Editors FethiAhmed OnisimoMutanga ElisabethZeil … 2015/AARSE2014...Polinsar coherence optimisation for deformation measurement ... Dry season biomass estimation as an indicator of

EditorsFethi Ahmed

Onisimo MutangaElisabeth Zeil-Fahlbusch

10th International Conference ofthe African Association of RemoteSensing of the Environment

27 - 31 October 2014University of Johannesburg, South Africa

First Edition Copyright: October 2014

Printed Congress Proceedings ISBN: 978-0-620-63066-5Electronic Congress Proceedings ISBN: 978-0-620-63067-2

Conference SecretariatAARSE54 Motor StreetWestdeneJohannesburgGauteng, 2092Rep. of South Africa

+27 83 628 4210

[email protected]

www.africanremotesensing.org

10th International Conference Of The African AssociationOf Remote Sensing Of The Environment

Space Technologies For Societal Benefits In Africa

University of JohannesburgSouth Africa27 -31 October 2014

Edited by

FETHI AHMED

University of the Witwatersrand,Johannesburg, South Africa

ONISIMO MUTANGA

University of KwaZulu Natal,Pietermaritzburg, South Africa

ELISABETH ZEIL-FAHLBUSCH

Member of AARSE Executive Council,AARSE Editor, South Africa

Space technologies for societal benefits inAfrica

AARSE 2014October 27 - October 31, 2014

AARSE2014 Scientific Review Committee

Name AffiliationProf Onisimo Mutanga University of Kwazulu-Natal,

South AfricaDr Mohamed M A Abd Elbasit University of Johannesburg,

South AfricaDr Elfatih Abdel-Rahman University of KwaZulu-Natal,

South AfricaProf Islam Abou El-Magd National Authority for Remote Sensing

and Space Sciences, EgyptDr Khaled Abutaleb University of Johannesburg,

South AfricaDr Elhadi Adam University of Witwatersrand,

South AfricaDr Samuel Adelabu University of Kwazulu-Natal,

South AfricaDr Clement Adjorlolo South African National Space Agency,

South AfricaProf Fethi Ahmed University of Witwatersrand,

South AfricaProf Harold Annegarn University of Johannesburg,

South AfricaDr Yazidhi Bamutaze Makerere University,

UgandaDr George Chirima Agricultural Research Council,

South AfricaDr Moses Azong Cho Council for Scientific and Industrial Research,

South AfricaProf Serena Coetzee University of Pretoria,

South AfricaMr Timothy Dube University of Kwazulu-Natal,

South Africa

Proceedings of the 10th International Conference of AARSE, October 2014 iii

Prof Hamisai Hamandawana University of Fort Hare,South Africa

Dr Riyad Ismail University of Kwazulu-Natal/ South African Pulp&Paper Industries, South Africa

Dr Nicky Knox South African National Space Agency,South Africa

Prof Jide Kufoniyi Obafemi Awolowo University, Ile-Ife,Nigeria

Prof Kamal Labbassi Chouaib Doukkali University,Morocco

Mr Romano Lottering University of KwaZulu-Natal,South Africa

Dr Melanie Lück-Vogel Council for Scientific and Industrial Research,South Africa

Dr Munyaradzi Manjoro North West University,South Africa

Dr Renaud Mathieu Council for Scientific and Industrial Research,South Africa

Dr Paida Mhangara South African National Space Agency,South Africa

Dr Walter Musakwa University of Johannesburg,South Africa

Mr Laven Naidoo Council for Scientific and Industrial Research,South Africa

Mr Andre Nonguierma United Nations Economic Commissions for Africa,South Africa

Dr John Odindi University of KwaZulu-Natal,South Africa

Ms Mercy Ojoyi University of KwaZulu-Natal,South Africa

Dr Zakariyyaa Oumar KwaZulu-Natal Department of Agriculture,Environmental Affairs and Rural Development,

South AfricaProf Lobina Palamuleni North West University,

South AfricaDr Abel Ramoelo Council for Scientific and Industrial Research,

South AfricaDr Gilbert Rochon University of New Orleans,

United States

Proceedings of the 10th International Conference of AARSE, October 2014 iv

Prof Khabba Said Cadi Ayyad University,Morocco

Prof Julian Smit University of Cape Town,South Africa

Dr Solomon Tesfamichael University of Johannesburg,South Africa

Ms Heidi van Deventer Council for Scientific & Industrial Research,South Africa

Prof Adriaan Van Niekerk University of Stellenbosch,South Africa

Dr Michel Verstraete South African National Space Agency,South Africa

Proceedings of the 10th International Conference of AARSE, October 2014 v

Preface

Space technologies for societal benefits in Africa:an introductory review

The African Association of Remote Sensing of the Environment (AARSE) wasfounded in 1992 and was incorporated in 2008 as an international NGO un-der Section 21 of the South African Companies Act 61 of 1973. AARSE is a non-political scientific organization. Its primary aim is to increase the awareness ofAfrican governments and their institutions, the academia, industry, the privatesector and the society at large, about the empowering and enhancing benefitsof developing, applying and responsibly utilizing the products and services ofEarth Observation Systems and Geospatial Technology.

The biennial conference of AARSE usually brings together a large number ofscientists, practitioners, vendors of geospatial solutions and government offi-cials from various countries across the globe with the objective of promotingthe advancement of knowledge, research, development, education and trainingin space science and geospatial technology including photogrammetry, remotesensing and geospatial information sciences and their applications in sustain-able development.

The 10th biennial International Conference of AARSE - AARSE2014 was held inthe University of Johannesburg, Johannesburg, South Africa from 27th to 31stOctober 2014 with the South African Space Agency (SANSA) and the Universityof Johannesburg as co-hosts. The theme of the conference, ‘Space technologiesfor societal benefits in Africa’ has been carefully chosen to contribute to the en-hancement of the human and institutional capacity of African nations and citi-zens in the utilisation of space technologies for sustainable development activi-ties at national and regional levels.

The week-long conference featured technical plenary and parallel sessions con-sisting of oral and interactive paper presentations; pre-conference workshops;scientific and commercial exhibitions of services and latest equipment; socialevents and a tour; as well as the AARSE General Assembly. 37 different countrieswere represented at the conference, with papers specifically describing the useof space technologies for societal benefits in Africa.

Proceedings of the 10th International Conference of AARSE, October 2014 vii

The conference provided insights into a diverse range of themes from benefitsand application of Earth Observation, through policy and economy to technol-ogy.

The collection of papers assembled in this volume provides a selection of paperspresented at the conference. These contributions are loosely arranged aroundfive sub-themes. The first concerns the Environment, Environmental Hazardand Disaster Risk Reduction. The second sheds light on Biodiversity, Ecosystemsand Climate. The third focuses on Technology as applied in Image processing,EO systems, Infrastructure and Spatial Data modelling. The fourth provides acollection of case studies dealing with Agriculture, Forestry, Geology, MineralExploration and Water and Wetlands. Finally, a range of issues related to Educa-tion, Capacity Building, Policy and Economy.

We would like to take this opportunity to extend our thanks to the many peopleinvolved in the AARSE2014 Conference and the reviewers of the papers assem-bled in this publication.

The papers contained in this proceeding have been subjected to an academicreview process. Each paper was reviewed by two anonymous referees. As an ed-itorial collective we would like to thank colleagues who assisted in reviewing thecontributions in this volume.

The Editors

Proceedings of the 10th International Conference of AARSE, October 2014 viii

Table of Contents

Preface ivTHE EDITORS

Part 1: Environment, Environmental Hazard and Disaster Risk ReductionD 2

Assessing the impact of land-cover chnage on surface water sources in SWNigera: The role of communities’ local experts’ knowledgeA. AYENI 3

Monitoring multitemporal vegatation change using landsat TM: Wadi Dek-ouk natural reserve in Southern TunisaM. OUESSAR 13

Categorization of flood prone areas in the lower Niger river channel usingshuttle radar terrain mission (SRTM) and Nigeriasatx ImageryA. ONWUSULU 25

Assessment and analysis of wildfires with the aid of remote sensing and GISW. VORSTER 32

Polinsar coherence optimisation for deformation measurement in an agri-cultural regionJ. ENGELBRECHT 41

Estimating leaf area index (LAI) by inversion of prosail radiative transfermodel on spot 6 imageryM.A CHO 50

Proceedings of the 10th International Conference of AARSE, October 2014 ix

Monitoring soil erosion features using a time series of airborne remote sens-ing data: A case study Wild Coast, South AfricaR. SINGH 56

Software engineering for Fybos Fire ManagementR. VAN DEN DOOL 67

Campaign measurements in the North West Province using the CSIR MobileLidarL. SHIKWAMBANA 75

Maximising Automation in Land Cover Monitoring with Change DetectionK. WESSELS 84

Validation of satellite-based rainfall measurements over arid and semi-aridregions of SudanM.A. ELBASITL 93

Evaluating flood hazards for land use planning in cross river basin, South-erneast Nigeria using remote sensing and GIS techniquesR. OLABANJO 102

Monitoring bilge oil dumping in the ocean using SAR image processingtechniquesR. VAN DEN DOOL 118

Flood Risk and Vulnerability Analysis in Kuduna Metropolis, Kaduna, Nige-ria 122D.N. JEB

Part 2: Biodiversity, Ecosystems and Climate 135D

Mapping the impact of anthorpogenic activities on the vegation of the Sout-pansberg Region of South AfricaP. KEPHE 136

Proceedings of the 10th International Conference of AARSE, October 2014 x

Land use and land cover changes in selected areas of Nigeria that are threat-ened by desertificationT. GARBA 147

Classification of sub-tropicl indigenous forest species using field spectroscopyand linear discriminant analysisV. SITHOLE 159

Vegatation density assessment in the Upper Molopo River Catchment, SouthAfricaL. PALAMULENI 168

Dry season biomass estimation as an indicator of rangeland quatity usingmulti-scale remote sensing dataA. RAMOELO 177

Multi-Temporal spatial analysis of the Mikea Dry forest landscape (South-west Madagascar)H.R. RAVONJIMALALA 186

Identifying the best season for mapping evergreen swamp and Mangrovespecies using leaf-level spectra in an estuarine system in Kwazulu-Natal,South AfricaH. VAN DEVENTER 196

Analysis of different empirical relations for the computation of the regres-sions coefficients for the climatic condition of Western Cape Province ofSouth AfricaE. MALUTA 205

Comparison of summer and winter carbon dioxide vertical and spatial dis-tribution over South AfricaX. NCIPHA 215

Investifation of urban heat island using landsat dataK. ABUTALEB 223

Proceedings of the 10th International Conference of AARSE, October 2014 xi

Validation of radar data using a ground based parsivel disdrometerJ. CAN LOGGERENBERG 233

Part 3: Technology: Image Processing, EO Systems, Infrastructure and Spa-tial Data Modeling 241D

Variations in urban growth in the different political dispensations in OsunState, NigeriaO. TAIWO 242

The soil moisture active passive missionE. NJOKU 252

Synthetic aperture radar for maritime domain awareness: Ship detectionin a South African contextC. SCHWEGMANN 257

Adding temporal data enhancements to the advanced spatial data infras-tructure platformB. SIBOLLA 266

Comparison of pixel-based and object-oriented classification approachesusing landsat-8 OLI and TIRS spectral bandsR. ISHIMWE 276

different entropies and Polsar landcover classification schemes using themA. MISHRA 285

Pansharpening methods based on contourlet transform applied to urbanareasS. OURABIA 291

Hyperspectral data redution based on wavelet transformS. CHOUAF 300

Potential utility of the worldview-2 multipsectral data and support vectormachines algorithm to classifying land use/cover in Dukuduku landscape,

Proceedings of the 10th International Conference of AARSE, October 2014 xii

Kwazulu-Natal, South AfricaG. OMER 309

Multi-Frequency SAR For Land Cover Classification Of Semi-Arid Aand ForestedRegions Of Africa, Using Random ForestsB. SPIES 320

Part 4: Agriculture, Forestry, Geology, Mineral Exploration and Water andWetlands 330D

Integrating image texture derived from high resolution worldview-2 im-agery and neural networks to predict Thaumastocoris Peregrinus (bronzebug) damage in plantation forestsZ. OUMAR 331

Crop and Rangeland Monitoring for End-Users: Operational Analysis Pro-tocols using Remote Sensing DataA. ROYER 342

Identifying crops using Landsat 8 thermal infrared bandsI. ROSELYNE 350

Random Purposive sampling for quality assessment of remote sensing landcover classification in South AfricaL. MAROPENG 359

Determining the availability of, and access to, fresh fruit and vegtables inArcadia and Eastwood, PretoriaM. PHAPHANA 368

Discrimination of maize cultivars using hyperspectral remote sensingA. NGIE 378

System implementation and capacity building for satellite based agricul-tural monitoring and crop statistics in Kenya (SBAM)G. LANEVE 388

Proceedings of the 10th International Conference of AARSE, October 2014 xiii

Major food crops yeilds response to climate change and variability in RwandaM. INNOCENT 396

Comparison of weights of evidence and rule-based classification for min-eral prospectivity mappingC. MUSEKIWA 409

Geologic mapping of parts of the Benue Trough, Nigeria using remotelysensed satellite data and GIS techniquesO. OMO IRABOR 417

The landscape of post mining communities in Ijesa Land, Osun State, Nige-riaN. ADEOYE 425

Mapping lithology using the group endmemberK. CAWSE-NICHOLSON 436

Water clarity mapping of Lekki Lagoon using remote sensing and least squaresregression modelO.D. NIHINLOLA 442

Validating Modis Imagery for monitoring water quality on Lake VictoriaA. GIDUDU 453

Estimation of potential crop evapotranspiration using remote sensing tech-niquesM. EL-SHIRBENY 460

Part 5: Education, Capacity Building, Policy and Economy 469D

Remote sensing education and research situation in Africa-Nigeria: An overviewtowards enchancing capacity buildingR. ASIYANBOLA 470

Challenges in capacity building and education in geospatial technology inAfrica

Proceedings of the 10th International Conference of AARSE, October 2014 xiv

M. KEITA 493

Application of geographic information systems (GIS) and remote sensing(RS) in monitoring and evaluation: A case study of national planning com-mission (NPC) Abuja, FCT, NigeriaA. ADEYEMI 501

List of PresentersD 509

Proceedings of the 10th International Conference of AARSE, October 2014 xv

Part 1Environment, Environmental Hazard and Disaster Risk

Reduction

2

Proceedings of the 10th International Conference of AARSE, October 2014 3

ASSESSING THE IMPACT OF LAND COVER CHANGE ON SURFACE WATER SOURCES IN SW NIGERIA:

THE ROLE OF COMMUNITIES’ LOCAL EXPERTS

Amidu Ayeni1, Moses Cho2, Alabi Soneye1, Renaud Mathieu2, 4, Jimmy Adegoke3

1Department of Geography, University of Lagos, Lagos – Nigeria

2Earth Observation Group, Natural Resources & Environment, CSIR Pretoria, South-Africa

3Department of Geo-Sciences, University of Missouri, Kansas City, United States 4Department of Geography, Geoinformatics, and Meteorology, University of Pretoria, South Africa

[email protected], [email protected]

KEY WORDS: Local experts, surface water, woodland and forest vegetation, SW-Nigeria

ABSTRACT

Land cover change (LCC) detection is essential for land use planning and crafting adaptation

measures to global change including global warming, water stress and security. In this study, we

investigated the impact of LCC on water stress in the woodland savannah and rain forest zones of

South-western, Nigeria as observed by the rural communities’ local experts. LCC was conducted using

orthorectified Landsat multi-temporal imagery for 1972, 1987, 2002 and 2007 using maximum

likelihood classification and change detection techniques. The results showed a decrease in the

forest area and an increase in built-up and cultivation or other areas such as open space, bare land

and grassland. Between 1972 and 2006, forest was reduced by about 50% while built-up areas

increased by about 300%. A social survey (Participatory Learning Approach PLA) involving local

experts between the ages of 50 to 70 was conducted to assess their observations in the region on (i)

LCC and (ii) the causes of water stress, and (iii) the associated risk and adaptation/recommendation.

