Category: Geospatial Projects

OpenStreetMap and the OSMnx add-on for viewing streets as a network can be used to find the steepest streets in a neighborhood. Loosely based on their tutorial, we find the steepest streets in my home San Francisco Bay Area suburb of Millbrae, make a map of street grades that turned out to be useful for walking the area, and show views looking up the streets via the Google Street View API. The same code can be applied to any named area present in the well-known Nominatim geocoding tool.

• GitHub

Why the steepest streets?

The main motivations for finding the steepest streets in Millbrae are

• Curiosity: I live there, and there are a lot of hills to walk or cycle on. Which is steepest?
• Use OpenStreetMap, OSMnx, and Street View
• The resulting code is applicable to any named neighborhood

OpenStreetMap + OSMnx

OpenStreetMap is the well-known and widely used open source project to provide street maps of the world. Here is provides a suitable base map for us, and is integrated with the second tool we need, OSMnx (docs, citation).

OSMnx allows you to work with the streets, or more accurately street segments between intersections, to be worked with as a NetworkX MultiDiGraph of nodes and connections, each of which can have properties.

When the locations of nodes are combined with elevation data, the grade of each street segment can be obtained using OSMnx’s function for the grade along the line segments corresponding to the street segment between 2 nodes. This gives us grades for every street segment in the network, and segment with the largest value is the steepest street.

OSMnx has an extensive set of tutorials, one of which includes finding steepest streets. The code in this project is loosely based on it, but is mostly original. Some of the lines are:

G = ox.graph_from_place(place_query, network_type="drive")
G = ox.elevation.add_node_elevations_google(G, api_key=api_key)
G = ox.elevation.add_edge_grades(G)

grades_gdf = gpd.GeoDataFrame([d for _, _, d in ox.convert.to_undirected(G).edges(data=True)])

grades_gdf_sorted = grades_gdf.sort_values(by='grade_abs', ascending=False)

G = ox.graph_from_place(place_query, network_type="drive")
G = ox.elevation.add_node_elevations_google(G, api_key=api_key)
G = ox.elevation.add_edge_grades(G)

grades_gdf = gpd.GeoDataFrame([d for _, _, d in ox.convert.to_undirected(G).edges(data=True)])

grades_gdf_sorted = grades_gdf.sort_values(by='grade_abs', ascending=False)

I found some detail improvements in OSM that could be submitted for Millbrae, such as speed limits information and presence of sidewalks.

Finding the steepest

OSMnx allows a map of the street network with the street segments colored by grade to be produced straightforwardly:

Interestingly, when first running this project, I was considering doing more walking around Millbrae after dropping our 1-year-old at daycare, and seeing the areas nearby to the northwest of the neighborhood in the above plot with graded but not-too-steep streets encouraged me to do so.

According to this analysis, the top 10 steepest streets in Millbrae are:

Now, though, comes a reminder that your analysis is only as good as your data. The steepest street is not, in fact, Mullins Court. The street segment in question happens to be one of very few in the neighborhood that at the time of writing are mislabeled on Google Maps. Having now walked there, I can confirm that, while it is definitely steep, it is part of Aura Vista. This makes sense, as Mullins Court is the no-through street to the west of the intersection. The rest of the top 10 are correct as far as I know.

We can see the location of the Aura Vista segment as well, either on the street network:

or with an OSM basemap (with the Mullins Court again mislabeled at the time of writing):

See it on Street View

To view the steepest street on Google Street View, we need two steps

• Access the Street View API at the location of the street
• Rotate the view direction and angle to be looking up or down the street in question

After setting up the required API key, accessing the Street View location is achieved by constructing the correct URL:

URL = (f'https://maps.googleapis.com/maps/api/streetview?size=600x300&location={lat_origin},{lon_origin}'
       f'&heading={heading:.0f}&pitch={pitch}&key={api_key}')

URL = (f'https://maps.googleapis.com/maps/api/streetview?size=600x300&location={lat_origin},{lon_origin}'
       f'&heading={heading:.0f}&pitch={pitch}&key={api_key}')

which for Millbrae looks like this (API key not shown)

'https://maps.googleapis.com/maps/api/streetview?size=600x300&location=37.5911979,-122.4058984&heading=254&pitch=8.527488864617514&key=<API key>'

'https://maps.googleapis.com/maps/api/streetview?size=600x300&location=37.5911979,-122.4058984&heading=254&pitch=8.527488864617514&key=<API key>'

To rotate the view direction and angle, we need to extract and transform the direction and grade of the street segment from the OSMnx network into direction and angle above horizontal to pass to Street View. This is done by considering latitude and longitude for the direction, along with the x and y distances and elevations for the viewing angle.

Here we always select the view of the bottom of the street looking up. This gives us our view looking up the steepest street segment in Millbrae, as at the start of the text.

