Processing Nightlights Data – Part 1

What is the Nightlights data?

This is a picture of India at night. It’s made up of images taken during September this year, by the Suomi National Polar-orbiting Partnership (Suomi NPP) satellite mission. The white spots on the map are the parts of the country which radiate light to the satellite’s sensors.Mapblog

The idea is simple : the brightly lit areas correspond to the country’s main urban centres. Areas which are less well-lit are those of smaller towns and settlements, and the darkest areas are those which have the lowest levels of human density. Broadly speaking, different levels of intensity of light correspond to different levels of human activity (including economic activity). The brightest areas correspond to the densest urban settlements, while the least well-lit areas are those with minimal or no human activity (e.g. forests).

Here for instance is the same map as the one above, but focused on North India, and colored according to the different levels of light received by the satellite sensor. We’ve kept it simple and not added a legend.

The deep red blob in the middle is the Delhi National Capital Region, the biggest urban centre in North India and a major locus of economic activity. The grey lines indicate the boundaries of different tehsils. Those deep red lines radiating out from Delhi are major roads and highways.


Notice the contrast between Delhi and surrounding areas, and the part of India to the top right which is Uttarakhand, which is hilly / mountainous and much less densely populated. It’s a light yellow colour.

And here is the Mumbai-Pune-Nasik belt (deep red again indicates highest values of light and hence,densest human settlements). Mumbai is on the left, Pune is on the bottom right, and Nasik is toward the top:


Like any photograph, the satellite images are composed of pixels, each of which gets assigned a value by the sensor, depending on the ‘amount’ of light reaching it (the more the light, the higher the pixel value). Further, each pixel of the satellite image corresonds to a roughly 750×750 – metre area on the ground.

What about the variation of light within cities? The plot below compares the distribution of pixel values ( or rather their log values) within six big cities in India (note that the y-axis, indicating the number of pixels in that city of a certain value, is customised to individual cities, whereas the x-axis is common).


So what use is it?

Apart from a pretty picture of the earth at night, there have been a number of studies which have found a correlation between measures of night time lights in different areas, and the level of economic activity and/or population density.

Let’s examine this relationship between the measured value of night time lights in an area, with other indicators of well-being. The Indian census in 2011, gives us data on the percent of households in each district who own different consumption goods – a tv, a four-wheeler, a computer and so on. Do districts which have higher ownership of such consumption goods also have ‘higher’ values of night time lights as measured by the Suomi-NPP satellite?

In the graphs below we plotted the ‘density’ of households with different consumer goods in each district1 against the ‘average’ value of the pixels covered by that district2.

For instance, here’s a plot, along with a regression line, of the density of households of TV sets per district, against the average pixel values of the district. (Note :In the following graphs, most of the districts are squeezed into the bottom left of the graph. Use your mouse to focus the graph to that area, if needed)


Another plot of households with four wheelers and pixel values


A third plot of households who own a four/two-wheeler, a TV set, a computer and a phone, as of 2011


And a fourth plot of households who have a toilet at home


Now that we’ve established the value of the night lights data, we will look at the practical stuff in future posts. Where to get the data from? How do we process it? and so on. Stay tuned.

  1. The measure used is the number of households owning a given consumption good (i.e. TV set) in a district, divided by the total surface area of that district. We do this because districts differ from each other in terms of total surface area.
  2. We use satellite data averaged over 11 months between April 2012 and February 2013. The average pixel value of a district is the total sum of all pixel values which ‘cover’ a given district, divided by the total number of pixels (Each pixel covers the same surface area of 750x750m)

Leave a Reply

Your email address will not be published. Required fields are marked *