See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/256307366

Remote sensing and digital image processing

Chapter · January 2002

CITATIONS READS

5 9,208

1 author:

Ulrich Walz
Hochschule für Technik und Wirtschaft Dresden
261 PUBLICATIONS 2,545 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Analyse Landschaftswandel und –planung View project

Wirkungen des Ausbaus von Verkehrsinfrastruktur auf Natur und Landschaft View project

All content following this page was uploaded by Ulrich Walz on 31 July 2014.

The user has requested enhancement of the downloaded file.

 U Walz 283
Today, it is almost impossible to deal with topics in Iandscape ecology in
its broadest sense without the use of data collected by remote sensing. Troll
(1939), a German geographer recognized this fact stating “Familiarity with
ecological connections in the landscape as a result of terrestrial investiga
tions . . . makes it possible . .. to draw boundaries on maps from aerial pictures
and aerial maps“. Maximum use can be made of aerial pietures provided
enough is known about the causal relationships between the different ele
ments (climate, rock forms, water and soil erosion, and plant cover) and their
typical structure within the area.
The advantages of being able to see extensive interrelations from above
arc obvious. Naveh and Lieberman (1994) concluded that: ‘The fields of
remote sensing and information science have a significant role to play in ho
listic landscape evaluation. [. .] High and bw-resolution sensors provide
specific applications for many ecological systems.“

6.3.2 Information from remote sensing data

The information from remote sensing data for landscape ecology comes
from different regions of the spectrum, depending on the backscattering
properties of different surface materials such as soil, vegetation type or areas
of water (Figure 6.3-1) and the sensor used. Green vegetation, due to the
pigmentätion of the leaves, generally absorbs a high level of red and biue
parts of the visible spectrum, hence, vegetation looks green because of a
lower level of absorption of the green spectral range. Leaf surfaces disperse
a high degree of infrared light, which makes this wavelength very suitable
for distinguishing between different vegetation types. A leafthat is dry or ill,
and whose structure has been damaged, can be recognized by means of bw
reflection in near infrared part of the spectrum. There is also a high level of
backscatter of short wave infrared as a result of the bw water content.
The spectral areas mainly used are those of visible light, infrared and
the thermal area. The human faculty of perception lies between 0.4 and
O.7im and the remaining wavelengths can only be recorded and used by
means of remote sensing devices. Digital recording of the radiation reflected
by the earth‘s surface does not take place in a continuous spectrum hut rather
in separate spectral channels which have a width of about 0. 1—0.2im. Most
satellite sensors record in near infrared (NIR = near infrared O.7—O.9m), a
region that can also be recorded by cobor infrared (CIR) aerial pictures.
Whilst visible light consists of near and short wave infrared radiation that
has been reflected, thermal infrared (TIR) results from the characteristic
temperature of an object. Optical sensors can only pick up radiation in spec
tral ranges that are emitted from the earth‘s surface where there is no absorp
tion or dispersion of the radiation by ozone, oxygen, steam, carbon dioxide,

1
284 Chapter 6

methane or nitrogen oxides. Such an “atmospheric window“ lies between 0.3

and 1 jim and there is a second winclow in the thermal infrared range. Clouds
and haze cause non-selective dispersion of the radiation and so appear to be
gray-white.

Figure 6.3-1: Reflection ofselected land zise

categories and spectral bands ofsensors

6.3.3 Remote sensing systems

According to the carrying platform and to the recording technique it is

possible to distinguish between the following systems:
a) airplane systems
— black to white aerial pictures (analogue),
— CIR aerial pictures (analogue),
— digital scanner data, and
b) satellite systems
— photographic imagery (analogue),
— digital scanner data.
Whilst most of the earth observation sensors are passive, i.e. they only
receive solar radiation reflected by the earth‘s surface or heat radiation emit
ted. A satellite in polar orbit can record every point on the earth but it cannot
do so at any given time. As a result, the chances of being able to record in a
cloud-free situation in Central Europe are greatly reduced. Clouds pose a
great problem when using remote sensing in the range of visible light to col
lect information about the earth. lt is often difficult to record enough data for
a particular area when there are no clouds, which makes it difficult when it
 U Waiz 285
cornes to choosing satellite data. Only one to four recordings per year are
usually free of clouds (Kühbauch et al. 1990).
However, radar sensors (Table 6.3-1) are active sensors. They carry a
source of radiation, which is ideal as it is constant, has a defined wave1ngth
and radiates active radar waves.

