Detecting land cover changes through remote sensing and GIS techniques
F. Catani , L. Ermini , M. Kukavicic , S. Moretti , G. Righini
Department of Earth Sciences, University of Firenze, Firenze, Italy I-50121, gaia.righini@geo.unifi.it
Abstract – This note concerns the updating of the land juxtaposition of small parcels of diverse annual crops, pasture
cover map of the Arno River basin (9.000 km2) developed and permanent crops. Several towns and human settlements
in the frame of a research project aimed at landslide risk are widespread along the river and in the floodplain.
assessment. The land cover map has been produced
through the analysis of panchromatic/multispectral
Landsat ETM+ images and TERRA-ASTER data; those
data were integrated and compared with B/W and color
ortho-photos. To improve photointerpretation and to
obtain a higher resolution of the multispectral data, some
image fusion techniques have been applied to Landsat
data, such as Hue Saturation Value (HSV), Color
Normalized (CN) and Principal Component Substitution
(PCS). The land cover data base of the CORINE Land
cover Project have been updated and improved to a scale
1:50.000.
Keywords: land cover, satellite images, photointerpretation,
landslide hazard.
1. INTRODUCTION
A land cover map represents the physical description of the
environment regarding several aspects such as
geomorphology, pedology, vegetation and anthropic activities Figure 1. Location of the Arno River Basin.
(such as agriculture, urban areas, infrastructures); information
on land cover, together with information on relief and
drainage systems is essential for the management of the 2. METHODOLOGY
environment and natural resources. For landslide hazard
assessment, data on land cover are fundamental and represent 2.1 Background and procedure
also an input to define elements exposed to landslide risk. The land cover data base was classified according to the
The aim of this work was the realization of an updated land CORINE (Coordination of information on the environment)
cover map at scale 1:50.000 using remote sensing data Land cover Project legend (Cumer, 1984; Heyman et al.,
validated through ground surveys in the Arno River Basin, 1994).
Italy. In fact the capability of detecting land cover changes One of the aims of the CORINE programme of the European
trough remote sensing analyses can allow low cost spatial and Commission was to compile information on the state of the
temporal updating. environment with regard to certain topics which have priority
for all the Member States of the Community; in the CORINE
1.1 Study area system, information on land cover and changing land cover is
The Arno River Basin, with a spatial extension of about 9.000 directly useful for determining and implementing
km2, represents one of the widest hydrographical basins of the environment policy and can be used with other data (on
Italian territory (Fig. 1). It can be considered a relatively climate, topography, soil, etc.) to make complex assessments
homogeneous area from the viewpoint of its general (e.g. mapping erosion risks and landslides risk).
geological characteristics, being located in correspondence to The legend developed for this purpose is a hierarchical
the Northern Apennines, in Tuscany. In fact, this mountain numeric three level legend widely adopted in Europe: the first
chain is mainly made up of arenaceous and calcareous level (5 items) indicates the major categories of land cover on
turbidite sequences and chaotic argillaceous units of the planet; the second level (15 items) is for use on scale of
sedimentary and tectonic origin. 1:500.000; the third level (44 items) was used in the Project
The type and frequency of mass movements are primarily on a scale 1:100.000.
controlled by lithological and structural factors, secondarily In this work the third level was applied improving the scale to
by the high relief and the rather severe meteorological 1:50.000 due to the improved Landsat ETM ground resolution
conditions. at 15 meters.
Regarding land cover, in the mountainous part of the area The study was focused on the analysis and use of optical data:
coniferous forest prevails, taking place to broad leaves forest panchromatic and multispectral Landsat images from the last
or sparsely by open meadows, while the valley bottom is ETM+ satellite have been acquired dated 20 June 2000 and 15
occupied by complex cultivation patterns resulting from the February 2001; B/W ortho-photos (1:10.000 scale) dated 1998
(from AIMA archive) and color ones of 1998 to 2000 were After replacement, the merged result is converted back into
obtained for the whole basin; visible-near-infrared TERRA- RGB color space.
ASTER image of October 2001 was free downloaded in In order to merge the data sets using the Principal Component
Internet for a part of the basin. Substitution model, the multispectral data set is subjected to a
Remote sensed data have been processed by means of ENVI® Principal Component Analysis. When a PCA is implemented,
software and thus analyzed in a G.I.S. environment for the the first principal component contains information related
photointerpretation phase. Specific ground truth surveys were mainly to intensity or brightness. The panchromatic data is
carried out for the definition of photointerpretation keys on substituted for the first principal component and an inverse
satellite images and for results validation. Fig. 2 shows the PCA is performed on the combined data set.
procedure applied in this work. The Color Normalized technique separates the multispectral
image space into color and brightness components. It works
by first normalizing the band to be displayed by the intensity
of the RGB image.
