining Urban Data Science” (download free PDF). It arises from a panel discussion I participated in at the 2019 AAG Annual Meeting in DC. Vanessa Frias-Martinez, Song Gao, Ate Poorthuis, and Wenfei Xu joined me on the panel, which was organized and moderated by Wei Kang, Taylor Oshan, and Levi Wolf. From the abstract:
I recently teamed up with an international group of public health researchers and spatial analysts to co-author an article, An Introduction to Software Tools, Data, and Services for Geospatial Analysis of Stroke Services, that has been accepted for publication at Frontiers in Neurology (download free PDF).
My new article, “Street Network Models and Measures for Every U.S. City, County, Urbanized Area, Census Tract, and Zillow-Defined Neighborhood” has been published in Urban Science. This paper reports results from a broader project that collected raw street network data from OpenStreetMap using the Python-based OSMnx software for every U.S. city and town, county, urbanized area, census tract, and Zillow-defined neighborhood boundary. It constructed nonplanar directed multigraphs for each and analyzed their structural and morphological characteristics.
The resulting public data repository contains over 110,000 processed, cleaned street network graphs (which in turn comprise over 55 million nodes and over 137 million edges) at various scales—comprehensively covering the entire U.S.—archived as reusable open-source GraphML files, node/edge lists, and ESRI shapefiles that can be immediately loaded and analyzed in standard tools such as ArcGIS, QGIS, NetworkX, graph-tool, igraph, or Gephi.
Containerization is the way of the future present. I’ve heard feedback from some folks over the past few months who would like to play around with OSMnx for street network analysis, transport modeling, and urban design—but can’t because they can’t install Python and its data science stack on their computers. Furthermore, it would be nice to have a consistent reference environment to deploy on AWS or elsewhere in the cloud.
So, I’ve created a docker image containing OSMnx, Jupyter, and the rest of the Python geospatial data science stack, available on docker hub alongside additional usage instructions. If you’re starting from scratch, you can get started in four simple steps:
Jobs, establishments, and other amenities tend to agglomerate and cluster in cities. To identify these agglomerations and explore their causes and effects, we often use spatial clustering algorithms. However, urban space cannot simply be traversed as-the-crow-flies: human mobility is network-constrained. To properly model agglomeration along a city’s street network, we must use network-based spatial clustering.
The code for this example can be found in this GitHub repo. We use OSMnx to download and assemble the street network for a small city. We also have a dataframe of points representing the locations of (fake) restaurants in this city. Our restaurants cluster into distinct districts, as many establishments and industries tend to do:
OSMnx is a Python package for quickly and easily downloading, modeling, analyzing, and visualizing street networks and other spatial data from OpenStreetMap. Here’s a quick round-up of recent updates to OSMnx. I’ll try to keep this up to date as a single reference source. A lot of new features have appeared in the past few months, and people have been asking about what’s new and what’s possible. You can:
OSMnx is a Python package for easily downloading and analyzing street networks anywhere in the world. Among other analyses, we can use it to explore street network orientation. That is, what are the bearings and spatial orientations of the streets in the network? All of the code for this example is in this GitHub notebook. First we use OSMnx to download the street network of Moraga, California, a small town in the hills just east of Berkeley:
OSMnx automatically calculates all of the streets’ bearings. Specifically it calculates the compass bearing from each directed edge’s origin node u to its destination node v. Now we can visualize these bearings, binned together as a histogram to get a sense of the relative frequency of the streets’ spatial orientations. Or better yet, we can project that histogram as a polar plot to match the compass bearings:
Check out the journal article about OSMnx.
We can measure and visualize how “important” a node or an edge is in a network by calculating its centrality. Lots of flavors of centrality exist in network science, including closeness, betweenness, degree, eigenvector, and PageRank. Closeness centrality measures the average shortest path between each node in the network and every other node: more central nodes are closer to all other nodes. We can calculate this easily with OSMnx, as seen in this GitHub demo. For example, here is the node closeness centrality for Piedmont, California:
My life in one chart:
This chart comes from Google’s Ngram viewer. In computational linguistics, an n-gram is a consecutive set of n items in order in some sequence of text. Here, we are comparing the frequency of the 1-gram “Geoff” to the 1-gram “Jeff” across Google’s corpus of books. As you can see, Jeff has been more popular than Geoff since, well, ever… leading to my endless problems trying to get people to spell and pronounce my name correctly. The good news is Jeff has recently taken a downturn. Score one for the Geoffs!
Check out the journal article about OSMnx.
OSMnx can now download street network elevation data for anywhere in the world. In one line of code it downloads the elevation in meters of each network node, and in one more line of code it can calculate every street (i.e., edge) grade. Here is the complete street network of San Francisco, California, with nodes colored according to their elevation:
