• Keine Ergebnisse gefunden

MASTER’S THESIS

N/A
N/A
Protected

Academic year: 2022

Aktie "MASTER’S THESIS"

Copied!
173
0
0

Wird geladen.... (Jetzt Volltext ansehen)

Volltext

(1)

UNIVERSITÄT SALTZBURG

MASTER’S THESIS

The Application of Satellite Imagery in Rural African Land Development

Prepared for the Department of Geoinformatics at the University of Saltzburg

and submitted by Gerrie van Tonder

S40490

Supervisor: Ann Olivier

March 2013

(2)

i

ABSTRACT

Satellite imagery employs numerous applications and has many uses. Very high resolution imagery has captivated the market and is usually an integral part of any information system or process. This thesis will discuss how a company, in an African setting, applied satellite imagery and its analysis of the results with a view to improving its operations. The procedures and results are compared with other relevant studies and findings, and overall conclusions drawn.

(3)

ii

DECLARATION

The results presented in this thesis are based on the author’s own research at the Salzburg Universität in Austria. All assistance received from other individuals and organisations has been acknowledged and full reference has been made to all published and unpublished sources used. Any views and conclusions expressed in this study are those of the author. This thesis has not been previously submitted for a degree at any institution.

(4)

iii

ACKNOWLEDGEMENTS

The author expresses his gratitude towards Sun Biofuels Pty Ltd, and particularly towards P.W.

Whitehead, the Operations Manager, for granting permission to use their imagery, data and the work done by the author for this company in this regard.

(5)

iv

CONTENTS

ABSTRACT ……….……….. i

SCIENCE PLEDGE ……….….… ii

ACKNOWLEDGEMENTS ……… .……….……… iii

CONTENTS ……… iv

LIST OF FIGURES ………..……….….. vi

LIST OF TABLES ……….……..……….. viii

LIST OF APPENDICES ……….……....…… ix

GLOSSARY OF TERMS ……….……….……... x

1. INTRODUCTION ………..………..………….…. 1

1.1 Motivation ……….……….………...……….. 1

1.2 Approach ……….………...……….. 3

1.3 Expected results ……….……….……...……. 3

1.4 Issues not discussed ……….………..……….…… 3

1.5 Intended audience ……….………..….. 4

1.6 Thesis structure ……….. 4

2. LITERATURE SURVEY ………....……. 5

2.1 Change-over-time analysis ……… 5

2.2 Population estimation ……… 6

2.3 Land-use classification ………..……….… 7

2.4 Stratification and canopy calculation ………..………….…. 9

2.5 Certification ……….……. 10

(6)

v

3. APPROACH ………...…………. 11

3.1 Theoretical foundation ………...…….……….... 11

3.2 Applied applications ……….……..………..…… 12

3.3 Tools ……….…….. 13

3.4 Test area and data sets ……….……….………….….…… 13

3.4.1 Mozambique ………..……… 15

3.4.2 Tanzania ……… 16

3.4.3 Conclusion ……… 16

4. EVALUATED PROCEDURES AND OUTCOMES ……….……….. 18

4.1 Change-over-time analysis ………..…………..……. 18

4.2 Population estimation ……….……... 24

4.3 Land-use classification ………..………..….. 31

4.3.1 High resolution imagery ………..………. 31

4.3.2 Fire damage to AOI immediately prior to image acquisition ……….……… 33

4.3.3 Phases of the supervised classification process ……….……….….. 35

4.3.4 Procedure followed ………..……… 36

4.4 Stratification and closed-canopy calculation ………..………...….………….. 39

4.5 Certification ……….... 42

4.6 Conclusion ……….. 45

5. CRITICAL ANALYSIS OF THE RESULTS ………..……….………… 46

5.1 Change-over-time analysis ……….………. 46

5.2 Pattern recognition ……….………. 47

5.3 Land-use classification ……….……….….…... 47

5.4 Stratification and closed-canopy calculation ……….………..……….…. 49

5.5 Certification ………..………… 49

5.6 Conclusion ……….. 50

(7)

vi

6. FUTURE WORK ………... 52

6.1 Southern Sudan ……….. 52

6.2 Solomon Islands ……….………. 52

6.3 Cambodia ……….……….……….. 52

6.4 Tanzania ………..………. 52

6.5 Conclusion ………..… 55

7. CONCLUSION ……….………….…… 56

REFERENCES ………..……….….…….... 60

APPENDICES 1–4

(8)

vii

LIST OF FIGURES

Figure 1: A schematic of the thesis structure ……….……….. 4

Figure 2: Kisarawe ……….……….….…………...… 14

Figure 3: The location of Kisarawe in Tanzania ……….………….. 14

Figure 4: Quinta [west] and Nhamatanda [east] ……….…….…….…... 15

Figure 5: The location of Quinta and Nhamatanda in Mozambique …….……….. 15

Figure 6: Charcoal activities at Kisarawe, Tanzania ……….………... 18

Figure 7: Nhamatanda AOI ………..….. 22

Figure 8: Kisarawe 1991 ………..….…..……. 24

Figure 9: Kisarawe 2009 ……….….…. 24

Figure 10: Dry grass: the main material used for roofing in the Quinta area .….. 25

Figure 11: Sengupta et al. (2003, p.2) Structures discovered via pattern recognition 26 Figure 12: A typical village: inconsistent shapes and materials used in structures 27 Figure 13: The same village at a different scale ……….. 27

Figure 14: Wei and Prinet (2005,p.3) Structures discovered via pattern recognition 28 Figure 15: Various machambas (informal farming areas) ………..……… 29

Figure 16: Changes in machambas observed within one year ……….….…... 29

Figure 17: The visibility of structures: square and round in shape ……….. 30

Figure 18: The visibility of the same structures at a different scale ……...……… 30

Figure 19: The complexity of a land-use class ……….…..…….…… 32

Figures 20 and 21: Burned areas that complicated the analysis process …….…… 32

Figure 22: Diverse vegetation, making classification of individual trees difficult… 33 Figure 23: The landscape of Kisarawe, Tanzania ……….………..…… 34

Figures 24 and 25: Riverine forest as classified from World-View 1 imagery ..….. 36

Figure 26: The analysis process ………...…….... 38

Figure 27: Variance and complexity of vegetation complicates stratification ... 40

Figure 28: Plot size ………....…... 40

Figure 29: Visual results of classification ……….…….. 40

(9)

viii

Figure 30: A breakdown of the height classes ……….…..…. 41 Figure 31: An example of results of a search of archive material ..………..………. 53 Figure 32: Makaani AOI (pink) and Biga West (yellow) in Tanzania ……..….……. 54 Figure 33: The price and product structure of archive material ………..….….. 54