The communities’ local experts generally reported that changes in climatic condition (e.g. decreasing

rainfall), continuous deforestation in the last 30 years and diversion of rivers and streams into

surface storages (earth dams and reservoirs) are the major factors responsible for water stress and

scarcity in the region. There is thus, a good correlation between the results of remotely sensed data

of LCC assessment and the communities’ local experts’ observations of land cover changes and

changes in surface water resources in the region. The study therefore inferred that LCC map products

could be used in a participatory approach involving the communities to assess the impact of

environmental change on an important service of forest ecosystems such as fresh water resources.

INTRODUCTION

The changes in land cover and land use in tropical forest environments are negatively impacting

on their ability to provide essential services to human communities, e. g. by threatening surface

water. This could be exacerbated by climate change, i.e. increasing frequency and severity of

extreme climatic events and long-term shifts in temperature and rainfall patterns as well as by rapid

mailto:[email protected]



population growth and fast-paced urbanization (Ren et al., 2012). This will increase the risk of leaving

many of the vulnerable rural communities in a fragile economic situation due to severe water stress.

An important question is whether inhabitants of tropical forest biomes in Africa are aware of the

changes occurring to the environment and the consequences thereof. This makes it particularly

challenging for people at the grassroots to craft common adaptation and mitigation measures to land

cover change (LCC, and land use associated) and climate change. The question arises whether local

perceptions of LCC can be correlated with quantitative and objective assessments of LCC using

remote sensing data and whether inhabitants of forest environments see these and climate change

as the causes to resource depletion, more specifically fresh water. Several studies have argued that

people’s perceptions and attitudes towards depletion of natural resources are instrumental to wise

use and management of natural resources (Mogome-Ntsatsi & Adeola, 1995; Pavlikakis and

Tsihrintzis, 2003; Kerr and Cihlar, 2005).

The aims of this study are to: (i) assess local experts’ perception of land cover change (LCC),

surface water and climate change in a deprived savannah area in Nigeria, (ii) assess local experts’

perception of the causes of water stress in the region, and (iii) investigate whether LCC information

derived from remote sensing data correlate with local experts’ perceptions of LCC and climate

changes.

STUDY AREA

The study area lies between longitude 3°2' and 6°0'E, and latitude 6°5' and 8°20'N, in the

Southwest corridor of the Niger River (Figure 1).

Fig 1: Study area


The northern region (see Fig. 2) is located within the woodland area of the southernmost part of

the woodland and tall grass savannah vegetation. Its annual rainfall ranges between 1000 – 1250 mm

while annual temperature is between 26° to 28°C. The average temperature is about 320C with

humidity as high as 95%. On the other hand, the southern region (see Fig. 2) is located at the fringe

of wooded savannah and rainforest zones, and extends to the extreme south of the rain forest

vegetation zone. Rainfall is between 1250mm and 1500mm at the fringe, and ranging between 1500-

1800mm in the extreme south of the region while temperature ranges from about 260C to 290C

(Barbour et al., 1982; Ayeni, 2012). The area is characterized by undifferentiated igneous and

metamorphic rocks, mostly granite, schist, gneisses, and metasediment (Barbour et al, 1982). The soil

is classified as ferric acrisols with relatively higher cation profiles (Nwachokor and Uzu, 2008).

MATERIALS AND METHODS

The analysis was based on Landsat scenes, freely available at the Global Land Survey (GLS)

(http://glovis.usgs.gov/). In this study, selected scenes were taken at four time periods over the last

40 years: 1972, 1987, 2002, and 2007. Four Landsat scenes were required to cover the study area.

The scenes were acquired by three Landsat sensors: the Multispectral Scanner (MSS) in the 1970s,

the Thematic Mapper (TM) in the 1990s, and the Enhanced Thematic Mapper Plus (ETM+) in 2000

and 2006. We targeted images captured between November and February when the region is at the

peak of the dry season. During this season the images are cloud free and the spectral contrast

between investigated classes is high, reducing the possibilities of classification errors. In addition, the

images provide information at the most water stressed period, for example when the extent of the

water bodies is smallest.

Image Classification

With the exception of the 1986/87 period, all images used for a given period where acquired

within a four month interval during the dry season. Based on the objective of the study, we identified

four major land cover types: built-up, forest, cultivation and other areas such as open space, bare

land, rocky outcrop, road networks, and water bodies. Each mosaic was trained and classified using

training data surveyed in the field. Training data were collected from four site in November 2011 and

used for each land cover class. The training data were generated from the centre of selected classes.

This study used the post-classification change detection technique for quantifying class changes

and comparing land cover between 1972 and 2007. It separately classifies multi-temporal images

into thematic maps and subsequently implements comparison of the classified images pixel by pixel

(Lu et al., 2004).

Based on information from local experts on changes observed over the years, the same training

data were used for the earlier classes. The confusion matrices’ result for each year shows that there

is correlation between the observed changes and land use pattern of the area based on result of

adjusted estimates for all classes at 95% accuracy level at 95% confidence level.

Acquisition of local experts’ knowledge of land cover and climate changes, and impacts on communities’ water access

http://glovis.usgs.gov/


Local experts referred to in this study are community members who use environmental resources

for various purposes and include farmers, herbalists/traditional healers, traditional priests

(custodians of surface water and forest shrines), loggers/retired forest guides. Interviews were

conducted to obtain information on water resources resulting from environmental changes (climate

and land cover) between 1972 and 2007. The information collected through interviews includes

perception of surface water and land cover change trends, change rate, and when the change was

first observed. The local experts selected are experienced elderly people above 60 years of age who

have the ability to describe and/or narrate the observed changes/situation of at least a 50km radius

of their environment in the last 45 years (i.e. 1971 - 2010). They have in-depth knowledge of their

immediate environment since their livelihoods are closely linked to natural resources; they spent

much of their time in the community and within their local setting, farming and breeding livestock,

collecting medicinal plants, protecting sacred forest and water and land resources. Local experts

should be identified through discussions with the traditional authorities or local institutions since

they know who the specialist users are within their community and/or environment. In this study,

local experts were identified by the Community Development Association Committees (CDAC) which

gathers senior members and decision makers in each community visited.

Interviews were conducted in sixteen of the 102 Local Government Areas (LGAs) present in the

study area. The 16 selected LGAs cut across the two vegetation zones (woodland savannah and

secondary rain forest) of SW Nigeria (Figures 1). Semi structured interviews were conducted with

three to seven local experts in each LGA. In total 36 experts were consulted in the North and 29 in

the South. The interviews were designed with a mix of open- and close-ended questions in order to

get the relevant information from the respondents, without the use of leading or prompting

questions.

RESULTS AND DISCUSSION

The land cover analysis revealed that in 1972, forest was the most extended class and covered

almost 42,800 km2 or 76% of the total area (Fig. 2a thru 2d and 3). This was followed by cultivation

and other areas which occupied an area of 13,100 km2 (23%), while built-up areas and water-bodies

together represented only 110 km2 (0.20%). In 1987, forest was reduced by 50% of its size observed

in 1972 (from 76% to 39%), while built-up, cultivation and other had increased by nearly 300% (from

0.2% to 0.6% and 23% to 60% respectively. There was also an increase (from 0.05% to 0.1%) in water

bodies resulting from the construction of the Oyan medium-size surface earth dam (ca. 40km2) and

Asa small surface earth dam (10.8km2).


Fig. 2a: Classified images of 1972 Fig. 2b: Classified images of 1987

Fig. 2c: Classified images of 2002 Fig. 2d: Classified images of 2007

Fig. 3: Land-cover classification between 1972 and 2007

2002/02)


Table 1 shows the representation of changes detected for each pair of years. The table indicates

the balance between the loss and gain for a specific class - built-up, forest, cultivation/others, and

water bodies. Between 1972 and 1987 forests had lost almost 55% of its 1972 extent, mostly to

cultivation and other areas. During the same period about 21% of cultivation and other areas

reverted back to forest.

Table 1: Change matrix between 1972 and 2007 (in Km2)

Pair of years

Built-up Forest Cultivations &

others Water bodies

197

2 &

198

7 Class Total (1972) 83 (100%) 42,819 (100%) 13,144 (100%) 27 (100%)

Class loss (1987) 12 (15%) 23,535 (55%) 2,743 (21%) 20 (73%)

Unchanged (1987 71 (85%) 19,284 (45%) 10,400 (79%) 7 (27%)

Class gain (1987) 236 (74%) 2,572 (6%) 23,438 (64%) 65 (71%)

Class Total (1987) 307 (100%) 21,856 (100%) 33,838 (100%) 72 (100%)

Image Difference 224 (270%) -20,964 (49%) 20,695 (158%) 45 (168%)

198

7 &

200

2 Class Total (1987) 307 (100%) 21,856 (100%) 33,838 (100%) 72 (100%)

Class loss (2002) 87 (28%) 10,941(50%) 6,993 (21%) 22 (31%)

Unchanged (2002) 220 (72%) 10,915 (49.9%) 26,846 (79%) 50 (69%)

Class gain (2002) 583 (66%) 6,390 (22%) 10,956 (24%) 112 (61%)

Class Total (2002) 803 (100%) 17,305 (100%) 37,802 (100%) 162 (100%)

Image Difference 497 (162%) -4,551 (21%) 3,964 (12%) 91 (127%)

200

2 &

200

7 Class Total (2002) 803 (100%) 17,305 (100%) 37,802 (100%) 162 (100%)

Class loss (2007) 126 (16%) 5,052 (29%) 4,335 (12%) 28 (17%)

Unchanged 2007 678 (84%) 12,253 (71%) 33,466 (89%) 134 (83%)

Class gain (2007 456 (36%) 3,895 (18%) 5,151 (12%) 40 (20%)

Class Total (2007) 1,134 (100%) 16,148 (100%) 38,617 (100%) 174 (100%)

Image Difference 330 (41%) -1,156 (7%) 815 (2%) 12 (7%)

The general pattern shown in Table 1 demonstrates the continuous loss of forest to other classes,

mainly cultivation/others. The rates of change calculated per year reveals that forest reduction was

twice as intensive between 1972 and 1987 (3.5%/yr.) compared to the two other periods (1.4%/yr.

each). Within 35 years more than 60% of forest in the area was lost.

Local Experts’ Knowledge about Land-cover Change

Table 2 shows the perception of the respondents in the sampled rural communities to changing

surface water, land cover, and climate between 1972 and 2007. On the average, 86.1% of the rural

inhabitants agreed that there were changes in surface water bodies and resources within their

communities over the last 40 years. In the northern LGAs (woodlands), 72.2%, 75.1%, and 77.8% of

experts observed changes in surface water availability, land cover, and climate (rainfall &

temperature) respectively, and 93.1%, 86.2%, and 79.3% of experts in the southern LGAs (secondary

rainforests).


Table 2: Summary of Local experts’ perception on changes in surface water, land-cover and climate

Change in Surface water Change in Land cover Change in climate

North South North South North South

Freq (%) Freq (%) Freq (%) Freq (%) Freq (%) Freq (%)

Ch

ange

Yes 26 (72.2%) 27 (93.1%) 27 (75.1)

25 (86.2%)

28 (77.8%)

23 (79.3%)

No 10 (27.8%) 2 (6.9%) 9 (25.0%0 4 (13.8%) 8 (22.2%) 6 20.7%) Total 36 (100%) 29 (100%) 36 (100% 29 (100%) 36 (100% 29 (100%)

Rat

e

Decrease 20 (55.6%) 23 (79.3%)

12 (33.3%)

7 (24.1%) 19

(52.8%) 17

(58.6%) Increase

2 (5.6%) 3 (10.3%) 10

(27.8%) 6 (20.7%) 6 (16.7%) 2 (6.9%)

Fluctuating 4 (11.1%) 1 (3.4%) 8 (22.2%)

12 (41.4%)

3 (8.3%) 4 (13.8%)

No change option

10 (27.8%) 2 (6.9%) 6 (16.7%) 4 (13.8%) 8 (22.2%) 6 (20.7%)

Total 36 (100%) 29 (100%) 36 (100%) 29 (100%) 36 (100%) 29 (100%)

Year

of

ob

serv

atio

n

> 30 years ago 3 (8.3%) 4 (13.8%) 7 (19.4%) 4 (13.8%) 7 (19.4%) 3 (10.3%) 30 - 15 years ago

12 (33.3%) 15 (51.7%) 17

(47.2%) 15

(51.7%) 13

(36.1%) 15

(51.7%) < 15 years 11 (30.6%) 8 (27.6%) 6 (16.7%) 6 (20.7%) 8 (22.2%) 5 (17.2%) No change option

10 (27.8%) 2 (6.9%) 6 (16.7%) 4 (13.8%) 8 (22.2%) 6 (20.7%)

Total 36 (100%) 29 (100%) 36 (100%) 29 (100%) 36 (100%) 29 (100%)

Considering all three categories (surface water, land cover, climate), in average 47.2% and 16.7%

of the northern experts claimed to have observed decrease and increase changes respectively, while

experts who believe there were fluctuating changes, accounted for 13.9% (Table 2). The results in the

south revealed a higher variation as 54.0% and 19.5% of the experts observed decrease and

fluctuating changes (Table 2). In the north and south, the majority of experts (average 38.9% and

51.7% respectively) started observing changes between 30 and 15 years ago (Table 2). On average,

less than 20% had no opinion in their observations about the changes (Table 2).

