Potential of remote sensing and GIS as landscape structure and biodiversity indicators [Elektronische Ressource] : methodological study relating field data to visually interpreted and segmented landscape objects and image grey values / vorgelegt von Eva Ivits-Wasser

Ivitseva

Informations

• Publié par: albert-ludwigs-universitat_freiburg
• Publié le: 01 janvier 2004
• Nombre de lectures: 14
• Langue: English
• Poids de l'ouvrage: 16 Mo

Extrait

Potential of Remote Sensing and GIS as Landscape
Structure and Biodiversity Indicators

Methodological Study Relating Field Data to Visually Interpreted and
Segmented Landscape Objects and Image Grey Values




Inaugural-Dissertation zur
Erlangung der Doktorwürde
der Fakultät für Forst- und
Umweltwissenschaften der
Albert-Ludwigs-Universität
Freiburg i. Brsg.


vorgelegt von

Eva Ivits-Wasser

Freiburg im Breisgau
2004





























Dean: Prof. Dr. E. Hildebrand

Supervisor: Prof. Dr. Barbara Koch

Co-supervisor: Prof. Dr. Dr. h.c. Dieter R. Pelz
ACKNOWLEDGEMENT

There are many persons I owe to thank for their support and encouragement during my PhD thesis. This
was a long and difficult period, which I might not have managed without them.

Firstly, I would like to thank Prof. Dr. Barbara Koch for accepting me as a PhD student and for
supporting me along the way that led to my PhD work. I also thank her for the possibility to be able to
work in the BioAssess project which was a great experience about how science functions in real life. I
also would like to thank Prof. Dr. Dr. D. R. Pelz for taking the task of co-referent of this work.

My thank also goes to the Swiss Federal Institute of Forest, Snow and Landscape research for
providing the digital elevation model, topographic maps and aerial as well as orthophotos for this
research. In particular, I thank Lars Waser and Dr. Christian Ginzler for their expertise and cheerful
contribution to my work.

This list would not be complete without mentioning all the members of the BioAssess project. During the
three years of the project I had many opportunity to work with “our biologists”, who learned me a lot and
made my work easier in times when it seemed impossible to come over the obstacles. Although I cannot
list all names, I would like to express my special thank to Prof. Allan Watt, who was the co-ordinator of
the project. He gave me a lot of credit and always encouraged my work concerning the project as well
as my PhD thesis.

Of course, a lot depended on the working environment. I thank all the members of the Department of
Remote Sensing and Landscape Information Systems at the Freiburg University. Although we all
worked on different projects we still built one group that made the atmosphere cheerful on the
department. At times when problems arose with software and computers as well as theoretical
questions remained unsolved I could always count on their help.

Although it was “long” ago, my Master thesis was a stepping stone in my scientific life. Therefore I would
like to thank Dr. Carolien Krouze for introducing me into the scientific life and teaching me how research
functions. Basics are very important. I also wish to express my gratefulness to Dr. Niall McCormick at
the Space Aplication Institute (ISPRA, Italy) for his neverending patience and increadible helpfulness in
explaining the functionalities of SILVICS.

Last but not least I need to thank my family. First of all I thank my husband Tobias Wasser for he knows
how difficult it is having a wife writing a PhD thesis. He always gave me strength and encouragement as
well as practical advises and was always there for me. I thank my mother for accepting a long-distance
daughter and my parents in-law for believing in me. Thank you all.
i


ii
EXECUTIVE SUMMARY

The present research was conducted within the framework of the “Biodiversity Assessment Tools”
(BioAssess) project, co-founded by the European Commission under the Global Change, Climate and
Biodiversity Key Action of the Energy, Environment, and Sustainable Development Programme (Project
number: EVK2-CT1999-00041, www.nbu.ac.uk/bioassess). Raw remote sensing and field data were
provided by the project for the purpose of this research.

