3103-450

Spatial data analysis with

Geographical Information
Systems (GIS)
Thilo Streck
Biogeophysics

Landcover mapping
In this lesson you will get to know

popular satellite systems

the characteristics of their sensors

techniques of landcover classification and some

applications

Some properties of satellite

systems

Spatial properties

Spectral properties

Radiometric properties

- Orbit geometry
- Coverage area
- Spatial resolution
- Revisit time
- Number of bands
- Width of bands
- Position of bands

- Radiometric resolution
- Gain setting
- Signal-to-noise ratio

Lillesand et al. (2004)

Orbit of Landsat-4 and -5

Lillesand et al. (2004)

Landsat-1,-3 configuration

Lillesand et al. (2004)

Landsat-1 to -7 missions
Orbit
(days)

RBV

Bands
MSS

1 1972-1978

18

1-3

4-7

2 1975-1982

18

1-3

4-7

3 1978-1983

18

A-D

4-7(8)

4 1982-1987

16

1-4

1-7

5 1984-

16

1-4

1-7

No

Use

6 Launch failure 16
7 199916
8 2013
16

TM

1-7/ETM
1-7/ETM+
1-9/OLI
Lillesand et al. (2004)

Sensors used in Landsat missions I

Sensor
RBV

Mission
1,2

MSS

3
1-5

TM

3
4,5

Sensitivity
(m)
0.475-0.575
0.580-0.680
0.690-0.830
0.505-0.750
0.5-0.6
0.6-0.7
0.7-0.8
0.8-1.1
10.4-12.6
0.45-0.52
0.52-0.60
0.63-0.69
0.76-0.90
1.55-1.75
10.4-12.5
2.08-2.35

Resolution
(m)
80
80
80
30
79/82
79/82
79/82
79/82
240
30
30
30
30
30
120
30

Lillesand et al. (2004)

Sensors used in Landsat missions II

Sensor
ETM+

Mission
7

OLI

TIRS

Sensitivity
(m)
0.45-0.52
0.53-0.61
0.63-0.69
0.78-0.90
1.55-1.75
10.4-12.5
2.09-2.35
0.52-0.90
0.43-0.45
0.45-0.52
0.52-0.60
0.63-0.68
0.85-0.89
1.56-1.66
2.10-2.30
0.50-0.68
1.36-1.39

Resolution
(m)
30
30
30
30
30
60
30
15
30
30
30
30
30
30
30
15
30

10.6-11.2
11.5-12.5

100
100
NASA (2014)

Landsat daily orbit pattern

Lillesand et al. (2004)

Landsat-4 orbit tracks

Lillesand et al. (2004)

Landsat-4,-5 coverage tracks

Lillesand et al. (2004)

RBV system configuration

Lillesand et al. (2004)

MSS system configuration

Lillesand et al. (2004)

Landsat MSS scene, band 5

Lillesand et al.
(2004)

Spectral sensitivity of the

Landsat MSS bands

Lillesand et al. (2004)

Thematic Mapper bands

No

Name

0.45-0.52

Blue

0.52-0.60

Green

0.63-0.69

Red

0.76-0.90

Near-IR

Vegetation types, soil

moisture

1.55-1.75

Mid-IR

Vegetation and soil

moisture, snow-clouds

10.4-12.5

2.08-2.35

Example applications
Coastal water mapping
Vegetation discrimination
Plant species
differentiation

Thermal-IR Vegetation stress analysis

Mid-IR

Rock type discrimination

Lillesand et al. (2004)

Surface water temperature

of Lake Michigan

Lillesand et al. (2004)

SPOT off-nadir viewing range

Lillesand et al. (2004)

SPOT stereoscopic imaging

Lillesand et al. (2004)

Sensor characteristics of SPOT-5

Sensor

Band
(m)

Resolution
(m)

Swath
(km)

panchromatic

2.5/5

60-80

B1, B2, B3

10

B4

20

HRS

panchromatic

5/10

120

Vegetation

B0, B2, B3, B4

1000

2250

HRG

B0: blue; B1: green; B2: red; B3: near-IR; B4: Mid-IR
Lillesand et al. (2004)

SPOT-4 vegetation images

Lillesand et al. (2004)

Summary of sensing systems

Lillesand et al. (2004)

Some more satellite systems

Name
IRS
RESURS
ADEOS
CBERS
EO
IKONOS
EROS
Quickbird
Orbview

Launch
1988
1985
1996
1999
2000
1999
2000
2001
2003

Country, remarks
India
SU
Japan
China, Brazil
USA, hyperspectral
Commercial
Commercial
Commercial
Commercial
Lillesand et al. (2004)

Selected
high-resolution
satellite systems

Lillesand et al. (2014)

The RapidEye satellite system

Options To Suit Your Individual Needs

Depending on the task at hand, two different levels of RapidEye Standard Image Products will be commercially available.