(10)

ix

LIST OF TABLES

Table 1: A result summary of a change over time analysis ………..……… 24 Table 2: Summaries of the data set results of a land-use classification ……..…... 38 Table 3: Summary of the data set results of a land-use classification

(analysis data) ………..……. 39 Table 4: Suggested plot sizes (Zhongming et al., 2010) …….………....….. 41

(11)

x

LIST OF APPENDICES

APPENDIX 1:

Baseline and land-change mapping

APPENDIX 2:

Vegetation stratification

APPENDIX 3:

RSB screening for SBF Tanzania

APPENDIX 4:

Land classification and Population count – Biga West area

APPENDIX 5:

Baseline and Land Change Mapping Nhamatanda

(12)

xi

GLOSSARY OF TERMS

AOI: area of interest DEM: Digital Elevation Model MAXLIKE: Maximum likelihood MINDIST: Minimum distance

EIA: Environmental Impact Assessment SIA: Social Impact Assessment

GPS: Global Positioning System

NDVI: Normalised Difference Vegetation Index LIDAR: Light Detection and Ranging

(13)

1

1. INTRODUCTION

This introduction outlines the motivation, the expected results, the approach, the issues not discussed and the intended audience of this thesis. It concludes with a diagrammatic representation of the thesis structure.

1.1 Motivation

Rural land development is a continuous process on the African continent. Business entities acquire land in order to start developments that will hopefully create jobs, improve food and energy production and result in financial gain. Satellite imagery is one option that can be utilised to assist in the various phases of the land-development process. There are several alternatives to satellite imagery that enable researchers to gain spatial knowledge of an area of interest (AOI), such as aerial photography and use of Google Earth. Where operations are conducted in remote areas, such as countries like Mozambique and Tanzania in Southern Africa, the options are very limited. Owing to the remoteness of the AOIs in this thesis, aerial photography was not considered as this type of data collection would have exceeded satellite imagery in both time and cost. Google Earth imagery is updated regularly as new data sets are acquired. However, such updates do not occur frequently in African as high-resolution imagery and acquisition are driven by demand and profitability on this continent.

Betsy Kenaston, the Technical Accounts Representative of LANDSAT INFO, commented in a personal communication (Monday, February 04, 2013) that current satellite imagery is largely acquired on a scheduled manner in developed parts of the world, such as North America and Europe. In the third world, new imagery is only captured when specifically requested. High-resolution imagery is still fairly new technology today and is therefore very expensive. It is logical, though, that its cost can be more readily justified in densely populated areas. In addition to this, imagery captured by certain sensors (for instance, World-View) can only be purchased by entities residing in Northern America as of 2012 (Kenaston, personal communication, February 04, 2013).

(14)

2

If businesses want to acquire imagery in rural Africa, they very rarely find partners to assist in carrying the cost. Therefore, the acquisition of high-resolution imagery needs to be properly motivated, the results properly explained and the wide array of applications highlighted.

These applications include the following:

• Assistance in land classification

• Assistance in land acquisition

• Assistance in ensuring environmental responsibilities are met

• Assistance in monitoring operational activities

• Assistance in population calculations

• Assistance in calculating population compensation

• Assistance in vegetation stratification

• Assistance in carbon sequestration

• Assistance in production monitoring.

Projects conducted by the author can be classified into two groups, namely acquiring new data sets and ordering archived data sets. These are the results of the needs as expressed by a client. Various techniques were considered in an effort to satisfy these needs. In every case, satellite imagery and the resultant analysis and study were deemed the most efficient and cost-effective manner of obtaining results. Owing to the fact that the acquisition process can be slow, certain permutations were encountered with the acquired data sets.

The main permutations are cloud cover in the LANDSAT imagery and burned areas in the GeoEye and WorldView data sets.

The research question in this thesis is as follows: “Is satellite imagery and its resultant analysis the most effective means of guiding the land-development process?”

The aims of the thesis may be stated thus:

• To evaluate the contributions made by satellite acquisition and analysis with respect to various phases of the land-development process

(15)

3

• To research alternative methods, procedures and principles which might prove more effective

• To describe how various permutations were overcome in order to obtain accurate results.

1.2 Approach

In order to answer the research question of this thesis, a list of criteria according to which satellite imagery will be evaluated for effectiveness should be offered. The value of other conventional means and resources must be measured against the same criteria.

These criteria will represent several key objectives that will need to be achieved in order to deem the land-development effort a success. These objectives are aimed at environmental, social and financial responsibility, and long-term viability.

The criteria that will be applied in this thesis are:

• Change-over-time analysis

• Population estimation

• Land-use classification

• Stratification

• Certification.

These five criteria represent vital objectives that are all integral to responsible land development as they are measures of change, improvement and accountable management of the land in question.

1.3 Expected results

It is expected to prove that satellite imagery is indeed an integral tool in the land- development process and that land-use development and planning cannot be conducted in an environmental and socially responsible manner without the use of satellite imagery. This will be achieved through the examination of recent projects, which will prove that satellite imagery has several advantages over other conventional resources.

(16)

4

1.4 Issues not discussed

The limitations of the thesis have been identified as follows:

• The strengths and weaknesses of the various software packages will not be discussed

• An integral part of the biofuel production process from a land-development point of view, namely carbon calculation and sequestration, cannot be discussed here as not enough use was made of satellite imagery in this regard prior to the dissolving of the company Sun Biofuels

• Geographic information systems (GIS) and satellite imagery analysis can play a role in the process of land-use development, but were not utilised in this instance and hence have not be addressed in this thesis.

1.5 Intended audience

The intended audience of this paper is any organisation contemplating development in Africa, as the paper aims to highlight the possibilities accompanying the start of such a venture, along with the encountered permutations.

1.6 Thesis structure

Figure 1: A schematic of the thesis structure

(17)

5

2. LITERATURE SURVEY

The literature consulted in this paper can be classified into several main topics. These closely resemble the main criteria chosen to evaluate the effectiveness of satellite imagery and analysis in rural Africa:

• Land classification

• Change-over-time analysis

• Population calculations

• Certification

• Vegetation stratification and closed-canopy calculations.

The literature was consulted in order to evaluate whether the criteria as listed earlier have been met in a satisfactory manner. Therefore, the literature has been grouped to relate explicitly to the specific criteria in question. The literature the author has referred to covered one or more of the following:

• It offered alternative procedures, results and opinions

• It supported the procedure that was employed

• It described a procedure/process

• It explained key words/concepts.

2.1 Change-over-time analysis

The relevance and value of satellite imagery was researched, particularly as studied by Coppin, Jonckheere, Nackaerts, Muys, and Lambin (2004).

