This article was downloaded by: [University of New Mexico]On: 22 November 2014, At: 09:34Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: MortimerHouse, 37-41 Mortimer Street, London W1T 3JH, UK

Geocarto InternationalPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/tgei20

Monitoring vegetation cover over diverselandscape types in semiarid South EasternBotswanaOpha P. Dube a ba Department of Geographical Sciences and Planning , University of Queensland ,Temporary address (until December 1998), Qld, 4072, Australiab Department of Environmental Science , University of Botswana , Private Bag 002,Gaborone, BotswanaPublished online: 17 Sep 2008.

To cite this article: Opha P. Dube (1998) Monitoring vegetation cover over diverse landscape types in semiarid SouthEastern Botswana, Geocarto International, 13:2, 43-51, DOI: 10.1080/10106049809354641

To link to this article: http://dx.doi.org/10.1080/10106049809354641

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”)contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensorsmake no representations or warranties whatsoever as to the accuracy, completeness, or suitabilityfor any purpose of the Content. Any opinions and views expressed in this publication are the opinionsand views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy ofthe Content should not be relied upon and should be independently verified with primary sources ofinformation. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands,costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly orindirectly in connection with, in relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantial orsystematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distributionin any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found athttp://www.tandfonline.com/page/terms-and-conditions

Monitoring Vegetation Cover over Diverse Landscape Types inSemiarid South Eastern Botswana

Opha P. Dube*Temporary address (until December 1998)Department of Geographical Sciences and PlanningUniversity of Queensland, Qld 4072, Australia

Abstract

Vegetation cover from six landscape types was related to Landsat MSS and SPOT multispectral (SPOT XS) singlebands and indices derived from red (RED) and near-infrared (NIR) (NDVI), and RED and green (GREEN) (NDGR)bands at three levels: individual landscapes, across landscapes and between landscapes. Individual landscapesexhibited different relationships with the SPOT XS bands and indices. Relatively weak relationships were observedacross landscapes with better results on the RED band, r2 of 0. 75 and 0.71 (P>0.001) for MSS and SPOT XS,respectively. Higher r2 of up to 0.93 (P>0.01) in MSS NDVI based on the RED and NIR band 3, and 0.85 (P>0.01)in SPOT XS RED band were established between landscapes. Variation in vegetation cover within landscapes wassignificantly different from variation across landscapes at P> 0.01. The study demonstrated the importance ofstratifying landscape at the appropriate scale for vegetation studies in semi-arid lands of Botswana.

Introduction

Botswana is dominated by woody vegetation with largespatial and temporal variability in cover due to land use, fireand rainfall. Woody vegetation is a source of browse andbiomass energy for 80% of the population in the country(Kgathi et al, 1994). Over 70% of the 586 000 km2 surfacearea of the country is used for livestock production (GRM,1987). Satellite data is the most cost effective way ofproviding repetitive information on vegetation cover in therangelands of Botswana.

Studies on the relationship between vegetation and digitaldata over Botswana have reported inconsistent findings(Vujakovic, 1985; Ringrose and Matheson, 1987; Jacobbergerand Hooper, 1991; Van Heist and Kooiman, 1992). Poorrelationships between woody vegetation and Landsat MSS,TM and SPOT XS data in different parts of Botswana havebeen reported by Ringrose and Matheson (1987), Van Heistand Kooiman (1992) and Sekhwela (1995). Van Heist andKooiman (1992) concluded that it was not feasible to monitorwoody vegetation using digital Landsat MSS data. In contrast,Vujakovic (1985) and Jacobberger and Hooper (1991), foundstrong relationships between woody cover and Landsat MSSand TM data. Nellis and Bussing (1989) observed thatseparation between bush savanna and dense grass occurredon the "tasseled cap" green vegetation index. Studies based

on NOAA AVHRR NDVI have reported strongerrelationships (Prince and Tucker, 1986; Choudhury andTucker, 1987). However, Nicholson and Farrar (1994) foundthat rainfall fluctuations over Botswana were not consistentlymanifested by changes in NOAA AVHRR NDVI but changeon the NDVI was soil dependent.

The lack of consistency in the relationship betweenvegetation in the semi-arid lands and a number of vegetationindices derived from RED and NIR is well documented(Tueller, 1987; Huete, 1988; Richardson and Everitt, 1992;Pickup et al., 1993). Nevertheless, strong relationships foundbetween vegetation and NOAA AVHRR NDVI in semi-aridlands, for example the Sahel, have facilitated regionalmultitemporal primary production studies (Tucker et al.,1985 and 1986). The relation of high spatial resolution datato vegetation cover needs to be outlined not only forcalibrating regional NOAA based studies but also for resourceplanning at local level in different ecoregions.

