Für neue Autoren:
kostenlos, einfach und schnell
Für bereits registrierte Autoren
Veröffentlichen auch Sie Ihre Arbeiten - es ist ganz einfach!Mehr Infos
Doktorarbeit / Dissertation, 2002, 98 Seiten
Doktorarbeit / Dissertation
1.1. Definition of monitoring and habitat
1.2. Birds as indicators of environmental change
1.3. Geographic Information Systems
1.4. Satellite-based Remote Sensing
1.5. The use of birds, GIS and satellite images in previous studies
1.5.1. Objectives of previous studies
1.5.2. Sites of previous studies
1.5.3. Methodology of previous studies
1.6. Objectives of the study
2. Study site
3. Data and methods
3.1. Bird census and terrestrial habitat mapping
3.1.1. Point-Stop Counts
3.1.2. Territory Mapping method
3.1.3. Measurements of habitat structures
3.2. Satellite image data
3.2.1. Image interpretation and classification
3.2.2. Ground truth and accuracy assessment
3.3. Analytical methods
3.3.1. Spatial data analyses using GIS
3.3.2. Preference Index
3.3.3. Bivariate statistics
3.3.4. Multivariate statistics
4.1. Terrestrially mapped structures
4.3. Habitat map
4.4. Habitat preferences of birds
4.4.1. Electivity Index
4.4.2. Statistical significance
4.4.3. Preferences on selected bird species
4.4.4. Buffer scale
4.5. Species diversity and individual density
4.6. Bird assemblies of the Serengeti Plains
4.6.1. Territory Mapping
4.6.2. Cluster analyses
4.6.3. Principal component analysis
4.7. Reduction of data
4.7.1. Analysis of environmental variables
4.7.2. Analysis of bird species
5.1. Verification and evaluation of the method
5.1.1. Evaluation of variables
5.1.2. Spatial scale
5.1.3. Temporal scale and seasonality
5.2. Sources of errors
5.2.1. Satellite image classification
5.2.2. Birds and bird census
5.2.3. Position determination
5.3. Results of grassland birds
5.3.1. Bird species density and bird competition
5.3.2. Future research suggestions on grassland birds in SNP
10.1. Overview of studies using GIS, satellite-based RS and bird data
10.2. Distribution maps of specific bird species
10.3. Habitat preferences
10.4. Bird species recorded using PSC
Figure 2: Location of SNP
Figure 3: Flow chart of the methodology
Figure 3.1.2: Sketch of a study plot Territory Mapping was conducted on
Figure 18.104.22.168: View of North Tanzania from a height of 705 km from a Landsat 7 ETM+ Image of 12.02.2000
Figure 22.214.171.124: Location of the satellite image within the Republic of Tanzania
Figure 126.96.36.199: Full scene and sub-scene characteristics
Figure 188.8.131.52: Subset of the Landsat Image
Figure 184.108.40.206: Scatterplot of band 5 against band 3
Figure 3.3.1: Buffer no. 74
Figure 4.1: Two north-south profiles of grass height (average)
Figure 4.2.1: Climatograms for the south-east Serengeti
Figure 4.2.2: Map of interpolated rainfall average values of the study area based on 14 rain gauges
Figure 4.3.1: Competition and abundance of the habitat types in the study area
Figure 4.3.2: Seasonally flooded riverbeds
Figure 4.3.3: Bare ground / short grass at point 125
Figure 4.3.4: Short grass 3 at point 42
Figure 4.3.5: Intermediate grass 2 at point 200
Figure 4.3.6: Intermediate grass 3 close to point 229
Figure 4.3.7: Bushed grassland at point 274
Figure 220.127.116.11: Habitat preferences of Mirafra africana
Figure 18.104.22.168: Habitat preferences of Oenanthe pileata
Figure 22.214.171.124: Habitat preferences of Cisticola juncides
Figure 126.96.36.199: The percent cover of the habitat types found in each buffer type
Figure 188.8.131.52: Influence of different buffer scales on habitat preferences of the Capped Wheatear
Figure 4.5.1: Birds per point and per hour
Figure 4.5.2: Relative abundance of bird individuals based on interpolated records of PSC at 280 points
Figure 4.5.3: Diversity map of bird species based on interpolated records of PSC at 280 points
Figure 184.108.40.206: Scree plot to determine the optimal number of clusters by using the Elbow Criterion
Figure 220.127.116.11: Assembly dendrogram of Euclidean Distances between 50 bird species
Figure 4.6.3: Principal component analysis of the habitat use
Figure 18.104.22.168: Cluster analysis based exclusively on remotely sensed habitat types.
Figure 22.214.171.124: Cluster analysis based exclusively on terrestrially measured variables
Figure 10.3: Electivity values
Map 2: Study area and study plots
Map 4.1: Grass height (average) of the study area
Map 4.2: Percent cover of bare ground of the study area
Map 4.3: Habitat map
Map 10.2.1: Distribution of the Rufous-naped Lark Mirafra africana
Map 10.2.2: Distribution of the Capped Wheatear Oenanthe pileata
Map 10.2.3: Distribution of the Zitting Cisticola Cisticola juncidis
Table 1.3: The main features which characterize GIS and non-GIS studies
Table 1.5.1: Classification scheme of wildlife habitat studies
Table 1.5.3: Census techniques in 35 analysed studies.
Table 2.1: Plant species found in the four study plots
Table 3.1.1: Periods of time PSC were conducted
Table 3.2: Technical summary of Landsat 7 ETM+ sensor
Table 3.2.2: Error matrix
Table 3.3.1: Data of point number 155
Table 4.1.1: Mean and maximum of two different grass height measurements taken on the Serengeti Plains
Table 4.1.2: Percent cover of woody vegetation
Table 4.3: Description of the habitat types
Table 126.96.36.199: Added numbers of statistical significances
Table 188.8.131.52: Number of significances found on each of the 50 bird species
Table 184.108.40.206: Number of statistical significances between occurrence of environmental variables and bird species
Table 4.4.3: Results on three bird species using PSC
Table 4.4.4: The number of habitat types of different buffer sizes
Table 4.5: Numbers of palaearctic and non-palaearctic bird individuals and bird species recorded at 280 points using PSC
Table 220.127.116.11: Percent cover of habitat types of three study plots
Table 18.104.22.168: Breeding territories and abundance of bird species on short grass habitats
Table 22.214.171.124: Breeding territories and abundance of bird species on intermediate grass habitats
Table 4.6.3: Component Matrix. Loadings and proportion of variance for principal components analysis of bird species for the 21 environmental variables
Table 4.7.1: Statistical significances of terrestrially measured variables versus remotely sensed habitat types
Table 4.7.2: Comparison of significanct relationships found for each of the 50 bird species
Table 10.1 a: Literature review of 35 studies (part 1)
Table 10.1 b: Literature review of 35 studies (part 2)
Table 10.1 c: Literature review of 35 studies (part 3)
Table 10.1 d: Literature review of 35 studies (part 4)
Table 10.1 e: Literature review of 35 studies (part 5)
Table 10.3.1: Ivlev’s Electivity Indices of 14 habitat types calculated for 50 bird species and for number of bird species and individuals..
Table 10.3.2: Statistical significances of occurrence on 21 environmental factors on records of 50 bird species
Table 10.4: Bird species recorded using PSC
illustration not visible in this excerpt
Over 10 % (1186 species) of the bird species in the world are threatened with extinction in the near future, almost all of them due to habitat change or loss by man (Birdlife International 2000). Likewise, 1130 mammals, 296 reptiles, 146 amphibians and 5611 plants have been identified as endangered species (IUCN 2000). The destruction of natural habitat is the major factor contributing to the global species extinction event (for example Collar and Stuart 1985, Groombridge 1992, Bibby 1995). The increasing loss of biodiversity has centred on conducting inventories and monitoring species and habitats, especially in identifying areas of high species richness, threatened species and species of restricted or local distribution (for example Collar and Stuart 1988, ICBP 1992, Stattersfield et al. 1998, Heath et al. 2000). In 1992 the UNCED-Conference in Rio de Janeiro pointed out the need for monitoring the environment, leading to the Convention on Biological Diversity (United Nations 1992) and the Agenda 21. Article 7 of the Convention on Biological Diversity deals with identification and monitoring, which are to be undertaken with sampling and other techniques. New methodologies with a view to undertaking systematic sampling and evaluation of the components of biological diversity are to be developed (Article 15 b of Agenda 21).