The local experts in the northern part of the study area attributed the causes of changes observed

to seven factors: a depreciation of cultural norms and values in most communities, intensive farming

activities and bush burning, charcoal production activities amongst the youths, low rainfall over the

years, rapid urbanization and population increase, poor conservation of water and forest resources,

and illegal logging, bush burning and deforestation activities.

In the south, the drivers of change were attributed to various factors including depreciation of

cultural values, farming, local charcoal production, low rainfall, urbanization and population increase,

and logging, bush burning and deforestation activities.

The majority of respondents argued that changes in climatic condition (reduced rainfall),

continuous forest degeneration in the last 30 years, and diversion of rivers and streams into surface

storages (earth dams and reservoirs) are the major factors responsible for water stress and scarcity


in most rural communities of South-western Nigeria. In addition, these have further negatively

impacted peoples’ physical environment. Coupled with an increase in temperature, the land use

changes are continuously affecting surface water catchment, socio-economic activity (farming),

access to natural resources, and availability and accessibility of water sources (Karl et al., 2009; Lai et

al., 2011). This is also in line with the view of Fairhead and Leach (1998), who found that land use and

settlements had a major effect on the distribution and expansion of forests, but contrary to the

conventional perceptions of forest trends in communal areas (Nigel and Fabricius, 2007).

The study also revealed that the local experts in virtually all LGAs/communities acknowledged

change within their immediate environment. The local experts expressed their contribution to the

understanding of the ultimate drivers of the changes. These are mentioned as factors/causes that are

intensifying changes in their environment, particularly water stress and scarcity in the rural area

which currently has important economic, social, environmental and management implications on

ecological biodiversity, sustainable forest resources, and livelihoods of communities around and

within the area (Nigel and Fabricius, 2007).

ASSOCIATED RISK AND ADAPTATION MEASURES / RECOMMENDATIONS

The threats of water stress on populations of the derived savannah of Nigeria are set to increase

in magnitude and scope due to the direct impact of uncontrolled deforestation that has resulted in

the drying up of freshwater resources, an increased frequency of drought and dust storms, and a

greater occurrence of flooding due to inadequate drainage or poor irrigation practices. For instance,

uncontrolled land cover change coupled with climate variability/change and the rapid economic

growth in recent years has resulted in a situation where the increased water demand is fast

exceeding supply, and this has led to severe groundwater over-exploitation in many regions of the

world (World Bank 2014). This portends to high risks in the region and will continue to have negative

impacts on the availability, quantity and quality of water resources.

This study therefore recommends several adaptation / mitigation measures that would promote

sustainable environment in the derived savannah region of SW Nigeria, namely

Changing to adaptable cropping patterns as well as promoting less water-demanding and more

drought resistant crops are encouraged as one option for adaptation policies.

Land use changes should be considered where current agricultural practices are no longer

sustainable in terms of water consumption (Merrey et al., 2005; Combest-Friedman et al., 2012).

Developing risk awareness in rural populations should be promoted by governments and

environmentalists / stakeholders by means of rural public events or face-to-face communication

using framing images, posters, public presentations, etc.

Environmental events involving authorities, experts, stakeholders and also affected citizens

should be organized for school children and those that are not privileged to go to school.

Water harvesting techniques should be initiated and encouraged among the water users through

the construction of small dams and pans, storage tanks etc.


Lastly, the use of state-of-art remote sensing technologies should be encouraged for forest

monitoring and management, and assessment of surface water extent and availability in the area

(Ayeni 2013).

CONCLUSION

The rural communities are at extremely high risk of climate – land cover change triggered impacts.

The impacts would affect socio-economic activities, access to natural resources and future growth,

and reduce the availability and accessibility of water sources and as well as food security. As

response capacity to threatening climate change is still a serious problem in the area, the impacts will

be worsening in the near future and may put the communities at extremely high risk of water stress.

In conclusion, local experts’ knowledge has often been ignored in environmental management

due to scientific ability to use technologically acquired data/information (e.g. rainfall, temperature,

remotely sensed data etc.) to monitor variables (climate change, land-use/landcover changes etc.)

that are difficult to detect through mere observation. This study shows a high consistency between

indigenous people’s perception of LCC, remotely sensed LC products, climate and surface water

situations. Therefore, local experts’ knowledge could be used in a participatory approach for

assessing the impact of environmental change on important services such as forest ecosystems and

fresh water provision.

REFERENCES

Ayeni A. O., 2013. Forestry in Nigeria: A brief historical overview, phases of development and present challenges. http://www.africanremotesensing.org/Default.aspx?pageId=1524987/ Published online on 08/11/13

Ayeni, A. O., 2012. Spatial Access to Domestic Water Sources in South-western – Nigeria: Assessment of domestic water sources, distribution and quality in rural – urban communities. Germany: Lambert Academic Publishing, pp 177

Barbour, K.M., Oguntoyinbo, J.S., Onyemelukwe, J.O.C., Nwafor, J.C., 1982. Nigeria in Map. London: Hodder and Stoughton, pp 209.

Combest-Friedman, C., Christie, P., Miles, E., 2012. Household perceptions of coastal hazards and climate change in the Central Philippines. Journal of Environmental Management, 112 (15): pp.137-148

Fairhead, J. and Leach M., 1998. Reframing deforestation. Global analysis and local realities: studies in West Africa. London: Routledge

Kerr, J.T., Cihlar, J., 2005. Land use mapping. In Encyclopaedia of Social Measurement, 2, pp. 441-449

Li, A., Jiang, J., Bian, J., Deng, W., 2012. Combining the matter element model with the associated function of probability transformation for multi-source remote sensing data classification in mountainous regions. ISPRS Journal of Photogrammetry and Remote Sensing 67: 80–92

Merrey D.J., Drechsel, P., Penning, F.W.T., de Vries, C., Sally, H., 2005. Integrating ‘‘livelihoods’’ into integrated water resources management: taking the integration paradigm to its logical next step for developing countries, Regional Environmental Change, 5, pp. 197–204

http://www.africanremotesensing.org/Default.aspx?pageId=1524987/


Mogome-Ntsatsi, K., Adeola, O.A., 1995. Promoting environmental awareness in Botswana: the role of community education. The Environmentalist, 15(4), pp. 281-292.

Nigel, C. and Fabricius C., 2007. Expert and Generalist Local Knowledge about Land-cover Change on South Africa’s Wild Coast: Can Local Ecological Knowledge Add Value to Science? In Ecology and Society 12(1), pp. 10

Nwachokor, M.A., Uzu, F.O., 2008. Updated Classification of Some Soil Series in South-western Nigeria. Journal of Agronomy, 7 (1), pp. 76-81

Pavlikakis, G.E., Tsihrintzis V.A., 2003. A quantitative method for accounting human opinion, preferences and perceptions in ecosystem management. Journal of Environmental Management, 68 (2), pp. 193-205

Ren, Y., Yan, J., Wei, X., Wang, Y., Yang, Y., Hua, L., Xiong, Y., Niu, X., Song, X., 2012. Effects of rapid urban sprawl on urban forest carbon stocks: Integrating remotely sensed, GIS and forest inventory data. Journal of Environmental Management, 113 (30), pp. 447-455


MONITORING MULTITEMPORAL VEGETATION CHANGE

USING LANDSAT TM: WADI DEKOUK NATURAL RESERVE IN SOUTHERN TUNISIA

Bouajila Essifi, Mohamed Ouessar

Eremology & Combating Desertification Laboratory, Institut des Régions Arides (IRA), Medenine, Tunisia. [email protected]

KEYWORDS: Land Cover/Land Use, Remote Sensing, Landsat TM, Wadi Dekouk, Arid Tunisia

ABSTRACT

Arid ecosystems are undergoing accelerated change due to natural and anthropogenic

disturbances. In pre-Saharan Tunisia, remote sensing datasets have been used as a tool for

monitoring desertification, land degradation and landscape management activities. The Wadi Dekouk

Natural Reserve was created in 1994 at 40 kilometres from the city of Tataouine in southern Tunisia,

and stretched over 5750 hectares. We applied a change detection technique to monitor and map

land cover changes in Wadi Dekouk natural reserve during the time period 1984-2010. This paper

assesses the protection effects on plant cover and soil surface based on satellite images. Landsat

data availability for the studied reserve, in conjunction with diverse ecological and socioeconomic

data sources has helped the evaluation of ecological indicators from a spatial perspective and over a

time scale of decades. The results, illustrated in thematic maps, have shown the changes along with

the trends associated with land cover and land use.

RÉSUMÉ

Les écosystèmes arides subissent des changements accélérés dus aux perturbations naturelles et

anthropiques. Dans la Tunisie présaharienne, les données de télédétection ont été utilisées comme

un outil efficace pour la surveillance de la désertification, la dégradation des terres et la gestion des

ressources naturelles. La réserve naturelle d’Oued Dekouk a été créée en 1994 à 40 kilomètres de la

ville de Tataouine dans le sud Tunisien, et s’étend sur 5750 hectares. Nous avons appliqué une

technique de détection de changements afin de suivre et cartographier l’évolution de l’occupation de

sol et l’utilisation des terres dans la réserve naturelle d’Oued Dekouk durant la période 1984-2010.

Ce papier évalue les effets de la protection sur le couvert végétal et la surface du sol basés sur les

images satellitaires. La disponibilité des données Landsat pour la réserve étudiée, en conjonction

avec les diverses données écologiques et socioéconomiques a contribué à l’évaluation des

indicateurs écologiques d’une perspective spatiale et à l’échelle de décennies. Les résultats, illustrés

dans des cartes thématiques, ont montré des changements avec des tendances associées à

l’occupation de sol et l’utilisation des terres.

MOTS-CLES: Occupation de sol/Utilisation des terres, Télédétection, Landsat TM, Oued Dekouk,

Tunisie Aride

INTRODUCTION



Drylands cover about 41% of Earth’s land surface and are home to more than 38% of the total

global population of 6.5 billion. Some form of severe land degradation is present on 10 to 20% of

these lands, the consequences of which are estimated to directly affect some 250 million people in

the developing world, an estimate likely to increase substantially in the face of climate change and

population growth (Reynolds et al., 2007). Tunisia lies along Africa’s Mediterranean coast, and the

country’s coastal Atlas Mountains gradually give way to the northern margin of the Sahara Desert.

The dry prone areas cover almost more than 4/5 of the total area where desertification related

problems are of major importance. In this arid land, agriculture and livestock grazing can easily push

the landscape toward desertification, and Saharan sands encroach upon 8,000 hectares each year.

The processes of land degradation affect the conservation of soil and water resources, because they

are strongly linked to unfavourable changes in the hydrological behaviour affecting soil water

balance and soil moisture regime. They are related to soil and climate characteristics, but

inappropriate land use and management is the main factor responsible of those processes (Samaali,

2011). As a consequence, restoring native grasslands in Tunisia takes concerted effort. To this end,

Tunisia has established a number of national parks and natural reserves in different bioclimatic

zones. Among them is the Natural Reserve of Wadi Dekouk which is located in the southern part of

the country and has an upper-Saharan bioclimate (Jauffret, 2001). Such reserves can be

complementary to the efforts made for ex-situ conservation of threatened species (e.g. gene banks).

Although abiotic and biotic conditions have not changed much in the past decades, traditional

subsistence oriented, migratory pastoralism has virtually disappeared as a land use system

throughout the Old World Dry Belt. In Tunisia, the southern rangelands are largely depopulated

because many pastoralists have opted for livelihood opportunities in other sectors of the economy.

Since the emergence of earth observation satellites, the use of remote sensing for the study of

arid zones was facilitated by the greater abundance of records over these regions, rarely covered

with clouds. It has developed rapidly because this technique meets a real need in environments

where topographic and/or thematic mapping is generally low, especially in developing countries

(Escadafal and Pouget, 1985). To identify the areas of biomass, visible red and infrared bands were

used as an indicator to monitor and quantify biomass through vegetation indices are a product of the

amount of chlorophyll and green growing material in plants. The subsequent diachronic comparison

of landcover maps should yield a representation of total growing plant biomass. Any reduction of

biomass over time may be attributed to a combination of factors such as plant phenology (growth

stages), plant moisture content, and removal by livestock grazing. However, effectiveness of

vegetation indices varies according to the aridity of the region (Essifi et al., 2008). The current study

seeks to analyse the spatiotemporal changes and trends of vegetation in Dekouk Natural Reserve

from 1984 to 2010. Our aim is to create a tool that could map landcover/landuse categories with a

focus on vegetation spatial distribution on rangeland within the park zone, and demonstrate the

potential of using satellite data and remote sensing applications in assessing significant variability in

landcover.

STUDY AREA

The Natural Reserve of Wadi Dekouk is situated at 621599.99 m E and 3610581.09 m N (southern

Tunisia, 10°32’ E; 32°08’ N), 37 km to the south of the town of Tataouine (Fig. 1). The area has an

upper-Saharan Mediterranean climate. According to nearest weather stations, rainfall varies


between 90 and 138 mm per year (Ennajeh et al., 2010). The region is also subject to at least 37 days

of south-west dry and hot winds (called Sirocco). The predominant soil type is a raw mineral sands.