This study investigated the relationship between percentvegetation cover with wet season Landsat MSS and SPOTHigh Resolution Visible multispectral (SPOT XS) bands,and normalised difference vegetation indices based on REDand NIR, and RED and green (GREEN) wavebands. Therelationship was studied at three levels; within individuallandscapes, across different landscapes, and betweenlandscape types found in the area.

* Permanent addressDepartment of Environmental Science,University of Botswana, Private Bag 002, Gaborone,Botswana

Geocarto International, Vol. 13, No. 2, June 1998Published by Geocarto International Centre, G.P.O. Box 4122, Hong Kong.

43

Dow

nloa

ded

by [

Uni

vers

ity o

f N

ew M

exic

o] a

t 09:

34 2

2 N

ovem

ber

2014

Study area

The study area falls between latitude 24° 11' to 24° 25'south and longitude 25° 25' to 25° 41' east in the KwenengDistrict, over an area of 400 km2 (Fig. 1). It extends over thetwo major ecological regions of Botswana; the hardveld,characterised by medium textured soil in the east, and theKalahari Sandveld, with deep sandy soils in the west(Government of Botswana, 1990). The area receives a meanannual rainfall of 450 mm with 35% variability (Bhalotra,1985). Rainfall declines west towards the Kalahari Sandveld.

Small scale crop production and livestock rearing are majorland uses in the area.

Methods

Field data were collected during the peak of the growingseason and corresponded to the satellite overpass dates,March and April 1988 for SPOT XS and Landsat MSS,respectively. The study area was stratified using a hardcopyApril 1986 SPOT XS image at 1:100 000 supplemented byinformation from soil, vegetation and landsystems maps

23°3O'S

26°00 S-

— Tropic of Capricorn

KWENENG DISTRICT

r! • " * * - * • * - ! _ • / • •

MahetlweLeritsweietau

STUDY ^ . —3AREA

«. SPOT XSScene GfíS

127-399

50. kms

LOCALITY SKETCH

SSTUDY SITE

LEGEND

Road network

International Boundary

District Boundary

Sandveld & Hardveld Boundary

25°OO E 26«00 E .

Figure 1 Study area

44

Dow

nloa

ded

by [

Uni

vers

ity o

f N

ew M

exic

o] a

t 09:

34 2

2 N

ovem

ber

2014

(Timberlake, 1980 and Government of Botswana, 1990).Six landscape types defined in terms of topography, soil, andvegetation cover type, were produced and used for samplingin the field; Rocky Hills, Floodplains, Sandstone Hills,Kalahari Sandveld, Plains and Dry/Fossil Valleys (Fig. 2and 3).

The size and layout of the sample plots in the field tookinto account the spatial resolution of the sensors (Curran andWilliamson, 1986), location errors on the images and theheterogeneity present in each landscape. A sample frameformed by four MSS pixels, 25 600 m2, with two plots, 80 mapart, each the size of four SPOT XS pixels, 1 600 m\ nestedinside, was used. Herbaceous and tree/shrub canopy covermeasurements in the sample plots were made along threebelt transects of 3 m x 160 m and 3 m x 40 m for MSS andSPOT XS, respectively (Muller-Dombois and Ellenberg,1974). Average vegetation and soil cover over each plotwere derived from the transect measurements. Soil colourwas determined using a Soil Munsell Colour Chart. A totalof 16 plots for MSS and 32 for SPOT XS were established(Fig. 3). Variability in measured cover over the landscapeswas assessed using one way analysis of variance (ANOVA)(Kleinbaum et al, 1988).

Sample plots were located on the images using linearcoordinate transformation equations with root mean squareerror of 0.49 pixels in both axis for MSS, and 0.95 and 1.11pixels in the X and Y axis respectively, for the SPOT XSimage. To take into account possible location errors, a windowof 3 x 3 pixels was taken over each plot. Average digitalvalues from each window were converted to radiance andreflectance values (Markham and Barker, 1987; Price, 1987and 1988). Two normalised difference vegetation indiceswere produced for both systems (Tucker, 1979): NDVIdetermined from,

NIR - RED / (NIR + RED)

and NDGR, derived from the difference between GREENand RED divided by their sum,

NDGR = GREEN - RED / (GREEN + RED).