Table 6.3-1: Operational radar satellites (after Schneider-Sliwa et al. 1999)

satellite operating since spatial resolution (in m) width (in km)

ERS-l/2 SAR 1991/1995 25 100
JERS-1 1992 18 75
Radarsat 1995 10—100 50-500
Envisat 1999 30—1000 56-400

The ground resolution of operational radar satellites lies between 1 8 and

30 m, which makes them interesting for questions in landscape ecology.
However, the radar sensors only deliver information about the three
dimensional structure of the surface, not about the reflectance of materials.
The data can only interpreted by the heip of mathematical processes and
gives less impressive images than optical sensors (Table 6.3-2).
One of the advantages of using airplanes as a platform lies in the fact
that it is possible to react quickly and flexibly to different weather situations.
In spite of the new high-resolution satellite sensors, the resolution of air
borne remote sensing data is not yet reached. Indeed high resolution images,
both aerial and satellite, needs to include a digital elevation model for geo
metric correction.
The main advantage of satellite sensors compared to airborne is that
large areas can be covered, which makes the time-consuming task of proc
essing each individual aerial picture and putting them together bit by bit un
necessary. Furthermore, there is no need to organize individual flights,
which are also expensive.
Continuous recording of data by remote sensing from space started with
Landsat—NASA satellites. Landsats 1—3 had a resolution of 60m. The latest
Landsat 7 manages a resolution of 1 5m, using the newly added panchromatic
channel. The three channels ofthe SPOT Satellites record radiation in visible
and near infrared range with a spatial resolution of 20 by 20m. An additional
panchromatic operational mode has a resolution of 10 by lOm.
Since 1995 the panchromatic sensor of the Indian satellite IRS-1C dcliv
ers a resolution of 5.8m, which is a completely new quality compared to all
of the previous systems. lt was mainly designed to be used for inventories
and planning for the Indian State.
286 Chapter 6
Table 6.3-2: Important optical reinote sensing Systems (B-blue, G-green, R-red, NIR-near
infrared, SWIR-short wave infrared, TIR-thei‘mal infrared, Pan-panchromatic)
satellite operat- recording spectral bands spatial swath- alti- revisit
ing System (im) resolution width tude rate
since (m) (km) (km) (days)
Landsat 1972 MSS, 0.5-06. 79 x 79 185x 930 18
1-3 RBV 0.6-0.7 170
0.7-0.8
0.8-1.1
10.4-12.6
Landsat 1982, MSS, 0.45-0.52(B) 30x30 185x 705 16
4-5 1984 IM 0.52-0.60 (G) 120 x 120 175
0.63-0.69 (R)
0.76-0.90 (NIR)
1.55-1.75 (SWIR)
2.08-2.35
10.4-12.5 (TIR)
Landsat7 1999 IM, 0.5-0.9(Pan) 15x15 183x 705 16
ETM 0.45-0.52(B) 30x30 170
0.52-0.60 (G)
0.63-0.69 (R)
0.76-0.90 (NIR)
1.55-1.75 (SWIR)
2.08-2.35
10.4-12.5 (TIR) 60 x 60
SPOT 1, 1986, HRV 0.50-0.73 (Pan) 10 x 10 60 x 60 830 26
2 und 3 1990, 0.50-0.59 20 x 20
1993 0.61-0.68
0.79-0.89

SPOT4 1998 HRV

1.58-1. 75
0.61-0.68 (Pan) lOxlO 60x60 822 26
1
0.50-0.59 20x20
0.61-0.68
0.79-0.89
1.58-1.75 (SWIR)
IRS 1C 1995 PAN 0.50-0.75 (Pan) 5.8 x ‚8 70 x 70 817 24
IRS 1D 1997 LISS-Ill 0.52-0.59 (G) 23 x 23 142 x
0.62-0.68 (R) 142
0.77-0.86 (NIR)
1.55-1.70 (SWIR) 70 x 70
IKONOS 1999 PAN 0.45-0.90 (Pan) 1 x 1 11 x 11 680 1-3
0.45-0.53 (B) 4x4
0.52-0.61 (G)
0.64-0.72 (R)
1
0.77-0.88 (NIR)
EROS A 2000 PAN 0.50-0.9 (Pan) 1.8 x 1.8 12.5 x 480 1-4
12.5
 U Walz 287

IKONOS (Figure 63-2) has been delivering panchromatic data with a

resolution of im since 1999. lt can be assumed that more systems with this
degree of resolution will be available in the near future and that they will
open new areas in environmental planning and landscape ecology. However,
the price level of the data delivered by these systems is very high and the
advantage of being able to cover extensive areas with satellite data is lost
because ofthe small swath width.