The United Nations Institute for Training and Research (UNITAR) workbook on time series analysis (Eastman, McKendry, & Fulk, 2007b), Coppin et al. (2004) and the literature provided by Eastman (2012) in the IDRISI Selva manual were used as a reference to the time-series-analysis procedure. The IDRISI Selva manual explained the difference between the two main supervised classifications, namely minimum distance (MINDIST) and maximum likelihood (MAXLIKE). It also discussed the scenario in which each is more appropriate. The basic requirements for performing the analysis are also listed here, as well as the difference

(18)

6

between time-series analysis and change-over-time analysis: time-series analysis utilises more than two data sets as opposed to change-over-time analysis utilising only two.

Roy et al. (2002) described the biggest challenges encountered when an analysis of this nature is undertaken, of which seasonal variation is the most noteworthy. It is indeed of paramount importance that data sets of the same season are selected, as vegetation can vary greatly from season to season. This is especially true of sub-Saharan Africa, because of the annual rainfall patterns. This was also mentioned by Coppin et al. (2004).

Myneni, Hall, Sellers, and Marshak (1995) and Verbesselt, Hyndman, Newnham, and Culvenor’s (2009) findings regarding the use of the Normalised Difference Vegetation Index (NDVI) ratio for classification purposes were of particular interest. The NDVI ratio played an important part in the change-over-time analysis.

The NDVI ratio is the most common vegetation index in use today, although others have been developed. Myneni et al. (1995) briefly explained the rationale behind the NDVI ratio, which revolves around the photosynthesis process of vegetation. The NDVI ratio is absolutely crucial to this thesis as it represents the main means of classification of and discrimination among land-use classes.

Verbesselt et al. (2009) described a procedure for reducing seasonal variation and noise.

According to them, the procedure could be applied to LANDSAT data sets, although they used Moderate-resolution Imaging Spectroradiometer (MODIS) data sets. While this procedure appears to be a scientifically sound method, it was not needed because of the seasonal similarity of the LANDSAT data sets used in the projects described in this thesis.

2.2 Population estimation

Prior to acquisition, research was conducted to establish whether pattern recognition could reduce time and resource requirements while performing population estimation. Based the research conducted by the author, it was decided to continue utilising a manual method.

The procedures for pattern recognition as described by Orun (2004); Sengupta, Kamath,

(19)

7

Poland, and Futterman (2003); and Wei and Prinet (2005) were therefore analysed. This was in order to research whether this procedure (pattern recognition) might have been viable with regard to the focus of this paper, namely Quinta AOI.

The research of Orun (2004) indicated that only one band could be used successfully in the process called pattern recognition, with the additional advantage of saving time and resources. Furthermore, he concluded that his results could be classified into numerous classes of which four were man-made structures. Whether one band or all four available were used, this procedure still needed the purchase of the individual bands. It also focused on man-made structures of a formal nature, built with brick and mortar (Orun, 2004) as opposed to the in-situ material of branches and grass as were present at the Quinta AOI.

Sengupta et al. (2003) added a new dimension to the research in that they utilised the corners of structures in order to locate man-made structures in their data sets. While this is noteworthy, it was not applicable to the challenge examined in this thesis.

Wei and Prinet (2005) used contour extraction to conduct feature extraction or pattern recognition. This was not applicable to the Quinta setting as it would have necessitated the purchase of yet another data set. Additionally, the structures which needed to be recognised at Quinta were too small for this procedure to be effective.

Owing to lack of compatibility of the procedures as mentioned above, it was decided that a manual procedure would be utilised. The size of the AOI was also significantly smaller than the study areas mentioned in the research employing these procedures.

2.3 Land-use classification

The work of Anderson, Hardy, Roach, and Witmer (1976) was used to indicate the importance of satellite imagery in land-use classification.

Corona, Lamonaca, and Chirici (2008) and Petropoulos, Kontoes, and Keramitsoglou (2011) offered techniques on classifying burned areas. One of the biggest permutations

(20)

8

encountered in this particular project was that some areas of the AOI had been damaged by fire. The work of Corona et al. (2008) suggested that data-set capturing should be conducted as soon as possible after the fire. However, this is not practical as companies such as Landinfo only guarantee capturing data within eight weeks after payment. Furthermore, Corona et al. (2008) assumed that only one fire would be encountered in the research project. Unfortunately, in the cases discussed in this thesis, numerous fires occurred over an extended period of time.

Petropoulos et al. (2011) were also researched in order to gauge whether it would be possible to extract the various land-use classes from the burned areas. The value of their work is reflected in the fact that they acknowledged the limitations of utilising a single data set when attempting to derive such land-use classes.

Pereira et al. (1999) and Kiang, Siefert, Govindjee, and Blankenship (2007) commented on the merit of utilising the near infrared band (NIR), as it offers a superior response and reflectance of leafy vegetation that has been burned.

The work of Scholes (1997) was used to describe the vegetation that was present in the areas that burned. The vegetation in the case studies is fairly homogenous and can be classified as savannah, if categorised according to the definition supplied by Scholes (1997).

The observations of Hudack et al. (2004) are relevant to this thesis as they warn about possible misclassification of burned grass areas as being of high intensity. Such areas may appear intense when the image is classified, but, in reality, the fire that sweeps through an area covered by grass is totally different and, indeed, less intense than that which burns in a wooded area.

The procedure as described by Eva and Lambin (2000) was considered as an alternative classification method with regard to burned areas. Unfortunately, the thermal band was not offered in the image data-set packages that were acquired for the purpose of classification.

At the time of acquisition, GeoEye and WorldView sensors only offered four bands, which did not include a thermal band.

(21)

9

Samaniego and Schultz (2009) and Perumal and Bhaskaran (2010) described the various phases that make up the process of classification. The classification process is crucial to proving that satellite imagery is superior to aerial photography because the cost implication between these two sources of data is becoming smaller as time progresses.

Yet, satellite imagery and the advances made in its application value far surpass those of aerial photography.

Smits, Dellepiane, and Schowengerdt (1999) discussed the value and importance of the traditional classifiers such as the K-nearest neighbour algorithm and MAXLIKE. They also commented on the subjectivity of training-site development (Smits et al., 1999). This is very important to this thesis as training-site development was seen as the most important phase in the classification process. The rationale applied here is that if the training sites are of high quality, unique and accurate, the result of the classification will be of a high accuracy, irrespective of which classifier has been used. Furthermore, it must be noted that the classification process itself is fairly short and, therefore, it is relatively quick to do comparative classifications, using various classifiers.

2.4 Stratification and canopy calculation

Although stratification and canopy calculations were executed by means of a manual method, research was conducted to determine the value and application value of using Light Detection and Ranging (LIDAR) imagery as an alternative.

In order to verify whether LIDAR imagery would have yielded superior results, the work of Baltsavias (1999) and Dubayah and Drake (2000) was researched.