The area falls within the mixed steppic ecosystem in a rangeland landscape. The reserve is

approximately 57.5 km2 divided by the Wadi Dekouk flowing in a NW-SE direction from the Dekouk

source. The reserve has been under protection from livestock grazing for almost 20 years. Various

eco-edaphic groups can be distinguished indicating the influence of soil type and the aeolian deposits

degree.

Figure 1. Wadi Dekouk, Google Earth, CNES 2013

DATA AND METHODS

Data acquisition

In order to investigate long-term variation in land cover type in the study area over several years,

we selected the two representative years of 1984 and 2010. The month selected was May, since the

vegetation reaches a maximum at that time of year and provides an opportunity to accurately

discriminate between land cover types. For better spatial resolution of the land cover, two Landsat 5

TM images taken in May 1984 and May 2000 were utilized, with a spatial resolution of 30 m. The

acquired images cover the whole area and its surrounding zone, as indicated in Figure 1. Despite it is

difficult to discriminate between bare soil and low vegetation cover by using coarse to moderate

resolution systems, the time-span of Landsat data have good temporal coverage of the hydrosphere-

biosphere interface to provide reliable information as a basis for environmental assessment,

management and decision support. The perennial and seasonally changing vegetation components

can be captured by satellite observation in the visible, near infra-red and thermal domains, given that

continuous satellite observation over a long period can be provided (Weissteiner et al., 2011).

Classification method

We chose the Iterative Self-Organizing Unsupervised classifier (ISOCLUST) based on a specific

implementation in IDRISI Selva® (Eastman, 2001). The six TM bands were specified, and then a


histogram was produced representing clusters that express the frequency with which they occur

across the bands. After examining the graph and looking for significant breaks in the curve which

represent major changes, the number of clusters to be created was specified. The cluster seeding

process leads to a far more efficient and accurate placement of clusters than either random or

systematic placement. The iterative process makes use of a full Maximum Likelihood procedure. This

provides a very strong cluster assignment procedure.

Post-classification: Change analysis and change maps

To undertake the change detection, the Land Change Modeler (LCM), an application within IDRISI,

was used. We focused on Change Analysis to analyze past landcover change. LCM has the following

requirements for the input land cover maps used for change analysis: the legends and the categories

in both maps are similar and sequential, backgrounds in both maps are the same and have a value of

zero and the spatial dimensions including resolution and coordinates are the same (Eastman, 2001).

Change is assessed between time 1 and time 2 between the two land cover maps. The identified

changes are transitions from one land cover state to another. It is likely that with many land cover

classes, the potential combination of transitions can be very large. The LCM change analysis tab

provides a set of tools for the rapid assessment of change, and generates evaluations of gains and

losses, net change, persistence and specific transitions both in map and graphical form. A particular

landcover category is selected to map gains and losses; this includes whether there is persistence

(i.e., no change) or not. However, two landcover categories are selected to map transitions or

exchanges.


Classification of steppic vegetation variables in Dekouk Natural Park

Based on previous research addressing phytoecological aspects of Wadi Dekouk Natural Reserve

(Ennajeh et al., 2010, Ould Sidi Mohamed, 2003), we have distinguished the following vegetation

groups: a first group is made of some psammo-halophyte species found on sandy soils moderately

fixed, and having a very salty water supply where species such as: Aeluropus littoralus, Limoniastrum

monopetalum, Linaria aegyptiaca and Lotus creticus are found. Based on the edaphic demand, a

second group encompasses species, which essentially colonise wadis and thalwegs like

Cymbopongon schoenanthus, Moricandia arvensis, Periploca leavigata, etc., or rocky soils to a certain

distance to watercourses (Anthyllis sericea, Atractylis serratuloides, Helianthemum kahiricum, etc.). A

third group is represented by gypseous or halophyte species which are developed on silt-sandy soil

on calcareous-gypseous encrusting such as: Echium pycnathum, Erodium glaucophylum, Erucaria

uncata, etc. These species generally occupy sites that have a relatively positive hydric balance:

terraces of hydrographic network, cliffs, deep sandy substratum in convenient topographic positions,

spreading zones, etc. A fourth group encompasses fleshy halophyte species linked to salty terrains

(Suaeda mollis and Traganum nudatum). Several change detection studies have shown that interdate

changes in vegetation properties are best identified when image data are enhanced using vegetation

indices prior to image differencing. In addition, changes in landcover are most often associated with

a combination of indices rather than any single index or change feature (Rogan et al., 2002).

Therefore, five landcover classes are deduced highlighting phyto-ecological aspects coupled with


edaphic characteristics of the study zone: Sparse Vegetation/Bare Soil (SCBS), Psammophetic

Vegetation (PV), Psammo-Halophyte Vegetation (PHV), Halophyte Vegetation (HV) and Sandy Bare

Soil (SBS) (Fig. 2 and 3).

Figure 2. Landcover map of Dekouk Natural Reserve, 1984

Figure 3. Landcover map of Dekouk Natural Reserve, 2010

By contrast, a large part of the landscape revealed a mixed distribution of vegetation categories in

the reserve. The SVBS and SBS classes are characterised by a very low vegetation amount. These

classes have almost identical mosaic shape of the vegetation in the entire region, which is a sparsely

vegetated area. HV and PHV classes representing salty soils, are a dominant feature along the Wadi.

But they have different spatial extents, with HV exceeding at different occurrences.

Change Analysis

The change analysis provides three graphs of land cover change between the two land cover maps

of 1984 and 2010. These graphs can be viewed in a variety of units. We chose to use hectares,

percentage of change and percentage of area. The first graph shows a rapid quantitative assessment

of change by graphing gains and losses by landcover category. Net change by category: shows the


result of taking the earlier landcover areas, adding the gains and then subtracting the losses.

Contributors to net change: examines the contributions to changes experienced by a single

landcover, selected by the user from the different landcover categories.

Gains and losses by category The first map derived from the change analysis in LCM based on gains and losses by category,

showed that two landcover classes experienced very little change, both in almost equal gain and loss:

the Sparse Vegetation/Bare Soil with ≈ 850 ha and ≈ 370 respectively (Fig. 4 and Tab. 1). However, we

noticed a clear gain in terms of Halophyte Vegetation quantity with high amplitude of positive

change, and slight gains of the Psammophetic-Halophyte Vegetation class. The area colonized by the

psammophetic vegetation was reduced between 1984 and 2010.

Figure 4. Dekouk Natural Reserve: Gains and losses by landcover categories between 1984

and 2010

Net change by category

Figure 5 shows that two landcover classes have known the highest rate of net change: a positive

change of the HV class with 300 ha and a negative change of the PV class reaching little less than 400

ha (Fig. 5 and Tab. 1).

Figure 5. Dekouk Natural Reserve: Net change by landcover categories between 1984 and

2010

The vegetation is likely influenced by protection and rainfall response. PHV and HV classes show

maximum positive change and pronounced seasonality of vegetation growth, although HV class has a

greater magnitude than PHV. PV vegetation category has the highest negative change, with a rapid

decline of area reaching 379 ha.

Table 1. Dekouk Natural Reserve: change analysis statistics of landcover categories between

1984 and 2010

Gains by category Losses by category Net change by category

Hectares % Change Hectares % Change Hectares

0 200 400 600 800-200-400-600-800

Sparse Vegetation/Bare Soil

Psammophetic Vegetation

Psammo-Halophyte Vegetation

Halophytes

Sandy Bare Soil

Gains and losses between 1984 and 2010Gains and losses between 1984 and 2010

0 100 200 300-100-200-300-400




Halophytes

Sandy Bare Soil

Net Change between 1984 and 2010Net Change between 1984 and 2010


Sparse Vegetation/Bare Soil 865 42,70 -835 -41,86 29 Psammophetic Vegetation 271 18,44 -650 -35,18 -379 Psammo-Halophyte Vegetation

171 22,83 -127 -18,04 44

Halophyte Vegetation 501 52,97 -193 -30,27 308 Sandy Bare Soil 372 100,0 -374 -100,0 -2

Contributors to net change experienced by landcover category

Table 2. Dekouk Natural Reserve: change analysis statistics of landcover categories

between 1984 and 2010

SVBS PV PHV HV SBS

Sparse Vegetation/Bare Soil (SVBS)

0 -385 68 180 108

Psammophetic Vegetation (PV) 385 0 0 76 -81 Psammo-Halophyte Vegetation (PHV)

-68 0 0 -27 52

Halophyte Vegetation (HV) -180 -76 27 0 -80 Sandy Bare Soil (SBS) -108 81 -52 80 0

Figure 6. Dekouk Natural Park: Contributions to Net change by landcover category between

1984 and 2010

Climatic (long dry season, important hydric deficit), edaphic (inherited and fragile soils) and

floristic (poor steppic vegetation) characteristics seem at first to have contributed to desertification

in this part of the Tunisian territory (Talbi, 1997). In addition, the re-settlement of people has led to

the disappearance of many large nomadic herds, replaced by small units of less than 50 head grazing

almost all year round in the immediate vicinity of houses or villages, accelerating overgrazing of

0 100 200 300 400-100-200




Halophytes

Sandy Bare Soil

Contributions to Net Change in Sparse Vegetation/Bare SoilContributions to Net Change in Sparse Vegetation/Bare Soil

0 50 100-50-100-150-200-250-300-350-400




Halophytes

Sandy Bare Soil

Contributions to Net Change in Psammophetic VegetationContributions to Net Change in Psammophetic Vegetation

0 10 20 30 40 50 60 70-10-20-30-40-50




Halophytes

Sandy Bare Soil

Contributions to Net Change in Psammo-Halophyte VegetationContributions to Net Change in Psammo-Halophyte Vegetation

0 20 40 60 80 100 120 140 160 180-20




Halophytes

Sandy Bare Soil

Contributions to Net Change in HalophytesContributions to Net Change in Halophytes

0 20 40 60 80 100-20-40-60-80




Halophytes

Sandy Bare Soil

Contributions to Net Change in Sandy Bare SoilContributions to Net Change in Sandy Bare Soil


rangelands. These changes profoundly alter the landscape and ecological systems of southern

Tunisia. Steppes that covered the floors of the glaze silty foothills of the mountains, are now

completely cleared, and water erosion is becoming serious. Steppic sandy areas are currently the

most attractive to dry farming (cereals), and each year new areas are cultivated restricting traditional

grazing areas. This is all the more serious because these soils are particularly sensitive to aeolian

erosion, which gradually reduces shallow sandy horizons and causes the formation of dunes. Wind

erosion, in addition to water erosion, leads to an overall decrease in the ability of these soils to store

rainwater. Rainwater now runs off the barren land, creates episodic wadis, fills depressions and even

causes localized flooding. Simultaneously, pressure by pets is mounting on steppes that have

superficial soils and are not suitable for cultivation, either because of the presence of limestone or

gypsum crusts. The plant cover is decreasing with, at first, good woody species (Chamaephytes,

Nanophanerophytes) cut off for firewood or cooking needs of people. There is an overall areal

decrease of pastoral steppes in good condition. Some areas have reached degradation levels at

which it is difficult to imagine a possibility of re-vegetation. The management of natural resources is

then unbalanced: the annual harvest exceeds the renewal capacity (Floret et al., 1981).

Change maps

To examine the edapho-climatic factors affecting vegetation variation, we analyzed the evolution

of vegetation landcover categories in growing season on a diachronic basis. A change analysis was

applied on the various landcover classes to yield the following maps of vegetation (Fig. 7-12). The

figures show the spatial distribution of vegetation categories in terms of gain, loss and persistence. A

visual inspection of these maps revealed a source of significant variation within the different classes,

helping to understand the intrinsic variability in the studied landscape. Fig. 7 shows that the biggest

change in terms of gains, losses and persistence was encountered within the SVBS landcover class

but the overall class area is nearly the same while the PV landcover class has known the greatest

negative change, as shown in red colour in Fig. 8. The main positive change was attributed to the HV

class (Fig. 10). Contrarily, we didn’t notice any persistence as far as the SBS class is concerned which

can be explained by the violent wind regime from February to October contributing to the

redistribution of sands in the studied zone (Fig. 11).

Figure 7. Dekouk Natural Reserve: Vegetation change map of Sparse Vegetation/Bare Soil

landcover category between 1984 and 2010


Figure 8. Dekouk Natural Reserve: Vegetation change map of Psammophetic Vegetation


Figure 9. Dekouk Natural Reserve: Vegetation change map of Psammo-Halophyte

Vegetation landcover category between 1984 and 2010


Figure 10. Dekouk Natural Reserve: Vegetation change map of Halophyte Vegetation


Figure 11. Dekouk Natural Reserve: Vegetation change map of Sandy Bare Soil landcover

category between 1984 and 2010

In this study, the results illustrate an overall increase in biomass quantity in Wadi Dekouk Reserve.

The analysis also has implications for planning with regard to protection strategies that support

vegetation recovery. Spontaneous shrubs are more vigorous and appear more adapted to aridity.

This advantage could be explained by more efficient defence mechanisms against environmental

constraints (Ennajeh et al., 2010). The results indicated that although the potential for a proliferation

of the halophyte vegetation in areas when sparse vegetation species on bare soil are present, exists,

they also show that with the sandy bare soil, very little change or even increases in psammophetic

vegetation could occur. According to Ould Sidi Mohamed (2003), the recorded species in Wadi

Dekouk protected area are characterized by the predominance of halophytes, some psammophetic

species and annual species colonising different environment types. It shows a high heterogeneity

engendered by the protection. The effect of edapho-climatic characteristics of the zone can be

explained by a progressive dynamic under protection effect and a regressive one under the effect of

anthropogenic factors. It can be attributed to a regeneration of vegetation cover, in particular the

appetized species on one hand, and to the disappearance of these species under the uncontrolled

grazing effect on the other hand. It can be attributed to the regeneration of annual species in Wadi

Dekouk (recently protected and presenting diversified environments) which benefit from the

protection. Moreover, the edaphic conditions of Wadi Dekouk (sand accumulation, salty soil)

constitute specific biotopes for psammophetic, psammo-halophyte and halophyte species.