While the number of identified threatened species has increased dramatically, a huge gap in knowledge of ecosystems and their fauna and flora remains. Distribution, status and ecology of species are mostly unknown in many countries, as is the degree they are endangered. In view of the immense unknown ecosystems in the world, a great number of which are located in developing countries, conventional survey and mapping methods cannot deliver the necessary information in a timely and cost-effective fashion. Nature conservation will require large volumes of Remote Sensing (RS) data if the quality of planning is to improve. With RS technology, we may be able to make real progress in understanding why more species occur in some places than in others and in identifying the most critical places that must be protected to preserve the maximum number of species into the 22nd century and beyond (Stoms and Estes 1993). As current air photos are often not available, satellite images are the sole source of data for many regions of the world.
Fortunately, computer technology has improved enormously in the last years, mainly processing time, storage requirements as well as programme features and possibilities. Concurrent declining costs of computer hardware have favoured the design of new techniques for special data processing and combining remotely sensed information with other extensive data sources.
In the last 20 years Geographic Information Systems (GIS) have been widely accepted and used as a tool for a host of applications in planning processes, in storing, analysing and maintaining data. Ground survey information together with RS imagery by using GIS techniques offers a huge potential for quick identification of areas of high biodiversity.
The approach of this study is to combine the potential of bird data, GIS and satellite-based RS in view of using these components to monitor the environment. After defining the terms monitoring and habitat in chapter 1.1, chapter 1.2 to 1.4 mainly focus on the potential using GIS, bird and satellite data. Chapter 1.5 continues with a detailed literature review of previous studies, which used GIS, bird and satellite image data to focus attention on issues considered to be the most important for effective use of the three components. Therefore, the main characteristics, especially methodology, satellites image analysis, bird census, scale and accuracy requirements were analysed and compared. Chapter 1.6 focuses on weaknesses of these studies and specifies objectives for the present study corresponding to the weaknesses.
As the two terms monitoring and habitat are widely used in this study precise definitions are given in the following.
Monitoring has become more and more important in assessing nature and its natural and human-induced changes (Bischoff 2000). It is a very important information-tool for decision-making in conservation policy (Bischoff and Dröschmeister 2000). The term monitoring has been defined by several authors (for example Furness et al. 1993, Hellawell 1991, Maas 1999, Peithmann 1996). Dröschmeister (2000) favours the following definition of Hellawell (1991), as it is the most useful and the most unambiguous one: “Monitoring - Intermittent (regular or irregular) surveillance carried out in order to ascertain the extent of compliance with a predetermined standard or the degree of deviation from an expected norm”.
Hellawell (1991) gives three reasons for monitoring:
1. assessing the effectiveness of policy or legislation
2. regulatory (performance or audit function)
3. detecting incipient change (‘early warning’).
The latter is of greatest interest to ecologists and conservationists. Further information, aims, strategies, techniques and programmes on monitoring can be found for example in Bischoff and Dröschmeister (2000), Blaschke (1999), Furness and Greenwood (1993), Goldsmith (1991) and Spellerberg (1991).
The use of biological monitoring (for example birds as biotic key indicators) in addition to non-biological monitoring (for example measurements of physical parameters) has the advantage of disclosing reactions of living organisms to environmental changes, which may otherwise be left undiscovered. Monitoring merely by non-biological methods has the disadvantage of its specificity on single environmental variables and often of its higher frequency of sampling occasions, which may not include significant events. Biomonitors are usually selected to complement physical monitoring, but in some instances provide the only available means of monitoring (Furness et al. 1993).
Garshelis (2000) mentions two distinct usages of the word habitat. According to the dictionary definition habitat is the definable place where an animal usually lives or, more specifically, the collection of resources and conditions necessary for its occupancy. On the other hand habitat is often characterized by a dominant plant form or physical characteristic. The various meanings and uses of habitat in literature are shown and discussed in detail by Corsi et al. (2000). According to Block and Brennan (1993) the notion of habitat as the place containing resources needed for survival and reproduction is clearly central to any theoretical consideration of the habitat concept. Habitat use refers to the manner in which a species uses a collection of environmental components to meet its life requisites (Block and Brennan 1993). For definition of the term habitat selection see Jones (2001).
Bird habitats may be defined in terms of elevation, topography, climate, soil types and dominant vegetation types. Other bird specific key factors could include the presence of essential resources, such as food, water, and nest-building materials as well as heterogeneity of the environment and weather conditions. The factors which affect bird occurrence are often specific to the biotope. To give an example of forests: Morrison et al. (1992) list about 29 different variables such as woody foliage profile density or tree stump size for measuring habitat structure. In addition, many non habitat-related phenomena influence habitat selection in birds (Cody 1985a) such as intra- and interspecific competition, food availability, predator abundance and diseases or intraspecific attraction.
Birds can be very useful bio-indicators of the state of the environment and also of changes in complex ecosystems (Baillie 1991, Bezzel 1974), as they reflect the condition of the many aspects of an ecosystem. Furness et al. (1993) showed that birds play a valuable role in monitoring several aspects of environmental change and Brooks et al. (2001) gave an example of birds which were used in a conservation planning programme to represent the majority of other terrestrial vertebrate diversity. A detailed description of indicator species is given in Caro and O'Doherty (1999) and a further explanation and differentiation of bio-indicators and biomonitors in Furness et al. (1993). Of the animal indicators, birds have received the most attention (Morrison 1986).
The following features of birds, taken from Baillie (1991), Bennun (2000), Bezzel (1974), Furness et al. (1993), Morrison (1986) and Steele et al. (1984) demonstrate why in comparison to other animals the avifauna is a suitable indicator group for survey and monitoring:
- Birds are well-known organisms, as their taxonomy, ecology and behaviour are far better known than most comparable taxa.
- Birds are ubiquitous as bird species inhabit a wide range of habitats and with varying degrees of specification are dependent on a wide variety of food resources.
- Birds show high mobility and thus are able to (re-) colonize appropriate habitats quickly.
- In many cases birds can be more sensitive or vulnerable to environmental changes and contaminants as they hold top positions in food chains.
- Birds, especially passerines, have a relatively short generation time, consequently they may exhibit quick response to environmental change.
- Data on birds are relatively easy to collect, because they are quite large, often conspicuous or vocal, mainly diurnal, and rather easy to identify, capture and mark.
- Birdwatching has a wide following: by involving amateurs, large amounts of information can be collected with great cost-effectiveness.
The problem with using birds as indicators lies in separating the many factors that can cause changes in bird populations. According to Morrison (1986) birds are usually not good indicators of primary environmental changes, but they may be good indicators of more subtle secondary changes. For example, the effects of “El Niño” on fish-eating birds in the Pacific Ocean resulted in a change in the pattern of food available to the birds which had been reduced by a change in ocean currents. Moreover, climate change has had an effect on the egg-laying dates of birds in the UK (Crick and Sparks 1999). Birds have been used to detect and monitor the effects of environmental contaminants most successfully as they are high in food chains (Morrison 1986, Baillie 1991). Bird species can be influenced directly by human activities used as well as indicators of an overall change in habitat quality. For example, diversity, distribution and status of European woodpeckers are affected by anthropogenic landscape changes (Mikusiński and Angelstam 1997). Bird communities are affected by forest fragmentation (Lynch and Whigham 1984); furthermore, noise resulting from off-road driving can disrupt the behaviour and movement of various bird species (Berry 1980). Many environmental changes lead to a gradual change of communities and the disappearance of species (Hellawell 1991). Several examples of effects of human-induced change on birds are given in Morrison (1986), Furness and Greenwood (1993) and Bairlein (1996).
