First trials on Sentinel-1 performance for mapping built-up areas Kaupo Voormansik 1,2 , Anni Sisas 2,3 , Jaan Praks 1 Aalto University, Tartu Observatory, University of Tartu
First trials on Sentinel-1 performance for mapping built-up areas
Kaupo Voormansik1,2, Anni Sisas2,3, Jaan Praks1
Aalto University, Tartu Observatory, University of Tartu
Outline
1. Motivation – why urban monitoring? 2. How to extract built-up areas from SAR imagery,
in theory? 3. Sentinel-1 input data and processing 4. Performance figures 5. Copernicus Urban Development Analyser - CUDA
2
Urbanisation – a global megatrend
• Increasing number of people live in urban areas. • Cities grow, covering more and more land. • Taking place all around the world, but fastest growth
in Asia and Africa. There are cities, which population grows up to 10% in year!
• From 2011 to 2050 world’s urban population is expected to grow from 3.6 billion to 6.3 billion.
• 83 % of governments are concerned about their population distribution in the country.
3
According to United Nations, Department of Economic and Social Affairs, Population Division (2011): World Urbanization Prospects, the 2011 Revision.
Percentage of urban population and agglomerations by size class: 1960, 1980, 2011, 2025
4
1960 Source: United Nations, Department of Economic and Social Affairs, Population Division (2011): World Urbanization Prospects, the 2011 Revision.
5
1980 Source: United Nations, Department of Economic and Social Affairs, Population Division (2011): World Urbanization Prospects, the 2011 Revision.
Percentage of urban population and agglomerations by size class: 1960, 1980, 2011, 2025
6
2011 Source: United Nations, Department of Economic and Social Affairs, Population Division (2011): World Urbanization Prospects, the 2011 Revision.
Percentage of urban population and agglomerations by size class: 1960, 1980, 2011, 2025
7
2025 Source: United Nations, Department of Economic and Social Affairs, Population Division (2011): World Urbanization Prospects, the 2011 Revision.
Percentage of urban population and agglomerations by size class: 1960, 1980, 2011, 2025
How to extract built-up areas from SAR imagery?
• Several methods: intensity thresholding, InSAR coherence, polarimetry…
• One of the most robust method ideas: exploiting local area statistics!
• Due to the speckle effect homogeneous areas distribution function is known.
8
Figure source: Lee and Pottier „Polarimetric Radar Imaging“ 2010
1-look intensity data Exponential distribution 1-look amplitude data Rayleigh distribution 4-look amplitude data Chi distribution
How to extract built-up areas from SAR imagery?
• 𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍_𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍_𝒎𝒎𝒔𝒔𝒍𝒍𝒎𝒎
= 𝒍𝒍𝒍𝒍𝒎𝒎𝒔𝒔𝒔𝒔𝒍𝒍𝒎𝒎𝒔𝒔.
• The relation is broken in urban areas.
• It is possible to measure the deviation from the expected ratio - „speckle divergence“ method suggested by T. Esch et al in 2010.
• But there is more than just the width of the distribution!
9
Figure: Lee and Pottier „Polarimetric Radar Imaging“ 2010
1-look intensity data Exponential distribution 1-look amplitude data Rayleigh distribution 4-look amplitude data Chi distribution
Input data
• Sentinel-1 in orbit since April 2014.
• Most common is the Interferometric Wide (IW) swath mode.
• GRD data products available, VV+VH polarisation.
Sentinel-1 IW mode GRDH data
Swath width 250 km
Resolution (rg X az) 20 m X 22 m
Pixel spacing (rg X az) 10 m X 10 m
Equivalent Number of Looks (ENL)
4.9
Incidence angle 29°-45°
10
Source: Sentinel-1 User Handbook
Input data
• Sentinel-1 IW VV/VH GRDH
• Distribution of an homogeneous forest area in Estonia.
• 5 looks, Chi distribution, appears close to Gaussian.
11
Sentinel-1, Oct. 23, 2014, Tallinn and Viimsi in Estonia
12
R: VV, G: VH, B: VV+VH Sentinel-1, European Space agency 2014
13
Sentinel-1, Oct. 23, 2014, Tallinn and Viimsi in Estonia
VV-channel, mean-median Sentinel-1, European Space agency 2014
How we processed?
• Compared against Estonian Building Registry data about Tallinn and the surrounding Harjumaa county.
• Mean-median performance.
• 5x5 window size, 50 m by 50 m on ground.
14
Sentinel-1, VH amplitude and buildings of Tallinn
Sentinel-1, Oct. 23, 2014, Tallinn and Viimsi in Estonia
VV VH Natural areas 90% range -12.7 .. 15.5 -5.8 .. 6.5
Sparse built-up areas inside natural areas range
70% 74%
- lower 10% 9%
- higher 20% 17%
Dense built-up areas inside natural areas range
33% 30%
- lower 20% 18%
- higher 47% 52%
15
Weather in the area: dry, -6° C
Building orientation effects
• Sentinel-1, Oct. 23, 2014, Tallinn and Viimsi in Estonia
16
VV VH
Natural areas 90% range -12.7 .. 15.5 -5.8 .. 6.5
Buildings orientation respect to SAR flight path
90° 45°
90° 45°
Built-up areas inside natural areas range
29% 51% 45% 37%
- lower 23% 14% 15% 19%
- higher 48% 35% 40% 44%
Weather effects
Comparison of two datasets: • Wet conditions:
Jan. 8th, 2015, 5° C, 3.6 mm percip. 6 h prior the data take
• Dry conditions: Dec. 27th, 2014, -6° C, dry
17
Weather effects
18
VV VH
Weather Wet Dry Wet Dry
Natural areas 90% range
-17.9 .. 20.2 -15.1 .. 17.5 -9.15 .. 11.3 -6.88 … 8.01
Built-up areas inside natural areas range
49% 38% 52% 37%
- lower 15% 17% 14% 17%
- higher 36% 45% 34% 46%
Window size considerations
19
VV amplitude
VV amplitude, local_mean-local_median
Ortophoto aboout the area from Estonian Land Board
Conclusion
• Built-up areas could be well extracted from Sentinel-1 IW mode imagery.
• Use images from dry conditions. • Dense built-up areas easier to detect than
detached houses. • Having dual pol. gives rather significant
improvement for detecting buildings at different orientation angles.
20
What is CUDA? • Copernicus Urban
Development Analyser (CUDA) - a complex information system for monitoring urbanisation, related infrastructure and population changes.
• Using Copernicus Sentinel satellite data and anonymous mobile Location Based Services (LBS) data.
21
CUDA
Satellite data for infrastructure
mapping.
Anonymous LBS data for population statistics.
User interface concept
22
Thank you! Questions?
Photo: Villem Voormansik
23
First trials on Sentinel-1 performance for mapping built-up areasOutlineUrbanisation – a global megatrendPercentage of urban population and agglomerations �by size class: 1960, 1980, 2011, 2025Slide Number 5Percentage of urban population and agglomerations �by size class: 1960, 1980, 2011, 2025Percentage of urban population and agglomerations �by size class: 1960, 1980, 2011, 2025How to extract built-up areas �from SAR imagery?How to extract built-up areas �from SAR imagery?Input dataInput dataSentinel-1, Oct. 23, 2014, �Tallinn and Viimsi in EstoniaSlide Number 13How we processed?Sentinel-1, Oct. 23, 2014, �Tallinn and Viimsi in EstoniaBuilding orientation effectsWeather effectsWeather effectsWindow size considerationsConclusionWhat is CUDA?User interface conceptThank you! Questions?