Sentinel-1 for Urban Areas

Comparison between Automatically Derived Settlement Layers from Sentinel-1

Data and Copernicus High Resolution Information Layers

Arthur Lehner

1,2

, Vahid Naeimi

and Klaus Steinnocher

AIT – Austrian Institute of Technology, Energy Department, Giefinggasse 2, 1210 Vienna, Austria

Department of Geoinformatics –Z_GIS, University of Salzburg, Schillerstraße 30, 5020 Salzburg, Austria

Department of Geodesy and Geoinformation, Vienna University of Technology, Research Group Remote Sensing E120.1,

Gußhausstraße 27–29, 1040 Vienna, Austria

Keywords: Remote Sensing, Sentinel-1, Impervious Surface, Copernicus.

Abstract: This work deals with the development of automatically derived settlement layers from Sentinel-1 data. The

produced layers allow differentiation between built-up and non-built-up area. The results are visually

compared with a standardized product of the Copernicus earth observation program, the Copernicus High

Resolution Layer Imperviousness Degree. For evaluation of the accuracy, the European Settlement Map 2016

was chosen as a reference data set. In this study the ISODATA unsupervised classification algorithm is used

for generating two layers that allow differentiation between built-up and non-built up area. The results reveal

the suitability of Sentinel-1 data for urban areas mapping. The quality of the produced settlement layers are

comparable to standardized products that are based on data from optical sensors e.g. Copernicus High

Resolution Layer Imperviousness Degree or European Settlement Map 2016 respectively.

1 INTRODUCTION

Land monitoring within Europe and the European

Union based on Earth Observation systems is

fundamental regarding its scientficial, cultural and

environmental impacts. With the agreement in 1985

to produce the first CORINE (EEA, 2000) land cover

(CLC), the foundation for a European land

monitoring program was laid. The programme

maintains a number of databases including an

inventory of land cover\land use (LCLU), produced

operationally for most areas of Europe on a 6 to 10

year cycle (Ben-Asher, 2013). The CLC program

later was incoporated in the European earth

observation program “Copernicus”. Founded in the

year 1998, Copernicus, previously known as GMES

(Global Monitoring for Environment and Security),

represents the European Programme for the

establishment of a European capacity for Earth

Observation.

Pan-European High Resolution Layers (HRLs)

provide information on specific land cover

characteristics, and are complementary to LCLU

mapping such as in the CORINE land cover (CLC)

datasets. The HRLs are generated from 20 m

resolution satellite imagery through a combination of

automatic processing and interactive rule based

classification (Sannier et al., 2016). Out of five layer

themes, the imperviousness layer presents the degree

of sealed soil as it captures the spatial distribution of

artificially sealed areas, including the level of sealing

of the soil per area unit. The layer represents in this

way a base map for various fields of research

(Ciobotaru et al., 2016; Lefebvre et al., 2015; Mücher

et al., 2015). Different applications using the

Copernicus Imperviousness Layer (Hennig et al.,

2015; Steinnocher et al., 2011) reveal its basic

necessity and usability.

Another available data product is the European

Settlement Map (ESM) 2016 that represents

percentage of built-up area coverage per spatial unit

using SPOT5 and SPOT6 satellite imagery from the

year 2012 (European Commission, 2017). The ESM

is a map expressing the proportion of the pixel area

covered by buildings, and it was produced in

2013/2014 (Florczyk et al., 2016). In addition to these

developments, the European commission recently

announced the provision of an information layer on

built-up presence. This information layer will be

Lehner, A., Naeimi, V. and Steinnocher, K.

Sentinel-1 for Urban Areas - Comparison between Automatically Derived Settlement Layers from Sentinel-1 Data and Copernicus High Resolution Information Layers.

DOI: 10.5220/0006320800430049

In Proceedings of the 3rd International Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2017), pages 43-49

ISBN: 978-989-758-252-3

derived from Sentinel-1 image collections (2016) and

resolution will be about 20 meters. Since these data

has not been available yet, it could not be considered

for a visual comparison nor an accuracy assessment

in this study.