Two approaches are possible to inventory the status and monitor the change of species abundance and distribution. The first approach is direct monitoring of the species in question or an indicator species that may signal habitat degradation or loss that affects other species. For a species to be a valuable indicator, the variation of habitats occupied should be small, or at least understood (Bock and Webb 1984). Therefore, a good understanding of the status, distribution, ecology and biology of potential indicator species is necessary to establish and interpret changes in numbers of the species (Spang 1992). The second approach is to monitor changes of crucial habitats. Species change can be inferred by using well-documented causal relationships between species and environment. The sensitivity and reliability of such a monitoring programme depends on the relationship between bird data and the environmental factor being monitored (Furness et al. 1993), in particular, how close the relationship is and how well it is understood. It is important that the monitoring scheme should have high statistical significance. Only if direct statistical relationships between birds and their environment are established, can a species be expected to be a direct indicator of a specific environmental change (Morrison 1986). Also a description of the resources selected more often than others is of particular interest (Manly et al. 1992).
As many environmental changes often become visible when species composition and density change, bird communities are an important component in a study on monitoring the environment.
The use of GIS technology in ecosystems research is a recent phenomenon which has rapidly become part of the mainstream in research (Stow 1993). GIS is now a leading tool in the development and application of contemporary urban and regional research for planning purposes (Michalak 1993). It is designed to assemble, analyse and display spatially distributed data. Storing, retrieving and analysing ecological data using geographic coordinates and spatial data structures are a powerful basis for establishing GIS for multiscale studies of ecosystems (Marble et al. 1984). A central step in these GIS analyses is a logical or arithmetic map overlay operation which is employed to merge the different GIS environmental layers to yield the combined effect of all environmental variables (Corsi et al. 2000).
A review of utilization of GIS in natural resources and ecology can be found in Johnson (1990). The studies presented in chapter 1.5 show how GIS can be a helpful tool in analysing RS images and bird species habitats. Information on advantages and disadvantages of the use of GIS in habitat studies can be found for example in Shaw and Atkinson (1990), Schuster (1992), Homer et al. (1993) and Corsi et al. (2000). Some characteristics of GIS and non-GIS studies are shown in Table 1.3. Corsi et al. (2000) suggest that the use of GIS spatial analysis tools can increase the ability to assess effects in species-environment relationships. But they also mention that the use of GIS is not central to a better understanding of causal effects in a species-environment relationship, especially if the quality does not support both high-resolution and large-extension analyses.
Abbildung in dieser Leseprobe nicht enthalten
Table 1.3: The main features which characterize GIS and non-GIS studies
Despite the obvious advantages of GIS, limitations exist such as the lack of available digital data, the lack of standards for digital data quality (Shaw and Atkinson 1990), settling-in periods in more and more extensive GIS-software packages and also costs of computer technology, GIS software and digital data.
Satellite imagery has been applied with considerable success in monitoring and analysing environment, for example, in analysing the dynamics of African vegetation (Townshend and Justice 1986), in examining the rate and extent of deforestation in the tropics (Nelson and Holben 1986, Malingreau et al. 1989, Westman et al. 1989), in examining the ecological conditions of peat, bog lands and pasture in northern Germany (Ehlers and Rhein 1996), in detecting changes (Mather 1992, Sader and Winne 1992, Kennedy 1989, Woodcock 2001), in analysing wetlands in relation to climate (Roshier et al. 2001) or taking inventory of waterfowl habitats (Barnard et al. 1981).
Stoms and Estes (1993) have shown the capabilities and possibilities of RS for mapping and monitoring biodiversity. Many of the factors of resource quality and quantity and of dynamic processes associated with biodiversity can be estimated from remotely sensed data including habitat types, vegetation structure, landscape geometry, habitat fragmentation, leaf area index (LAI), net primary production (NPP), biomass, actual evapotranspiration (AET) or digital elevation models (DEM).
Satellite-based RS in ecological studies provides several advantages (for example Sabins 1997, Budd 1991):
- Image acquiring compared to methods using aerial photos is cost effective as large areas can be imaged quickly and repeatedly
- Image interpretation is faster and less expensive than conducting ground surveys
- Periodic images acquisition records changes, such as urbanization, deforestation, and desertification
- Images are available in digital format suitable for a direct input to computers and for the following processes and analysis
- Images are often the only available source of information for many parts of the world
- Images contain a wide variety of spectral bands.
Some of the disadvantages are:
- Some types of land use may not be distinguishable and some landscape features are not discernible on images.
- Most images lack the horizontal perspective that is valuable for identifying many categories of land use
- Cloud cover and satellite track routes can limit image acquisition in regions of the world
- Image processing is expensive as it requires computer facilities, special software, knowledge and can be labour-intensive.
Since 1972, when satellite images became available for public use, extensive research has been conducted in which animals, GIS and satellite images have been used. In a literature review, Corsi et al. (2000) found 82 papers, which contain the keywords: GIS, RS, wildlife habitat and distribution. 25 and 42 of these studies include RS and GIS, respectively.
Much has been done for example on habitat modelling of mammals:
- Pronghorn Antelope Antlocapra americana in Kansas, USA (Martinko 1978)
- Caribou Rangifer tarandus in Keewatin, Canada (Thompson et al. 1980)
- Rufous Hare-wallabies Lagorchestes hirsutus in central Australia (Saxon 1983)
- Elk Alces alces (Leckenby et al. 1985)
- Grizzly Bear Ursus arctos (Craighead et al. 1986, Agee et al. 1989)
- White-tailed Deer Odocoileus virginianus (Ormsby and Lunetta 1987)
- Giant Panda Ailuropoda melanoleuca (Wulf et al. 1988)
- Musk Ox Ovibos moschatus (Pearce 1991)
In the present study, a literature review of papers using GIS, satellite image data and bird data was conducted. The Tables 10.1 a –10.1 e (see appendix) show details on the main characteristics of 35 studies which were investigated and compared. Some of the studies were published in more than one paper and thus were analysed in this review as one. The studies vary enormously depending on method, scale and accuracy requirements. This review should help to focus attention on issues considered to be the most important for the effective use of GIS, satellite image and bird data in species habitat studies and monitoring programmes. Thus, the following chapters focus on specifics, techniques and methods applied. While in the chapters 1.5.1 - 1.5.3 a descriptive analysis and comparison are given, in the last chapter 1.6 weaknesses on studies using GIS, bird data and satellite data are specified.
Combining wildlife survey data and satellite image data with the help of a GIS allows a variety in habitat analyses. The objective of all these studies was to find relationships between habitat categorization based on spectral reflectance pattern and bird species records. Some of the specific objectives of the studies are:
- Predicting distribution (for example Herr and Queen 1993, Andries et al. 1994, Austin et al. 1996, Dettmers and Bart 1999, Osborne et al. 2001)
- Analysing minimum habitat patch size (Palmeirim 1988) and fragmentation (Lauga and Joachim 1992, Klaus et al. 2001)
- Habitat change detection (Hodgson et al. 1988, Sader et al. 1991)
- Monitoring available foraging cover during different seasons (Hodgson et al. 1987, 1988)
- Density maps (Palmeirim 1988)
- Avian diversity and biodiversity map (Nøhr and Jørgensen 1997)
- Estimates of species abundance and their population size (Avery and Haines-Young 1990)
- Distances for example to forest edge (Palmeirim 1988), to highways (Herr and Queen 1993) or to agricultural land (Griffiths et al. 1993)
- Selection of release sites (Palmeirim 1985, Stoms et al. 1992)
Herr and Queen (1993) distinguish three different ways of applying GIS in order to assess wildlife habitat. This classification is used in Table 1.5.1 and supplemented by possible map outputs and a fourth category of “modelling”.
Abbildung in dieser Leseprobe nicht enthalten
Table 1.5.1: Classification scheme of wildlife habitat studies
All analysing possibilities are interrelated: to identify habitats and to model bird distribution, habitats have to be characterized at least on a minimum level and to monitor them habitats have to be identified rather than characterized in detail.