CONCLUSION

The loss of plant biodiversity is the subject of increasing concern worldwide. In general, drylands

have not benefited so far from the necessary attention in terms of their contribution to national and

international strategies for the preservation, conservation and enhancement of biodiversity. The

period during which increasing aridity conditions have developed in these areas, combined with old


human practices, resulted in the process of adaptation of original gene pools and the evolution of a

matching mosaic of homes. The properties of eco-physiological and genetic adaptation to drought

encountered in many dryland species, and the diversity of ecosystems make these areas centers of

valuable resources for future use (ROSELT/OSS, 2004). In southeastern Tunisia, desertification is a

process driven by various factors that makes a territory originally fertile, sterile and unable to

produce. Anthropogenic influence remains the key factor, more important than physical and climatic

constraints in most cases. Knowledge of the causes and mechanisms of desertification, however,

should not only serve to explain the why and how of the phenomenon. Research (including

geographic) is called to translate its results into development efforts, and place man and survival

among its priorities (Talbi, 1997). Species well adapted to arid environments, are tightly linked to

precipitation events. Water availability appears to be the most determining factor controlling species

development (Ennajeh et al., 2010).

Landsat satellite imagery were used in this study to investigate temporal patterns and dynamic

characteristics of the vegetation change for 1984–2010. Satellite imagery is timely, has minimal costs

and has proven to give tangible benefits. Satellite data images provided an objective, permanent

data set for comparative analysis of vegetation change in the Natural Reserve of Wadi Dekouk

including ecosystem dynamics and monitoring desertification processes. In order to inspect the

effect of protection on natural vegetation, vegetated areas, grass, and soil features were included to

obtain full classification of the Wadi Dekouk landscape. The research consisted of two parts: the first

was the development of general land cover maps. The second part involved the calculation and

analysis of change statistics to depict patterns and trends, and to understand land cover changes

under protection. The produced change maps illustrate the importance of assessing landscape

change history, especially when developing biodiversity conservation and land use management

plans.

ACKNOWLEDGMENTS

The authors acknowledge financial support through an IEEE-GRSS/AARSE TRAVEL FELLOWSHIP.

REFERENCES

Eastman, J. R. 2001. IDRISI Release 2: Guide to GIS and Image Processing. Volume 2. Manual Version 32. . In: CLARK LABS, C. U. (ed.).

Ennajeh, M., Cochard, H. & KHEMIRA, H. 2010. Re-introduction success of an autochthonous plant species, Periploca angustifolia, in the Natural Reserve of Oued Dekouk, Tunisia. Spanish Journal of Agricultural Research, 8, 1005-1011.

Escadafal, R. & Pouget, M. Luminance spectrale et caractères de la surface des sols en région aride méditerranéenne (Sud tunisien). Symposium de l'AISS Groupe de Travail Pédologie et Télédétection, 1985/03/04-08 1985 Wageningen. AISS, 12 p. multigr.

Essifi, B., Ouessar, M. & Rabia, M. C. 2008. Surveillance de la désertification autour des points d’eau dans les parcours d’El Ouara (Tataouine - Tunisie) par des séries temporelles Landsat MSS/TM/ETM+. Revue Française de Photogrammétrie et de Télédétection, France, 190, 49-59.


Floret, C., Le Floc'h, E., Romane, F. & Pontanier, R. 1981. Dynamique de systèmes écologiques de la zone aride : application à l'aménagement sur des bases écologiques d'une zone de la Tunisie présaharienne. Acta Oecologica.Oecologia Applicata, 2, 195-214.

Jauffret, S. 2001. Validation et comparaison de divers indicateurs des changements à long terme dans les écosystemes mediterraneens arides : Application au suivi de la désertification dans le Sud Tunisien. Docteur de l’Université de Droit, d’Économie et des Sciences d’Aix-Marseille, Discipline : Ecologie Thèse de Doctorat, Universite de Droit, d’Economie et des Sciences d’Aix-Marseille (Aix-Marseille III).

Ould Sidi Mohamed, Y. 2003. Biodiversité et suivi de la dynamique des phytocénoces en Tunisie présaharienne: cas des observations de Sidi Toui et de Oued Dekouk. Thèse de Doctorat en Biologie Doctorat, University Tunis ElManar, Faculty of Sciences.

Reynolds, J. F., Stafford Smith, D. M., Lambin, E. F., Turner Ii, B. L., Mortimore, M., Batterbury, S. P. J., Downing, T. E., Dowlatabadi, H., Fernández, R. J., Herrick, J. E., Huber-Sannwald, E., Jiang, H., Leemans, R., Tim Lynam, Maestre, F. T., Ayarza, M. & Walker, B. 2007. Global Desertification: Building a Science for Dryland Development. Science 316, 847-851.

Rogan, J., Franklin, J. & Roberts, D. A. 2002. A comparison of methods for monitoring multitemporal vegetation change using Thematic Mapper imagery. Remote Sensing of Environment, 80, 143-156.

ROSELT/OSS 2004. Indicateurs écologiques ROSELT/OSS: Une première approche méthodologique pour la surveillance de la biodiversité et des changements environnementaux, collection ROSELT/OSS, document scientifique n° 04, . Montpellier, France: Observatoire du Sahara et du Sahel (OSS).

Samaali, H. 2011. Etude de l'évolution de l’occupation et de l’utilisation du sol dans le Delta de Mejerda par télédétection et Systèmes d’Informations Géographiques. Docteur en Géographie, Option: Cartographie, Université de Tunis, Faculté des Sciences Humaines et Sociales de Tunis.

Talbi, M. 1997. Action anthropique et dégradation de l'environnement aride : La désertification en Tunisie du Sud-Est. Méditerranée, 86, 25-31.

Weissteiner, C. J., Strobl, P. & Sommer, S. 2011. Assessment of status and trends of olive farming intensity in EU-Mediterranean countries using remote sensing time series and land cover data. Ecological Indicators, 11, 601-610.


CATEGORIZATION OF FLOOD PRONE AREAS IN THE LOWER NIGER RIVER CHANNEL USING SHUTTLE

RADAR TERRAIN MISSION (SRTM) AND NIGERIASATX IMAGERY

Onwusulu Andrew Uzondu, Efron Nduke Gjere, Dashan Titus, Dr. David Nyomo Jeb, Omomoh Emmanuel

National Centre for Remote Sensing (NCRS) Jos, National Space Research and Development

(NASRDA), Abuja, Nigeria Agency; Onwusulu Andrew Uzondu [email protected]

KEYWORDS: DEM, SRTM, Flood Inundation, Topography, Slope

ABSTRACT

This article examined how various flooded sections of the lower Niger River channel was influenced

by the topography during the 2012 flood event. The study area spreads from Lokoja in Kogi State to

Onitsha in Anambra State between Latitudes 50 52’ 10.825” N and 70 59’ 18.052”N and Longitudes 120

11’ 29.91” E and 130 22’ 17.101”E. The study was carried out by processing and interpretation of

remote sensing data and integration of field survey data. A Shuttle Radar Topographic Mission (SRTM)

data of the area of study was used to generate the Digital Elevation Model (DEM) and Painted Relief

using ERDAS IMAGINE software. Elevations ranging from 15M above sea level to 657M above sea level

were characterized in the output raster map of the DEM and the painted relief showing the topography

of the area. Additionally, a slope map was produced, with values ranging from 0 – 48; from lower to

higher slope values which represent flatter to steeper terrain, respectively. Flood level data for 2012

obtained from the field survey, were used to run flood inundation models using Integrated Land and

Water Information System (ILWIS) in which areas that are characterized as 30M, 35M, 45M above the

sea level, were highly flooded. Settlements affected by the 2012 flood include Idah down to

Onitsha/Asaba within flatter terrain, whereas settlements from Agenebode up to Ajaokuta were less

inundated. Supervised classification of NigeriaSat-X image using ERDAS Imagine 9.2 software was used

to generate land use and land cover maps of the area. Training samples sets were acquired from the

image to run the supervised classification of five classes, namely vegetation (782,021hectares), flood

plain agriculture/farmlands (799,601hectares), settlements (129526hectares), water body and bare

surface (9747hectares) using maximum likelihood method.

INTRODUCTION

Between July and October 2012, Nigeria witnessed an unprecedented flood that affected about 33

States with varying degrees of intensities and effects. Flood waters pushed rivers over their banks and

submerged hundreds of thousands of acres of farmland. By mid-October, floods had forced 1.3 million

people from their homes and claimed 431 lives, according to Nigeria’s National Emergency

Management Agency (IRIN, 2012). Flood phenomena are considered one of the worst hazards in terms

of magnitude, occurrence, geographical spread, loss of life and property, displacement of people and

social- economic activities (Myers and White, 1993).



Generally, floods occur more in the low-lying areas or the areas below sea level. One of the main

reasons is that rivers flow slowly in these areas. This is controlled by the slope of the surface through

which water travels. Thus it is important to note that the area of inundation for a flood depends not

just on factors like river stage but also on the slope of the ground around the water. According to Eric

M. Baer (2014), the impact of a flood that raises the water level of a stream by 10 feet is very

dependent on the slopes of the valley. In a steep canyon, there might be no significant impact, while

on a broad flood plain the water could cover a large area and do tremendous damage. So flood damage

does not only dependent on the discharge, but also the topography around a river. This article will

examine how various flooded sections of the lower Niger River channel were influenced by the

topography during the 2012 flood event. The study area spreads from Lokoja in Kogi State to Onitsha

in Anambra State between Latitudes 50 52’ 10.825” N and 70 59’ 18.052”N and Longitudes 120 11’

29.91” E and 130 22’ 17.101”E. A thorough understanding of the topography and slope of land areas

adjacent to the river bank will contribute much knowledge in the management of flooding through

construction of floodwalls, levees, channel excavation and modification which can substantially reduce

flood damages.

The objectives of this study are to generate a Digital Elevation Model (DEM) and Painted Relief, a

flood inundation model, and a classification of the slope of the area under investigation. Another

objective is to use NigeriaSat – X to map landuse and land cover of the inundated areas.

Remote sensing data were processed and interpreted, and field survey data were integrated.

Shuttle Radar Topographic Mission (SRTM) data of the area were used to generate the Digital Elevation

Model (DEM) and Painted Relief using ERDAS Imagine software, flood inundation model, and

classification of the slope of the area. Supervised classification of NigeriaSat-X image using ERDAS

Imagine software was used to generate land use and land cover map of the area. Training samples sets

were acquired from the image to run the supervised classification of six classes, namely vegetation,

flood plain agriculture, farmlands, settlements, water body and bare surface using maximum likelihood

method. The Remote Sensing approach helped to explore the general flood situation of the flooded

sections of the lower Niger River channel.

DATA AND MATERIALS USED

Data and Materials

Shuttle Radar Terrain Mission (SRTM), Nigeriasat-x and existing settlement maps were used for this

study. Their georeference projection is universal transverse Marcator (UTM), Datum – Minna, Datum

area, Nigeria, Ellipsoid: Clerk 1880 and Zone 32. The Shuttle Radar Topography Mission (SRTM) is an

international research effort obtaining digital elevation models on a near-global scale from 56° S to

60° N, to generate the most complete high-resolution digital topographic database of Earth

downloaded from EROS data Centre. NigeriaSat-X is a 22m imaging capability, multispectral Earth

observation satellite with red, green and near infrared spectral bands. Other materials used were a

Global Positioning System (GPS), digital cameras, and software.

METHODS

These includes field work and data analysis.

http://en.wikipedia.org/wiki/Digital_elevation_model


Field Work

The field data gathering phase took place in July, 2013. The following primary data were collected:

the 2012 flood level, landuse and landcover data within the flooded area and oral interview of the

affected communities. Data sheets were designed for data collection in the field. The following

parameters were obtained from the field: GPS coordinates, landuse types, nearest settlements, 2012

flood level, etc. at each location. Also, the research teams interacted with different communities’

leaders, Local Government officials and other relevant Government agencies to get some useful

primary as well as secondary data about the 2012 flood and other floods that occurred in previous

years.

Data Analysis

Analysis includes the generation of a digital elevation model, painted relief, slope map, flood

inundation model, and landuse/ landcover map of the study area.

Digital elevation model (DEM): SRTM elevation data were used to generate the Digital elevation

model (DEM) in ILWIS. DEMs are sampled arrays of elevation values representing ground positions at

regularly spaced intervals. Digital Elevation Model (DEM) is the terminology adopted by the USGS to

describe terrain elevation data sets in a digital raster form. The standard DEM consists of a regular

array of elevations cast on a designated coordinate projection system. The DEM data are stored as a

series of profiles in which the spacing of the elevations along and between each profile is in regular

whole number intervals.

Painted relief: Painted relief of the study area was obtained from SRTM using the topographic

analysis operations in ERDAS Imagine to show the topography of the area at a scale of exaggeration of

5m in z-direction.

Flood inundation model: The DEM was subsequently used to generate flood inundation models in

ILWIS domain through slicing operations. A domain of type ‘class-group’ was created with upper bound

of two classes, namely flooded and non-flooded areas using the 2012 flood level data obtained from

the field. The elevation data used are Lokoja at 45m above sea level, Ibaji at 42m above sea level and

Onitsha at 32m above sea level. The areas above the highest elevations obtained from the field, were

the areas that are not flooded. The elevations below the highest point obtained from the field, were

flooded.

Slope map: The slope map was generated from the DEM. The Slope tool is most frequently run on

an elevation dataset (http://webhelp.esri.com). The flooded areas were later overlaid on it to show

the relationship between the nature of the topography and flooding in the area.

Landuse and landcover map: The landuse/ landcover map was produced through supervised

classification of NigeriaSat-X image. ERDAS Imagine 9.2 software was used for this analysis. Training

samples sets were acquired from the image to run the supervised classification of six classes

(vegetation, flood plain agriculture, farmlands, settlements, water body and bare surface) using

maximum likelihood method. The area of each class extracted was converted from square meters to

hectares, and bar charts were generated to graphically illustrate the landuse and landcover of areas

affected in the 2012 flood disaster.


Figure 1: Digital Elevation Model Figure 2: Painted Relief


All the maps generated in this study namely the digital elevation model (figure 1), painted relief

(figure 2 ), slope map (figure 3 ), painted relief with 2012 flood level (figure 4) have settlement layers

overlain on them to relate to terrain characteristics, flooded areas and settlements.

