09 Jun Stefan Steiniger/Has Santiago less street space than other cities?
** [update on 8. July 2015]: After some discussion of this blog entry on Twitter it appears that the original publication by Vasconcelos (2001) does not included street space for Santiago but instead for Shanghai. The value of 8% was calculated and added by De Grange. We apologize to Eduardo Vasconcelos for this error.
Por Stefan Steiniger / Cluster Acceso y Movilidad
In a report of the newspaper La Segunda from 2013, titled «Louis de Grange, políticamente incorrecto: Sus argumentos para decir que la bicicleta NO es una opción al transporte público«, it is written that Santiago has so few street space, that there is no space available to accommodate for the surge of bicycle use. Dr. De Grange has, however, his number from a publication by Vasconcelos (2001) – which contains the following graphics:
As we see, it is estimated that only 8 percent of Santiago’s area is dedicated to streets. And even more so, it is suggested that this is significantly less than for other cities in the world. The number of only 8% seemed to us pretty low and we therefore wanted to do an analysis of Santiago’s street space ourselves. For a better comparison, and to exclude gross errors in the estimation, we will also evaluate how much street space is allocated to Sao Paulo.
Data needed for the analysis
To estimate the space used by streets we need two basic datasets:
– a street dataset, and
– a dataset outlining the municipal limits, or something similar.
A street dataset we can obtain from OpenStreetMap, e.g. the ready country shapefiles from here: Chile + Brazil. To not bother with the huge amount of street data for the whole country, its better to actually use the metro extracts from Mapzen.
If we overlay the street data with the municipal boundaries, we may be aware that the comuna contains also areas that are not urban areas, such as mountainous, forests or agricultural areas. Including such areas in our analysis would greatly influence, i.e. lower, the percentage of city space designated to streets. For example in the figure below we display Sao Paulo’s administrative boundaries in green (obtained from OpenStreetMap as well) and the road dataset. As it becomes clear we should select for Sao Paulo a representative, central region to calculate the percentage of street space. For our analysis this will be the area within the yellow box.
A bit different situation we encounter for Santiago: If we want to perform the street space analysis for Santiago, then we actually do not want to do the analysis for the fairly small «comuna» Santiago, which has only 200k inhabitants. Instead we want to perform the analysis for Gran Santiago, which consisting of many more comunas and has about 5.1 Million inhabitants.
The limits of the urban area for Gran Santiago, and other urban areas of Chile, are available in a dataset from the Biblioteca Nacional de Chile (BNC). This dataset is not the newest (its from 2009), but as the population of Santiago grows, it is a good estimate. This means, to calculate the percentage of street area for Gran Santiago we will not select a representative area, as done for Sao Paulo, but will use the urban limits provided in the BCN dataset. The figure below shows the urban area for Gran Santiago from the BNC datasaet in yellow and the road network obtained from OpenStreetMap.
From road centre lines to street surface area
The street network data downloaded from OpenStreetMap contain only the center lines for each street or path. Hence, if we would calculate the street area from the lines, then we would got a big fat zero. So how to get the street area? Fairly easy from a GIS perspective: (1) first we should check what road classification was used in the OpenStreetMap dataset (e.g. residential, motorway, etc.) (2) then we measure in an aerial image the width for each street type, and (3) we apply a GIS buffer operation to each road, in which the buffer size will correspond to the street width with respect to the road type. Below I will outline the steps a bit more detailed:
1. Road types in the OpenStreetMap dataset
For all urban areas of Chile I analyzed the road type tag for each street segment. Then I calculated the segment length and ordered the types by length to see which seem to be most important. I got then the following list:
– residential – 21’319 km*
– tertiary – 3’066 km
– living_street – 2’610km
– primary – 2’500 km
– secondary – 2’235 km
– service – 836 km
* Note, the length in km is calculated for all way segment of this type within urban areas of Chile, taken from BCN dataset. In total these were 165’433 street segments in 400 urban areas.