The approach fails if the street segment has a sharp curve between the two nodes, because you are looking directly towards the next node and not necessarily the direction the street immediately heads. This could be solved by decomposing the segment into its component linestrings and aligning with the first of them, but in practice this rarely comes up in Millbrae.

Other neighborhoods

The code is in principle generic to any neighborhood worldwide that can be named in the widely-used OpenStreetMap Nominatim geocoding tool. Simply swap out the name in the settings and explore. Here are some others nearby in the Bay Area:

Limits of the data: San Francisco

As is common in GIS, however, the real world has a way of bringing on limitations on workflows that naively seem straightforward, often in instructive ways. In this project, we see it if we set the place considered to be not a medium-sized suburb, but the whole of San Francisco.

Because of its many steep hills, streets navigating them, and large size, San Francisco is particularly prone to street segments with spurious grades:

• Limited latitude + longitude accuracy of the elevation data results in street node points on hillsides close to the street, not on the street
• Very short segments of only a few meters connecting one-way halves of streets at an intersection (common in San Francisco)

The grades plot itself looks good:

but it thinks that this wall is the steepest street (71% grade!):

Appropriately the name of this “street” is NaN.

There are some detail ways that these artifacts could be reduced:

• Filter out any segments of obviously spurious grade, say over 40%, and too-short length, say less than 10 meters
• Manually inspect remaining Street View images
• Manually adjust street node points so they correspond to the satellite image of the intersection (being sure that the network and raster are precisely aligned using a local CRS projection)
• Remove the OSMnx default consolidation of intersections, which may over-simplify some short steep segments

Ultimately addressing all of these in detail would constitute a full research project.

Summary & Future work

We found the steepest streets in my local neighborhood of Millbrae using OpenStreetMap and OSMnx, and viewed each one using Google Street View. The steepest street is Aura Vista, with a grade of 18%.

The code can in principle be run on any neighborhood worldwide that is named in the well-known Nominatim geocoding tool.

The main improvements to the results here would be

• Make a webapp so users can see their neighborhood without writing code
• Refine the analysis to be more robust in places such as San Francisco, using the ideas detailed in the main text above
• Share it directly to target communities, e.g., OSM User Diaries
• Add an option to use Open Topo data instead of Google elevation data so that the code can optionally be run without requiring the user to set up a Google API key

There are smaller improvements as well. For example, normalizing the grades colormap to be absolute and not calibrated to only the range of grades in that neighborhood. As currently coded, for example, somewhere very flat like Foster City would still show yellow for the streets with the highest grades, even though those grades are not high. The GitHub repository has the full set of ideas.

Interestingly, one consequence of these improvements might be, if a minimum threshold length for consistent grade is introduced to remove short street segments, such as the short ~ 10m segments of Hillcrest in Millbrae that place second and third, this might result in the 65m Aura Vista getting dropped, and the much longer (307m) but almost-as-steep Larkspur Drive taking the title. The whole segment between Pinehurst Court and Crestview Drive has an average grade of just under 16%, and in a way is a much worthier steepest street. Ultimately it depends on what we would like to highlight as a steep street.

My brother Andrew wrote a walking tour of the historical roots of the modern steel industry, Sheffield’s Don Valley for his blog. The open-source software uMap with an OpenStreetMap base can be used to turn this into a visual tour with a map, walking route, text and photographs. uMap can be used similarly for any piece of writing that involves traveling around an area. The map is available here .

• GitHub

Don Valley

Extending northeast from Sheffield City Centre, Don Valley is an area of global historic importance because of the history of the Sheffield steel industry. This encompasses its origins, the invention of stainless steel, and the manufacture of many products, the most well-known of which is cutlery.

Why map it?

Sheffield is a large modern city that has moved on since these industrial days, and so viewing such history in-place today requires knowing where to look, and what you are looking at. What you need, in other words, is a guide, and Andrew’s walking tour provides this.

The original tour is a text document that is nice to read and well researched. It is, however, only text, and so unless you know Sheffield very well already, or walk the tour in-person, there is a certain amount of abstraction to the history described.

For this reason it benefits from adding a map:

• Visualize the locations
• Show the walking route
• Incorporate text into the map
• Add pictures

Furthermore, the approach taken is generic: any text that involves describing places could be mapped in the same way.

uMap

uMap provides an open source solution that allows geospatial locations (points, lines, and polygons) to be overlaid on a street map base. It is built on top of the widely used Leaflet mapping software.

uMap is generally used in a no-code fashion, although GIS knowledge, use of markup, etc., is useful for more advanced features.

Making the map

We start by selecting the basemap, OpenStreetMap in English, and an appropriate center and zoom level. The center is the start of the walk at Sheffield Cathedral, and the zoom is to Don Valley.