Figure 6.3-2: Panchromatic IKONOS-Image (Saxony SwitzerlandSpace Imaging Europe

SA)

6.3.4 Application of remote sensing data for landscape ecology

Satellite images are primarily suited for purely visual interpretation.

Combined with overlaying geographie information they form an excellent
basis for the illustration of spatial facts. Multi-spectral ciassification of satel
ute data for drawing up maps of surface cover is a (semi-)automated evalua
tion method. Further information that can be gained is interesting as far as
ecology is concerned (for example):
— information on the structure and texture (of landscapes or landscape ele
ments),
— thermal information (on surfaces),
— biomass, e.g. vegetation index or leaf index,
— dynamics, changes in land use, and
288 Chapter 6

—
information on altitude.
Remote sensing can be applied within very different task areas of land-
scape ecology and planning, eg.:
—
urban ecology (identification of urban or landscape structure),
—
biotope mapping,
—
land use and landscape planning,
—
agricultural statistics,
—
analysis of current land use pattems,
—

cartographic processing ofplanning,

—
observation of the effects and, to a limited extent forecasting, of natural
risks (eg. desertification, flooding, landslides, rock falls, wind damage,
fire, water pollution), and
—
observation of forest damages (by air pollution, insect outbreaks, climatic
stress wind, snow, dryness).
Probably the most important area in which data from remote sensing can
be used is that of regional and global monitoring of changes in land use,
vegetation cover and environmental media (see Chapter 4.2). Earth observa
tion systems are most suited for delivering information for such monitoring
purposes as they cover extensive areas. They record regularly and guarantee
comparable information over long periods of time (e.g. from Landsat). lt is
absolutely necessary to consider and investigate the processes which take
place in the different dimensions of space and time from a landscape eco
logical or holistic approach.
In principle, remote sensing data should generally be considered as only
one source of data within a GIS (see Chapter 6.2).
Satellite data must be preprocessed before they can be used for different
applications. Errors in the data caused by geometric distortion as it was re
corded or where atmospheric factors had an effect should be corrected. Fur
thermore, sensor errors should be taken into consideration and corrected
where necessary.
Satellite information must be entered into a standardized system of coor
dinates so that it can be combined with other spatial information. Reference
points that are clearly recognizable and whose coordinates are known are
marked on the satellite image. The data is then converted into a geodetic sys
tem of coordinates by means of polynomial transformation.
The cultural landscape is changing very quickly and so there is a lack of
up-to-date information on land use. Topographical maps often do not contain
the information required, are frequently out of date, and do not distinguish
clearly enough between particular land use forms. lt is possible, by combin
ing high-resolution panchromatic images with multi-spectral data of a lower
U Walz 289

spatial resolution, to achieve a high geometric resolution without losing the

matic information.
By using such a composition and with the use of panchromatic data,
changes in land use can be quickly recognized visually. Very good general
maps can also be drawn up by superimposing other geographic data. Super
imposition of administrative boundaries, roads, railways, watercourses
(Hildebrandt 1996) could be considered when dealing with tasks, for exam
ple, in forestry, environmental planning, geography and cartography. The
person observing such illustrations can get additional information of about
the surrounding landscape (Figure 6.3-3).

Figure 6.3-3: Satellite map. Overlay ofvector data and IKONOS image data (Data source:
Space Image Europe, Officefor Environment and Geology ofSaxony)

Potential areas of satellite image products‘ application in landscape and

urban planning are:
— updating of land use maps,
— mapping of town structure types,
— determination of the extent of surface sealing,
— overall view of potential of land use conflicts,
— drawing up of general maps for terrain mapping,
— visual recording of very structured areas to determine for which areas
further on-site mapping is necessary, and
290 Chapter 6

—
updating of existing geographical data such as mapping of biotypes.
lt is possible to create three-dimensional representations of the landscape
by including digital elevation data, eg. from laser scanner recordings car
ried out from airplanes. Such perspective Iandscape presentations allow a
better understanding of geographic areas and so make deeper analysis of
landscape ecological conditions and interrelations possible. In addition, fur
ther levels of information such as planning concepts (buildings, settlements,
traffic systems, artificial lakes etc.) or the resuits of simulations can be in-
serted into a picture of the real landscape. Their effects on the landscape can
then be assessed from the varied viewpoints.