By launching the Sentinel mission in 2014, the

European Space agency ESA aimed to satisfy the

need of the Copernicus program. Sentinel-1 is the first

of five missions that ESA developed for the

Copernicus initiative. Sentinel-1 comprises a

constellation of two polar-orbiting satellites

(Sentinel-1A, Sentinel-1B), operating day and night

performing C-band synthetic aperture radar imaging,

enabling them to acquire imagery regardless of

weather conditions or light conditions (D’Aria et al.,

2016).

This work presents the comparison of Sentinel-1

data used for the development of two automatically

derived settlement layers differentiating between

built-up and non-built up area and the Copernicus

high-resolution layer ‘Imperviousness Degree’ and

the European Settlement Map (ESM) 2016. In

contrast to CLC data that are confined for Europe,

satellites of the Sentinel mission collect data globally.

The Sentinel-1 data coverage makes it possible to

establish a global settlement layer.

2 DATA

This study uses Sentinel-1 image data, collected from

the first 7 months of the year 2016, the Compernicus

HRL imperviousness for the year 2012 and European

Settlement Map 2016. Additionally a Sentinel-2A

scene (date of acquisition: 02.07.2016) is used for

visual interpretation of the results.

2.1 Sentinel-1 Data

The Sentinel program is the most comprehensive and

ambitious European Earth Observation program. The

Sentinel satellites provide unique operational sensing

capabilities across the whole measurement spectrum,

covering a broad range of applications. Thanks to

their advanced sensing concepts and outstanding

spatio-temporal sampling characteristics, the Sentinel

satellites will collect more data than any earth

observation program before (Attema et al., 2007).

The first of the Sentinel satellite series, Sentinel-1A

was launched on 3 April 2014. Seninel-1 (S-1) is a

Synthetic Aperture Radar (SAR) mission for ocean

and land monitoring. S-1 is the continuity mission to

the SAR instruments flown on board of ERS and

ENVISAT. The S-1 mission is implemented through

a constellation of two satellites. The S-1B was

launched on 25 April 2016. The S-1 data over the land

masses are mainly acquired in Interferometric Wide

swath (IW) mode. The S-1 Level-1 Ground Range

Detected (GRD) products, which are suitable for the

most of the land applications, consist of focused SAR

data that has been detected, multi-looked and

projected to ground range using an Earth ellipsoid

model such as WGS84. The IW GRD products are

provided in two High (20 m x 22 m) and Medium (88

m x 87 m) spatial resolutions resampled to 10 m and

40 m pixel spacing grids respectively (European

Space Agency, 2013, p. 1).

Despite all corrections from Level-0 up to Level-1

data, the GRD data still need to be processed further

before generating level-2 products. The S-1 Level-1

GRD data used in this study were pre-processed using

the TU Wien SAR Geophysical Retrieval Toolbox

(SGRT) (Naeimi et al., 2016). The pre-processing

workflow include calibration, noise removal,

georeferencing and terrain correction using a Digital

Elevation Model (DEM), shadow mask generation,

data conversions, and data resampling and tiling to a

regular grid using an appropriate cartographic map

projection. For the calibration, georeferencing and the

terrain correction, the ESA’s Sentinel-1 toolbox

(S1TBX) is employed. The S1TBX operators are

called via SGRT to perform the georeferencing using

the S-1 precise orbit files provided externally by ESA.

In this study the S1TBX Range Doppler algorithm

and SRTM digital elevation data are used for terrain

correction of the SAR scenes. After some further

preprocessing steps like thermal noise removal, data

format conversion and shadow mask generation the

geocoded SAR scenes are resampled to the TU Wien

Equi7 Grid. The TU Wien Equi7 Grid is designed to

minimize the oversampling rate of the high resolution

satellite data globally, while keeping its structure

simple (Bauer-Marschallinger et al., 2014). After the

pre-processing step, the S-1 backscatter time series

were used to generate composites of monthly mean of

backscatter for each polarization separately over the

test site. In this study high resolution S-1 image stacks

collected from 7 months (January – July) of the year

2016 are used.