Then we use the points datatype to add the locations mentioned in the text, formatting with various properties such as name, icon type, color, etc.

The walking route is added using the line datatype, following paths where they are marked, and placed in the road center where they are not. Sidewalks are not generally marked on the basemap used, likely because its maximum zoom level is not as high as the original OSM. The linestring resulting also has nodes in some places with more than two lines intersecting, because the route loops and returns to places already passed.

We did not use the polygon datatype on the current map. It could be added, for example to show the outlines of buildings of interest, but there was sufficient content already for our purposes.

Likewise we used the simple default symbology available in the uMap GUI, without, e.g., addons of icon types or symbols.

A nice feature of uMap is that it allows points to be labeled with media such as text, pictures, or video, activated in various ways such as clicking on the point. This allows pictures of the locations to be added, or even the entire original blog text appropriately spaced. Currently each point has a simple label and the text is the original blog entry.

Details, details…

So this is all fairly easy, right? Just read a few pages and add some map points. Well, nope. As you might expect with the real world (and especially GIS), some things come up.

A particular example from this walking route was on the Five Weirs Walk. Andrew’s original route had you following it for some distance, but since the text was written part of the walk between East Coast Road and Attercliffe Road has collapsed into the river. If you like climbing steep grass embankments with trees and old pieces of concrete it might still be possible, but for normal people we need a reroute round the streets.

The first step was to confirm that it was closed. Google Maps shows the path as still there, labeled Five Weirs Walk. Apple Maps shows the path, unlabeled. But the OSM base map shows closed gates and the segment of the path no longer present. On the satellite images, on both Google and Apple the collapsed section is visible, with the river extending close to the wall on the west bank, and pieces of concrete in the river. Not much is visible from Street View.

To reroute round the closed section, from Google Maps you might naively choose Windsor Street, but Apple Maps hints you may not want to because the street is marked narrowly as not a regular street, and indeed OSM and StreetView show that it passes through the premises of Thessco Ltd. At the other end there is a gate blocking the way to Princess Street, and it kind of goes near the Five Weirs Walk again but is behind the gate on that path. So maybe you can still walk through, but simpler is to route across the river around Faraday Road and Washford Road, which are regular streets.

The obvious step if you really wanted to go along Windsor would be to go there (or have Andrew go there), but it’s not necessary to do that here.

Addressing details like this routing ideally results in a route you can walk in-whole or in-part.

Some places along the route

The full route is on the map, but a few highlights are below.

Bower Spring: These are the best preserved remains of a mid-19th century cementation furnace, the process that enabled production of steel in quantity and put Sheffield on the map as a world center. It was recently restored and removed from the Heritage at Risk list in Yorkshire, which was reported on by the BBC in this article.

Kelham Island Industrial Museum: Showcases the history of the steel industry in Sheffield, along with other exhibits. The 12,000 horsepower River Don Engine that was used to roll steel plate is regularly run and has the shortest time to reverse direction of a machine its size anywhere.

Ball Street bridge: Aside from being my namesake, this is a scenic bridge over the River Don, giving good views of the river and former works on the banks such as Alfred Beckett and Sons.

Birthplace of stainless steel: This building used to house Brown-First Research, and was the world birthplace of stainless steel, one of the materials that revolutionized manufacturing in the 20th century due to its lack of rusting.

Terry Shellby, canal terrapin: Not officially part of the tour, but found along the way.

Can AI make this map?

The short answer is no, not yet. While the latest foundation models and coding tools (Google’s Gemini 3 and Claude Code at the time of writing) can help make maps from a prompt, and do other agentic work, the material here is too esoteric to work well. Places such as the sites of old steelworks not now named on the basemap need to be appropriately placed, and actually-possible routes like not taking the part of the Five Weirs Walk that has collapsed into the river need to be given, plus the suitable rerouting round a street mentioned above.

What AI could likely provide is a rough draft map or list of locations from the text, and help with creating a map with the uMap API if you prefer that route to the no-code one. It could also provide a handy summary of the text, an interface to answer questions about it, further pictures or links about places of interest, and how to travel there. But it is not going to outdate the full manual work to create this project just yet.

Summary & Future work

We have made a map of a walking tour of Don Valley showcasing the globally important history of the steel industry present in the area. The original walking tour text is here, and the map is here.

The map is made in a no-code fashion using the open source software uMap using an OpenStreetMap base. The emphasis is on the content and making a visualization of sufficient accuracy that it can be used as a guide to walk the route.

Some improvements to the work include

• Add pictures and text to the points
• More extensive symbology
• Have someone walk the route to confirm it is 100% correct

The uMap approach is generic, meaning that any text that describes travel around locations could be mapped in the same way.