The work of Lefsky, Cohen, Parker, and Harding (2002) and Lim, Treitz, Wulder, St-Onge, and Flood (2003) were consulted, as these explained the principles of LIDAR imagery and outlined the types of LIDAR imagery available, as well as commenting on the expected applications thereof.

(22)

10

With respect to stratification efforts, Zhongming, Lees, Feng, Wanning and Haijing (2010) were used as a source of reference in order to compare the procedure and suggested plot sizes.

2.5 Certification

Sun Biofuels decided to initiate preparations for certification via an organisation called the Roundtable for Sustainable Biofuels (RSB). According to their factsheet (“Making Real the World’s Commitment to Sustainable Biofuels,” 2013), they developed and maintained a Global Sustainability Standard for biomass and biofuels production. It is applicable to any region and any feedstock from first- or second-generation biofuels and it covers the entire supply chain from initial land-use acquisition to the end product and resultant clients.

Devereaux and Lee (2009) were researched as their document investigates the process of forming policy with regard to sustainable biofuel production.

Woods and Diaz-Chavez (2007) and Searchinger (2009) were quoted to explain the terminology involved in certification. The former afforded the definitions of certification terminology and provided an overview of the process (Woods & Diaz-Chavez, 2007). The latter came to the conclusion that the production of biofuels is accompanied by certain risks (Searchinger, 2009). This is included to highlight the negative sentiments that are harboured by the international community towards biofuels. It also underlines the pressure under which a company functions to ensure that its operations are above standard with regard to environmental and social expectations. This, in turn, serves to prove that the added advantages offered by satellite imagery highlight and uphold the integrity of a company whose operations are judged as less than ideal by the international community.

(23)

11

3. APPROACH

This section comprises an outline of the theoretical foundation upon which this document is based, the applied applications identified and the tools utilised. The subsection on the test area and data includes a description of Sun Biofuels’s operations in Mozambique and Tanzania. The section is rounded off by the conclusion.

3.1 Theoretical foundation

During the initial commencement of converting the AOIs into commercial farming areas, certain challenges arose from a management point of view.

Firstly, a development plan had to be drawn up. This plan had to include certain target dates, as well as tangible and measurable spatial references for production control and measurement. For example, the total number of hectares prepared for planting and those actually planted, as well as the total length of roads developed, had to be specified.

Secondly, a means of expressing environmental and social awareness and responsibility had to be recorded.

Thirdly, a means of preparation for certification by an industry body had to be implemented and maintained.

The last challenge was to visualise the project, its progress and development somehow and then convey these concepts to board members and investors in offices on another continent.

It was felt that data sets of satellite imagery would aid in meeting all these challenges, adding value to the project and, hence, justifying the expenditure of acquisition and analysis.

(24)

12

Such data sets were thought to have the following value:

• The imagery of the AOI could be visually studied and displayed as a common reference

• The AOI could be broken down into various land-use classes

• These land-use classes could be quantified, accurately defined and therefore properly managed

• Social/environmental awareness and responsibility could be proven

• A history, stretching over several decades, of each AOI could be compiled to indicate the state of the AOIs before the activities of the company

• A snapshot of the state of the AOI could be provided as a visual image of its condition just prior to commencement of development

• In future, carbon quantification could be conducted

• The imagery would assist in all subsequent environmental and social impact assessments (EIAs and SIAs).

If one took into account the vast improvement in the field of satellite imagery at the time, it was deemed a worthy expenditure. The resolution of the purchased imagery was 50 centimetres.

According to Jin and Sader (2005), satellite sensors would be appropriate for this procedure because they provide consistent measurements which are well suited to capturing and measuring the influence of change. Such changes may be natural, for example, insect attacks; or may be due to the actions of man, for example, deforestation and farming (Jin and Sader, 2005).

3.2 Applied applications

Apart from visual aid, the acquired high- and low-resolution imagery was used in the following applications:

• The division of the area in land-use classes, using supervised and unsupervised methods

• The division of vegetation into classes, using their NDVI values

• The changes in land-use classes over an extended period of time

(25)

13

• The division of vegetation into approximate height classes

• The digitising of all building infrastructure to calculate population density of the AOIs

• The digitising of informal farm lands present on the acquired AOIs. This was in order to calculate their size and impact on the project.

3.3 Tools

Two types of imagery were used: purchased high-resolution imagery and low-resolution imagery as sourced from government institutions through the World Wide Web. The high resolution imagery is the focus of this paper.

Two well-known software packages were used as an interface with the imagery data sets.

The one was used to compile and analyse the imagery and the other to convert the results into vector data for presentation.

Standard industry analysis techniques from the image processing software were applied to arrive at the expected end results. The following were used:

• NDVI calculation

• Training-site capturing

• Spectral-signature development

• Supervised analysis

• Unsupervised analysis

• Image calculations

• Calculation of slope classes and aspect values

• Raster to vector conversion.

3.4 Test area and data sets

The data sets and locations discussed in this thesis are all part of an enterprise called Sun Biofuels (http://sunbiofuels.leo.titaninternet.co.uk/index.html). Please refer to Figures 2 to 5.

(26)

14

Figure 2: Kisarawe (Google Earth Incorporated, 2013)

Figure 3: The location of Kisarawe in Tanzania (Google Earth Incorporated, 2013)

(27)

15

Figure 4: Quinta [west] and Nhamatanda [east] (Google Earth Incorporated, 2013)

Figure 5: The location of Quinta and Nhamatanda in Mozambique (Google Earth Incorporated, 2013)

3.4.1 Mozambique

In a description of their Mozambique initiative (“Projects: Mozambique,” 2013), it becomes clear that this development is Sun Biofuels’s flagship operation. It is located at Chimoio in the Manica Province of Mozambique. The company spent 18 months restoring the farms

(28)

16

and preparing the land for conversion to Jatropha farms. One thousand hectares were planted with Jatropha curcas, a tall, bushy plant producing inedible fruit, in January 2009.

Germination was over 90 per cent and the plants flourished. The first fruit was harvested from these trees in April 2010 (“Projects: Mozambique,” 2013).

3.4.2 Tanzania

According to their website, Sun Biofuels started operating in Tanzania early in 2006 and they were one of the first entities to enter into the biofuels production sector in the country. The operation in Tanzania was located in the Kisarawe district of the country, about 70 kilometres north-west of Dar es Salaam. The company secured a 99-year lease on 8,000 hectares of degraded coastal forest. No food cropping or community buildings were displaced and no communities were moved. The company planted the first 600 hectares of Jatropha at Kisarawe in November 2009 (“Projects: Tanzania,” 2013).