In the DEM and slope map, generally high elevation areas are shaded red (this red color is different

from that representing settlements which usually appear sharper and occur mostly along the river

channel). Steeper slopes with slope values of 9.5 - 48.2 shaded light red correspond to red shades on

the DEM which also corresponds to light gray color on the painted relief map. Yellow colors on the

slope map represent slope values of 4.7 to 9.5 whereas the corresponding color on the DEM and

painted relief is green and gray/cyan respectively. Areas shaded blue on the slope map have slope

values of 0- 4.7 which corresponds to blue and gray shades on the DEM and painted relief, in that

order. Every cell in the output raster has a slope value: the lower the slope value, the flatter the terrain;

and the higher the slope value, the steeper the terrain. The output slope raster can be calculated as

percentage of slope or degree of slope (http://webhelp.esri.com). The slope values for the area range

from 0–48. From the DEM, there are elevations ranging from 15M above sea level to 657M above sea

level in the output raster map. Inundation models based on 2012 flood levels (table 1) at 45M, 35M

and 30M were generated for different sections of the river channel based on terrain characteristics

(figure 4). At these levels, the areas affected were flooded.


Figure 3: Slope Map Figure 4: Flood Inundation on Painted Relief

There are various topographic slopes along the lower river Niger channel from Lokoja to Onitsha

that influenced the flooding. However, two are selected for discussion here. Lokoja area around the

confluence has a low slope value and low elevations and thus was extensively flooded. Terrain

characteristics change significantly from around Shinaku down to Ajaokuta and to Agenbode. All along

here there are steeper slopes, higher elevations and consequently deeper channels along the river.

This has no doubt accounted for the fact that these areas were not extensively flooded except for the

flood plain and adjacent areas. A constriction of the river channel at Agenebode/Idah no doubt

generated a push-back effect on the surging flood water thereby causing rise in the flood level

upstream. A similar physiographic attribute is found at Onitsha-Asaba River Niger Bridge which must

have generated a rise in flood level upstream. On the contrary, from Agenebode/Idah to Onitsha/A the

river section widens down the channel, especially towards Ibaji down to the northern part of Anambra

state. Ibaji LGA is a low lying area with varying elevations and Illushi/ Edo state on the opposite side is

generally higher, down to Illah basin and Asaba, except for settlements along the river bank and the

floodplain that were inundated. In this section the floodwater has little or no topographic control. Thus

it freely occupied every available space enabled by low slope and elevations, destroying houses,


farmlands and other economic activities. It was also retained for over three weeks before it gradually

drained down river at Onitsha-Asaba Niger Bridge where the channel is narrowed again.

The landuse/ landcover map of total areas inundated (Figure 5) in the 2012 flood was produced

through supervised classification of NigeriaSatX image using the flood inundation model to delineate

the areas. The statistical result of landuse and landcover show that 129526 hectares of settlement

lands and 799601 hectares of farmland/floodplain agriculture were inundated (Figure 6). Settlements

affected by the flood include Idah down to Onitsha/Asaba and are within flatter terrain. Settlements

from Agenebode up to Ajaokuta were less inundated.

Figure 5: Landuse/Landcover of Flooded Area


Figure 6: Affected settlements and farmlands at Illa basin in Delta State (author’s photograph)

CONCLUSIONS

Through the processing of SRTM data into a DEM, slope map and painted relief, the area of study

has been categorized into mainly higher topography and flatter topography. The former has DEM and

slope values ranging from 45m-657m above sea level and 4.7-48 respectively. The flood inundation

model was performed for different sub-basins: Lokoja at 45m above sea level, Ajaokuta 44m, Ibaji 42m,

and Onitsha 32m using terrain height as a major factor for basin delineation. The areas above the

highest elevations obtained from the field were the areas that are not flooded, but the elevations

below the highest point obtained from the field were flooded. The statistical results of landuse and

landcover show that 129526 hectares of settlement lands and 799601 hectares of farmland/floodplain

agriculture were inundated.

ACKNOWLEDGEMENTS

I will like to thank National Centre for Remote Sensing Jos and National Space Research and

Development Agency, Abuja, Nigeria for funding this project.

REFERENCES

Eric M. Baer, 2014. Geology program, Teaching quantitative concepts in floods and flooding.

http://serc.carleton.edu/quantskills/methods/quantlit/floods.html (accessed March, 2014)

http://webhelp.esri.com/arcgisdesktop/9.3/body.cfm (accessed 2014)

IRIN. (2012, October 10) Nigeria-Worst Flooding in Decades ReliefWeb. http://www.irinews.org/Report/96504/ (accessed October 15, 2012)

Myers, M.F and White, G.F. 1993. The challenge of the Mississippi flood. Environment 35, 6-35

http://serc.carleton.edu/quantskills/methods/quantlit/floods.html

http://webhelp.esri.com/arcgisdesktop/9.3/body.cfm


ASSESSMENT AND ANALYSIS OF WILDFIRES WITH THE AID OF REMOTE SENSING AND GIS

Willem A. Vorster1, Maarten Jordaan2

1. SANSA – South African National Space Agency

2. UNISA – University of South Africa

KEYWORDS: Wildfires, Post-wildfire, Remote Sensing, Veld fires, Landsat, AFIS, MODIS, National Veld and Forest Fire Act, Forensic, Disputes, Optical sensors

ABSTRACT

Wildfires destroy large tracts of veld and forest in South Africa every year. These fires can be

devastating, resulting in the loss of human lives, the destruction of property and the loss of income.

For example, a forest fire in 2007 in the Sabie district in Mpumalanga resulted in the loss of seven

lives and destroyed approximately 7% of South Africa’s forested areas. There are frequent legal

disputes with respect to the starting point of wildfires, the extent of fire damage, and the specific

land cover and property destroyed by the fires.

The starting point, extent and period of the fire event can be determined by combining the

images of different optical satellites. An analysis can be made with the aid of indices and

classification of medium resolution satellites to determine the above fire characteristics. Legal

disputes and court cases with regards to a fire often ensue a few years after the event. By the time a

legal dispute arises, little or no evidence can be found on the terrain where the wildfire occurred.

Remote sensing archives provide a reliable source of data that can be used to analyse events after

long intervals.

The forensic capabilities of remote sensing in detecting and analysing post-wildfire characteristics

have become an important contribution towards solving such legal disputes and in understanding

wildfire characteristics. Earth observation images can be used as evidence in court cases. In this

project we will present a case study showing the methodology of producing post wildfire products

with the use of remote sensing.

INTRODUCTION

Even though fire is a natural phenomenon in South Africa, man-induced fires destroy large areas

of veld and forest every year, causing damage to infrastructure and sometimes even loss of people’s

lives (Gordin, 2008). The forensic capabilities of remote sensing in detecting and analysing post-

wildfire characteristics have become an important contribution towards solving legal disputes and in

understanding wildfire characteristics.

The 1998 National Veld and Forest Fire Act (NVFFA) of South Africa was put in place from a socio-

economical and a legal perspective to account for losses due to wildfires. “The primary purpose of

this Act is to prevent and combat veld, forest and mountain fires throughout the Republic” (NVFFA,

1998:5). The Act states that firebreaks are needed around terrains, to prevent the spread of a fire

from the property where it originated. A firebreak is an area or strip where most of the flammable


material has been removed. Firebreaks do not always stop a raging fire but could minimize the effect

thereof. Based on the implementation of firebreaks as endorsed by this Act, in order to determine

legal responsibility for a wildfire, it is important to determine the starting point to prove negligence

by a landowner.

The objective of this research project is to highlight the satellite image interpretation methods

that help to generate post-wildfire analysis products that can be used in legal cases. The main focus

of the research is the following three objectives: the determination of the starting point; the

computation of the fire scar and the spread thereof; and the impact on the environment.

METHODOLOGY FOR FIRE ANALYSIS

Remote sensing can be used to detect and monitor disasters and hazards such as wildfires.

Deploying an aircraft for every fire is very expensive and also impractical. Satellite remote sensing is

more economical and reliable than aerial photography, especially in post-fire analysis.

Characteristics of satellite images such as spatial, temporal and spectral resolution play an important

part in the forensic capabilities.

Temporal resolution

Earth observation satellites revisit the same area on a regular basis and therefore have a good

temporal resolution, resulting in the area destroyed by the fire being scanned within a day to thirty

days after the event.

Temporal resolution plays a vital role in terms of detecting a fire as it burns. Satellite sensors with

low spatial resolution (e.g. MODIS) but with a high temporal resolution, can detect wildfires;

however the analysis thereof is constrained.

Spatial resolution

Apart from temporal resolution, spatial resolution also has an important role in the analysis of

wildfires. Figure 1 illustrates the difference in resolution. Higher spatial resolutions (e.g. Landsat

30m), will result in a better analysis of the wildfire event. The Landsat image on the left covers the

same area as MODIS on the right, but indicates a clearly visible fire scar in the centre of the image

(the scar is the area with a reddish colour).

Figure 1: A subset of Landsat on 10 March 2012 (left) with high resolution and a subset of a

MODIS image of 13 February 2012 (right) with low resolution


All satellite data are ortho-rectified so that the different datasets can be compared (Mather,

2004) and be used in a geographic information system (GIS). The accuracy of the position of the data

is crucial to determine on which side of a fence a fire started.

Spectral resolution

The spectral bands of the satellite sensor determine which indices can be calculated for the fire

analysis. The following indices can be useful in interpreting a fire event: the normalized differential

vegetation index (NDVI), which consists of the red band and the near-infrared band (NIR); the

normalized burn ratio (NBR), which consists of NIR and short-wave infrared bands (SWIR2.2) (Brewer,

et al, 2005; Carla, et al, 2011; Henry, 2008; Lentile, et al, 2006) and the normalized differential

infrared index (NDII), which consists of NIR and SWIR1.6 bands (Carla, et al, 2011). The results of the

different indices are added to the original bands of the sensor used. This increased dimensionality

enables a supervised classification to determine the extent of a fire scar more accurately.

Image analysis

The higher resolution data is classified through supervised classification with the use of samples

from the classes of a known identity (Campbell &Wynne, 2011). With supervised classification and

change detection, old fire scars can be separated from newer scars to eliminate the older scars

through image subtraction. The final classification is then vectorized in a GIS to be able to calculate

the area that has been destroyed by the fire. Using a combination of fire scar vectors and the farm

boundaries, the burnt area can then be calculated per farm and the starting point of the fire be

identified.

AFIS system

The Advanced Fire Information System (AFIS) monitors all active fires on a daily basis.

Information obtained from AFIS (point data) was used in the GIS to keep track of the dates and times

of the fires. The AFIS system is based on MODIS data that are obtained from sensors on-board the

two satellite platforms, Aqua and Terra. AFIS also obtains information from the Meteosat Second

Generation (MSG) satellite, a geo-stationary metrological satellite that provides weather information

over Africa and Europe. The image acquisition frequency of MSG is every fifteen minutes, which

makes it ideal to track the spread of fires, even though it has a very low spatial resolution of 3km.

The AFIS system detects the active flames of a fire, using bands from the mid-infrared range (Frost,

2012). The spatial resolution of the mid-infrared bands and the thermal infrared bands, which are

used in the AFIS system, is approximately 1km. The smallest flaming unit (single fire) that can be

detected is 50m x 50m (Frost, 2012). In the case of MSG, the smallest flaming unit is 500m x 500m.

The use of weather conditions in fire analysis

Climatic factors such as wind, temperature and humidity (heat index) play an important role when

a fire event occurs. Prevailing winds result in a fire burning in an ellipsoidal pattern (Figure 2). The

shape of the ellipsoid depends on the strength and direction of the wind (Finney, 1998; Steensland,

et al, 2005).


Figure 2: The ellipsoidal pattern of a fire scar resulting from a strong western wind

CASE STUDY – THE SABIE FIRE

2007 was a dry year, and this fire took place just after the middle of winter. Strong westerly winds

blew during this time (SA Forestry Magazine, 2012). On 27 July 2007, four large wildfires destroyed

approximately 9% of the forest plantations of Mpumalanga (Forsyth, et al, 2010). These four wildfires

occurred east of Lydenburg in the vicinity of Graskop and Sabie on the escarpment. The present

study refers to the Sabie fire.

Landsat 5 and MODIS data were used to analyse this fire. Landsat 5 images dated 15 May 2007

and 18 August 2007 respectively, were used to provide images before and after the wildfire. MODIS

data were used as a time series to keep track of the fire. Using point data from AFIS as well as the

image information from MODIS and MSG, the spread of the fire was determined.

The NDVI, NBR and NDII indices were calculated and the products were added to the original

Landsat dataset, on which a supervised classification was then executed with the aid of training sites.

All the red areas in Figure 3 indicate the fire scars. The two Landsat images were subtracted from

each other and the Landsat image of 18 August 2007 was classified. The rationale for this method is

to ensure that all the old fire scars are eliminated from the latest fire scars.


Figure 3: Supervised classification from the Landsat 5 image dated 18 August 2007

Figure 4 shows (by means of a pie-graph time series) the first few hours of the devastating fire

that burned in the Sabie region on 27 June 2007. To visualize a timetable in a geographical way,

triangles are used to track the progress of the fire. The red triangles indicate the area where the fire

was first detected at 19:45. The progress of the fire is indicated anti-clockwise on the pie diagrams.

The fire was then detected a second time at 20:00 and that is indicated by the orange triangle.

Figure 4: MSG representation of the first few hours of the fire. The backdrop is from the

Landsat of 18 August 2007. The legend in the pie diagram indicates the progress of the fire in

local time.

Figure 5 portrays the information from the MODIS data in AFIS, detected at 22:03 on 27 June

2007 not long after the fire started, and three hours later at 01:20 on 28 June 2007. It is indicated by

the spreading of the red dots. The figure on the right illustrates that the whole area around Sabie

was on fire when the MODIS sensor detected it again, just after midnight. At that stage two fires

were burning because the upper eight dots indicate a second fire.


Figure 5: MODIS information at 22:03 on 27 June 2007 (left) and at 01:20 local time on 28

June 2007 (right) on a backdrop of Landsat image of 18 August 2007. The red dots indicate the

burning areas detected by AFIS.

Figure 6: MODIS time series of the first three days of the fire – From 27 July 2007 to 29 July

2007.