6.3.5 Digital spectral land use ciassification

To detect and classify separate areas of land use, the different reflective
properties they have arc recorded in different channels of the satellite scan-
ner (Figure 6.3-2). There arc two basic methods of ciassification: supervised
and unsupervised. In an unsupervised ciassification process, the division of
ciasses is carried out automatically by a ciassification algorithm. The classes
that result from unsupervised ciassification must then be interpreted and as
signed to particular land uses.
At the start of supervised ciassification assignment of areas on which in-
formation exists either from inspections or existing geographic information
is interactively determined on-screen. A decision function is derived from
these “sample groups“. All the pixels in the whole image are then compared
to the characteristics of the sample groups and are assigned to an object
group by the decision function (Wieneke 1988).
Spectral separation of the individual classes and spatial resolution are
limitations in classification. The problem of separating classes according to
the spectrum becomes clear, for example, in land use classes such as grass-
land and germinating crops. Therefore, it is important to choose the time of
recording correctly or to use several recordings from a given year to distin
guish between fields that have not yet been ploughed and grassland.
lt is also important to note that only the type of surface cover can be re
corded and there is no information regarding function. A sealed area can, for
example, be recognized as such with a degree of certainty. However, it is not
as easy to determine whether it is a road or a parking lot.
The resolution of multi-spectral scanner systems, which is usually be
tween 4 and 30m, limits the ability to record small objects. An object that is
smaller than the given dimensions or is only partially contained in a pixel
will be received as a mixed signal including information from the environ
ment surrounding the object and it will not be possible to identify it clearly.
 U.Walz 29

Hyperspectral ciassification is used to identify individual minerals or

water components by their typical behavior when absorbing or reflecting the
spectrUm. lt is also used to distinguish between several surface materials in
urban regions (roofs). To do this, it is necessary to record the characteristic
absorption bands with sufficient spectral resolution. Such spectrometers or
hyperspectral sensors in airplanes arc currently used successfully, mainly in
dry areas where is no plant cover, for geological exploration. lt is conceiv
able that this equipment could be used for mapping biotopes, investigating
water quality or forecasting harvests. Such systems arc still at the develop
ment stage.
Methods of multi-spectral classification look at each pixel separately.
However, methods of texture analysis evaluate the gray scale values of the
neighboring pixel and their relationships by using texture filters. Texture is
defined by means of elements in a given form and size, and by the recur
rence of this pattern. This approach is only of limited use for the characteri
zation of natural structures, as they neither usually contain simple, clearly
defined elements nor occur in a rigid recurring pattern (Turner and Gardner
1991b). Structures that are determined by utilization have recognizable regu-
lar recurring patterns. Examples of this are the rows in corn cultivation or in
intensive cultivation of fruit, or the regular structure of individual elements
such as division of fields into typical sizes. In some cases it is possible to
improve multi-spectral classification by including such textural parameters :
(Kaifel and Straub 1990). For this aim, one can distinguish between charac-
teristics of texture such as homogeneity, non-homogeneity, contrast, mean
value, aberration from the standard and entropy. Characteristics of texture
and multi-spectral properties arc connected by means of the classification
polynomial, which includes a channel containing the characteristics of tex
ture.
New approaches in digital image processing are based on segmentation
of images into homogenous areas on different hierarchical levels (Blaschke
2001). Distribution of gray scale values (texture) and the form of individual
units of utilization play an important role here. Areas that have been divided
into segments can then be integrated into the ciassification process or used
directly for defining boundaries for objects.
Methods of digital image processing can also be used for the extraction
of linear landscape structures (Figure 6.3-4). Linear elements found in sat
ellite images are linear infrastructures such as roads, paths, railway lines or
power lines. Every boundary between two areas of different utilization is to
be regarded as a linear element. In ecology such areas are called ecotones.
The margins of every land use plot are dominated by species that are only
found or are mainly found near the boundary. Therefore, an ecotone is a
292 Chapter 6