As we can see, the most important road types for our analysis are «residential», «tertiary» and «living_street». We should therefore measure the width of these type more carefully. This means, we should actually measure the road width of these frequent street types (a) repeatedly, and (b) perhaps also perform an error analysis later on.
Since the OpenStreetMap (OSM) data are a live snapshot from the current OSM database, there may exist also some road type tags that appear non-standard or represent street types that we are not interested in, such as bridleway, raceway, construction, etc. For these cases the road segments that have such unwanted or non-standard tags are to be removed from the dataset.
As a further note I want to add that before I started analyzing the road data, I filtered the road network dataset so that only those line segments within urban areas are retained. This will later save a significant amount of processing time.
2. Measuring street type widths.
To measure the street widths there are three options, (a) going out and measuring it like a surveyor, (b) consult road construction manuals, or (c) measure the width in aerial images. Now, option (a) is quite time consuming and rather difficult if we want to measure the width of highways and motorways, as we cannot stop traffic. (b) is in our case also not so recommendable, since the average data creator of OpenStreetMap does not consult any road design and construction manual when assigning the road type tags. Hence, we are left with option (c).
So, fortunately we had some high resolution aerial images available (16cm/pixel) from Sectra for Santiago, because the satellite images from Google Maps, Here and Bing Maps do not seem to be sufficient for measuring road widths. The street furniture details and even sidewalks are hardly to see in image data from these providers. However, this also means that the widths measured are more or less all from Santiago. Hence, when we apply the same street widths to data from other cities, e.g. Sao Paulo, then we obtain some larger error, since cultural background (i.e city planning history) and street construction manuals will be different from country to country, region to region, and climate zone to climate zone.
For almost all street types I measured the street width in the aerial images in a repeated fashion. For some road types, such as links, e.g. «secondary_link» and «trunk_link», etc., I defined a default width after measuring only once. The street width measurements do include the sidewalks, and if existing, also the green spaces between road and sidewalk.
As I measured the widths repeatedly I am able to calculate some measurement statistics. In particular I calculated the meanand the standard deviation (std). With the help of the standard deviation I later on calculated a lower bound for the street width, i.e. width_min = mean – 1*std , and a upper bound: width_max = mean + 1*std. Both bounds enable to calculate an lower and a upper limit for the total street area, which helps to get an idea about the effect of variation in width measurements.
In the listing below we summarize the mean, min, and max width of the measurements, and indicate the number of repeated measurements in brackets:
road type (#) min (m) mean (m) max (m)
– default: 6m 6m 6m
– primary (15): 10.8 15.6 20.5
– secondary (11): 10.0 12.5 15
– residential (15): 7.1 10.3 13.6
– tertiary (15): 8.3 12 15.7
– motorway (3): 12.4 13.5 14.7
– path (7): 2.7 4.5 6.4
– service (8): 6.1 6.9 7.8
– cycleway (2): 2.9 3.3 3.6
– pedestrian (10): 4.8 9.8 14.8
– footway (6): 2.1 4.4 6.7
– living-street (10): 5.2 6.6 7.9
– motorway-link (6): 4.8 5.6 6.4
– track (11): 4.5 6.5 8.4
– unclassified (9): 6.8 9.2 11.6
– trunk (6): 8.0 8.9 9.8
3. Buffering of road center lines, and union of street segment
Having obtained the mean, min, and maximum street type widths, I added the information as an attribute to each street segment. With the help of a GIS buffer operation, where the widths are converted into a radius value, the area for each street segment is then obtained from its given center line. In the figure below we can see, overlaid on an aerial image, the street segment center lines in cyan and the obtained buffered areas in blue. Additionally the road type of each street segment is displayed, e.g. footway and primary road, so one can see that different road types will result in different buffer sizes.
Now that we have the area for each road segment one more GIS operation is necessary before we can get the street area: All street segments need to be union-ed, i.e. merged. This is necessary since the resulting buffer areas may overlap, as can be seen in the figure above on street crossings (emerging circle forms). If one would calculate the area with the overlaps, then the street area would be over-estimated. Important to note is, that the union operation is a very costly operation, with the number of operations dependent on the square of the number of vertices / street segments in the dataset. Hence, calculation can take a few hours for a dataset that contains all urban areas of Chile.