Level
1B
3A

Description
RapidEye Basic Product Radiometric and sensor corrections applied to the data.
On-board spacecraft attitude and ephemeris applied to the data.
RapidEye Ortho Product Radiometric, sensor and geometric corrections applied
www.rapideye.de
to the data. All products have been rectified using a DTED Level 1 SRTM DEM or
better, and with appropriate ground control can meet an accuracy of 6m 1-sigma (12.7
m CE90). The highest accuracy achieved by these products will meet 1:25,000 NMAS
standards

RapidEye
specification
Standard Image Product
Specification
Spectral Bands

Ground Sampling
Distance (Nadir)
Pixel Size
(orthorectified)
Swath Width
Revisit Time
Equator crossing
time
Image Capture
Capacity

Specification
Blue
Green
Red
Red Edge
NIR
6.5 m

440-510 nm
520-590 nm
630-685 nm
690-730 nm
760-850 nm

5m
77 km
DAILY
11:00 a.m. (approximately)
4 Million km DAILY
www.rapideye.de

5m Pixels

How Can RapidEye Work With You?

RapidEyes international team of specialists are ready to assist by providing any combination of products, services or
consulting solutions. Contact RapidEye today and find Sales Representatives ready to meet and discuss what solutions fit
best your needs.

Approximate costs of
satellite data

Get started now to create a competitive advantage and new possibilities for you and your organization.

Data
Landsat (MSS)

Euro/km2
0.01

RapidEye AG
Molkenmarkt 30
14776 Brandenburg an der Havel
Germany

Landsat (TM)

0.12

SPOT (HRV XS)

0.50

Phone:
Fax:
Toll Free (US):
Email:

+49 3381 8904 555

+49 3381 8904 101
1 800 940 3617
sales@rapideye.de

www.rapideye.de

Lillesand et al. (2004)

2008 RapidEye AG, Germany.

All rights reserved. 002 - 09/2009

Digital image processing

Rectification and restoration

Image enhancement
Image classification
Data merging and GIS integration
Biophysical modeling

after Lillesand et al. (2004)

Spectral reflectance curves

s: snow; c: cloud; rw, uw: ripe, unripe wheat; l: limestone;

ds, ms: dry, moist soil; tw, cw: turbid, clear water.

Rees (2004)

Summary of sensing systems

Lillesand et al. (2004)

Idea of vegetation indices

R

NIR

0.68

0.13

s: snow; c: cloud; rw, uw: ripe, unripe wheat; l: limestone;

ds, ms: dry, moist soil; tw, cw: turbid, clear water.

Rees (2004)

Vegetation indices
Difference Vegetation
Index

DV I = N IR
N IR
RV I =
N IR
R
N DV I =
SW
NIR
IR
+
N DW I =
NSW
IR IR
+ SW
+ IR
GV I =
R + M IR

Ratio Vegetation Index

Normalized Difference
Vegetation Index
Normalized Difference
Wetness Index
Green Vegetation
Index

Schultz & Engman (2000)

NDVI of Europe

1-Week Composite
04.07.2005 - 10.07.2005
DLR, Germany

NDVI of Europe

1-Week Composite
14.04.2014 - 20.04.2014
DLR, Germany

Green vegetation fraction

119

et al. / Agricultural and Forest Meteorology 206 (2015) 113123

GVF - ground measurement

Silage Maize

1.0 Gutman & Ignatov approach

GVF=1.11NDVI-0.06
EF=0.86
0.8
0.6
0.4
0.2
GVF=1.15NDVI-0.16
EF=0.91

0.0
0.0

0.2

0.4

0.6

0.8

1.0

NDVI RapidEye
Relationship
measured
GVF vegetation
and NDVI calculated from RapidEye
GVF: fractionFig.of6.ground
areabetween
covered
by green
Imukova et al. (2015)

225

images. The black line shows the relationship between both variables following the
Gutman and Ignatov (1998) approach. The red line shows the linear regression of
GVF from ground truth measurements on NDVI calculated from RapidEye images.
(For interpretation of the references to color in this figure legend, the reader is
referred to the web version of this article.)

