The slides are from Rishabh Srivastava's talk at the PyData meetup in Singapore on December 2, 2019

1. S AT E L L I T E I M A G E RY F O R T R A C K I N G
E C O N O M I C C H A N G E
R i s h a b h S r i v a s t a v a | L o k i . a i | @ r i s h d o t b l o g

2. O V E R V I E W
A. Why use satellite data
B. How to use satellite data
C. Drawbacks and limitations of using satellite data to
estimate growth

3. A . W H Y U S E S AT E L L I T E D ATA ?
W E S T E R N D E L H I , I N D I A

4. U s e - C a s e
• Todd, a financial analyst
specialising in the
construction industry
• Wants to identify fast
growing cities (in terms of
infrastructure) in Asia

5. O b j e c t i v e
E s t i m a t e re l a t i v e i n f r a s t r u c t u re g ro w t h r a t e s
a c ro s s p ro v i n c e s i n A s i a

6. C h a l l e n g e s
1. Few “official” metrics for measuring granular growth
2. The ones that do exist (like province GDP) often
cannot be compared across countries
3. Even when available, official figures are unreliable
4. Figures in developing countries can come with a
massive (up to 18 months) time-lag

7. S o l u t i o n c h a r a c t e r i s t i c s
1. Reliable and impervious to manipulation
2. Regularly updated
3. Relatively inexpensive or free

8. S o l u t i o n :
S a t e l l i t e
I m a g e r y !
• Free sources:
• NOAA Nightlight Data
(updated daily)
• Sentinel-2 Data
(updated every 5 days)
• Paid sources (not covered
in this talk)

9. B . H O W T O U S E S AT E L L I T E D ATA ?

10. N I G H T L I G H T
D ATA
• An effective proxy for
economic growth in
developing countries
• Produced by NOAA (USA)
• Updated daily, free to use

11. A c c e s s i n g t h e
D a t a
• Data available as raster (GeoTiff)
files, where each pixel
represents the brightness (in
nanoWatts/cm2/sr) of a roughly
450m x 450m area
• Files can be read by Python
(with GDAL doing the heavy
lifting), using the rasterio library
• The rasterstats library can be
used to calculate summary
statistics for different areas in a
shapefile
rasterio
rasterstats

12. C o m p u t i n g
s u m m a r y s t a t s
• You can convert a raster to
a CSV file with average
luminosity in predefined
areas using literally 10 lines
of code
• Shapefile (which is a
machine readable map) of
a region needed for this

13. E x a m p l e f o r S i n g a p o re
Water reservoirs
Central
Business
District

14. M e a s u r i n g c h a n g e s o v e r
t i m e
• CSVs generated for each
time-unit can be used to
track changes in average
nightlight intensity over
time
• Significant seasonal
variations can exist
• Interactive version for
Indian districts: https://
embed.loki.ai/nightlight/

15. G ro w t h o f n i g h t l i g h t i n t e n s i t y i n
A s i a f ro m 2 0 1 3 - 2 0 1 8
I N T E R A C T I V E V E R S I O N : H T T P S : / / E M B E D . L O K I . A I / N I G H T L I G H T / A S I A . H T M L

16. S E N T I N E L - 2
S AT E L L I T E
I M A G E RY
• Measures reflected solar
radiation across 13-bands of
wavelength (3 of which are
RGB)
• 1 pixel represents 10m x 10m
• Produced by the European
Space Agency
• Updated roughly every 5
days, free to use

17. A c c e s s i n g t h e
D a t a
• Rasters can be
downloaded from the
ESA’s Sentinelhub, or from
a Google Cloud Storage /
Amazon S3 bucket

18. D i ff e re n t b a n d s c a p t u re d i ff e re n t
w a v e l e n g t h s o f l i g h t
Sentinel-2 Bands
Central
Wavelength (Nm)
Resolution
(M)
Band 1 - Coastal Aerosol 443 60
Band 2 - Blue 490 10
Band 3 - Green 560 10
Band 4 - Red 665 10
Band 5 - Vegetation Red Edge 705 20
Band 6 - Vegetation Red Edge 740 20
Band 7 - Vegetation Red Edge 783 20
Band 8 - Near Infra-Red 842 10
Band 8A - Vegetation Red Edge 865 20
Band 9 - Water Vapour 945 60
Band 10 - Short-Wave Infra Red 1375 60
Band 11 - Short-Wave Infra Red 1610 20
Band 12 - Short-Wave Infra Red 2190 20

19. P ro c e s s i n g t h e d a t a
• Different bands can be combined to get meaningful
metrics — like amount of built-up area in a city, amount
of green space etc
• For instance, roads and buildings tend to absorb more
infra-red light (heat) and reflect more visible light
(specially in the blue spectrum)
• A simple, “built-up” index that is commonly used is
(Blue-NIR)/(Blue+NIR). Note: areas with water tend to
also have a high built-up index

20. Tr u e - C o l o r i m a g e o n l e f t , b u i l t - u p i n d e x b a s e d
i m a g e o n r i g h t ( d a r k e r = m o re b u i l t - u p a re a )

21. B u i l t - u p a re a b y re g i o n i n S i n g a p o re
( d a r k e r a re a s = m o re b u i l t u p )

22. Q u a n t i f y i n g
c h a n g e o v e r t i m e
• Changes in built-up
indices can be computed
over time. Extremely
effective at the micro-scale

23. C h a n g e i n H o C h i M i n h C i t y ’s D i s t r i c t 2 o v e r t i m e
( 2 0 1 9 o n l e f t , 2 0 1 6 i n c e n t re , c h a n g e - m a p o n r i g h t )

24. C h a n g e i n H o C h i M i n h C i t y ’s D i s t r i c t 2 o v e r t i m e
( 2 0 1 9 o n l e f t , 2 0 1 6 i n c e n t re , c h a n g e - m a p o n r i g h t )

25. C h a n g e i n H o C h i M i n h C i t y ’s D i s t r i c t
2 o v e r t i m e
Feb, 2019 Jan, 2016
Change Map
(Red= more built)

26. C . L I M I TAT I O N S A N D C AV E AT S
C L O U D S O V E R S I N G A P O R E

27. K e y
D r a w b a c k s
• Cloud-cover makes using
satellite imagery far more
difficult
• Extreme haze in India and
China can cause nightlight
and index-based approaches
to go awry
• Algorithm can sometimes
think that water-logged
neighbourhoods are highly
built-up

28. L i m i t a t i o n s
• Growth due to non-
infrastructure factors (like
Finance or Software) not
measured through satellites
• Assumes that increase in
supply of infrastructure,
nightlights, and buildings =
growth. Not always the case
(Japanese Roads to
Nowhere)

29. R i s h a b h S r i v a s t a v a
r i s h a b h @ l o k i . a i | l o k i . a i
@ r i s h d o t b l o g
Questions?