2.2 Copernicus High Resolution Layer

Imperviousness Degree

The HRL imperviousness is produced using an

automatic algorithm based on calibrated NDVI.

Similar to other HRLs of the Copernicus program, the

imperviousness HRL is derived from 20 m resolution

optical satellite imagery. The layer has 20 m

GISTAM 2017 - 3rd International Conference on Geographical Information Systems Theory, Applications and Management

geometric resolution and provides 101 classes of

imperviousness while:

 0: all non-impervious areas

 1-100: imperviousness values

 254: unclassifiable (no satellite image available,

or clouds, shadows, or snow)

 255: outside area (Langanke, 2013)

Figure 1: Coverage of the Copernicus HRL Imperviousness

degree, red = impervious surface, grey = no satellite image

available.

It is produced in a three years cycle and covers all

28 EU members (including United Kingdom) and 11

additional countries (Figure 1). The imperviousness

layer contains two products: a 2012 status layer

(degree of imperviousness 2012), as well as an

imperviousness density change layer (2009-2012),

based on the existing imperviousness product for

2009 (Langanke, 2013). In the presented study, we do

not consider the imperviousness density change layer.

2.3 European Settlement Map 2016

The European Settlement Map is a spatial raster

dataset with 10 meters resolution; an aggregated

version with 100 meters resolution is also available.

The ESM is mapping human settlements in Europe

based on SPOT5 and SPOT6 satellite imagery from

the year 2012. Similar to the Copernicus HRL it

covers all 28 EU members (including United

Kingdom) and 11 additional countries. For the ESM

a 95% accuracy for the built-up class is stated

(European Commission, 2017). Based on the

accuracy and its other properties, the ESM was

chosen as reference data for the accuracy assessment

of the automatically derived settlement layer (S-1

USC II).

2.4 Study Area

The selected test site is the city of Vienna

(coordinates: 48°12′N 16°22′E), the capital of

Austria. The terrain ranges from hilly in the west to

flat in the east of the city. Forest areas predominate

the west and agricultural areas are mainly found in the

east and southeast. The urban spectrum ranges from

single-family houses to high-rise buildings, covers

green houses and industrial areas.

3 METHODOLOGY

The objective of this study is to automatically derive

two settlement layers using Sentinel-1 image stacks.

The chosen method for this study is the ISODATA

unsupervised classification algorithm. The

ISODATA unsupervised classification algorithm

clusters many-variable data around points in the

data's original high- dimensional space and by doing

so provides a description of the data (Ball and Hall,

1965). In other words, the cluster analysis groups data

into objects – in our case pixel values – based only on

information found in the data that describes the

objects and their relationships. The objective is that

the pixel values within a group be similar (or related)

to one another and different from (or unrelated to) the

pixel values in other groups. The greater the similarity

(or homogeneity) within a group and the greater the

difference between groups, the better or more distinct

the clustering (Tan, 2006). In our case, a complete

clustering is performed since every pixel value is

assigned to a cluster.

The ISODATA unsupervised classification is

performed on a single image composite representing

the monthly mean of backscatter for a particular

polarization. Parameters used for the unsupervised

classification were:

 no. of iterations: 20,

 max. no. of clusters: 100,

 initial no. of clusters: 10,

 min. cluster size: 10 pixels.

This procedure is repeated for a stack of 5, 9 and 17

S-1 image composites. A final ISODATA

unsupervised classification is performed again on 17

S-1 image composites but changing the number of

iterations from 20 to 80 (S-1 USC). The result of the

classification/clustering does not lead to the creation

of a layer consisting of two classes, e.g. built-up and

Sentinel-1 for Urban Areas - Comparison between Automatically Derived Settlement Layers from Sentinel-1 Data and Copernicus High

Resolution Information Layers

non-built-up area, but instead an image layer is

created, where values correspond to a cluster ID. In

this way all pixels having the same value belong to

the same class/cluster (eCognition Developer 9.2,

2016).