It then multiplies the result by the panchromatic image data to
add the brightness (shadows or albedo information) of the
higher resolution image back into the color image that was
removed by the rationing.
In this work the first step in image processing was to geocode
the panchromatic image with the multispectral one projected
in UTM ED 1950 Zone 32. Then all three fusion method have
been applied to Landsat images and the data set obtained have
been studied.
All the methods have shown reliable results in improving the
geometric resolution and in producing highly readable color
composites; the HSV seemed to be the best technique at all
for visual interpretation even if for band 2 the PCS was the
best performing one. In Fig. 3 some examples of the results
obtained are shown.
Figure 2. Flow diagram of the procedure applied in this work.
2.2 Data Processing
The aim of data processing was to obtain higher resolution
multispectral images to improve photointerpretation for land
cover mapping, using satellite data; for this purpose some
methods for merging multispectral 30 m resolution images
with a 15 m resolved panchromatic image of Landsat ETM+
data have been tested in order to produce suitable images for
visual interpretation of land cover feature at scale 1:50.000.
Data fusion means the combination of two or more different
images to form a new image of high quality by using a certain
algorithm in order to obtain a high resolution multispectral Figure 3. Comparison between the results gathered applying
image (Wald, 1998; Pohl & Touron, 1999). A high quality of the three different fusion methods to the original image on
geometric information on satellite data is important to map June 2000.
different features of anthropic environment, both urban and
rural, while the multispectral characteristics are fundamental
to develop thematic maps (Fritz et al., 1999; Peccol & De The general methodology for Corine Land Cover updating
Luca, 2001). was based on the Technical and Methodological Guide for
We applied different methods of data fusion: Hue Saturation Updating CORINE Land Cover Data Base (Perdigão &
Value HSV, Principal Component Substitution PCS (Li et al., Annoni, 1997) guide lines.
1999), and Color Normalized CN (Pohl, 1996; Vrabel, 1996). The photointerpretation of the land cover changes was carried
The Hue Saturation Value method transforms the data from out by the integration of the whole information available in
actual color space (Red Green and Blue, RGB) into another order to obtain the new updated land cover map; land cover
space (HSV) and replace the value band with the more highly changes were digitized as lines (Fig. 4).
resolved panchromatic image, while the hue and saturation Uncertain areas were verified trough ground surveys and
bands are resampled to the high resolution pixel size. cross checks before setting the final polygons and data base.
high reflectivity in the visible and mid infrared and without
vegetation. In the field survey the area was recognized as a
construction site for a new little reservoir for drinkable water
supply of the surroundings villages. As shown in Fig. 6 the
area is bare and some sheds are evident.
Figure 4. Newly created areas 121: industrial or commercial
areas. Landsat ETM Band 7 HSV fusion technique: 15m
resolution.
2.3 Ground Surveys
Ground surveys were carried out for two different purposes:
the definition of interpretative keys on satellite images for Figure 6. Construction site.
photointerpretation and the solution of uncertainties during
the work. Some examples are shown hereafter.
An area with high reflectivity was detected in all composites 3. RESULTS AND ANALYSES
of the February and June scenes while the presence of
escarpments were noted in topographic maps. During the The updating work has interested 7320 polygons, 2560 of
ground survey a quarry of marble located near the village of which (corresponding to 913 km2) were involved in land
Casole d’Elsa (Siena province) was found the activity of cover changes. This means that the 10% of the whole
which started in 1999 (Fig. 5). coverage of the river basin has undergone changes in 6 years.
From the original 44 classes, 34 were implicated in total or
partial modifications, for example several classes were totally
converted in different codes and some others were splitted in
two or more new typologies. The reclassification and the
modification of the limits is closely connected to the better
resolution of the data set.
In the figure below (Fig. 8) 15 selected items are shown along
the interesting and consistent degree of variation.
100,00%
50,00%
Variation
(%)
0,00%
111 112 121 132 133 142 311 312 313 322 324 333 334 421 512
-50,00%
-100,00%
Figure 5. Quarry of marble (Casole d’Elsa, Siena province).
Classes
The second example concerns an area classified as olive Figure 8. The histogram shows the differences between the
groves (class 223) in 1996; in the color aerial photo dated situation in 1996 and the updated Corine Land Cover of 2002.
2000 it remains the same while in the Landsat ETM image of Positive values represent territorial increase and negative
2001 it appears clearly different from the surroundings with values represent territorial decrease (see table 1 for details).
Table 1 shows the percentage variation of the most significant and definition of elements exposed to landslides risk. In fact,
land cover types between 1996 and 2002 and the coverage statistic analysis showed that landslide hazard is particularly
difference of each class in the same period. affected by land cover type.