According to the classification scheme of Table 1.5.1 in 29 out of 35 analysed studies, bird species habitats were identified. 19 studies dealt with characterizing habitats, whereas modelling approaches to predict species distribution or abundance were used in 22 studies. Monitoring bird species habitats was the objective in only five studies.
Stoms et al. (1992) used a deductive-inductive categorization which was also used in Corsi et al. (2000) for categorization of GIS based species distribution models. This categorization focuses on the definition of the species-environment relationship. Whereas in the inductive approach these relationships were not known and were studied before an analysis of resource selection or a modelling process could be conducted, in the deductive approach species-environment relationship was known a priori (for example from literature). Five of the analysed studies used the deductive approach (Palmeirim 1985, Shaw and Atkinson 1988, Stoms et al. 1992, Tucker et al. 1997 and Klaus et al. 2001). A priori knowledge of species-environment relationship was considered to be sufficient in these studies.
Most of the 35 analysed studies were performed in North America (17 in USA, 2 in Canada), one was conducted in Central America (1 in Costa Rica), 11 were carried out in Europe (7 in Great Britain, 1 in Belgium, 1 in France, 1 in Spain, 1 in Austria), two in Africa (Senegal and East Africa) and just one in Asia (China) and in Australia (for more details see Tables 10.1 a-e in the appendix). There is a huge variation in habitats and in size of the study areas with the smallest covering 13 km2 in Kansas, USA (Palmeirim 1988) and one of the largest with 85,000 km2 in Maine, USA (Hepinstall and Sader 1997). The largest study sites comprise three African countries (Wallin et al. 1992), large parts of Australia (Roshier et al. 2001) or the Northwest Atlantic (Huettmann and Diamond 2001). Mean size of the study areas was about 201,000 km2 (n = 28) and without the large-scale study of Roshier et al. (2001) about 8,700 km2.
Although no exact percentages were specified for the different types of habitat, rough estimates can be given. While open habitats such as agricultural land, heath, pasture, grass-, wet- and moorland were predominant (ca. 64 %), forests as main habitats were listed in about 36 % of the studies. Marine habitats were studied only by Haney and McGillivary (1985a) and Huettmann and Diamond (2001).
Typically, the studies using GIS, satellite-based RS and bird data follow a common pattern. Most of them involve satellite image classification using ecologically meaningful classes. The inherent assumption is that the produced map has ecological relevance to the wildlife species of interest. Methodology of the studies differs depending for the most part on bird census techniques, RS data, additional environmental data and statistical analyses, which is shown in the following.
Bird census and bird species
Three main census techniques were used depending on the habitat or species being targeted. Mainly simple counts of sightings of bird species or their nests were conducted (see Table 1.5.3). Several of these counts of systematic and non-systematic origin were conducted with tools such as radio telemetry, aircraft or helicopter. Transect Counts and Point-Stop Counts (PSC) were performed five times. Furthermore, literature was used to model species distribution.
Abbildung in dieser Leseprobe nicht enthalten
Table 1.5.3: Census techniques in 35 analysed studies. In some studies more than one source of data was used.
Most of the research was done to predict distribution or to find habitat preferences either for a single species (57%) or for up to five bird species (6 %). Several of these bird species were rare or endangered. More than 20 bird species were used in the studies of Haney and McGillivary (1985a), Palmeirim (1988), Hepinstall and Sader (1997), Nøhr and Jørgensen (1997), Mack et al. (1997) and Debinski et al. (1999). Most of the listed species were passerines (114 species) and 107 were non-passerines.
Data on Remote Sensing and other environmental variables
Many of the analysed studies worked with Landsat satellite images (74 %). While 57 % of the studies were based on Landsat TM 4 or 5 images, Landsat MSS scenes were used in older studies (17 %), sometimes along with Landsat TM scenes. The higher resolution French Spot satellite was applied only five times by Miller and Conroy (1990), Andries et al. (1994), Thibault et al. (1998), Klaus et al. (2001) and Saveraid et al. (2001). In addition to Landsat images NOAA (National Oceanic and Atmospheric Administration) satellite data were the basis in the study of Nøhr and Jørgensen (1997). The Advanced Very High Resolution Radiometer (AVHRR) from NOAA was applied to calculate the actual and integrated Normalized Difference Vegetation Index (NDVI) to produce maps of primary production. Also Wallin et al. (1992) and Osborne et al. (2001) used AVHRR scenes to analyse spatial and temporal distribution of green biomass by NDVI values. NDVI was also calculated to detect green biomass (Andries et al. 1994, Hepinstall and Sader 1997 and Saveraid et al. 2001), to detect habitat change (Sader et al. 1991) or to map wetland distribution (Roshier et al. 2001). Huettmann and Diamond (2001) used the AVHRR/NOAA Climate Data Set incorporating values of atmospheric temperatures and sea surface temperature. The latter was also used in the study of Haney and McGillivary (1985a). A radar ERS image was used by Thibault et al. (1998) to discover ice-cover on a river.
The number of derived habitat classes can be used as a measure of the scale and it helps to show differentiation possibilities of the analysed habitats. Therefore the number of remotely sensed classes was counted, but unfortunately not always given in the published studies. It differs between 2 and 25 with a mean number of 9.6 habitat (or vegetation / land use) classes (n = 22). In the studies of Avery and Haines-Young (1990), Aspinall and Veitch (1993), Lavers et al. (1996), Hepinstall and Sader (1997), Osborne et al. (2001) and Huettmann and Diamond (2001) image classification was not done as raw data were used or only NDVI values were calculated.
If image classification, for example of a Landsat image, is not done, the obvious advantage is removing the otherwise time-consuming operation of producing a habitat map. The disadvantage, though, is that habitats can be identified but not described well in ecological terms.
Few authors included an accuracy assessment of the satellite image classification. In 11 studies, where details on accuracy of the classified image were given, it varies between 69 % and 98 % with a mean of about 82 %. In 14 studies accuracy assessment was not performed or at least not mentioned. In few of these studies existing land cover classifications were utilized, for example the ITE (Institute of Terrestrial Ecology) Landsat cover map of Great Britain, was used in Dettmers and Bart (1999) and in Mack et al. (1997), and thus details of the classification process were not explicitly mentioned. In all other studies raw data, without classifying the image or NDVI, were utilised.
In many of the investigated studies the bird census period does not fit with the time the satellite images were taken. The differences in time of acquiring a satellite image and performing the bird census period can span up to 20 years. Satellite images and bird census were from the same year in only eight studies (n = 25).
Most analyses were not based merely on satellite imagery, they also used different additional data. In 31 % of the studies aerial photographs - mainly colour infrared pictures - were included. In the majority of these studies they were used to verify image classification. Only in a few studies were terrestrial estimates of percent cover of vegetation types conducted. In nine of the 35 analysed studies, altitude data were used, mainly by generating a DEM. With the help of this data slope, aspect or land ruggedness were calculated. Other additional data comprised soils, water depths, different vegetation measurements such as vegetation height or cover value of a specific vegetation type and man-made features such as roads and buildings.
In several instances spatial texture measures, i.e. patchiness of a specific habitat, spatial configuration, fragmentation, area and boundary length of landscape elements and vegetation types were calculated using GIS. Also, distances between bird records and, for example, agricultural land, coast, shelf edge, roads or buildings were calculated. The additional data were used for further improvements of the birds habitat model. The extent of these data depends on availability, useable resources and ecology of the studied bird species. Especially habitat preferences and sensitivity to human disturbances were taken into consideration. The latter worked with indirect measures such as distance from roads or villages as surrogate measures of human impact.