3.4.3 Conclusion

The scope and scale of the operations of Sun Biofuels, in conjunction with the locations and nature of the individual AOIs, necessitated the use of satellite imagery. The AOIs are located far from urban areas and resources such as businesses specialising in aerial photography and land surveying are not readily available. Imagery of high and low resolution was acquired in order to satisfy a wide range of requirements regarding spatial and temporal dimensions. The high-resolution data sets consisted of red, green and blue (RGB) imagery for visual interpretation. In some cases, these sets included the individual bands for formal, orthodox satellite-imagery analysis.

In addition to imagery data sets, several software packages were used in order to analyse satellite-imagery data sets and to develop the resultant vector data sets. These software packages were utilised in order to perform several data manipulation techniques, such as:

• supervised and unsupervised classification

• NDVI calculation

• the digitising of training sites

• the development of all necessary spectral signatures in order to perform classifications.

(29)

17

Imagery data sets were acquired for all AOIs located in Tanzania and Mozambique. The end results of these techniques and processes were always vector data sets to be used as a guideline for land development, planning and management. The imagery data sets were still employed as a visual aid to management and operational teams. They also acted as a source of reference for certification purposes, as well as visual proof of environmental awareness and responsibility.

(30)

18

4. EVALUATED PROCEDURES AND OUTCOMES

This section comprises information on change-over-time analyses, population estimation, land-use classification, and stratification and closed-canopy calculations. An exposition of Sun Biofuels’s certification follows and the section is drawn together by the conclusion.

4.1 Change-over-time analysis

Change-over-time analyses were completed for all new AOIs where no previous commercial exploitation had been conducted. It was felt that a 20-year period would suffice to indicate the changes that took place over an extended period of time at these locations. The cycle commenced in 1990 and finished in 2009/2010 (see Appendix 1).

According to Coppin et al. (2004), satellite remote sensing has been used for a long time to locate and measure changes in areas of land-use change. At the time, it was believed to be the only application that could be utilised to measure these changes credibly (Coppin et al., 2004).

Owing to the time frame involved (commencement date in 1990), the only source of satellite imagery was LANDSAT. This procedure was used to quantify the volume and rate of degradation and to measure the change of the AOI, in particular, the indigenous forests prior to the new development. Indicating the general state of the AOI prior to development (for example, Figure 6) would also aid in certification processes at a later stage.

Figure 6: Charcoal activities at Kisarawe, Tanzania

(31)

19

The image-calculation procedure included in the software from Clark Labs proved not to be applicable, owing to the nature of the various data sets which did not exactly line up with one another. Therefore, a manual procedure was used.

The steps of the manual procedure were as follows:

• In-field teams digitised areas that, in their opinion, were representative of each of the land-use classes they had identified

• The imagery was independently classified, using supervised and unsupervised methods after in-field visits by the GIS consultant. He used his own training sites

• The training sites of the in-field staff were then used to prepare another supervised classification

• Spectral signatures were developed for all classifications

• Checks were performed to search for possible signature overlaps

• The results were evaluated in-field

• The best classification was used to subtract the 2010 data set from the 1990 data set for each land-use class

• The result was an area and percentage change for each land-use class.

According to the Time Series Analysis workbook from UNITAR (Eastman et al., 2007a), one data set should be deemed the independent variable and the other, the dependant. This was indeed the case. Because the latter 2009 data set could be ground truthed, it was used as the independent variable. The 1989 data set, however, would not be used. Therefore, the location and size of the training sites were derived from the 2009 data set, although the same locations were used in the 1989 data set.

The two data sets were classified separately, utilising the same training sites to ensure that the results and the comparison would be credible. The nature of the training sites was such that the MAXLIKE classification was used to derive the percentage change. Owing to the fact that the various land-use classes were defined independently (and only after comparing the classification), the implication is that they should have corresponding characteristics, even though scatter and atmospheric differences were likely to have been present.

(32)

20

The training sites in both data sets were captured on a composite derived from the respective data sets. This means the training sites and their unique characteristics were captured with their specific levels of scatter and atmospheric conditions. These factors apply to the whole data set and should therefore not be of concern when results are compared.

Forest remnants, for example, are compared as they are classified with the understanding that that is indeed what they represent and nothing else. The study was undertaken to determine changes in size of the land-use classes only; the condition of the land-use class in general has not been investigated. The fact that only four or five land-use classes were derived did make the classification easier. It was initially the intention to derive at least ten land-use classes, but the client insisted on fewer as these classes would form the basis for future planning.

Because of the constraints in finding applicable data sets dating from the early 1990s, it was necessary to consider using LANDSAT MS (multi-spectral) in conjunction with LANDSAT TM (thematic mapper). The minimal differences between these two sensors were not considered a threat to the validity of the conclusions and outcomes as the training sites were applied individually to each data set from a spatial point of view. This means that only the spatial location of the training sites is the same for the different data sets. Spectral signatures were derived from each set of bands for each data set. This ensures that the characteristics of the derived land-use classes are measured only in the data set from which they have been drawn.

In this sense, it is possible to compare results derived from multi-spectral data sets to results derived from thematic-mapper data sets. This is because the results reflect the land-use classes as found when manipulating the bands present in the data set. The differences encountered between the two sensor types are nullified. A very important factor that assists in an accurate comparison is that the resolution of the data sets is identical, namely thirty metres.

According to Eastman (2012), MINDIST performs very well when standardised distances are used. Indeed, it often outperforms a MAXLIKE procedure when training sites have high variability.

(33)

21

After in-field verification was conducted on the 2009 classification and, interestingly, contrary to the findings of Eastman (2012), the MAXLIKE classification was preferred to the MINDIST classification. The training sites were fairly uniform and were unique in their characteristics, hence, the choice of MAXLIKE instead of MINDIST, although both classifications were performed.

The following permutations were encountered during this process:

• The resolution of 30 metres complicated collection, recognition and mapping of the individual land-use classes. This was because some of the training sites eventually comprised only one or two pixels on the LANDSAT data set and proved too small for successful classification. Owing to the nature of the AOI, numerous training sites for each land-use class were identified, which led to reservations as to whether the data were homogenous enough to represent one land-use class.

• Owing to the weather patterns in the countries concerned, seasonal changes had to be eliminated by searching for data sets from the same months in the years 1989 and 2009/2010. Thus, the data sets that were used had been acquired in the same month, although two decades apart. According Coppin et al. (2004), change-in-time series is often masked by the seasonality-driven variations in yearly temperatures and rainfall.

Existing change detection techniques minimise seasonal variation by focusing on specific periods within a year (for example, the growing season). Seasonality was taken into account, mainly because the data sets acquired were cloudless.

In many cases, the infrared band can be used to analyse areas with cloud cover present in the data set, as this particular band is unaffected by clouds. However, this cannot be used alone where classification revolves around vegetation, as both red and NIR are needed to calculate NDVI values.

Unfortunately, clouds complicated this effort. According to Roy et al. (2002), two major challenges emerge. Firstly, methods must allow for the detection of changes within complete long-term data sets, while simultaneously accounting for seasonal variation.