250

on photographs taken
gau in 2012 and 2013.
legend, the reader is

Green vegetation fraction

on Kraichgau study sites

of the two-point approach (EF = 0.91). Regression results were, of

course, non-biased, RMSE was 0.12. Fig. 6 compares the result of the
regression analysis and the two-point approach. The linear regression was
validated against the second subset that had not been used
K. Imukova et al. / Agricultural and Forest Meteorology 206 (2015) 113123
in the regression (see Section 2.3.1.). The statistical criteria for the
regression equation (EF = 0.91, RMSE = 0.11, bias = 0.008) were sim2012
ilar to those of the calibration data. The regression equation was
2013
used to generate the gridded GVF data of the Kraichgau region. We
repeated the linear regression crop-wise. Mean slopes and intercepts are summarized in Table 3. Based on Students t-test, the
regression parameters of maize were significantly different from
those of the winter crops, while the differences between those of
the latter were not ( = 0.05).
The high intercepts in case of winter wheat and winter rape is
a result of senescence. NDVI of senescent plants was higher than
NDVI of bare soil, while GVF for both of them equals to zero. Shortly
before harvest, both crops lost their chlorophyll, thus, the greenness
cover was about zero and the soil was covered with dry vegetation
like it is shown on the photographs of Figs. 2 and 3 on DOY: 195 and
202 in 2012 and DOY: 204 in 2013. In this case, the average NDVI of
the winter wheat was 0.25 0.01 and that of the winter rape was
apidEye image
(RGB) of the image,
central study
site resolution
Katharinentalerhof taken on June 30, 2012 [DOY: 182]. (b) Green vegetation fraction (GVF) map derived from the
RapidEye
5m
0.23 0.01.
The mean NDVI of bare soil of sample plots remained
Eye image. The more green the pixel, the higher the GVF. Bluish pixels indicate areas almost void of vegetation.
the same in all three fields in both years. It was about 0.13 with a
standard deviation around 0.01.

Imukova et al. (2015)

Fig. 11. Comparison of ground truth data with satellite data for the municipality
will have a developed ground-covering canopy only as late as in
Sinsheim. The black line belongs to the GVF dynamics which is used in the NOAH
June or July. Leaves of these crops are still green in August and
model by default. In this case, GVF was derived from satellite images with low resSeptember, while the early covering crops are already harvested.
olution (15 15 m2 ) what leads to mixed GVF pixel values because the pixel cover
Both phenological types show very pronounced differences with
different land-use types. The red line is the areal weighted average GVF which was
regard to the Bowen ratio or rather energy partitioning in the
calculated with account of spatial distribution of croplands for the municipality Sinsheim. (For interpretation of the references to color in this figure legend, the reader
course of the season (Wizemann et al., 2014). Fig. 10 shows the
is referred to the web version of this article.)
GVF dynamics of the early- and late-covering crops based on the
7. (a) RapidEye image (RGB) of the central study site Katharinentalerhof taken on June 30, 2012 [DOY: 182]. (b) Green vegetation fraction (GVF) map derived from the
Gaussian fits shown in Fig. 9. The early-covering crops have quite
RapidEye image. The more green the pixel, the higher the GVF. Bluish pixels indicate areas almost void of vegetation.
high GVFs in April and May and reach their maxima in June. In July,
the default settings of NOAH-MP LSM (Fig. 11). The land use data
the GVF declines strongly until it is in the range of 0.1 in August
were taken from the Statistical Office of the Federal State of Badenand September. In contrast, the late-covering crops have low GVF
Wrttemberg (http://www.statistik.baden-wuerttemberg.de) as
in April and May, while the maximum is reached in July, and it
described above. In the default settings of NOAH-MP LSM, GVF valstays relatively constant until September.
ues are taken from the NESDIS/NOAA satellite. The NESDI/NOAH
Based on the modal values of the Gaussian fits, we derived the
data have a resolution of 15 15 km2 . Due to the mixing of cropGVF dynamics of early- and later covering crops for the Kraichlands, forest and urban areas in practically all pixels, the overall GVF
gau region (Fig. 10). These satellite-derived GVF dynamics were
is
too high. Moreover, the seasonal dynamics are strongly smoothed
close to the ground measurements. In the NOAH-MP LSM, crop(Fig. 11) in comparison with the actual GVF dynamic.
lands of our study region would be classified as one land cover type