The overall objective of the research was to evaluate the potential of remote sensing to (1) describe
landscape structure regarding land use intensity and (2) to monitor species diversity. In particular, the
human vision system, object based assessment, and image grey values were investigated in order to
extract habitats from remote sensing images. Furthermore, the influence of spatial resolution of remote
sensing images as land use and biodiversity indicators were studied.

Three taxa sampled within the BioAssess project provided reference to assess the value of remote
sensing: plants, birds and ground beetles. Sampling of species diversity was carried out in eight
European countries: Finland, France, Hungary, Ireland, Portugal, Spain, Switzerland, and UK. In each
country six so called Land Use Units (LUUs) were established along a land use gradient from old-growth
forest to intensive agriculture. Percentage of forest one LUU contained was the indicator for land use
intensity grade. Each LUU contained sixteen plots where species were sampled. Altogether forty eight
LUUs were sampled with ninety six sampling plots per country.

The LUUs sampled in the eight countries were covered with the “standard dataset” of remote sensing
images. This dataset included Landsat ETM and IRS images because of the good spectral- of the
former and the good spatial resolution of the latter image. To facilitate the analysis, the two images were
fused into a 5m spatial resolution dataset. Switzerland was defined as “super test site” where additional
remote sensing data was acquired. This was called “potential dataset” and contained Quickbird satellite
with 2.8m and colour infrared orthophotos with 0.6m spatial resolution.

A hierarchic classification system was defined based on the CORINE database. The first two
hierarchical levels gave land use classes for all LUUs across the eight countries while additional levels
were adjusted for local landscape conditions. A visual interpretation protocol was developed that local
specialist used to derive the defined land use classes from the standard dataset. In addition to visual
interpretation, segmentation-based classification was applied for all the images of the super test site.
Thus, visual interpretation was performed for the standard dataset of all countries which was completed
with object-based classification for the potential dataset of the super test site (Switzerland).

Classes extracted from visual interpretation were used to calculate landscape indices, which served as
land use intensity indicators. Six indices were enough to explain the main variation. These were:
iii
Modified Simpson’s Evenness, Aggregation, Patch Richness, Mean Contiguity, Interspersion and
Juxtaposition, and Mean Euclidean Nearest Neighbour Index. LUUs with similar land use intensity grade
were expected to evidence similar values on these variables. However, the six indices failed to identify
similar stages of land use intensity. Cluster analysis also failed to order similar LUUs into the same
cluster. It was evidenced, that land use intensity cannot be quantified across different European
countries based on landscape indices, if percentage of forest is the land use intensity criterion.

Land use intensity analysis was extended to the super test site based on the ninety six sampling plots.
Sixteen patch indices were extracted from visual interpretations and segmentations as well as image
grey values were computed. These variables were produced for all the images of Switzerland.
Furthermore, a digital elevation model was included together with information like slope, aspect,
curvature, and texture. Factor Analysis revealed, that spatial resolution of the images did not play a
crucial role when patch indices described land use intensity. With increasing spatial resolution however
segmented patch indices explained slightly more variance then visually interpreted indices. With
increasing spatial resolution however, grey values explained more variance between the sampling plots
then patch indices. Orthophoto image derivatives explained as much as 74 percent of the variation.
Elevation derivatives reached a considerably 62 percent.

Patch, grey value and elevation indices were related to biological data obtained from sampling plots in
Switzerland. In particular abundance and species richness of plant, birds and carabides together with
presence-absence of Sorbus aucuparia (plants), Erithacus rubecula (birds) and Pterostichus melanarius
(carabides) species were investigated. Canonical Correspondence Analysis of abundance data revealed
little difference between visually interpreted and segmented patch indices within and between spatial
resolutions. Carabides abundance was the best explained by patch indices followed by plants and birds
abundance. However, grey value and elevation derivatives were stronger indicators of species
abundance then patch indices. Furthermore, with increasing spatial resolution grey value derivatives
associated stronger to the species data. Grey values explained in carabides abundance the most
variation followed by plants and birds abundance.

Specie