The results obtained in this phase will be implemented in a
risk model available for urban planning purposes in areas
Area in '96 Area in '02 defined as prone to slope failures, starting from
Classes (ha) (ha) % geomorphological factors controlling landslide occurrences.
111 Continuous urban fabric 2.542 2.787 9,6%
Discontinuous urban 5. ACKNOWLEDGEMENTS
112 fabric 28.696 31.029 8,1%
Industrial/commercial
121 units 9.164 10.440 13,9% The research has been partially supported by the Arno River
Basin Authority and was carried out in the framework of
132 Dump sites 26 54 108,3% activity of CNR-GNDCI U.O. 1.46 (headed by Prof. Sandro
133 Construction sites 277 116 -58,1% Moretti)
Sport and leisure
142 facilities 342 450 31,6%
6. REFERENCES
311 Broad-leaved forest 247.632 219.588 -11,3%
312 Coniferous forest 21.887 26.001 18,8% A. Cumer, “Il progetto CORINE Land Cover in Italia: un
313 Mixed forest 73.504 101.742 38,4% modello da seguire” Documenti del territorio Anno VIII N.
28/29 giugno/dicembre 1994.
322 Moors and heathland 12.452 10.396 -16,5%
Transitional
324 woodland/shrub 22.326 20.522 -8% R. Fritz, I. Frech, B. Koch, and Chr. Ueffing, “Sensor fused
images for visual interpretation of forest stand borders”
333 Sparsely vegetated areas 509 216 -57,5%
Archives of Photogrammetry and Remote Sensing, 32/7-4-3
334 Burnt areas 1.499 339 -77,4% W6, Valladolid, Spain, 1999.
421 Salt-marshes 0 369 new
512 Water bodies 891 1.715 92,4%
Y. Heymann, C. Steenmans, G. Croisille, and M. Bossard,
“CORINE land cover project” Technical guide. European
Commission, Directorate General Environment, Nuclear
Table 1. Percentage variation for 15 selected classes.
Safety and Civil Protection, ECSC-EEC-EAEC, Brussels-
Luxembourg, 136 pp. 1994.
As it can be seen classes such as urban fabric or industrial
J. Li, Y. Zhou, D. Li, “PCA Wavelet transform for fusion
units present increasing values thus reflecting the increase in
panchromatic and multispectral images”. SPIE’s International
population density. Conversely, items such as 133 and 311
Symposium on Aerosense ‘Image exploitation and target
present negative values. The first one (133) is the obvious
recognition’ Orlando, Florida, USA Vol 3719, pp369-377,
result of construction activity, whilst the decrease of the
1999.
second one (311) is probably connected to the parallel
increase of 312 due to a different choice in forest
E. Peccol, A.De Luca, “Valutazione di tecniche di fusione ai
classification and also due to a better resolution of the images
fini del rilevamento di superfici a prato”. Atti 5° Conf. Naz.
used. More interesting is the decrease of burnt area and
ASITA ‘La qualità dell’informazione geografica’, Rimini,
transitional woodland and shrub that were converted into
Italia, 2001.
forests.
The high variation of the 512 class (92%) is justified by the
V. Perdigão, A. Annoni “Technical and Methodological
emplacement from the year 2001 of the Bilancino reservoir in
Guide for Updating CORINE Land Cover Data Base”, 1997.
the Mugello valley; it is an artificial lake still under
construction in 1996 that was completed at the beginning of
C. Pohl, Geometric aspects of multisensor data fusion for
2001; this lake is now covering more than 5 km2 and its
topographic map updating in the humid tropics. ITC
environmental impact is still under development.
publication, N0.39, ISBN 90 6164 121 7, pp.37-38, 1996.
The final map of land cover shows that more than 10% of the
total area has changed in 6 years. This variation is very
C. Pohl, H. Touron, Operational applications of multi-sensor
important for two reasons: the temporal interval of 6 years is
image fusion. International Archives of Photogrammetry and
short and the Arno River basin is a territory already greatly
Remote Sensing, 32/7-4-3 W6, Valladolid, Spain, 1999.
affected by human impact.
J. Vrabel, Multispectral imagery band sharpening study.
4. CONCLUSIONS
Photogrammetric Engineering and Remote Sensing, 62/9:
1075-1083, 1996.
The updated land cover map has been used as one of the input
parameters for the computing of landslides hazard and risk
L. Wald, An European proposal for terms of reference in data
assessment. This new map shows that the Arno River Basin
fusion. International Archives of Photogrammetry and
has experienced a lot of changes due to anthropic activities in
Remote Sensing, 32/7, 651-654, 1998.
the time investigated. These changes underline a situation of
warning because of their role for landslides hazard assessment