This allows distinguishing of the pixels that have

values with high variance (non-built-up) from those

with low variance (built-up). Depending on the

parameters and the input S-1 image composites,

different numbers of clusters are generated. The

resulting clustered pixels are visually assigned to

classes “non-built up”, “impervious” and “built-up”.

The interim results are compared with each other

and the classification results are compared with the

Copernicus HRL Imperviousness degree of the year

2012. In order to classify and to quantitatively

compare the results from the clustering process with

the European Settlement Map 2016, a threshold is

definded. A Multi-Threshold Segmentation algorithm

is chosen to split the domain based on pixel values

(eCognition Developer 9.2, 2016). Image objects are

created and classified as “built-up”- and “non-built-

up” area respectively (Figure 6). A final layer (S-1

USC II) is created. Further the ESM layer is

aggregated to a 20 meters’ resolution. 500 points

randomly distributed across the study area are used to

assess the accuracy of the classes “built-up” and

“non-built-up”. The authors are aware of the fact that

the comparison of layers from different acquisition

years can be seen critically. However, the areas that

were selected for this study did not face severe

changes over the last 10 years. Furthermore, changes

in the buildings type (e.g. loft/attic conversion) within

the urban landscape do not impact the comparison.

4 RESULTS

The objective of this research was to enhance the

differentiation between built-up and non-built-up

areas by including more S-1 image composites in the

analysis (Figure 2). In Figure 2., steady values are

represented by light gray while variant values are

represented by dark gray values.

By adding more input S-1 images the separation

between the built-up area class, impervious surface

and non-built area become more visable. The

ISODATA unsupervised classification (ICA) was

performed on an image stack consisting of 17 S-1 data

composites and resulted in a layer (S-1 USC) of 39

clusters (1-38). Each cluster was then assigned to a

certain class (built-up, impervious surface, non-built-

up) with a particular degree of imperviousness or

perviousness respectively using an optical Sentinel-

2A image as reference.

In Figure 3, the

produced layer S-1 USC and the

Copernicus HRL are visually compared. Figure 4 and

Figure 5 allows the comparison of the ISO-Clustering

results, the Copernicus HRL and the optical image

from Sentinel-2A. The results of the Multi-Threshold

Segmentation (S-1 USC II) are quantitatively

compared with the European Settlement Map 2016

(European Commission, 2017) that was used as

reference data. (Figure 6). The overall accuracy (OA)

of the final layer (S-1 USC II) is 78,2% (Table 1).

Table 1: Accuracy Assessment (S1- USC II).

Classification

Non Built-up Built-up

Totals

Non Built-up 272 69 341

Built-up 40 119 159

Totals

312 188 500

Producers

Accuracy

87,2% 63,3%

OA:

78,2%

Users Accuracy 79,8% 74,8%

Figure 2: Increasing of the contrast between steady (light

gray) and variant backscatter values (dark gray) by adding

more layers to the analysis. a) ISODATA unsupervised

classification (ICA) on 1 layer, polarization VH; b) ICA on

5 layers, mixed polarization (VH/VV); c) ICA on 9 layers,

mixed polarization (VH/VV); d) ICA on 17 layers, mixed

polarization (VH/VV).

GISTAM 2017 - 3rd International Conference on Geographical Information Systems Theory, Applications and Management

5 DISCUSSION

The results from the ISODATA unsupervised

classification reveal the possibility of mapping built-

up areas. In the produced layer S-1 USC, different

non-built up classes like urban parks, forest,

cemeteries and other spacious vegetated areas are

grouped and can particularly be distinguished from

built-up areas. Agricultural fields are also related

more acurately to non-built up areas, independent

from their degree of vegetation cover. Furthermore,

water bodies are classified as non-built-up areas

independent from its amount of algae, depth, water

quality or ground conditions. The “separation”

between built-up and non-built up areas is improved

by using more number of S-1 image composites.

Figure 3: Visual comparison of S-1 USC and Copernicus

HRL, Vienna, Austria 1) Sentinel-2A image, “true color”

image, R (4) G (3) B (2), 2) S-1 USC, 3) “False Color/Near

Infrared” image from Sentinel-2A, R (8) G (3) B (2), 4)

Copernicus HRL.