Secondly, estimating change from remotely sensed data is not straightforward, since any time series contains a combination of seasonal, gradual and abrupt changes, in addition to

(34)

22

noise that originates from remnant geometric errors, atmospheric scatter and cloud effects.

The data sets that were selected were free of burned areas, but unfortunately did include areas covered by clouds.

Data sets for these remote regions are less frequently captured than in first-world areas. This becomes evident when searches for data sets on the United States Geological Survey (USGS) website for Africa are compared with searches for data sets located in the USA or Europe.

Owing to this, choices are limited and certain compromises needed to be made when the selection of data sets was initiated. The 1990 LANDSAT data set for AOI Kisarawe, for instance, had 15 per cent cloud cover. It was discussed and accepted that another data set of the year 1991 would be used (see Figure 7). The values for the cloud-covered areas in the 1990 data set were replaced by cloudless values from the 1991 data set. The rationale here was simply that not much change is expected to occur in one year and especially not during the year immediately after the start of the analysis.

Figure 7: Cloudless view of Nhamatanda AOI (ESRI, 2013)

(35)

23

According to Myneni et al. (1995), the NDVI time series is the most widely used vegetation index in medium-to-coarse scale studies. The NDVI is a relative and indirect measure of the amount of photosynthetic biomass and is correlated with biophysical parameters such as green leaf biomass and the fraction of green vegetation cover. The behaviour of such parameters follows annual cycles of vegetation growth. This is true and particularly applicable to this study, as vegetation covers more than 90 per cent of the AOI. The NDVI values derived from the data sets played an integral part in the classification procedure as bands 1, 2, 3, 4, 5, 7 and NDVI were used in this process. One of the advantages of LANDSAT imagery is the number of bands on offer when compared with GeoEye and World-View, which, at the time of acquisition, only offered four. Although the NDVI indicator was used for this study, the UNITAR workbook for Forestry (Eastman, McKendry, & Fulk, 2007a) suggested that a ratio of bands 4 and 5 could be used to highlight the loss of foliage and changes in the canopy moisture of dying forests. This study, however, is focused on the disappearance of forests and not dying forests. Therefore, the NDVI was used.

Verbesselt et al. (2009) assessed Breaks For Additive Season and Trend (BFAST) for a large range of ecosystems by simulating NDVI time series with differing amounts of seasonal variation and noise and by adding abrupt changes with different magnitudes. They used Moderate Resolution Imaging Spectroradiometer (MODIS) 16-day image composites to detect major changes in a forested area in south-eastern Australia. The approach is not specific to a particular data type and could be applied to detect and characterise changes within other remotely sensed image time series (for example, LANDSAT). It could also be integrated within monitoring frameworks and used as an alarm system to provide information on when and where changes occurred. Owing to the nature of the project, only one data set per year was used. The study is concerned with mapping the difference between two data sets, not mapping a single area for a short period of time (see Figures 8 and 9). MODIS imagery was not considered, owing to cost and the fact that data sets from circa 1990 were needed.

(36)

24

Figure 8: Kisarawe 1991 (ESRI, 2013) Figure 9: Kisarawe 2009 (ESRI, 2013)

Table 1: A result summary of a change-over-time analysis

4.2 Population estimation

Pattern recognition would have been a viable option for the task at hand at the Quinta AOI located in Mozambique, during 2009 and 2010. Unfortunately, only imagery was purchased and not the individual bands involved, owing to financial constraints. Although not utilised, the procedures as described by Orun (2004) and Sengupta et al. (2003), as well as Wei and Prinet (2005), were researched and scrutinised for future application value.

(37)

25

According to Orun (2004), an automatic operation was used on a single band in order to save cost and time, as multispectral data would have increased labour and computational cost.

Orun felt using more bands would not add to the textural analysis result. This is based on the assumption that textural content remains almost the same in different bands (Orun, 2004).

The final classification results obtained fall into two categories:

• classification of two textural classes (man-made and natural)

• classification of six textural classes (four man-made and two natural).

Seeing that an image was purchased by the company without the inclusion of individual bands, the procedure as described above was not an option. The size of the area made it possible to inspect the data set visually as it is by far smaller than the AOI as described by Orun (2004). The aim of the investigation described above is to extract human fabric. It does not offer a solution for extracting human fabric/structures that are made of in-situ materials, as is the case in Quinta where grass is used for roofing (Figure 10).

Figure 10: Dry grass – the main material used for roofing in the Quinta area

The aim of this exercise was to determine population densities and change over a two-year period. This was needed to estimate the financial implications of relocating the population

(38)

26

rightfully occupying the area to new areas, with incentives for their efforts. When it became clear that the local population had been informed about the future development and subsequent financial gain should one occupy the land, two data sets were purchased. An influx of immigrants was noticed after the fact and a compromise realised with the local government as compensation for the delay in the process of acquisition.

Sengupta et al. (2003) attempted feature extraction on a data set located near the Mexican border. In a similar fashion to Orun (2004), they used a single band and claimed that their method was successful on a large scale operation (see Figure 11).

Figure 11: Sengupta et al. (2003, p. 2) Structures discovered via pattern recognition

In both Sengupta et al. (2003) and Orun (2004), the scale of the operation surpassed the scale of the operation at Quinta. They were extracting formal buildings and settlements. The structures were, however, located in a rural setting, although still of a formal nature, and cannot really be compared to the structures found at Quinta. Sengupta et al. (2003) do mention the use of corners as an identifier of settlements and structures. This would not have been successful in the Quinta exercise, as not all structures are square – some are round (see Figures 12 and 13).

(39)

27

Figure 12: A typical village: inconsistent shapes and materials used in structures (ESRI, 2013)

Figure 13: The same village at a different scale (ESRI, 2013)

Wei and Prinet (2005) used a system of contour extraction, which lends itself to extraction structures of a permanent nature, to execute feature extraction. A highly detailed data set would be needed to extract contours resembling the outline of buildings (see Figure 14).

(40)

28

Figure 14: Wei and Prinet (2005, p. 3) Structures discovered via pattern recognition

In addition, the structure size as extracted by Wei and Prinet (2005) varied to a large degree from the structures encountered at Quinta. The procedure followed by these researchers would have had a huge impact on computational resources. They used IKONOS imagery which has a higher resolution than GeoEye and World-View (Wei & Prinet, 2005).

Owing to these differences, the procedure selected for analysing the data at Quinta, was a manual one:

• A grid of 1000 metres by 1000 metres was superimposed over the imagery

• A systematic search was conducted block by block, during which all structures and

‘Machambas’ or informal farming land were digitised and classified

• Each structure was allotted a number of inhabitants, depending on its size

• Each machamba was measured for size

• The results were presented to the Country Manager of Sun Biofuels, who attached a value to individual machambas, as well as to each hectare of machambas (See Figures 15 and 16).