Mapping GVF in Kraichgau

Histogram of GVF map

GVF
Areal average
of GVF

(Dryland cropland and pasture). Our results imply that merging early- and late-covering crops in regional climate simulations
into one land-use type as recommended in the NOAH, LSM is an
oversimplification, because they will show distinct differences with
regard to energy and water fluxes over the vegetation period.
For the municipality of Sinsheim, we calculated based on the
ground measurements the area-weighted average GVF dynamics
on the basis of varying land use and compared with that of used in

24 May 2012

1.0

early-covering crops

0.8
0.6

late-covering
crops

0.4
0.2

5. Conclusions
For the first time, we presented high-resolution data (5 5 m2 )
on the GVF dynamics of croplands derived from RapidEye satellite
data at the field and regional scale. GVF is highly variable in time
and space. As to be expected, its seasonal dynamics are affected
by the prevailing weather conditions. At the field scale, the variability of GVF depends, among others, on management (drilling,
soil tillage etc.). At the regional scale, the variability was impacted
by the distinct differences in the phenological development of
crops. The observed bi-modal distribution of GVF suggests to distinguish at least between early- and late-covering crops (e.g., winter
wheat vs. silage maize), because seasonal dynamics of energy- and
water exchange between land surface and atmosphere are very
different between these. Concomitantly, we conjecture that splitting the land uses class croplands as used by NOAH-MP and
other land surface models into two classes (early-covering vs. latecovering crops) have the potential to improve simulations of land
surface processes in heterogeneous agricultural landscapes.

0.0

Acknowledgments
Apr
May
Jun
Jul
Aug
Sep
8. Map of the green vegetation fraction (GVF) of Kraichgau region calculated from RapidEye data on 24 May 2012. The inlet shows
linepresent study was supported by the German Research
Monththe histogram of the GVF. The redThe
ates the GVF used by the NOAH-MP land surface model in large-scale simulations.
Foundation (DFG) in the framework of the Research Unit (RU)
Fig. 10. Seasonal dynamics of the green vegetation fraction (GVF) of early covering
1695 Structure and function of agricultural landscapes under
crops (red line) and the later covering crops (green line) in Kraichgau region. Early
global climate change processes and projections on regional
covering crops are, e.g., winter wheat, winter rape, and summer barley. Late covering
scale and in part by Erasmus Mundus grant SGA 2010-2361. The
crops are mainly maize and sugar beet. (For interpretation of the references to color
authors acknowledge the German Aerospace Center (DLR) for the
this figure
legend, the
referred to the
webof
version
this article.)
High-resolution 2D GVF fields
2012,in we
observed
inreader
theiseastern
part
the ofEC2
field (silage maize)

Imukova et al. (2015)

1. Field scale
ig. 7 shows, by way of example, a RapidEye scene of the censtudy site acquired on 30 June 2012 and the derived GVF map.
bright green color indicates areas with a high green vegetation
er. The dense blue color is related to bare soil or sparsely veged areas. Buildings and asphalt roads were not excluded from
scene and show up in the graph as bluish pixels. During both
etation periods, we observed in most cases that the distribuof GVF in each field was bell-shaped (data are not shown).
most homogeneous spatial distribution of GVF was observed
C3. In June 2012, the distribution of the green cover was norat EC3 ( = 0.05). In 2012, we observed at EC1 (winter rape),
er contiguous zones of lower and higher GVF. In the north of
field, there was a strip with GVF lower than in the southern
of the field, which resulted in a bi-modal distribution. In June

a northsouth strip of lower GVF, while in July and September 2012,

the GVF distribution had a bell shape at this field. These differences
were mostly related to differences in management (see Section 4).

