India’s Urbanization- Viewed from 824 Kilometers above the Earth

Remote Sensing — Nightlight Analysis to Understand Human Activity

Satellites currently orbiting the earth collect a myriad set of information about the earth, ranging from temperature and climate to nighttime light emissions. In specific, policy makers and economists have extensively used data of nighttime light emissions for studying economic activity, migration, urbanization , and disaster monitoring, to name a few applications. Nighttime light emissions are a fingerprint of modern human activity, which is enabled by electricity and ubiquitous lighting.

To visualize urbanization in India, I use data on nightlight emissions captured by the Suomi National Polar-orbiting Partnership (Suomi-NPP) satellite,¹ orbiting the earth at an altitude of around 824 kilometers.² The satellite has a specialized instrument called Visible Infrared Imaging Radiometer Suite (VIIRS) which is very sensitive in low-light conditions and thereby captures information on nighttime lights (Day/Night Band).

The data is captured daily, but I use the monthly composites (values averaged through a month).³ This is to ensure that the estimation is not affected by idiosyncratic events in a given day.

How has India’s nightlight emissions changed from January 2014 to January 2021?

I start off by plotting areas across India that have a high-intensity of nightlight emissions (i.e., higher than the national average in January 2014).

Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The three thresholds represent the regional mean of average radiance in 2014, one standard deviation from mean in 2014, and two standard deviations from mean in 2014. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.

As expected, urban centers such as the National Capital Region, Chennai, Bengaluru, Kolkata, Mumbai, Hyderabad, and Pune have the highest intensity of emissions. Notice how regions outside the urban centers in states like Uttar Pradesh, Maharashtra, and Telangana have lit up in 2021, when compared to 2014.

The crucial question is, where do we see the highest rate of growth in nightlight intensity when comparing January 2021 with January 2014. The chart below holds the answer.

Notice how the fastest growth has happened in peripheral areas of the National Capital Region, Pune, Hyderabad, and Bengaluru. In the next set of images we will zoom in to a few states to have a closer look at this urbanization process.

Karnataka — Bengaluru’s Peripheral Urbanization

Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The three thresholds represent the regional mean of average radiance in 2014, one standard deviation from mean in 2014, and two standard deviations from mean in 2014. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.
Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The two thresholds represent the regional mean of rate of change between 2014 and 2021, and one standard deviation from mean. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.

Maharashtra — Mumbai and Pune’s Urbanization and Rural Electrification

Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The three thresholds represent the regional mean of average radiance in 2014, one standard deviation from mean in 2014, and two standard deviations from mean in 2014. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.
Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The two thresholds represent the regional mean of rate of change between 2014 and 2021, and one standard deviation from mean. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is using the Python API for Google Earth Engine.

Tamil Nadu — Chennai, Hosur and Coimbatore (to an extent) are Urbanizing but why are Madurai and Tiruchirappalli not urbanizing as rapidly?⁴

Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The three thresholds represent the regional mean of average radiance in 2014, one standard deviation from mean in 2014, and two standard deviations from mean in 2014. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.
Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The two thresholds represent the regional mean of rate of change between 2014 and 2021, and one standard deviation from mean. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.

Telangana — Hyderabad’s Expansion and Rural Electrification

Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The three thresholds represent the regional mean of average radiance in 2014, one standard deviation from mean in 2014, and two standard deviations from mean in 2014. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.
Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The two thresholds represent the regional mean of rate of change between 2014 and 2021, and one standard deviation from mean. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.

Uttar Pradesh — A Rural Electrification Success Story + Strong Urban Growth in Noida, Lucknow and Kanpur

Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The three thresholds represent the regional mean of average radiance in 2014, one standard deviation from mean in 2014, and two standard deviations from mean in 2014. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.
Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The two thresholds represent the regional mean of rate of change between 2014 and 2021, and one standard deviation from mean. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.

The Takeaway

Notice how much of the urbanization is centered around already large cities such as Bengaluru, Chennai, Hyderabad, Kanpur, Lucknow, and Mumbai. The excessive urbanization (specifically in the peripheral urban regions) is happening at a rapid pace, and rather haphazardly. For example, unregulated urbanization in Chennai was the prime reason for the horrific floods the city has witnessed since 2015.

Further, already crowded cities are getting more and more crowded, straining core public utilities. India should seriously start considering regulating the urbanization process, and focusing on the growth and prosperity of tier-two cities, instead of expanding existing metropolises.

Appendix — Tamil Nadu Data

Cloud cover is quite a concern for Tamil Nadu data. This is because of its relative proximity to the equator. Regions proximate to the equator typically have high cloud coverage, which substantially affects the quality of the nightlight emissions data captured. Hence, the results presented for January for Tamil Nadu, would need to be taken with a grain of salt.

Using data from March, we see a much stronger degree of urbanization (specifically in the Coimbatore-Salem belt), but in the rest of the cities (including Madurai and Tiruchirappalli), the urbanization is rather muted.

The following are the results for March 2021 vs. March 2014.

Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The three thresholds represent the regional mean of average radiance in 2014, one standard deviation from mean in 2014, and two standard deviations from mean in 2014. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.
Pixels with lesser than 3 cloud free observations are truncated as done in Elvidge et al., 2020. The two thresholds represent the regional mean of rate of change between 2014 and 2021, and one standard deviation from mean. To compute the regional means, I resample (down-sample) the data at one kilometer squared and compute the means within the respective boundary. The analysis is performed using the Python API for Google Earth Engine.

About Google Earth Engine

All core computations behind the visualizations in this article were performed using the Google Earth Engine platform. Geo-spatial datasets such as the ones used in this article are often extremely large and require serious computing power to produce such visualizations.

Endnotes

[1] The project is a joint partnership of National Aeronautics and Space Administration (NASA) and National Oceanic and Atmospheric Administration (NOAA).

[2] See the European Space Agency’s information page on the satellite for more details.

[3] I use the straylight corrected dataset. The dataset can be accessed via this link on Google Earth Engine.

[4] See Appendix about Tamil Nadu data.

[5] Full Citation for Elvidge et al., 2020 — Elvidge, C. D., Ghosh, T., Hsu, F. C., Zhizhin, M., & Bazilian, M. (2020). The dimming of lights in china during the COVID-19 pandemic. Remote Sensing, 12(17), 2851.

Author: Akshay Natteri Mangadu

All views are personal.

Applied Econometrician (M.A. from University of Chicago)