(41)

29

Figure 15: Various machambas [informal farming areas] (ESRI, 2013)

Figure 16: Changes in machambas observed within one year (ESRI, 2013)

The following permutations were encountered:

• Satellite imagery was used for the calculations. This necessarily implies that the classified objects were viewed from the top. During the in-field investigation/ground truthing, it was found that some of the structures were actually mess-halls or cooking areas and tool-sheds

• During the initial scouting exercise (and the process of calculation/visual inspection of the imagery), it was noted by the GIS consultant that it would be nearly impossible to derive accurate figures for structures because:

o the size and shape of the structures were not consistent enough for pattern recognition processes

o Some structures did not have occupants – they were still visible, but abandoned as the owners had moved away

(42)

30

o The material used for the roof is grass cut from the nearby fields. Thus, there might easily be confusion during classification between the grass-roofed dwellings and the fields from which the grass originated.

• In terms of time, a visual inspection would be less time-consuming than fine-tuning a pattern-recognition exercise

• The funds for individual bands could be utilised more valuably elsewhere, seeing that the area in question had not been successfully leased at the time

• In order to be as accurate as possible, yet have the results available quickly, the grid was set at 1000 metres. A smaller grid would have led to a more accurate count, but would have been more time consuming (see Figures 17 and 18).

Figure 17: The visibility of structures: square and round in shape (ESRI, 2013)

Figure 18: The visibility of the same structures at a different scale (ESRI, 2013)

(43)

31

Although this thesis focuses on urban areas, it does describe a process of detecting human settlements. In the case of the Quinta operation, this really was not an option as time was short and answers urgently needed. Furthermore, multiple imagery sets, which were at the disposal of the author at the time, were used.

4.3 Land-use classification

According to Anderson et al. (1976), any country or large business must be acquainted with the intricate network of processes that interact in order to make the best decisions and derive the best strategies possible. Anderson further explains that these countries or businesses need information about the land-use classes present to maintain the status quo with respect to environmental and human-resource conditions. This way, responsible planning and action can take place (Anderson et al. 1976).

4.3.1 High resolution imagery

During the time the author worked for Sun Biofuels, land-use classification was performed on newly tasked satellite-imagery data sets for various AOIs. This classification served to provide a ‘snapshot in time’ of the prevailing scenario on the ground. High-resolution imagery was acquired for this purpose; therefore, the classification process used here was far more complicated than with the change-over-time exercise. More detailed training sites were used in an effort to increase the accuracy of the outcome.

The advantages of high resolution (one metre by one metre) may be outlined thus:

• More land-use classes could be determined, owing to the high resolution

• The freshly tasked imagery was cloudless

• Because of the high resolution, very accurate land-use classes were defined.

Disadvantages are as follows:

• The high resolution meant some training sites had wide spectral signatures, owing to the variance in reflectance angles. Thus, a single big tree, for example, consisted of 25 pixels, all with various values. The crown shape, as well as the variation of vitality from the crown to the lower branches, also had an affect.

(44)

32

• The complexity of the vegetation complicated classification and signature development.

Owing to the high number of pixels per tree, the spectral signatures became very complex owing to the wide array of band values that was recorded for each tree.

Figure 19: The complexity of a land-use class

• During the two-month waiting period for the new tasking, the local population started their annual burning practices. During this time, an overwhelming portion of the AOI was burned down, rendering the burned areas unclassifiable.

Figure 20 Figure 21

Figures 20 and 21: Burned areas that complicated the analysis process

(45)

33

4.3.2 Fire damage to AOI immediately prior to image acquisition

GeoEye and World-View imagery only consists of four bands: red, green, blue and near- infrared, which hampers discrimination during the classification process. LANDSAT has seven possible bands that can be used (excluding the thermal band). This implies that land classification largely revolves around the strength, or lack thereof, of the NDVI or ‘red edge’, meaning that only vegetation can be successfully classified.

Figure 22: The diverse vegetation present at the AOI

The vegetation present at the AOI was located in small pockets, surrounded by burned areas. Although the software from Clark Labs does have the functionality to filter out reflectance from bare soil in order to highlight the values recorded from vegetation, the amount of bare soil and/or burned areas was simply too great.

According to Corona et al. (2008), the single image has to be acquired very shortly after the occurrence of a fire or the spectral signature of the burned areas becomes difficult to interpret. This was indeed the case: the author observed a fire still visible in the north- eastern corner of the GeoEye data-set image of Kisararwe. Numerous fires had occurred

(46)

34

over a period of time prior to the acquisition and the spectral signature of the burned area was indeed a wide one with values varying greatly. Fortunately, all the burned areas had a signature unique enough to isolate each from the rest of the land-use classes present. Still, according to Perreira et al. (1999), the NIR band depicts the spectral difference of the fire the best after burning. This is due to the destruction of high scattering, leafy vegetation and was therefore used in this study to isolate the burned areas (Perreira et al. 1999).

Petropoulos et al. (2011) mentioned a potential limitation of single-date burned-area mapping as it is not possible to determine which land-cover types have been affected by the fire. Thus, assessing the post-fire damage becomes difficult if a land-cover map of the affected site is not already available. As the areas were burned by quick-moving fires that did not burn very intensely, it was accepted that the burned areas consisted of a single land- use class with grass or savannah as its main vegetation. Scholes (1997) defined savannahs as a tropical vegetation type in which the biomass is shared by trees and grass. He further commented that this type of vegetation has at least a two-layered structure above ground:

a tree layer of between two and ten metres in height and a grass layer between half a metre and two metres high. This was indeed the case, as can be seen in Figure 20.

Figure 23: The landscape of Kisarawe, Tanzania

According to Hudak et al., burned grasslands, for example, are often wrongly interpreted as highly damaged, despite the fact that grasslands never burn intensely.

Eva and Lambin (2000) suggested using surface temperature distribution as a method of analysis by utilising the thermal infrared part of electromagnetic spectrum. Unfortunately, a thermal infrared band was not available.

(47)

35

According to Kiang et al. (2007), the ‘red edge’ is so named because plant photosynthetic pigments absorb strongly in the visible or PAR. This strongly contrasts with high scattering in the NIR, owing to refraction between leaf mesophyll cell walls and air spaces inside the leaf.

Kiang et al. (2007) further stated that the wavelength of the red edge is more accurately described as the inflection point in the slope of the reflectance between the red and NIR, sometimes referred to as the “red edge inflection point” (p. 234) or the “red edge position”

(p. 234). In addition, Kiang et al. (2007) stated that this is generally located around 700 newton metres, but the location and steepness may vary according to the organism’s abundance/thickness (if sensing over a large area) and physiological status. As shown later in this paper, there can be distinct differences between organism types (Kiang et al., 2007).