For the study of relationships between bird species distribution and environmental variables a bewildering array of statistical methods has been used in the 35 studies. Some of them have been specifically designed for these studies (for example Tucker et al. 1997, Aspinall and Veitch 1993), while others apply more general statistical techniques. Univariate and bivariate tests as well as multivariate statistics were carried out. Bayesian models in Aspinall and Veitch (1993), Griffiths et al. (1993), Tucker et al. (1997), Hepinstall and Sader (1997) and regression models in Avery and Haines-Young (1990), Lauga and Joachim (1992), Griffiths et al. (1993), Austin et al. (1996), Nøhr and Jørgensen (1997), Huettmann and Diamond (2001) and Osborne et al. (2001) were mainly used to predict species distribution. Chi-square test, t-test and Mann-Whitney U-Test were used to compare single habitat variables between study plots with and without the species, and between two study areas containing species.
When regarding the previous studies which used satellite-based RS and bird data depicted in chapter 1.5 four main weaknesses are obvious:
1. All but three of the studies (Sader et al. 1991, Wallin et al. 1992 and Nøhr and Jørgensen 1997) were conducted in temperate regions, where wildlife is relatively well-known and species diversity is low compared with the species diversity of the tropics.
2. Little research has been done on multi-species analyses. Most studies involve a single or a small number of bird species, only six studies used more than 20 species.
3. Although many bird species, especially the smaller passerines, show habitat preferences for fine-scale structural and compositional attributes of wildlife habitats, most of the studies derived relatively coarse habitat classes from satellite data to analyse species habitats.
4. In order to specify habitat preferences of birds other environmental factors were often supplemented to the satellite-based habitat model. There is little general information given regarding which of these factors are most powerful for describing species-environment relationships.
These weaknesses pose the following questions:
- How can bird species habitat analyses based on RS data be applied in tropical regions, where bird life has a much higher diversity and knowledge about habitat preference is weak?
- How can satellite-based RS be applied to bird community descriptions? Is it possible on the basis of images to separate and describe habitat preferences of the species within the bird community? Which variables and which methods are best to show species competition?
- How detailed can habitat structures, which are decisive for bird occurrence, be identified using Landsat TM imagery? For which bird species can the Landsat TM based habitat analyses be useful?
- Which additional environmental factors are relevant to species-environment relationships and are practicable taking time and costs into account? On which scale should they be recorded?
Research on the use and the potential of birds, satellite images and GIS for monitoring the environment as well as research on bird species-environment relationships and on bird assembly structure will help to answer some of these questions and enhance biodiversity conservation efforts. Thus the main aim of the present study is to develop and to test a model by combining satellite-based RS, direct survey of features of the ground and bird species census with the help of a GIS.
According to Corsi et al. (2000) species distribution modeling can be divided into two phases. The first phase assesses the species’ preferred ranges of values for the environmental variables taken into account, and the second identifies all locations in which these preferred ranges of values are fulfilled. In the present study bird species distribution will not be predicted over large areas rather it focuses on the first modeling phase, a detailed analysis of species-environment relationships. According to the classification scheme of Table 1.5.1 habitats will be identified and characterized in this thesis.
For this research the grasslands of Serengeti National Park (SNP) in Tanzania were selected exemplarily. As species-environment relationships in SNP are not known a priori they were studied in conjunction with the capabilities of satellite-based RS to characterize habitats. Therefore, mainly a focal-bird approach (Block and Brennan 1993) was followed, where habitat characteristics are mapped at the location where birds were observed.
Three main objectives result from the weaknesses depicted, from the topics discussed in the previous chapters and from the requirement to enhance basic knowledge of bird life in SNP, exemplarily for the grasslands. These objectives are:
1. Analysing species-environment relationships among a large number of species as well-documented relationships are the key points of a monitoring programme.
2. Showing diversity patterns of bird life within the study area, as these areas are of high interest for nature conservation and habitat management.
3. Describing the patterns in avian assembly within the study site.
Additionally, in order to investigate important aspects concerning size, specification and type of bird and environmental data, respectively, in this thesis three different hypotheses will be tested.
- The first hypothesis to be tested is that satellite-derived habitat types more effectively determine species presence and absence than ground-based data. Therefore, I compared the hypotheses that species presence and absence can best be described by establishing statistical relationships between (a) satellite-based habitat types and bird distribution data and (b) between terrestrially measured habitat structures and bird distribution data. Therefore, the number of statistically significant relationships between bird occurrence and terrestrially measured habitat structures as well as between bird occurrence and satellite-derived habitat types have to be compared. Additionally, the same approach will be followed focusing on bird assemblies instead of species presence and absence.
- The second major hypothesis to be tested is that more statistical relationships exist between bird occurrence and coarse scaled satellite-derived habitat types than between bird occurrence and finer scaled satellite-derived habitat types. This approach calls for a comparison of two habitat maps derived from a Landsat Image. One is based on a maximal number of possible habitat types and the second is based on a reduced number of habitat types. Therefore, the habitat types of the first map will be accumulated using several major types only.
- The third hypothesis to be tested is that species presence and absence can best be described if time-consuming bird census is involved above a reduced time effort on bird census. Thus the number of visits per point where bird census was conducted was reduced.
It has to be kept in mind that the intention of this study is to contribute to the development of a practical and efficient monitoring method, especially for remote areas as well as those difficult to access, in order to detect and document human-induced changes in the abundance of species and in the change and quality of the environment. Thus, the effort on mapping and analyses should be minimized as far as possible.
After a short description of the study area (chapter 2) in chapter 3 a description on data and methods used in this study follows. The result chapter is subdivided into two main parts. In the first three subchapters (4.1 - 4.3) some general outcomes (terrestrially mapped features, satellite-derived habitat map) of the study are presented which form the basis for the following subchapters. The second main part focus to the three objectives listed above. Therefore three series of investigations were undertaken with the bird data and environmental variables: the first analyses the responses of individual species of birds to the measured environmental characteristics and the habitat types (chapter 4.4), the second deals with species diversity and bird density (chapter 4.5) and the third investigates the composition of the different bird species on the grasslands (chapter 4.6). In the last chapter 4.7 the three hypothesis will be tested and therefore species-environmental relationships were analysed using different subsets of the data. Chapter 5 gives a critical evaluation of the results, the methodology and the used environmental variables and gives some answers to the questions depicted above. Chapter 6 focuses on validity of the three hypotheses, possible applications and further improvements of the methodology used. An English and a German summary can be found in chapter 7.
SNP in Tanzania was selected, as it best meets the prerequisites for the study purposes for the following reasons:
- The park is one of the largest intact ecosystems of the world, where species-environment relationships are less affected by human disturbance.
- Serengeti has a high diversity of bird species, which may implicate many species-environment relationships.
- Little research has been done on birds in the park, so the findings will be a contribution to scientific knowledge.
- In SNP a well-developed GIS database of several spatial cartographic features exists.
Few studies using satellite-based RS, GIS or birds have been done or published in Tanzania (e.g. Justice et al. 1986, Miller et al. 1989, Wallin et al. 1992 and Homewood et al. 2001). SNP is mentioned three times. Justice et al. (1986) used AVHRR/NOAA data from May 1983 to April 1984 and Campbell and Hofer (1995) from 1990 to calculate monthly NDVI values in Serengeti. Homewood et al. (2001) analyzed land-cover changes, in particular the expansion of arable land between 1975 and 1995 in the Serengeti-Mara ecosystem using Landsat scenes.
Although knowledge on many aspects of the birds of SNP is still poor, some studies are available on birds. Little is known about distribution, population density and dynamics, habitat structure, habitat preferences and migration of birds, as well as effects influencing bird species composition and their distribution. However, an annotated checklist of the Serengeti National Park does exist which was written by Schmidl (1982) and provides a short description of the distribution and status of 496 species. A list of species recorded in the Serengeti-Mara ecosystem was published in Sinclair and Arcese (1995). Information of records on several additional bird species is given by Stronach (1990) and Gottschalk (2001, 2002). Furthermore, a few studies have been done on food supply and breeding seasons (Sinclair 1978), plains birds (Folse 1976, 1981), scavenging birds (Houston 1972, Pennycuick 1971) and duetting birds (for example Wickler 1980). Woodworth et al. (1994) initiated a project monitoring of bird species at two lodge sites in SNP, but it was discontinued.
