Creating a Vehicle Density Map at Grid-level for a city !!

[Sarath Guttikunda]

In our group, we frequently perform extensive numerical analysis, particularly to understand emission intensities, both spatially and temporally. The spatial understanding of emissions is a significant component of our work, as it is crucial for accurately placing emissions before they are modeled and concentration maps are generated.

This process is technically known as the "gridding of emissions." For example, if we know there are one thousand trucks operating in a city, each traveling one hundred kilometers a day, we can multiply these figures by an emission factor for a specific pollutant to determine the total emission intensity of trucks moving within the city's airshed. The question then becomes: how do we distribute these emissions into various grids for a city? We typically work with one-square-kilometer grids on average, and you can see some examples below.

https://urbanemissions.info/india-air-quality/india-ncap-city-airsheds/

One of the proxies we use for trucks is highways. The assumption is that most trucks will travel on highways and spend the majority of their time there. Therefore, we assign a higher weight to the grids that intersect with highways. We also incorporate other layers of information with additional weights. For instance, industrial hubs, commercial hubs, malls, and markets are places where these vehicles are likely to go and spend some time. This methodical approach generates various weighting functions, and once we have the emission intensities, it produces a gridded emission file. So far this method of madness works and we have a good understanding of how the layers are behaving with some plus minus. We have an example tool to play with this method -- https://urbanemissions.info/tools/

We aim to improve this process. One of the layers we introduced in the past was speed information from the Google Maps API. We can download speed data, which also indicates congestion times. We utilized this as another proxy to understand where and for how long vehicles spend time, and accordingly, assign weights. See example image for Mumbai here - https://urbanemissions.info/india-apna/mumbai-india/

A new approach we want to explore, given some recently available information (and algorithms), is vehicle density. This would again be a static input. For example, if you take a satellite image and apply an algorithm, you could determine how many vehicles are visible within each grid. Because this is a static image for a specific time, we cannot use it as a layer for all-purpose gridding. However, it would serve as an additional layer of information that accurately reflects what is happening on the ground. It could also be used to extract information about official and unofficial parking lots where vehicles spend a significant amount of time on a given day. This would allow us to extract valuable insights.

There are many online examples of this being done using geostationary images in Europe and the United States and most of them require an image and rest seem to work (take it with a pinch of salt -- non-it-person speaking).

https://www.linkedin.com/posts/giswqs_geoai-geospatial-ai-activity-7309216621997281280-46yC/

https://up42.com/marketplace/analytics/detection-cars-oi

A commercial portal -- but seems to do exactly what we want at a price

So, the question to the group today is this: If there is a grid file, let us say for Bangalore -- has anyone done anything similar to create a vehicle density map, regardless of the vehicle type? or have any ideas on how to approach this for Indian cities?

Please keep in mind that the ultimate goal is not to identify individual vehicles or count vehicles from traffic cameras. The focus is on a static image: if we have one, can we, or has anyone, worked on creating a vehicle density map from it?

Any sights into making an example and hopefully scaling it up is appreciated.

Looking forward to the follow ups.

With best wishes,

Sarath

--
Dr. Sarath Guttikunda
https://urbanemissions.info

[Anupam Sobti]

The constraint here might be availability of usable high resolution imagery. I believe Google prohibits use of its imagery and the imagery that these tutorials use are not available in India. Some people use night time lights overlayed on street networks as a proxy for vehicle density but I think GPS based google data would be more reliable.

Regards

Anupam

[Sarath Guttikunda]

Dear Anupam,

Thanks for the response. The main bottleneck in this process is having access to high-resolution imagery. The codes are all open and many groups seem to be using them in places other than India. If you come across any useful applications, we would be open to exploring those opportunities.

On the second point, yes, GPS-based data would be much more useful, but again, that's not a free dataset. The Google Distance API actually uses cell signals to gather congestion-related information, but it's not a free dataset, and over the last three or four years, they've really increased the price tag for using that API.