Figure 6 portrays a time series of MODIS data from 27 July 2007 to 29 July 2007. In the morning

(at 10:00) on 27 July 2007, no fire was visible, not even smoke. At 14:20 on 27 July 2007, smoke was

detected. The wildfire started on the evening of 27 July 2007, before 20:30. Satellite MSG detected

the fire for the first time at 20:30. For MSG to detect a fire, it must be at least 500m in diameter. The

next day, 28 July 2007, the whole area was on fire and it was detected by MODIS at 10:40. At the

next three overpasses of the MODIS sensor, fires were burning over a large area.


Figure 7: Vectorized fire scar with the Landsat of 18 August 2007 as backdrop

Based on the methodology described in section 2, the fire scar extent was calculated. In this case

the pre-event image (15 May 2007) was used to detect old fire scars and the post image (18 August

2007) for the actual fire scar extent. The double Figure 7 shows the total fire extent (the total extent

was approximately 27914 hectares) on the left, and the extent of the farm portions destroyed by the

fire, on the right. For some portions the area that was destroyed per farm, is indicated in hectares.

Another very important factor in fire scar analysis is determining the area where the fire started.

This links back to the legal questions around the fire and the fire damage. Figure 8, left outlines the

fire extent and Figure 8, right indicates the probable location of the start of the fire. This location was

derived based on Figures 4 and 5.

Figure 8: Landsat image of 18 August 2007 with the mapped fire scar and area of probable

starting point (in zoom-in picture)

CONCLUSION

The analysis of wildfires with remote sensing proofed successful in the calculation of the total

area that was burnt as well as in determining the starting point. It is however, not always possible to

determine the exact starting point but one can arrive at a good estimation. In the case of the Sabie

fire, external information was needed to determine the exact starting point. This information came

from people on the ground.

The accuracy in determining the exact starting point of most fires is actually enhanced when using

additional external information. This also goes for the fire shown in Figure 1.


Successful tracking or following of a fire can be done if the fire has burnt over an extended period

of more than a day. When a fire has burnt between two passes of the MODIS sensor, the fire cannot

be tracked, but the final scar can still be calculated.

One shortcoming of optical data is that a fire scar cannot be determined after the rain season has

started because rain washes away the evidence and the vegetation starts to grow back quite rapidly.

The next step in fire analysis would be to do fire modelling (Finney, 1998), especially when more

than one fire occurred in the same area and these fires burnt into each other.

ACKNOWLEDGEMENTS

Thank you SANSA for supplying the imagery to do this study and for the financial support to do a

M.Sc. based on wildfire investigations. Thank you Dr Nicky Knox for the support given in writing this

article.

REFERENCES

Brewer, C.K., Winne, J.C., Redmond, R.L., Opitz, D.W. & Mangrich, M.V., 2005. Classifying and Mapping Wildfire severity: A comparison of methods. Photogrammetric Engineering & Remote Sensing. 71, November 2005. pp. 1311-1320.

Campbell, J.B. & Wynne, R.H., 2011. Introduction to Remote Sensing. Fifth Edition. New York: Guilford Press.

Carla, R., Santurri, L., Bonora, L. & Conese, C., 2011. Multitemporal burnt area detection methods based on a couple of images acquired after the fire event. The 5th International Wildland Fire Conference. Sun City, South Africa.

Finney, M.A., 1998. FARSITE: Fire Area Simulator – Model Development and Evaluation. http://www.landsinfo.org/ecosystem_defense/Federal_Agencies/Forest_Service/Region_1/Idaho_Panhandle_NF/Bonners_Ferry_District/Myrtle%20HFRA/Myrtle%20Creek%20HFRA%20Objection%20references%20disk%204/fireareaall.pdf [Access: 28 January 2010]

Forsyth, G.G., Kruger, F.J. & Le Maitre, D. C., 2010. National Veldfire Risk Assessment: Analysis of exposure of social, economic and environmental assets to veld fire hazards in South Africa. http://www2.dwaf.gov.za/webapp/Documents/Veldfire_Risk_Report_v11.pdf [Access: 10 February 2012]

Frost, P., 2012. Personal conversation about the AFIS system. CSIR. South Africa.

Gordin, J., 2008. Devastating forest fires 'worst ever'. http://www.iol.co.za/index.php?from=rss_News&set_id=1&click_id=79&art_id=vn20080127093506915C583203 [Access: 22 January 2010]

Henry, C.M., 2008. Comparison of single- and multi-date Landsat data for mapping wildfire scars in Ocala National Forest, Florida. Photogrammetric Engineering & Remote Sensing. 74, July 2008. pp. 881-891

Lentile, L.B., Holden, Z.A., Smith, A.M.S., Falkowski, M.J., Hudak, A.T., Morgan, P., Lewis, S.A., Gessler, P.E., & Benson, N.C., 2006. Remote sensing techniques to assess active fire characteristics and post fire effects. http://www.cnr.uidaho.edu/for570/Readings/2006_Lentile_et_al.pdf [Access: 13 January 2011]

http://www.landsinfo.org/ecosystem_defense/Federal_Agencies/Forest_Service/Region_1/Idaho_Panhandle_NF/Bonners_Ferry_District/Myrtle%20HFRA/Myrtle%20Creek%20HFRA%20Objection%20references%20disk%204/fireareaall.pdf



http://www2.dwaf.gov.za/webapp/Documents/Veldfire_Risk_Report_v11.pdf

http://www.cnr.uidaho.edu/for570/Readings/2006_Lentile_et_al.pdf


Mather, P.M., 2004. Computer Processing of Remotely-Sensed Images. West Sussex: John Wiley & Sons.

National Veld and Forest Fire Act (NVFFA) 101, 1998. South Africa

SA Forestry magazine: Fire storms rip through KZN, Mpumalanga, Limpopo & Swaziland. http://www.saforestrymag.co.za/articles/detail/fire_storms_rip_through_kzn_mpumalanga_limpopo_swaziland [Access: 13 October 2012]

Steensland, P., Henricks, J., Garvey, B., White, G., Carpenter, J., Heath, M., Ness, K., Herman, H., Dunn, N., Parker, C., Carlsom, qqw en, A., Hilton, G., & Nanamkim, J., 2005. Wildfire

http://www.saforestrymag.co.za/articles/detail/fire_storms_rip_through_kzn_mpumalanga_limpopo_swaziland

http://www.saforestrymag.co.za/articles/detail/fire_storms_rip_through_kzn_mpumalanga_limpopo_swaziland


POLINSAR COHERENCE OPTIMISATION FOR DEFORMATION MEASUREMENT IN AN

AGRICULTURAL REGION

Jeanine Engelbrecht1, Michael Inggs2 1. CSIR Meraka Institute, [email protected]

2. University of Cape Town, South Africa

KEYWORDS: Interferometry, Coherence optimisation, Surface deformation

ABSTRACT

Surface deformation due to underground mining poses risks to health and safety as well as

infrastructure and the environment. Consequently, there is a need for long-term operational

monitoring systems. Differential interferometry (dInSAR) techniques are well known for its ability to

provide centimetre to millimetre scale deformation measurements. The maturity of dInSAR has, in

principle, overcome the limitations associated with field-based techniques and has been extensively

used for its ability to monitor deformation over large areas, remotely. However, in natural and

agricultural areas, the presence of vegetation and the evolution of the land surface introduce a phase

noise component which limits successful interferometric measurement. This paper aims to address

the known limitations of traditional dInSAR in the presence of disturbances to reflected signals due

to agricultural activities, by testing the polarimetric interferometry (polInSAR) technique for its ability

to increase interferometric coherence and to detect surface movement in the areas of interest. The

results suggest that, although coherence optimisation algorithms result in a statistically significant

increase in interferometric coherence, the spatial heterogeneity of the scattering process and how it

changes over time caused random phase changes associated with temporal baseline effects and the

evolution of the land surface. These effects could not be removed from interferograms using the

polInSAR approaches. The heterogeneity of the scattering processes implied that different phase

centres were present in interferograms which introduced a spatially heterogeneous topographic

phase contribution. Consequently, the polInSAR techniques are considered to be unsuccessful in

enhancing the ability to extract deformation measurements in the area of interest.

INTRODUCTION TO DEFORMATION MEASUREMENTS IN VEGETATED REGIONS

Surface deformation due to underground mining poses risks to health and safety as well as

infrastructure and the environment. Consequently, there is a need for long-term operational

monitoring systems. Traditional field-based measurements are point-based, meaning that the full

extent of deforming areas is poorly understood. Field-based techniques are also labour intensive if

large areas are to be monitored on a regular basis. Differential interferometry (dInSAR) techniques

are well known for their ability to provide cm to mm scale deformation measurements. The maturity

of dInSAR has, in principle, overcome the limitations associated with field-based techniques and has

been extensively used for its ability to monitor deformation over large areas, remotely.



Although dInSAR techniques are successful in monitoring surface deformation, the phase

decorrelation effects due to temporal and geometric sources are described as the most limiting

factors (Prati et al., 2010; Reigber et al., 2007). Temporal decorrelation effects include decorrelation

due to a change in the position of the scatterer over time as well as a change in scattering

characteristics of the target (including a change in the shape, orientation and dielectric constant of

the scatterer) (Reigber et al., 2007). In a dynamic commercial agricultural region for instance, the

evolution of the land surface is quite pronounced with activities such as tilling, crop growth and

harvesting significantly altering the observed surface which leads to signal decorrelation (or an

increase in phase noise) depending on the wavelength and polarisation of the signal. Temporal

decorrelation effects can also occur over short time periods. For instance, the random movement of

leaves and twigs of vegetation implies that scattering elements are continuously re-arranged causing

signal decorrelation. The interferometric coherence, used as indicator of phase noise, will be low

(high noise) in areas with higher vegetation densities and will decrease rapidly with time (Grey and

Luckman, 2001). Consequently, temporal decorrelation effects makes dInSAR measurements

impractical over vegetated areas (Ferretti et al., 2001). The effect of signal decorrelation can be

partially overcome by using longer wavelength L-band data that penetrate through agricultural crops,

thereby increasing the interaction with the surface. However, a change in surface characteristics

induced by activities such as tilling and planting may still lead to signal decorrelation.

To minimise the phase noise for dInSAR measurement, several advanced processing techniques

have been developed. This paper aims to address the known limitations of traditional dInSAR in the

presence of disturbances to reflected signals due to agricultural activities, by testing the polarimetric

interferometry technique for its ability to increase interferometric coherence and to detect surface

movement in dynamic agricultural environments.

ADVANCED INTERFEROMETRY TECHNIQUES FOR DEFORMATION MEASUREMENT

To overcome the problems associated with temporal decorrelation, several advanced

interferometric processes have been developed (Berardino et al., 2002; Euillades et al., 2011; Ferretti

et al., 2001; Mora et al., 2003; Prati et al., 2010). Techniques focusing on pixels that remain coherent

over the entire stack of interferograms, include the persistent scatterers interferometry (PSI) and

Small BAseline Subset (SBAS) techniques (Ferretti et al., 2001; Prati et al., 2010). The coherent pixels

are stable natural reflectors usually corresponding to man-made structures or rock-outcrops (Ferretti

et al., 2001; Raucoules et al., 2007), meaning that in vegetated, non-urban areas, the density of

coherent pixels may be very low, limiting the viability of the SBAS and PSI techniques in these areas

(Galloway and Hoffman, 2007; Reigber et al., 2007). Therefore an alternative approach to overcome

temporal decorrelation effects is needed.

Since temporal decorrelation not only has an effect on the interferometric coherence, but also

leads to a different polarimetric responses in two SAR images (Cloude and Papathanassiou, 1998;

Perski and Jura, 2003) the introduction of SAR polarimetry into conventional interferometry has been

proposed (Stebler et al., 2002). These advanced dInSAR techniques have focussed on exploiting the

polarimetric properties of SAR signals to maximise interferometric coherence as opposed to merely

selecting the high coherence targets for further processing (Navarro-Sanchez et al., 2010; Pipia et al.,

2009). The ability to identify and separate scattering mechanisms (i.e. surface scattering vs.

vegetation scattering) using SAR polarimetry implies that the combination of interferometric and


polarimetric information can be used to infer the interferometric phase of any scattering mechanism

and, consequently, the vertical distribution of different scattering mechanisms (Colin et al., 2006;

Papathanassiou and Cloude, 1997; Papathanassiou and Cloude, 2001). The combination of radar

polarimetry and radar interferometry, known as polarimetric interferometry (polInSAR), enables the

development of coherence optimisation algorithms to improve the quality of interferometric

measurements (Cloude and Papathanassiou, 1998; Cloude and Papathanassiou, 1997; Colin et al.,

2006; López-Martinez et al., 2009; Navarro-Sanchez et al., 2010; Neumann et al., 2007; Neumann et

al., 2008; Pipia et al., 2009). Coherence optimisation is achieved by identifying the scattering

mechanism which leads to the highest possible coherence and, consequently, the scattering

mechanism providing best phase estimates (Colin et al., 2006).

In this investigation, coherence optimisation was achieved using the Multiple Scattering

Mechanism (MSM) technique, which relies on the identification of the scattering mechanisms that

provide the highest possible coherence, as well as those that provide intermediate and low

coherence values (Cloude and Papathanassiou, 1998). A potential drawback of the MSM technique is

that the scattering mechanism that provides the highest possible interferometric coherence, can vary

between neighbouring pixels and, different scattering mechanisms can contribute to the optimal

phase (Neumann et al., 2007; Neumann et al., 2008; Pipia et al., 2009; Reigber et al., 2007). In a

dynamic agricultural area, this implies that phase measurements at different phase centres can be

provided with the implication that topographic phase can be incorporated depending on the

scattering mechanism that provides the best coherence.

STUDY AREA AND DATA

The area under investigation is situated in the Mpumalanga Province of South Africa, an area

associated commercial agricultural activities as well as near-surface and underground coal mining.

The exact location of the area is not provided due to operational sensitivity. Surface subsidence is

associated with near-surface coal mining and, dInSAR techniques on both L-band and C-band data

have been successfully used to detect and monitor surface deformations (Engelbrecht et al., 2011;

Engelbrecht et al., 2013; Engelbrecht and Inggs, 2013). The commercial agricultural nature of the

area under investigation however meant that temporal decorrelation effects had a significant impact

on interferometric coherence.