An overview of the landscape, flora and fauna of the SNP and especially of the study area is given in this chapter. A more detailed description of the Serengeti-Mara ecosystem can be found in Bell (1971), Grzimek and Grzimek (1959), Schaller (1972), Sinclair (1977), Sinclair and Arcese (1995), Sinclair and Norton-Griffiths (1979) and Wit (1978).
With its herds of ungulates and their associated predators SNP is the last remnant of a Pleistocene large mammal ecosystem in all its complexity (UNEP and WCMC 1997). The whole Serengeti ecosystem as defined by Sinclair and Norton-Griffiths (1979) covers about 25,000 km2. The Serengeti was designated a protected area in 1940 and a National Park in 1951, the first in Tanzania. Since 1981 it has been a part of the Serengeti-Ngorongoro Biosphere Reserve which contains 23,051 km2 and includes SNP and Ngorongoro Conservation Area (NCA) in Tanzania and Maasai Mara National Reserve in Kenya. In 1981 it was placed on the World Heritage List.
SNP is located between Lake Victoria and the Great Rift Valley, about 130 km west-north-west of Arusha in north Tanzania, East Africa (Figure 2). The Park has a size of 14,763 km2 and an elevation increasing from 1,150 m in the west at Lake Victoria to 1,850 m in the northeast. A corridor extends westwards to within 8 km of Lake Victoria and a northern sector extends to the Kenya border.
To obtain extensive field data a large study area 1,100 km2 in size which contains 280 survey points was selected in the southeast part of the Serengeti Plains (Map 2). The east and south border on the Ngorongoro Conservation Area, the west is restricted to the longitude of E 35° and the northern extent reaches up to the transition zone of the woodlands but not further north than the latitude of S 02°30’. The south-west extent is restricted by the woodlands of the Oldupai Gorge, which begin south of the latitude S 02°58’. The southern part of the study area is crossed by the main road Arusha/Ngorongoro - Seronera and the northern part by the small road Loliondo - Seronera.
Additionally, four smaller study plots of grassland were selected for this study. These plots were mainly used to conduct data on abundance of the bird species. The plots were placed to represent the bird life of the different types of grassland and of the different geographical regions within the Serengeti Plains. Therefore two sites were placed on short grass types and two on intermediate grass types. Each site was located on a flat, treeless site. Each location was chosen for its topographic and vegetational homogeneity. While the plots “Maasai Kopjes”, Barafu Kopjes” and “Serengeti Plains” were located within the study area, “Simba Kopjes” was placed 18 kilometres west of it (Map 2). Except for the 21 ha large “Maasai Kopjes” all study plots comprised an area each of 25 ha (Map 2 and Figure 3.1.2 for details).
The topography of the study area is gently undulating with a few small, deeper river valleys. The largest and deepest is the valley of the Ngare Nanyuki River and Magungu River in the northern part of the study area, which extends from south-east to the north-west. In addition to the distinctive Naabi Hill several groups of Kopjes are well visible from afar. These are mainly Barafu Kopjes in the northeast and Gol Kopjes in the centre of the study area. Termite mounds occur in the northwest part of the study area, especially around the study plot “Maasai Kopjes”.
The largest river on the study area is the Ngare Nanyuki River, which is located in the northern part and extends over a total length of about 32 km from south-east to north-west. At point number 186 (see Map 2) a tributary which is called Magungu River flows into the Ngare Nanyuki River. Except for some small pools, both rivers contain water only after heavy rains. Furthermore, there are a few additional pools, which are located in deeper depressions, mainly in the central and northern part of the study area. Only the larger pools may hold water the whole year. Lake Ndutu is the largest lake in SNP and is located south of the study area. It is a shallow soda lake and characterized by very seasonal water-level fluctuation.
Rainfall is mainly restricted to the period from November to May with small peaks in December and March/April, although the exact timing varies from year to year. More details for the study area are given in chapter 4.2. Precipitation differs from year to year but the mean annual rainfall is around 515 mm of which 425 mm fall in the wet season from November to May (Norton-Griffiths et al. 1975). This season is followed by a severe dry season (June-October) with 90 mm mean rainfall. Rainfall tends to decrease from the north-west of the study area (mean annual rainfall about 600 mm) towards the south-east (500 mm) to the border of Ngorongoro Conservation Area. The temperature varies from about 29°C at day to about 14°C at night with a relatively constant year-round mean around 21°C. On the open plains of the study area there is an obvious prevalence of cold nights and hot days. Easterly winds prevail.
Approximately 60 % of SNP has a woody vegetation canopy ranging from open bushland to dense rich evergreen wood in some gallery forests. According to Herlocker (1974) Acacia -woodland savanna, where semi-deciduous to deciduous Acacia spec. dominate, cover most of the National Park woodlands and wooded grasslands. These woodlands comprise Acacia tortilis, A. lahai and A. seyal. North of the study area mixed C ommiphora -woodlands are common.
The grasslands of the Serengeti Plains, which are located in the southeast of SNP, are described by Anderson and Talbot (1965), McNaughton (1983) and Banyikwa (1976). Following Anderson and Talbot (1965) and Sinclair and Northon-Griffiths (1979) three main grassland types comprising several subtypes can be grouped:
1. Long grasslands dominated by Themeda triandra and Pennisetum mezianum
2. Intermediate grasslands dominated by Andropogon greenwayii
3. Short grasslands dominated by several species such as couch grass Digitaria macroblephora, Sporobolus marginatus and Michrochloa kunthii.
McNaughton (1983) identified 17 grassland communities, of which six were short grasslands and eight were intermediate grasslands, by using a numerical clustering.
The distinct vegetation near and on the kopjes is more luxuriant. Themeda triandra long grassland occurs and several species of bushes and trees, like Euphorbia candelabrum or Ficus spec., grow close or up the kopjes.
The study area is mainly treeless, except at the drainage valley some trees and small woodlands can be found, especially in the valley of the Ngare Nanyuki River. For more details see chapter 4.3.
According to Justice et al. (1986) highest NDVI values using AVHRR data were calculated on the Serengeti Plains in January (0.37) and April and the lowest in August (0.04). High values of this index are obtained for areas covered by green vegetation and low values for vegetationless areas.
The four smaller sized study plots comprised two short grass plots “Maasai Kopjes” and “Barafu Kopjes” as well as the two intermediate grass plots “Serengeti Plains” and “Simba Kopjes”. Each site consisted of relatively homogeneous vegetation. The main plant species found in the study plots are shown in Table 2.1. While 25 plant species were recorded in all study plots, in the plots of the short grass type species differ between six and nine and in the plots with intermediate grass type 11 and 12 species were found, respectively. “Barafu Kopjes” was mainly characterized by Cynodon
Abbildung in dieser Leseprobe nicht enthalten
Table 2.1: Plant species found in the four study plots (w = woody plant, h = herb, g = grass, se = sedge, s = seed plant, X = present, X = dominant species). English names are known for only the more common species.
Over 550 bird species have been recorded in SNP. The Park is one of the endemic bird areas of the world (Stattersfield et al. 1998) including the endemics Grey-breasted Spurfowl Francolinus rufopictus, Fischer´s Lovebird Agapornis fischeri and Rufous-tailed Weaver Histurgops ruficaudus. Prominent birds of Serengeti include several species of eagles and vultures, Kori Bustard Ardeotis kori, Ostrich Struthio camelus and Saddle-billed Stork Ephippiorhynchus senegalensis.
Three different data sets were investigated for the study. These comprise bird data, a habitat map resulted from a Landsat 7 ETM+ imagery classification and terrestrially measured structures of the grassland at 280 points (see Figure 3).
To build a model of bird species-environment relationships using remotely sensed data and GIS methods, bird census was conducted using two different techniques (chapter 3.1.1 and 3.1.2). Additionally, vegetation and several habitat features were measured or estimated (chapter 3.1.3). The satellite image used in this study had to be interpreted and classified (chapter 3.2.1) and after that ground truth and accuracy assessment of the classified image were done (chapter 3.2.2). Terrestrially measured habitat features and a soil map were utilized in the image classification process to find distinct habitat classes. Before statistical analyses could be carried out the different thematic layers comprising grid and vector data had to be prepared with the help of GIS tools (chapter 3.3.1). Analyses comprise calculations of habitat preferences derived from a specific index (chapter 3.3.2), bivariate (chapter 3.3.3) and multivariate statistics (chapter 3.3.4).