In our emissions distribution step, having granular information is much more useful -- granular in space and time. Using an image provides a static dataset on where vehicles are on the road and also where parking lots are (space dimension). This, again, is a proxy that could align with just road density, which is the theory we've been using for the longest time. If we can add another layer of vehicle density to it, the spatial analysis of emissions at higher resolutions, especially in big cities, will improve, along with some of our other analyses.

Hoping others on the list can add to this discussion.

With best wishes,

Sarath

--
Dr. Sarath Guttikunda
https://urbanemissions.info

--
Datameet is a community of Data Science enthusiasts in India. Know more about us by visiting http://datameet.org
---
You received this message because you are subscribed to the Google Groups "datameet" group.
To unsubscribe from this group and stop receiving emails from it, send an email to datameet+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/datameet/c0afb013-c9b2-4f13-8b92-570b3d79b916n%40googlegroups.com.

[Amit Shirsath]

Hi Dr. Sarath,

There is another approach, though not very accurate and too complex.

• There is Vahan data available at RTO level. These RTO offices usually serve the whole district or municipal corporation or a few taluka, etc. GIS boundaries for these entities are generally available online. So, RTO can be mapped to a geographical boundary.
• Also, there are .tiff images available for all India on Bhuvan. Using Zonal Histogram with these images, one can derive Land Use Land Cover (LULC) for all India. I had made LULC at 100m x 100m grids for all India a year back. So, 1 Sqkm is completely feasible.
• Once LULC is available, we have to identify areas where high density of "Builtup Cover" is available. In theory, it is the area where some kind of construction is there like buildings, bungalows, informal settlements, etc. There can be different approaches to identify this, as classes available with Bhuvan files may not directly correspond to what we want.
• I am making the assumption that the "Builtup Cover" areas will have vehicle ownership or usage. This assumption has to be refined further based on your requirement.
• Once RTO boundary and "Builtup Cover" is available, the total vehicle sales can be distributed to "Builtup Cover" area proportionally.
• So, finally you get the number of vehicles per grid.
• One advantage of using Vahan data, is that the vehicles are categorised by Vehicle Class, category, Fuel, Norm (BS IV, BS III, etc.) and Maker. So, you can assign weight

Notes:

• The vehicle classifications on Vahan portal may not be available for all the years.
• Here, we are assuming that the vehicle purchased within an RTO boundary is operating within the same RTO, which is not true in many cases. For example, Nowadays, I am observing many high end consumer vehicles with "DD" registration numbers operating in Mumbai. Also, some large commercial vehicles like tourist buses with "NL" registration numbers operate on Mumbai, Pune, Bangalore routes.
• However, If this type of migration percentage is lower compared to the total number of vehicles sold in the last 15 years, then this method can be useful.

To view this discussion visit https://groups.google.com/d/msgid/datameet/CAAj%2BwWo%3DT5siPy4red1eLVexAMoeKPE3V3d69WrLAt8tN9%2BQow%40mail.gmail.com.

--

Regards,

Amit Shirsath

+91 9970944537

www.amitshirsath.com

It's easier to invent the future than predict it. -- Alan Kay

[Sarath Guttikunda]

Dear Amit,

Good morning. All excellent suggestions and glad you summarized them here. 

These are in default use -- built-up area, LULC fields, road network, population density, commercial activity density, and road activity points density (like stops). https://urbanemissions.info/india-air-quality/india-ncap-cities/

We have a template tool for playing with these combinations.

https://urbanemissions.info/tools -- look for the gridding tool.

This has been the default setup for a long time and the process works (not just in India, but anywhere).

Question is not if we can grid the emissions. We want a new layer -- vehicle density from the images at the grid level -- a new proxy layer. Given the buzz around AI and Satellites, this is an open question to the community -- is this something we can make for Indian cities? If yes, how.

With best wishes,

Sarath

--
Dr. Sarath Guttikunda
https://urbanemissions.info

To view this discussion visit https://groups.google.com/d/msgid/datameet/CALVzC30LGr84vHm4ZU4Cz%3D3HQRWcnM_TX9z0STu%3DvWbTzzL6tg%40mail.gmail.com.