Image classification for

landcover mapping
3.3.2. Regional scale

At the regional scale, the GVF data showed bi-modal distributions (Figs. 8 and 9). In the scene of 24 May 2012, for example, the
histogram shows two clearly distinct peaks in the low (blue color)
and the high (green color) range. The dense dark blue color covers
around 17% of the total area. Highest GVF values (80100%; dark
green) contribute 38% of the total area. At 28% of the total area, GVF
was between 0.6 and 0.8. To each histogram, we fitted a bi-modal
Gaussian distribution (red and green lines in Fig. 9). In nearly all
months, both distributions were clearly separated from each other.
Only in June, both distributions distinctly overlapped.

Supervised classification

(first identify, then classify)

Unsupervised classification
(first classify, then identify)

Hybrid classification

Supervised classification

Lillesand et al. (2004)

Observation and minimum-distanceto-means classification

Lillesand et al. (2004)

Parallelepiped classification

Lillesand et al. (2004)

Maximum likelihood classification

Lillesand et al. (2004)

Unsupervised classification

C: Conifers
D: Deciduous trees
S: Stressed trees

Lillesand et al. (2004)

Classification error matrix

Lillesand et al. (2004)

Identification of field crops using

multitemporal RapidEye data
RapidEye
By use of the temporal interaction potential (Equations 5-9)
these results are set in temporal context and the overall accuracy
is increased to 84.2%. Results of the CRFmulti-classification and
the corresponding reference can be seen in Figure 5 e) and f).

7. CONCLUSION
Reference

InCRF
this
work we presentedCRFtwo
CRF-based approaches for
SVM-classification
all-classification
multi-classification
crop other
type analysis and tested them on RapidEye
Legend: grassland corn winter crop multitemporal
rapeseed root crops
data of 4 epochs. Even with using just very few simple features,

Confusion
matrix accuracy of far over 80% for six
we achieved a classification

Figure 3. Reference data and classification results in comparison

6.2 Description of manual reference

grassland

A manual reference is used that was made in the same year 2010
like the satellite images. It consists of 121 separate fields of an
area of about 322ha and has been acquired by field walking.
The portion of the individual crop types of the whole area is
12% grassland, 21% corn, 28% winter crop, 11% rapeseed,
11% root crops and 17% other crops. In the process an existing
GIS was used to define the borders of single fields. The manual
corn winter
crop
rapeseed
rootsample
cropsforother
reference
builds the
training
and evaluation
the test
classifications.

Figure 4. CRF
-classification superimposed by GIS cropland
6.3multi
CRF vs. SVM
object borders (black borders)
6.5

For our tests we applied the cross-validation method by

learning sample into two equal parts. Tables 1-4
Detailedseparation
results the
of CRF
show the results for multi
the two CRF-approaches and the SVM-

crop
Cla type classes (grassland, corn, winter crop, rapeseed, root
Gra
Cor
Win
Oth
crops and
other
crops)
withRapboth Roo
approaches.
Both of them
Ref
63.2 better
11.1 than
13.8a SVM-classification
7.1
3.2
1.6 that served as a
performed
Gra
benchmark
classic
CRF
0.1 with
90.6the 0.1
4.5approach
4.0
0.7 all being slightly
Cor
better
than
CRF
. Nevertheless
2.1
0 multi94.9
0
0 the 2.9CRFmulti approach
Win
generally
higher0.2potential
0.8has a
17.7
81.2 for 0any kind
0.2 of multitemporal
Rap
analysis.
its flexibility
0.3Because
13.5 of 0.1
0
83.5in the
2.7 definition of the
Roo
temporal
potential,
also applicable
for tasks of
1.5interaction
7.4
3.4
0.3 it is 7.1
80.4
Oth
change
detection
multi-scale
analysis..
Table 2.
Confusionor
matrix
for CRFmulti
-classification

Mller
& Hoberg
(2011)
Cor
Win
Rap
Roo
Oth

Cla
Ref
Gra

70.9

0.5

Cor

1.8

76.6

Gra

ACKNOWLEDGEMENT
8.9

12.9

0.3

6.6

0.1

12.3

0.4

8.9

138

U. Schuler et al. / Catena 81 (2010) 137171

Mapping the Mae Sa Mai area

enhance agricultural sustainability and help to resolve misunderstandings or conicts between different stakeholders and authorities.
Consequently, detailed information on soils and their properties are
required, but intensive soil mapping of all of northern Thailand is
very costly and time-consuming. The mountainous areas of northern

Thailand represent a high priority target for the development and

application of alternative quick, cheap, but sufciently accurate
mapping methods.
The main objective of this study was to develop an efcient soil
mapping approach, facilitating the regionalization of soil information

Available data
Terrain data
- Elevation
- Slope
- Curvature
- Aspect
Geological map
Remote sensing data
- Landsat7 (bands 1-8)
- SPOT5 (bands 1-4)

Schuler et al. (2010)

Fig. 1. Physiography of the Mae Sa Mai area (topography based on Royal Thai Survey Department (1976)).