Depending on the habitat or species being targeted many survey techniques can be used. Advantages and disadvantages of survey techniques of birds are discussed in detail by Baillie (1991), Bibby et al. (2000) and Flade (1994). Following Palmeirim (1988) for broad scale analyses, PSC were selected rather than Transect Counts or the Territory Mapping Method. PSC are less time-consuming when collecting count data over a large area. As bird records with this method are taken at fixed locations, accurate assignment of the locations to the classified imagery is easily possible. Furthermore, Transect Counts are walking intensive, which can be risky, especially in tall grass areas because of the occurrence of big cats in SNP. Terborgh et al. (1990) suggest that a combination of different bird census methods must be used to achieve the best possible accuracy in counts of species that possess different social and territorial systems. Therefore, PSC at 280 survey points and Territory Mapping on four study plots, which were located on representative sites of different grassland types, were conducted (see Map 2). Territory Mapping was done to compare results of PSC and to evaluate avian assembly patterns derived from the multivariate statistical analyses.
Abbildung in dieser Leseprobe nicht enthalten
Figure 3: Flow chart of the methodology
Along with the PSC, terrestrial habitat mapping was done on the 280 selected points. The location of the points within the study area was based on the grid of the World Geodetic System (WGS-84). The points were placed on every minute of the grid which corresponds to a distance of about 1.85 km between two neighbouring points. These points and also the corners of the study plot were located in the field using Global Positioning System (GPS). The computed latitude and longitude on the GPS were based on the geodetic datum of WGS-84.
The primary census method was PSC, where I mapped all birds seen and heard at the points within five minutes. Therefore, the area was searched for birds by viewing slowly the entire horizon using a pair of binoculars (Leica 10x42 BA). Bird counts were done throughout the whole day and each point was visited three times during the period of October 1999 to June 2000 (Table 3.1.1).
Abbildung in dieser Leseprobe nicht enthalten
Periods of time PSC were conducted at 280 points on the Serengeti Plains Night surveys and recording birds by playback responses, i.e. playing a tape recording of birds songs and calls and documenting responses, was not used as a census method.
Results of Territory Mapping of four square study plots (see Figure 3.1.2) were used in this research. According to the methodology used for Territory Mapping (Bibby et al. 2000), the locations of all birds present in the plot were mapped on different days. Between eight and 14 visits were made within one year between July 1999 and June 2000. Census work was repeated on each plot at least two weeks after the previous visit. To record as many birds as possible in a study plot, visits were placed in the early morning when birds are most active and proof of presence is most easily obtained. The plots were walked along auxiliary lines, which were spaced in 100 m intervals running from north to south (see Figure 3.1.2). Marking was done with a vehicle driven through the grass a few days or at the latest in the afternoon prior to the bird census. In tall grass areas, where large cats or buffalos were difficult to detect, a vehicle drove behind me keeping a distance of a minimum of 100 m. Most birds singing or showing any breeding behaviour were recorded during the wet season from December 1999 until May 2000, when they were defending territories. The records from all visits were plotted for each species on a species map. Birds exhibiting territorial behaviour tended to appear on the map as clusters. When these locations were combined, individual territories could be identified. A territory was defined, if at least three records were made of a bird singing or showing any breeding behaviour. Assuming that each territory was occupied by a pair of that species, the number of territories was equivalent to the number of breeding pairs in the plot. So the Territory Mapping method provided estimates of population densities for each species present in the study plots.
Abbildung in dieser Leseprobe nicht enthalten
Figure 3.1.2: Sketch of a study plot Territory Mapping
Habitat differences rather than food differences seem to be the main means of “niche segregation” on the Serengeti Plains (Folse 1981). Generally many bird species were more likely to respond to the structure of vegetation than to floristic composition (Cody 1985a, Wiens 1969, Wiens and Rotenberry 1981). Therefore, the main structures of the grassland were mapped. However, vegetation structure also determines the food resources. Characteristics of the habitat features were sampled by taking measurements and estimates of vegetation structure parameters at all 280 points in May and June 2000.
I collected ground cover of woodland, shrub and bare ground within the grassland on each point by visually estimating percentage cover. Woodland cover included all trees and bushes higher than a metre and shrub cover all woody plants less than a metre. Bare ground includes all unvegetated areas (soil, leaf litter and rock). For a few sites - which were burned - the cover of burned area was estimated as well.
Vertical height measurements were taken using a ruler. Two grass heights were sampled: an average grass height and a maximum grass height. These heights were determined by measurements of about 10 different representative grasses or forbs at each point. Measurements were weighted by estimating the coverage of the different grass strata.
An image of the Landsat 7 satellite was used in this study. This satellite was launched on April 15, 1999 by the U.S. National Aeronautics and Space Administration (NASA). It is equipped with an improved imaging system called the Enhanced Thematic Mapper Plus (in the following called ETM+). Landsat 7 extends the design of the earlier Landsat 4 and 5 TM with several improvements, including increased spectral information content, improved geodetic accuracy, reduced noise and reliable calibration (Masek et al. 2001). Furthermore, the spatial resolution of the thermal infrared band has been improved from 120 to 60 m and the ETM+ in use now is equipped with a panchromatic channel of 15 metres resolution (Table 3.2).
Abbildung in dieser Leseprobe nicht enthalten
Table 3.2: Technical summary of Landsat 7 ETM+ sensor
The acquisition of the satellite image focused on three main objects:
1. Date of acquisition should be within the period of time of bird census
2. The image should be cloud free
3. Vegetation should have a maximum of spectral reflectance
A Landsat ETM+ image of February 12, 2000 was selected for this study as later ones were not cloudless and earlier ones show less reflectance of vegetation. The acquisition date of the satellite image did not fit with the peak of the breeding season and the time period in which vegetation measurements were taken, but the date fell in the middle of the period the bird census was conducted in. Figure 126.96.36.199 shows the image location inside of Tanzania.
Image distortions were restored and the image was geometrically rectified to a Universal Transverse Mercator (UTM) grid by the satellite data supplier. This was done by using an affine transform and nearest neighbour resampling algorithm. WGS-84 was used as the reference grid. The location accuracy of the corrected image was given as 15 m.
A combination of Landsat TM bands 5, 4 and 3 assigned to the colours red, green and blue (RGB) was used in this study, as these bands are most sensitive to vegetation structures und characteristics (see Figure 188.8.131.52 and 184.108.40.206) and land-use categories can often be distinguished quite well on the basis of this combination (Janssen 2000, Horler and Ahern 1986, Griffiths et al. 1993). While band 3 is important for plant species differentiation, the near infrared band 4 is useful for vegetation type discrimination and as an indicator, for example, for plant cell structure, biomass and plant vigour. The mid-infrared band 5 indicates vegetation and soil moisture content (see Campell 1996, Sabins 1997).
The objective of image classification is to construct spectral classes from the Landsat scene corresponding to the unique signatures of the land cover categories on the Serengeti Plains. According to Anderson et al. (1976) classification systems based on RS images should fulfill the following requirements:
- The minimum level of accuracy in identifying land-cover categories from RS should be 85 percent.
- Repeatable results should be obtainable from one interpreter to another and from one time of sensing to another.
- The accuracy of interpretation for all categories should be approximately equal.
- The system should be applicable over extensive areas and usable for different times of the year.
Abbildung in dieser Leseprobe nicht enthalten
Figure 220.127.116.11 (above):
View of North Tanzania (Path: 169, Row: 62) at a height of 705 km from a Landsat 7 ETM+ Image of 12.02.2000. The view includes large parts of SNP, Ngorongoro Conservation Area and Lake Manyara National Park. Band 5, 4 and 3 of the satellite image are displayed as red, green and blue. Red line and the 280 dots show the study area and the black line the boundary of SNP.