transition zone between neighboring ecological systems that has its own
characteristics (Hansen and Di Castri 1992, see Chapter 2.5).
Linear infrastructures often cause disruption, ecotones however are as a
rule regarded as positive. Whilst existing geographic data can be used in the
evaluation of infrastructures, these data sources contain very little informa
tion on ecotones. In particular, boundaries between grassiand or arable land
and the edges of fields bordering other fields or roads can not be distin
guished. Modern remote sensing methods are necessary as they can fiu the
gap in information. Johnston et al. (1992) see great potential in determination
of ecotones from panchromatic satellite images: “Satellite imagery is useful
for boundary detection at landscape to global scales, and it provides an ob-
jective means of identifying and quantifying ecotones which can be applied
to large areas.“ Smaller landscape elements (eg. undergrowth) can also be
picked out in high-resolution satellite data but the automatic cut-off detec
tion processes that are based on the evaluation of differences in gray scale
values between neighboring pixels arc not able to distinguish whether such
an area is the boundary between fields and forest or between a tarred fore
court and the roof of a house. The cut-off areas that arc detected should,
therefore, either be combined sensibly with existing data or should be
rnasked. Another possible approach is the evaluation of classified data or
existing data, which has been used for another purpose, for example, bio-
mass calculated using NDVI data. Johnston and Bonde (1989) use NDVI
values from Landsat TM data for the determination of ecotones.
Landsat TM recordings with channel 6 contain thermal information.
The gray scale value of these heat images is closely connected to the surface
temperature of the object displayed. Conclusions on the absolute surface
temperature cannot be made without calibration using the heip of terrestrial
measurements. In comparison, it is possible to make reliable statements on
the relative differences of the surface temperatures of different objects on the
surface of the earth (Hildebrandt 1996). Thermal information becomes im
portant when the connection is made between energy balance and the land
scape. Ripl (1995) assumes that the structure of the earth‘s surface will be
more long-lasting the more efficient an area is at dissipating the energy im
pulse from the sun‘s radiation. Dissipation of energy is achieved by a combi
nation of local evaporation, dissolving and biological production processes.
In terms of evaluation this means that areas with a high level of reflection
have a bw level of landscape efficiency and those with a bw level of reflec
tion a high level of landscape efficiency.
U Walz 293

Figure 6.3-5: Extraction of linear features from IRS-JC panchromatic dato (Dato source:
ANTRIX, SIE, Euromap Neustrelitz)

The different characteristic levels of reflection of living green vegetation

in particular spectral ranges mean that spectral indices can be developed,
which make it possible to distinguish between areas with living vegetation
cover and areas where there is no vegetation or the vegetation has died. As
living plant populations reflect more in the near infrared range than inani
mate surfaces and less in visible infrared range it is possible to form a vege
tation index using these two spectral ranges. The Normalized Difference
Vegetation Index (NDVI) has proved to be the index most used: As Netz-
band (1998) was able to show the values of the vegetation index correlate
with the extent to which the land is sealed in urban areas.
The Green Component (Hildebrandt 1996) offers a further possibility for
biomass evaluation. lt is calculated using the so-called Tasseled Cap. Fur
ther Tasseled Cap coefficients are brightness components, areas that are in
dicated with bw vegetation and high level of reflection and moisture com
ponents that indicate water or humidity.
294 Chapter 6

6.3.6 Summary and outlook

Remote sensing data are an important basis for dealing with questions in
landscape ecology. lt makes it possible to get current information on large
areas of land. Used alongside visual evaluation and superimposing other
geographical data it is possible to classify land use areas as weil as carry out
a whoie range of other thematic evaluations. Technical development shows a
trend that is going into two directions. Firstiy, the resolution of sateiiite data
is improving with every mission; it is expected that in the near future a spa
tial resolution of O.5m will be possible. This will mean that satellite remote
sensing will have achieved the same resolution as aerial photography. Sec-
ondly, surveys carried out using airplanes arc becoming more important, for
example, when gathering highly accurate data on terrain by means of laser
scanning. Moreover, use of radar sensors will undoubtedly play a more im-
portant role in the future as it is the oniy method which can make recordings
regardless ofthe weather situation possible.
lt can also be ascertained that the new generation of satellites (eg. IRS
ic, IKONOS) will not really be able to make a significant improvement in
the classification of land use areas. Landsat TM data is still the most suitable
for classification with its seven spectral channels, one of which of course is
bw resolution. New digital image processing technologies are under devel
opment. In future the combination of spectral and textural characteristics will
be important for an object-oriented ciassification, instead of single pixel
based ciassification algorithms.
For the purpose of application-oriented visualization, remote sensing
provides a method that can be processed and used with a relatively bw
amount of effort. If data is externally processed in advance, it is possible to
include remote sensing data in an anabogue work process or on a PC, which
is not equipped with a GIS. Provision of current information on land use
covering whole areas is important for planning authorities and offices in par
ticular.
Integration of high-resolution satellite remote sensing data into the basic
data and information systems of the survey offices and individual specialist
institutions will be decisive for its future use. lt can be hoped and expected
that the availability of graphic geographic information on actual surface
cover can make a contribution in supplementing the sometimes abstract in
formation that is found in planning documents and maps to achieve a better
understanding of the consequences in intervention in the decision making
process.
View publication stats