To test polInSAR techniques to decrease signal decorrelation due to evolving land surfaces, 12

RADARSAT-2 (Fine Quad Polarization, Beam mode FQ16) scenes acquired between 26 January and 28

December 2011 were used. In general, scenes were captured at 24 day intervals with the exception

of the period between 2011/05/02 and 2011/08/06 during which no scenes were captured due to a

satellite scheduling conflict. The dates of image acquisition and incidence angle are presented in

Table 1. DInSAR and PolInSAR processing was performed using each consecutive image as master

image with successive scenes as slaves. The processing was performed using SARscape software. To

determine the effectiveness of coherence optimisation algorithms, the interferometric coherence

from traditional dInSAR techniques were compared to the coherence achieved after coherence

optimisation algorithms were applied.

Table 1. Incidence angle ranges and dates of image capture

Date of image capture Incidence angle range (degrees)


2011/01/26 35.4 - 37 2011/02/19 35.4 – 37 2011/03/15 35.4 – 37 2011/04/08 35.4 – 37 2011/05/02 35.4 – 37 2011/08/06 35.4 – 37 2011/08/30 35.4 – 37 2011/09/23 35.4 – 37 2011/10/17 35.4 – 37 2011/11/10 35.4 - 37 2011/12/04 35.4 - 37 2011/12/28 35.4 - 37


The result of the coherence optimization algorithm is 3 interferograms, representing high,

medium, and low coherence products and their associated differential interferograms (Figure 1). The

interferograms were constructed by identifying the dominant scattering mechanisms and separating

those representing the highest coherence values from the scattering mechanisms exhibiting

intermediate and low coherence values. The theoretical optimum coherence will have a value of 1,

meaning that no phase noise is present. However, an increase in phase noise is associated with

coherence values < 1. To compare the coherence optimisation (hereafter called polInSAR) results

with the results obtained using traditional dInSAR techniques, the coherence values obtained using

traditional dInSAR techniques were compared to the maximum, intermediate and low coherence

data obtained by polInSAR. The frequency distribution of the average scene coherence for dInSAR

and polInSAR coherence is presented in Figure 1. The results indicate a significant increase in average

scene coherence for both the maximum and intermediate polInSAR coherence products compared to

the dInSAR coherence.

Figure 1: The frequency distribution of average scene coherence values for dInSAR

processing and the maximum (Max), minimum (Min) and intermediate (Med) coherence

optimisation results.

When comparing the average scene coherence achieved using polInSAR with the average scene

coherence achieved using the traditional dInSAR techniques, correlation coefficients (Table 1) reveal

a strong correlation between the coherence of traditional dInSAR (dInSAR Cc) and polInSAR

(maximum coherence = Max Cc, intermediate coherence = Med Cc and minimum coherence = Min

Cc) results. This suggests that parameters affecting interferometric coherence obtained by traditional


dInSAR techniques, will affect polInSAR coherence as well. To determine the effect of evolving land

surfaces, enhanced vegetation index (EVI) data were considered. Table 1 indicates the correlation

between the polInSAR results and the change in land cover conditions (as simulated by the change in

EVI over time (|δEVI|)) as well as the day difference between image acquisitions (temporal baseline

(Btemp)). A decreased sensitivity to a change in land cover conditions as described by |δEVI| is

observed for polInSAR results compared to dInSAR results. However, polInSAR coherence appears to

be affected most significantly by an increase in temporal baseline with correlations of p = -0.65, p = -

0.66, and p = -0.64 for Max Cc, Med Cc and Min Cc respectively being observed.

Table 1: The Pearson correlation coefficients of the coherence using dInSAR and coherence

obtained using polInSAR (Max Cc, Med Cc and Min Cc). Correlations with |δEVI| and temporal

baseline are also indicated.

Max Cc Med Cc Min Cc |δEVI| Btemp

dInSAR Cc 0.80 0.89 0.94 -0.59 -0.67

Max Cc 0.98 0.93 -0.51 -0.65

Med Cc 0.98 -0.51 -0.66

Min Cc -0.50 -0.64

The polInSAR results revealed that the maximum and intermediate coherence results showed a

significant improvement in average scene coherence compared to the average dInSAR coherence

(Figure 2). However, in traditional interferometric processing, interferometric measurement on a

single pixel is not sensible since single pixels can incorporate phase noise in an unpredictable way.

The noise effects can cause pixels to undergo random phase changes that visually manifest as

randomly coloured speckles in interferograms. The random nature of the phase noise component of

the interferograms was estimated by calculating the coefficient of variation (CV), of the coherence

values for each interferogram. The CV is defined as

CV

Where is the standard deviation of coherence and is the average coherence. Lower CV values

indicate more homogenous data (lower variance) whilst higher CV values indicate more

heterogeneous data (higher variance). Subsets of 11 x 11 pixel interferograms were used to illustrate

the effect of homogeneous and heterogeneous areas on interferograms with high and low average

coherence in Figure 2 A and B respectively. Figure 2 A shows interferograms over relatively

homogeneous areas (CV = 0.094 and 0.132 respectively) associated with high average coherence

(Figure 2 A i) as well as low average coherence (Figure 2 A iii). Additionally, interferograms created

over heterogeneous areas (CV = 0.431 and 0.457 respectively) with high average coherence (Figure 2

B 1) and low average coherence (Figure 2 B iii) are shown in Figure 2 B. The results indicate that, in a

more homogeneous area with a high average coherence, there is a better agreement between

neighbouring pixels implying a higher confidence in interferometric measurements. However, if the

area is homogeneous but the average coherence is low, phase noise remains problematic (Figure 2 A

iii). Similarly, high average coherence in a heterogeneous region (Figure 2 B i and ii) is associated with


some speckle effects although to a lesser extent than low average coherence in a homogeneous

region (Figure 2 A iii and iv). The results demonstrate that high coherence values, although needed

for reliable interferogram generation, is not the only requirement and that homogeneity of

coherence values over several pixels (estimated here by coefficient of variation) is needed for high

quality interferogram generation.

A

B

Figure 2: A: The interferograms over homogeneous areas (low CV) and high average scene

coherence (i and ii) vs. interferograms over homogeneous areas with a low average scene

coherence (iii and iv) and B: The interferograms over heterogeneous areas (high CV) and high

average scene coherence (i and ii) vs. interferograms over heterogeneous areas with a low

average scene coherence (iii and iv).

The coefficient of variation of coherence values was calculated for dInSAR coherence as well as

the maximum coherence result of the coherence optimization algorithm (Figure 3). The results

indicate that a consistently higher CV of coherence is achieved after coherence optimization is

applied compared to a lower CV achieved prior to coherence optimization. This suggests that,

although average coherence values may be higher after polInSAR, the phase noise manifested by

random phase changes is not minimised using polInSAR on C-band data and, consequently, the phase

agreement between neighbouring pixels that is needed for successful deformation measurement, is

not met.


Figure 3: The coefficient of variation (CV) of coherence values calculated for dInSAR and

polInSAR (max) results

CONCLUDING REMARKS

Phase noise introduced into interferograms causes pixels to undergo random phase changes

which visually manifest as randomly coloured speckles and, consequently, destroy the organised

fringe pattern needed for deformation measurements (Massonnet and Feigl, 1998). The coherence

optimisation algorithms implemented in this investigation attempted to optimise the interferometric

coherence by identifying the phase contribution from scattering mechanisms that lead to the highest

possible coherence. Although optimization of interferometric coherence was statistically achieved,

the optimization did not translate to the differential interferometric phase. A strong correlation

between the polInSAR coherence values and coherence values obtained by traditional dInSAR was

observed. This implied that the parameters affecting the phase noise component of interferograms

equally affected both the dInSAR and polInSAR interferograms. Although the polInSAR coherence

was found to be slightly less affected by the evolution of the land surface (as approximated by

|δEVI|) than the dInSAR coherence, the correlation between the polInSAR coherence and temporal

baseline was similar to the correlation between the dInSAR coherence and temporal baseline. It was

therefore concluded that, although coherence could be statistically optimised using polInSAR

techniques, the techniques were not successful in significantly decreasing the phase noise due to

evolving land surfaces and temporal baselines in the agricultural area under investigation.

The spatial heterogeneity of the scattering process and how it changed over time, caused random

phase changes associated with temporal baseline effects and the evolution of the land surface. The

random nature of phase changes between neighbouring pixels was recognised by calculating the

coefficient of variation (CV) for dInSAR and polInSAR coherence data. It was observed that the CV for

polInSAR results was consistently higher than the CV for traditional single polarisation dInSAR results.

This was indicative of a decrease in the spatial homogeneity of the phase noise contribution. The

decrease in the spatial homogeneity of the phase noise was associated with an increase in the

random speckle effect on the derived interferograms. In this investigation, the Multiple Scattering

Mechanism technique for coherence optimisation was tested and is therefore considered to be

unsuccessful in enhancing the ability to extract deformation measurements in the area of interest.

Future investigations should focus on Equal Scattering Mechanism approaches for coherence

optimisation to increase the spatial homogeneity of the optimal scattering mechanisms selected.


REFERENCES

Berardino, P., Fornaro, G., Lanari, R. & Sansosti, E., 2002. A new algorithm for surface deformation monitoring based on small baseline differential SAR interferograms. IEEE Transactions on Geoscience and Remote Sensing, 40, 2375-2383.

Cloude, S. R. & Papathanassiou, K., 1998. Polarimetric SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing, 36, 1551-1565.

Cloude, S. R. & Papathanassiou, K. P., 1997. Coherence optimisation in polarimetric SAR interferometry. IEEE International Geoscience and Remote Sensing Symposium, Singapore.

Colin, E., Titin-Schnaider, C. & Tabbara, W., 2006. An interferometric coherence optimization method in radar polarimetry for high-resolution imagery. IEEE Transactions on Geoscience and Remote Sensing, 44, 167-175.

Engelbrecht, J. & Inggs, M. R., 2013. Differential interferometry techniques on L-band data employed for the monitoring of surface subsidence due to mining. South African Journal of Geomatics, 2, 82-93.

Engelbrecht, J., Inggs, M. R. & Makusha, G., 2011. Detection and monitoring of surface subsidence associated with mining activities in the Witbank Coalfields, South Africa, using differential radar interferometry. South African Journal of Geology, 114.1, 77-94.

Engelbrecht, J., Musekiwa, C., Kemp, J. & Inggs, M. R., 2013. Parameters affecting interferometric coherence - the case of a dynamic agricultural region. IEEE Transactions on Geoscience and Remote Sensing, 52,

Euillades, L. D., Euillades, P. A., Pepe, A., Blanco, M. H. & Barón, J. H., 2011. On the generation of late ERS deformation time series through small Doppler and baseline subsets differential SAR interferograms. IEEE Geoscience and Remote Sensing Letters, 8, 238-242.

Ferretti, A., Prati, C. & Rocca, F., 2001. Permanent scatterers in SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing, 39, 8-20.

Galloway, D. L. & Hoffman, J., 2007. The application of satellite differential SAR interferometry derived ground displacements in hydrogeology. Hydrogeology Journal, 15, 133-154.

Grey, W. & Luckman, A., 2001. Deriving urban topography from multi-baseline SAR interferometric phase coherence images. http://earth.esa.int/pub/ESA_DOC//gothenburg/309grey.pdf. Accessed 4-1-2012.

López-Martinez, C., Fàbregas, X. & Pipia, L., 2009. PolSAR and PolInSAR model based information estimation. IEEE International Geoscience and Remote Sensing Symposium, Cape Town, South Africa.

Massonnet, D. & Feigl, K. L., 1998. Radar interferometry and its application to changes in the Earth's surface. Reviews of Geophysics, 36, 441-500.

Mora, O., Mallorqui, J. J. & Broquetas, A., 2003. Linear and nonlinear terrain deformation maps from a reduced set of interferometric SAR images. IEEE Transactions on Geoscience and Remote Sensing, 41, 2243-2253.

Navarro-Sanchez, V. D., Lopez-Sanchez, J. M. & Vicente-Guijalba, F., 2010. A contribution of polarimetry to satellite differential SAR interferometry: increasing the number of pixel candidates. IEEE Geoscience and Remote Sensing Letters, 7, 276-280.


Neumann, M., Ferro-Famil, L. & Reigber, A., 2007. Multibaseline PolInSAR coherence modelling and optimization. IEEE International Geoscience and Remote Sensing Symposium, Barcelona, Spain.

Neumann, M., Ferro-Famil, L. & Reigner, A., 2008. Multibaseline polarimetric SAR interferometry coherence optimization. IEEE Geoscience and Remote Sensing Letters, 5, 93-97.

Papathanassiou, K. P. & Cloude, S. R., 1997. Polarimetric effects in repeat-pass SAR interferometry. IEEE International Geoscience and Remote Sensing Symposium, Singapore.

Papathanassiou, K. P. & Cloude, S. R., 2001. Single-baseline polarimetric SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing, 39, 2352-2363.

Perski, Z. & Jura, D., 2003. Identification and measurement of mining subsidence with SAR interferometry: Potentials and limitations. Proceedings of the 11th FIG Symposium on Deformation Measurements, Santorini, Greece.

Pipia, L., Fabregas, X., Aguasca, A., Lopez-Martinez, C. & Mallorquí, J. J., 2009. Polarimetric coherence optimization for interferometric differential applications. IEEE International Geoscience and Remote Sensing Symposium, Cape Town, South Africa.

Prati, C., Ferretti, A. & Perissin, D., 2010. Recent advances on surface deformation measurement by means of repeated space-borne SAR observations. Journal of Geodynamics, 49, 161-170.

Raucoules, D., Colesanti, C. & Carnec, C., 2007. Use of SAR interferometry for detecting and assessing ground subsidence. C.R Geoscience, 339, 289-302.

Reigber, A., Neumann, M., Erten, E., Jäger, M., & Prats, P., 2007. Multi-baseline polarimetrically optimised phases and scattering mechanisms for InSAR applications. IEEE International Geoscience and Remote Sensing Symposium, Barcelona, Spain.

Stebler, O., Meier, E., & Nüesch, D., 2002. Multi-baseline polarimetric SAR interferometry- first experimental spaceborne and airborne results. ISPRS Journal of Photogrammetry and Remote Sensing, 56, 149-166.