In comparison to Copernicus HRL

Imperviousness Degree, the produced layer S-1 USC

shows generally a finer representation of the urban

inventory. Green and (pervious) open spaces are

visually easier to distinguish from built-up areas such

as buildings, bridges and railroads (Figure 4, Figure

5).

Figure 4: Danube River, Vienna, and crossing bridges

depicted in (1) S-1 USC, (2) Copernicus HRL (some of the

bridges captured by S-1 are missing in this image), (3)

Sentinel-2A “true-color” image.

The shape of buildings appear sharper. The

outlines of rivers and channels are more exact and

bridges are depicted as built-up area in detail (Figure

5).

Figure 5: Schönbrunn palace and surrounding castle

grounds (1) S-1 (2) Copernicus HRL (3) Sentinel-2A “true-

color” image.

Sentinel-1 for Urban Areas - Comparison between Automatically Derived Settlement Layers from Sentinel-1 Data and Copernicus High

Resolution Information Layers

In comparison to European Settlement Map 2016,

bridges crossing the Danube River are represented in

the final produced layer S-1 USC II. Urban structures

such as courtyards or street canyons are represented

in greater detail within the ESM layer (Figure 6). This

richness of detail leads back to fact that the ESM 2016

itself has a higher geometric resolution. The data set

used for compiling the ESM 2016 has a higher

geometric resolution (SPOT 5 and SPOT 6 data of

2.5m pixel size) and includes use of ancillary datasets

(e.g. OpenStreetMap) (Florczyk et al., 2016).

Figure 6: Vienna Downtown, Comparison of (1) ESM

2016, Copyright European Commission, European

Settlement Map 2016, (2) ESM 2016 aggregated to 20 m

and (3) S-1 USC II, (4) Overlay between ESM 2016 and S-

1 USC II.

Although the results are overall promising, some

challenges remain regarding the accuracy of the

classification. High pixel values that indicate built-up

area are found within forest areas, mainly in steep

slopes. We attribute this to the fact that the

backscatter measured by S-1 sensor is influenced by

the terrain. Such areas with complex topography

could be masked using a digital elevation model.

Alternatively, the introduction of an object-based

approach allows for defining certain areas as

unsuitable for built-up area.

6 CONCLUSIONS

This study could demonstrate the value of Sentinel-1

data for mapping built-up areas. The preliminary

results revealed that pixel based unsupervised

classification of S-1 stacked backscatter composites

allows to differentiate between built-up and non-

built-up area within the urban landscape. The quality

of the automatically derived settlement layers (S-1

USC and S-1 USC II) is comparable to standardized

products based on optical sensors e.g. Copernicus

HRL Imperviousness Degree or European Settlement

Map 2016 respectively. The S-1 data available for this

study were limited to a half year from 2016.

Differentiation between built-up and non-built-up

area based on the presented methodology can be

enhanced by using more data covering the full year or

multiple years.

The introduction of object-based methodologies

where classification will not only depend on single

pixel values but also on relation between objects may

help to correct misclassifications. Certain terrain may

be defined as inappropriate for built-up area or

distance to other built-up area may be (more) relevant

for classification. Finally, another proposal for a

solution is the use of additional input data (composite

monthly mean single band images) in order to

enhance the differentiaton between built-up area (low

variance) and non-built-up area (high variance).

REFERENCES

Attema, E., Bargellini, P., Edwards, P., Levrini, G., Lokas,

S., Moeller, L., Rosich-Tell, B., Secchi, P., Torres, R.,

Davidson, M., 2007. Sentinel-1-the radar mission for

GMES operational land and sea services. ESA Bull.

131, 10–17.

Ball, G.H., Hall, D.J., 1965. ISODATA, a novel method of

data analysis and pattern classification. DTIC

Document.

Bauer-Marschallinger, B., Sabel, D., Wagner, W., 2014.

Optimisation of global grids for high-resolution remote

sensing data. Comput. Geosci. 72, 84–93.