Abbildung in dieser Leseprobe nicht enthalten
Figure 18.104.22.168 (left):
Location of the satellite image within the Republic of Tanzania
To focus on the study area a sub-scene was extracted from the Landsat Image (see Figure 22.214.171.124 and 126.96.36.199), which corresponds to 3.2 % of the full scene. To enhance the image a linear contrast stretch was made.
Abbildung in dieser Leseprobe nicht enthalten
Figure 188.8.131.52: Full scene and sub-scene characteristics
The middle infrared band displayed in Figure 184.108.40.206 in red colour shows vegetation which almost dried out. As the image was taken after the short wet season, extensive areas are discernible in red. Vital vegetation in Figure 220.127.116.11 is visible in green and brown tones. Woody structures are depicted by a dark, bright green. The blue and violet areas show regions with no vegetation containing chlorophyll, plants are either missing, or they have become more or less dry.
The southeast part of the sub-scene shows extensive green, probably because of local rainfall patterns.
Digital image classification
Image interpretation was done by using the programme package of Earth Resources Data Analysis System (ERDAS Inc., Atlanta), Version 8.4. Although the whole sub-scene was used for classification, the accuracy of the interpretation of the areas outside the park was not assessed and is perhaps scant. A preliminary definition of habitat classes was not possible, as different grass types or subtypes on the Serengeti Plains were not known in detail a priori. Therefore, habitat class separation depended mainly on class separability by its specific reflectance.
Following Palmeirim (1988) an optimal habitat class selection is one in which classes are perceived by the species in consideration as distinct from each other but internally homogeneous. This is difficult in practice as
1. The same habitat map was used for a variety of species
2. Factors which control the distribution of the species were not always known
3. Class separability was limited by the information the imagery supplied.
Multispectral classification was done in two stages, first by a procedure known as unsupervised classification and second by a supervised classification (see Albertz 1991). For unsupervised classification, a clustering algorithm automatically finds and defines a number of classes. For supervised classification, clusters are defined during a training process (Janssen 2000). In the beginning of the classification the number of classes could not be clearly defined. A first idea of class separability and patterns of classes was given by applying the unsupervised classification. Therefore, an Iterative Self-Organizing Data Analysis (ISODATA) clustering algorithm (Tou and Gonzalez 1974) was applied to the image file in order to identify spectrally similar pixels. The unsupervised clustering approach was used to develop the spectral signatures for the habitat classes.
A supervised classification was performed in the second stage. In this stage spectral characteristics of the classes were defined by identifying training areas. Prior knowledge of the locality, based on hundreds of hours which I had spent on the Serengeti Plains, was incorporated to classify the image. Apart from detailed knowledge of the survey area, additional information such as a soil map of the plains from Wit (1978) and measurements and estimates of the vegetation structures at the 280 different points were used (see chapter 3.1.3).
A set of pixels that adequately represents spectral variation within each informational region was identified. This was done with the help of the Region Growing Seed Tool of the ERDAS software. Pixels of similar values to the seed pixel were found by calculating Euclidian distance from the mean of the seed pixel. Areas of similar characteristics were collected to a signature set. To differentiate not only between main grassland types (such as high grass, intermediate grass, short grass), but also among a gradient of, for example, short grass types more than 20 initial spectral classes were selected.
In order to compare signatures of these classes they were visualised as ellipses on a two-dimensional scatterplot (Figure 18.104.22.168). Data file values of band 5 were displayed against the values of band 3. Each ellipse is based on the mean and standard deviation of one signature (Erdas 1999). When the signatures represented similar pixels, the ellipses overlapped and I tried to modify, merge or delete them. Finally, to create a map of high spectrally distinct vegetation classes, the number of classes was reduced to 16.
Information from the training areas was incorporated for use in the classification procedure. The pixels of the image were sorted into classes based on the signatures by using the Maximum Likelihood (ML) classifier algorithm. Instead of other classification decision rules, such as Box classifier or the Minimum Distance to Mean classifier, ML not only considers the mean, or average, values in assigning classification but also the variability of brightness values in each class (Campell 1996). The pixels are assigned to the class (cluster) to which they have highest probability. The basic equation assumes that these probabilities are equal for all classes, and that the input bands have normal distributions (Erdas 1999). A detailed description of the classes is given in chapter 4.3 and the distribution of the habitat classes is shown in Map 4.3.
Three areas were identified, where the classified image showed differences in reality or the classification did not reflect ecological conditions of the area. Two areas lie outside or at the margin of the area in question for which the bird data were analysed (see patterns of the 280 points on Map 2). These areas were reclassified into the right classes by using GIS tools. In the following the areas are described.
Area 1) In the north part of the sub-scene an area was post-classified as “bushed grassland 1”, because intermediate grass containing acacia bushes and shrubs with coverage up to 5 % were in fact present. This area had previously been classified as “intermediate grass 1”. It was not possible to separate intermediate grass with scattered acacia trees from treeless intermediate grass areas on the basis of specific vegetation reflectance, as small acacias show less vegetation reflectance. Therefore, areas in the north part with more than 1 % bush coverage were separated from the “intermediate grass 1” - class and merged to a new generated class: “bushed grassland 1”.
Area 2) Parts of the “intermediate class 3”, a few kilometres on either sides of the northern central part of the small valley of Ngare Nanyuki River, were also reclassified as “bushed grassland 1”. These areas clearly show more than 1 % bush cover, while the class “intermediate grass 3”, usually shows less than one percent bush or shrub-cover.
Area 3) In the central east part of the sub-scene within the “short grass 4” pattern, “intermediate grass 1” was identified with the help of the classifying algorithm. Soil background effects may have caused this. In the field it was not possible to notice any visible differences inside “short grass 4”. Therefore, after selecting the “intermediate grass 1” of this area, a “recode” was done to merge it with “short grass 4”.
The final step of the mapping process was an assessment of accuracy. As in this study only the areas around the 280 points were used for further analyses the accuracy of image classification should have been high especially for these sites. Several sites within the study area in SNP were visited in January 2001. Although timing of field observations fit to the season the satellite image was acquired, the conditions in 2001 were different than in 2000. Compared to the short wet season of 1999/2000, much more rain had fallen in 2000/2001. Random sampling was not possible, as unusual heavy showers in Serengeti prevented visits to all regions within the study area. However, for the ground truth a route was designed to observe as many classes and regions of the study area as possible. A total of 56 samples were collected which represent 20 % of the 280 points.
Comparison of the ground truth and the interpretation of the results were the basis for calculating the accuracy of the image classification. Values of the reference points observed by ground truth were compared to the values of the classified image. The reference points were mainly located on habitat types covering more than 0.5 percent of the study area. To summarize the results of the accuracy assessment, an error matrix was compiled (Table 3.2.2) and the percent of correctly classified pixels was calculated. 51 out of 56 of the reference pixels were classified correctly. The overall accuracy of the classified image, which is determined as the number of correct classifications divided by 56, is 91 %. This high value is greater than the accuracy given in Anderson et al. (1976) as requirements for image classification (see chapter 5.2.1 for further comments). It is therefore sufficient for further analyses.
Spatial data analyses were done using GIS tools (chapter 3.3.1) of the programme package of ArcView 3.2 from Environmental Systems Research Institute (ESRI). Bivariate (chapter 3.3.3) and multivariate statistic analyses (chapter 3.3.4) were performed on SPSS 10.0 (Superior Performing Software Systems, Chicago, IL) statistical software.
 Granite rock outcrops, also known as inselberge
Bachelorarbeit, 100 Seiten
Doktorarbeit / Dissertation, 146 Seiten
Masterarbeit, 108 Seiten
Bachelorarbeit, 109 Seiten
Bachelorarbeit, 61 Seiten
Masterarbeit, 53 Seiten
Diplomarbeit, 94 Seiten
Seminararbeit, 207 Seiten
Diplomarbeit, 102 Seiten