Reference and transect-based soil maps

U. Schuler et al. / Catena 81 (2010) 137171

pp. 141144

pp. 141144

U. Schuler et al. / Catena 81 (2010) 137171

Schuler et al. (2010)

U. Schuler et al. / Catena 81 (2010) 137171

pp. 141144

Maximum-Likelihood approach

Fig. 4. Mae Sa Mai: (A) Reference soil map, (B) Transect sampling-based map (C) Maximum Likelihood-based-soil map, and (D) probability map: Umbrisol.

Probability
of Umbrisol

soil map, (B) Transect sampling-based map (C) Maximum Likelihood-based-soil map, and (D) probability map: Umbrisol.

ML-based map

pp. 141144

Schuler et al. (2010)

U. Schuler et al. / Catena 81 (2010) 137171

Fig. 4. Mae Sa Mai: (A) Reference soil map, (B) Transect sampling-based map (C) Maximum Likelihood-based-soil map, and (D) probability map: Umbrisol.

soil forming process. Accordingly, soils were mainly classied as

Umbrisols comprise less than 1% (Fig. 6).

Maximum-Likelihood approach

Table 2
Error Matrix for transect sampling-based map resulting from classifying training set data in Mae Sa Mai.
Classication
data [ha]

Training set data (known cover types) [ha]

Acrisol

Anthrosol

Cambisol

Chernozem

Gleysol

Leptosol

Technosol

Umbrisol

Water

row

Acrisol
Anthrosol
Cambisol
Chernozem
Gleysol
Leptosol
Technosol
Umbrisol
Water
column

885.7
0.0
0.1
0.0
0.0
0.0
0.0
0.8
0.0
886.6

0.1
1.5
0.0
0.0
0.0
0.0
0.0
0.0
0.0
1.6

14.8
0.0
75.8
0.0
0.0
0.0
0.0
6.5
0.0
97.1

0.3
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.3

0.4
0.0
0.3
0.0
0.0
0.0
0.0
0.0
0.0
0.6

0.0
0.0
0.7
0.0
0.0
5.4
0.0
0.0
0.0
6.1

0.0
0.0
0.0
0.0
0.0
0.0
16.0
0.0
0.0
16.0

19.8
0.0
0.0
0.0
0.0
0.0
0.0
21.6
0.0
41.4

0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
2.6
2.6

921.1
1.6
76.8
0.0
0.0
5.4
16.0
29.0
2.6
1052.4

Type 1 error [%]

Acrisol
Anthrosol
Cambisol
Chernozem
Gleysol
Leptosol
Technosol
Umbrisol
Water

Type 2 error [%]

3.8
2.2
1.4

0.0
0.0
25.4
0.0

Acrisol
Anthrosol
Cambisol
Chernozem
Gleysol
Leptosol
Technosol
Umbrisol
Water

0.1
8.0
22.0
100.0
100.0
11.9
0.0
47.8
0.0

Error ratings: very low: 020, low: 2040, medium: 4060, high: 6080, very high: 80100.
Overall accuracy: ((885.7 + 1.5 + 75.8 + 0.0 + 0.0 + 5.4 + 16.0 + 21.6 + 2.6) / 1052.4) 100 = 95.8%.

Type 1 error (complementary to users accuracy):

Percentage of pixels assigned to a given soil category although they belong into another soil category
Type 2 error (complementary to producers accuracy):
Percentage of pixels belonging to a given soil category but classified as belonging to another soil category

Schuler et al. (2010)

Landsat subscene of Black Forest

1990, Landsat-5 (TM)

2000, Landsat-7 (ETM+)

DLR, Germany

CORINE landcover classification

2000

Change
yellow: shrubs; magenta: industrial areas; red: urban areas DLR, Germany

DLR, Germany

1990