The final step for each city is to calculate (a) the size of the areas for the buffered streets, and (b) the size of the urban area; or for Sao Paulo the size of the study area bounding box (in yellow above).
Result of estimation of urban area dedicated to street
Below are listed in detail the different estimates for the areas dedicated to streets: For Santiago I obtain an average street area of 19.7% from the total urban area, and for Sao Paulo I obtain that only 17.5 % of the urban area is dedicated to streets. As outlined above we applied here the same street type width, measured in Santiago, to the street center lines that we have for Sao Paulo. Hence, we need to account for a certain amount of error when we compare the street density of both cities. However, it needs also be added that these estimates are fairly conservative since the OpenStreetMap dataset that we use does not contain parking lots, e.g. parking areas of shopping malls (they would be modeled as polygons and are therefore not part of the center line dataset), and the dataset does only in a few cases contain private roads.
1. Area of Santiago dedicated to streets:
– urban area of Santiago (from BNS dataset) : 582.1 km2
– street area estimates:
mean: 114.5 km2 => 19.7%
lower bound: 83.7 km2 => 14.4%
upper bound: 143.6 km2 => 24.7%
2. Area of Sao Paulo dedicated to streets
– study area in bounding box: 443.3 km2
– street area estimate within bounding box:
mean: 77.5 km2 => 17.5%
lower bound: 57.4 km2 => 12.9%
upper bound: 97.2 km2 => 21.9%
More than 8% of Santiago’s urban space is street space
Given the numbers above, we can clearly say that for Santiago more than 8 percent of its urban area is used for streets. I am even quite convinced that we can safely say that Santiago’s urban streets area occupies by more than double of that, i.e. 19 percent of the area is occupied by streets. If one has followed how we calculated the street area, then one will also notice that this includes space for sidewalks too – and not only space for cars (and bicycles).
Looking at the estimate for Sao Paulo, then we see that our value of 17.5% is much closer to the 21% presented by Vasconcellos (2001) than for Santiago’s street values (19% vs. 8%). This seems to indicate to me that there is something wrong in the calculation of Vasconcellos (2001) for Santiago’s estimate. Perhaps he included the huge municipal area of Colina in the north of Santiago, which contains large areas of agronomic uses, into the calculation? Or he added the eastern parts of the comunas in the «oriente», that cover mountainous areas too? However, one may also need to account for the fact that since 2000 there has been quite a bit of road construction activity in Santiago. But this should not be so much that street surface goes up by more than 2 percent, perhaps.
When we compare the calculated street surface estimate for Santiago with the numbers for the other metropolitan cities in the first figure (20-25%), then with the 19% Santiago seems to align more or less well with them – but still taking the same spot in the density ranking. If we instead chose our optimistic upper bound value of 24.7 % for the comparison, then Santiago seems to have more space available for cars and pedestrians than cities such as London and New York. Calcutta claims the last rank in the original listing from 2001 with a surprising 7% of street space. Now with the availability of a global road network dataset from OpenStreetMap.org, a map project that gained momentum in 2007, it should be not so difficult to re-evaluate the street density for Calcutta. I assume a strong corrections will be necessary here.
But returning to the original question of «Is there space available for cycling infrastructure in Santiago?» – Well, we don’t even need to look at the numbers. Just walk around in Santiago and you will see that there is plenty of space; especially when one sees all the roads with cars parked on the sides taking up a full lane. Such space could easily be converted into cycling infrastructure if the residents can be convinced (which seems to be a problem in the comuna Providencia). Also, if there seems to be street space available for conversion to BRT lanes, then there is also space that can be converted for use by bicycle.
Vasconcellos, Eduardo A. (2001), Urban Transport, Environment and Equity – the case for developing countries, Earthscan, UK, table 2.1, página 12.