4.3.3 Phases of the supervised classification process

According to Samaniego and Schultz (2009), a supervised classification algorithm can be subdivided into two phases:

(i) the learning or ‘calibration’ phase. The algorithm identifies a classification scheme based on spectral signatures of different bands obtained from ‘training’ sites with known class labels (for example, land cover or crop types)

(ii) the prediction phase. The classification strategy is applied to other locations with uncertain land-use class values.

Perumal and Bhaskaran (2010), however, define three steps in the supervised classification process: (i) defining training sites; (ii) developing signatures; (iii) classifing the image.

The main distinction between the algorithms mentioned above is the procedure to identify relationships (for example, rules, networks or likelihood measures) between the input and the output. Here, the input is spectral reflectance at different bands, also called ‘predictor space’

and the output is the land-use class label. This is so that either an appropriate discriminant function is maximised or a cost-function accounting for misclassified observations is minimised.

As stated above, it was indeed the case that two phases were involved. Firstly, more than ten land-use classes were identified. In-field personnel and management scrutinised the initial

(48)

36

findings and requested fewer land-use classes. This can be seen as a calibration effort, as well as a verification of the land-use classes. It has to be noted that only training sites and their resulting land-use class were employed at this stage, as no formal classification had been done.

4.3.4 Procedure followed

The procedure that was used consisted of the following steps:

• In-field personnel recorded training sites, using a Global Positioning System (GPS)

• These were superimposed over the imagery bands

• Training sites were utilised to develop spectral signatures

• The spectral signatures were checked for overlaps

• The imagery was classified, using a wide array of techniques

• The various results were presented

• Ground-truth operations were performed to select the most appropriate classification.

Figure 24 Figure 25

Figures 24 and 25: Riverine forest as classified from World-View 1 imagery (Mapmart: Global Mapping Solutions, 2013).

(49)

37

According to Smits et al. (1999), traditional classifiers such as K-nearest neighbour and MAXLIKE are more than adequate in LANDSAT-TM data sets, but may not be as effective with other data sets. MAXLIKE was performed in this research and was compared to MINDIST. Both classifications were tested in-field and MAXLIKE was the preferred data set. In addition, Smits et al. (1999) claimed that there might be subjectivity involved in the development or decision making with respect to the training sites.

Subjectivity with respect to the location, size or content of a training site is always a danger. This danger is not limited to MAXLIKE classifications only and it must be noted that training sites are usually defined before the decision on a classification method is made. Training sites can be subjective; therefore, it is the responsibility of the analyst to ensure that it is not the case. In the research discussed here, the training sites were communicated to the in-field personnel for evaluation. The in-field personnel and management of the company actually suggested fewer land-use classes. This request was honoured by the analyst.

The fact that it was decided to use fewer land-use classes had advantages and disadvantages.

The major advantage was that fewer land-use classes made the classification easier and saved computing time. The main disadvantage was that additional stress was placed on the accuracy and level of representation of each training site. It was felt at the time that the fewer the land- use classes, the more training sites for each land-use class would be needed to ascertain a high level of representation.

Figure 26 illustrates the classification process and Tables 2 and 3 are examples of data from reports entitled Baseline and land change mapping: Kisarawe 2090 (see Appendix 1) and Baseline and land change mapping: Nhamatanda 2010 (see Appendix 5).

(50)

38

Figure 26: The analysis process

Table 2: Summary of the data set results of a land-use classification

(51)

39

Table 3: Summary of the data set results of a land-use classification (analysis data)

4.4 Stratification and closed-canopy calculation

In order to fulfil criteria for certification (refer to Appendix 2), a stratification of the vegetation was needed as certification rules stipulate that forests of a certain height and size may not be converted to commercial use.

This was required for the Kisarawe Estate in Tanzania, the only estate where land-use conversion took place. The other estate, Tabaccos de Manica (TDM), consisted of old tobacco plantations. The outcome would be the determination of the percentage of trees higher than five metres in any given pocket, along with the percentage area they represent out of the total area. Unfortunately, it was not be possible to achieve a comprehensive result with the satellite-imagery data sets available at the time. Then, additional challenges were found when exact specifications were requested from various certification bodies, as the results were extremely vague and varied to a large degree.

It was decided to do stratification by means of examining the existing land-use classes for possible homogenous height characteristics. This was possible as the various land-use classes were at a unique height, owing to the fact that the topology of the land is relatively flat. Zhongming et al. (2010), encountering a totally different set of circumstances, reported that the area on which they were working could be characterised as hills and gullies, with elevations varying between 495 and 1794 metres above sea level.

(52)

40

This definitely made the procedure impossible to adopt in the AOI upon which the author focused, as, visually, height cannot be judged in this manner.

Figure 27: Variance (and complexity) of vegetation complicates stratification

Closed-canopy calculations were also a requirement for certification (see 5.4 and Annexure 4).

Although no written confirmation could be found of this, the rule was understood to apply to forest pockets bigger than 1 hectare, higher than five metres and with a closed-canopy percentage of 30 per cent or higher across the total area.

Figure 28: Plot size (ESRI, 2013) Figure 29: Visual results of classification (ESRI, 2013)

Referenzen

ÄHNLICHE DOKUMENTE

Juni 1979 bestimmt, dass die katholischen Kirchendiener die Sakramente der Busse, der Eucharistie und Krankensalbung den (Christ-)Gläubigen der nicht in voller Gemeinschaft mit

Daraus lässt sich schließen, dass eine signifikante Steigerung der Werte in Selbstmitgefühl mit einer signifikanten Steigerung der Werte in RS einhergeht, was zu

Wenn die schreibende Seite die lesende hingegen passiv hält, indem sie diese nicht selbst erfahren und empfinden lässt, sondern dies nach der Art eines allwissenden Erzählers oder

Hätte sie länger überlegt, hätte sie die korrekte Antwort später gegeben, oder gar nicht, dann wäre die Reaktionszeit unendlich lang (was auch nach t liegt). Dieser Umstand wird

Schorb (2008) hält einleitend fest, dass „im Gegensatz zur Bedeutung, die den Medien für die politische Sozialisation zugemessen wird, es nur wenige über die Jahrzehnte

Bei der Analyse gehe ich anhand von zentralen Analysekategorien vor, um die Werke besser vergleichen zu können (Ich werde diese Analysekategorien weiter unten noch genauer

In the crowddelivery concept, arrived delivery request is to be assigned or matched to a suitable crowd carrier from a pool of available crowdsources, with regard

Mitgebrachter Ärger ohne Bezug zum Kind oder Kindergarten und der Umgang der Pädagogin damit: Wenn Eltern das Entwicklungsgespräch zum Abladen von anderen Themen