Ben-Asher, Z., 2013. HELM-Harmonised European Land

Monitoring: Findings and Recommendations of the

HELM Project. Tel-Aviv, Israel.

Ciobotaru, N., Laslo, L., Matei, M., Muşat, C., Lupei, T.,

Boboc, M., Deák, G., 2016. Mapping Romanian

Wetlands – a Geographical Approach. Presented at the

Conference: 3rd International Conference Water

resources and wetlands, Tulcea, Romania, p. 14.

D’Aria, D., Piantanida, R., Valentino, A., Riva, D., 2016.

Freesar, an innovative SAR data processing framework,

in: Geoscience and Remote Sensing Symposium

GISTAM 2017 - 3rd International Conference on Geographical Information Systems Theory, Applications and Management

(IGARSS), 2016 IEEE International. IEEE, pp. 1214–

1216.

eCognition Developer 9.2, T., 2016. eCognition®

Developer Version 9.2.1 Reference Book.

EEA, E.E.A., 2000. Corine Land Cover. Eur. Environ.

Agency Cph. 163.

European Commission, E. S. M. 2016 (2017) ‘European

Settlement Map’. Copernicus. Available at:

http://land.copernicus.eu/pan-european/GHSL/EU%20

GHSL%202014. Copyright European Commission,

European Settlement Map 2016

European Space Agency, 2013. Sentinel-1 User Handbook,

GMES-S1OP-EOPG-TN-13-0001,

https://sentinel.esa.int/.

Florczyk, A.J., Ferri, S., Syrris, V., Kemper, T., Halkia, M.,

Soille, P., Pesaresi, M., 2016. A new european

settlement map from optical remotely sensed data.

IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 9,

1978–1992.

Hennig, E.I., Schwick, C., Soukup, T., Orlitová, E.,

Kienast, F., Jaeger, J.A., 2015. Multi-scale analysis of

urban sprawl in Europe: Towards a European de-

sprawling strategy. Land Use Policy 49, 483–498.

Langanke, T., 2013. GIO land (GMES/Copernicus initial

operations land) High Resolution Layers (HRLs) –

summary of product specifications, version 7 of 2013-

10-14, European Environment Agency, GIO land team

at the EEA, 2013.

Lefebvre, A., Picand, P.-A., Sannier, C., 2015. Mapping

tree cover in European cities: comparison of

classification algorithms for an operational production

framework, in: 2015 Joint Urban Remote Sensing Event

(JURSE). IEEE, pp. 1–4.

Mücher, C.A., Hennekens, S.M., Schaminee, J.H.J.,

Halada, L., Halabuk, A., 2015. Modelling the spatial

distribution of EUNIS forest habitats based on

vegetation relevés and Copernicus HRL. European

Topic Centre Biological Diversity.

Naeimi, V., Elefante, S., Cao, S., Wagner, W., Dostalova,

A., Bauer-Marschallinger, B., 2016. Geophysical

parameters retrieval from sentinel-1 SAR data: a case

study for high performance computing at EODC, in:

Proceedings of the 24th High Performance Computing

Symposium. Society for Computer Simulation

International, p. 10.

Sannier, C., Gallego, J., Dahmer, J., Smith, G.,

Dufourmont, H., Pennec, A., 2016. Validation of

Copernicus high resolution layer on imperviousness

degree for 2006, 2009 and 2012, in: Proceedings of the

International Symposium on Spatial Accuracy

Assessment in Natural Resources and Environmental

Sciences, Montpellier, France. pp. 5–8.

Steinnocher, K., Köstl, M., Weichselbaum, J., 2011. Grid-

based population and land take trend indicators–new

approaches introduced by the geoland2 core

Information Service for Spatial Planning, in: NTTS

Conference, Brussels.

Tan, P.-N., 2006. Introduction to data mining. Pearson

Education India.

Sentinel-1 for Urban Areas - Comparison between Automatically Derived Settlement Layers from Sentinel-1 Data and Copernicus High

Resolution Information Layers