The MSS NDVI derived from the RED and NIR band 3 isreferred to as ND to differentiate it from the MSS NDVIcalculated using NIR band 4.

The reflectance of the different landscapes over thewavebands was observed using mean reflectance values.Simple linear regression was performed at three levels: (i)for each separate landscape using SPOT XS data wheresufficient measurements were available, (ii) with data pointsacross the six landscape types combined, and (iii) usingaverages of cover measurements of each landscape type(Kleinbaum et al., 1988). The analysis was carried out atthree levels to observe the strength of the relationship betweenvegetation cover and MSS and SPOT XS data at differentscales and degrees of variability in vegetation and soil. Astepwise regression analysis was run using soil colour rankedin terms of brightness, living and dead, tree/shrub and herbcover, to establish factors explaining variability in the MSS

and SPOT XS bands and their normalised indices acrosslandscape types.

Results

Field Vegetation MeasurementsTree/shrub cover accounted for about half of the total

average cover of most landscape types (Table 1 ) and onaverage 12 species were recorded at the SPOT XS level. Themost frequently occurring species were; Grewia flava foundin all landscapes, Acacia tortilis on Plains, Acacia melliferaon Floodplains, Terminalia sericea on the Kalahari Sandveld,and Combretum apiculatum on Sandstone Hills, and RockyHills (Fig. 2). Vegetation cover from sample plots rangedfrom 19% in Dry/Fossil Valleys to 85% on Rocky Hills, withgreater variance at the SPOT XS level (Table 1). Acomparison of variation of cover within each landscape withthat of cover measurements across all landscapes, using oneway ANOVA, showed that variability in each landscape wassignificantly different (P>0.01 ) from all landscapes combined(Table 2).

Landscape types and spectral dataFigure 4 shows the reflectance curves of the landscape

types. Dry/Fossil Valleys with sparse cover had the highestreflectance in all bands. Landscapes with higher vegetationcover had low reflectance. The MSS plot show betterseparation of landscapes than SPOT XS (Fig. 4b).

The regression models for individual landscapes usingSPOT XS show differences over the different bands andindices (Table 3). On the Kalahari Sandveld, relatively lowwoody cover with dead plant material on a bright soilbackground, reduced sensitivity in all cases. Dry/FossilValleys had positive strong relationships with all SPOT XSbands due to high reflectance from light eroded soils. Shadoweffect resulted in negative relationships in all bands forFloodplains with dense cover on dark soil and SandstoneHills with moderately dense cover on sandy soil. RockyHills had a strong positive relationship, r2 = 0.91 (P>0.05)with the NIR band and weak relationships in all other cases.Plains had better relationships with indices.

Vegetation cover across landscapes show negativerelationships with all wavebands for both systems (Table 4).The regression models of MSS bands had strongerrelationships, lower variance and standard error of Y estimate(SE) compared to SPOT XS bands. The RED band had thestrongest relationship with cover; r2 of 0.75 and 0.71 (P>0.001) for MSS and SPOT XS, respectively. The SPOT XS NIRband had the weakest relationship, r2 = 0.14 (P>0.05).Normalised indices had relatively weaker relationships withvegetation compared to the RED band, with better results inNDGR index, r2 of 0.67 (P>0.001) in both systems.

The regression models between landscapes, based onaverage values of each landscape, show stronger relationshipsin all cases with lower variance (Table 5). Negativerelationships were also recorded with all bands. Among thesingle bands, the SPOT XS RED band had the highest r2of

45

Dow

nloa

ded

by [

Uni

vers

ity o

f N

ew M

exic

o] a

t 09:

34 2

2 N

ovem

ber

2014

2(b)

2(e)

Figure 2 Examples of vegetation and soil types on the landscapes found in the study area. (a). Rocky Hills and footslopes at 1200 m above sea level:broadleaved Combretum apiculatum on rocky soil; (b). Floodplains: dense mixed woodland on gray soil; (c). Low Sandstone Hills: Combretumapiculatum woodland on bright sandy soil; (d) Kalahari Sandveld: broadleaved woodland with dead trees on bright sandy soil; (e). Plains: woodyshrubland on red soil and (f). Dry/Fossil Valleys: Acacia tortilis shrubs on bright soil

0.85 (P>0.01) and in MSS it was the GREEN band with r2 of0.82 (P>0.02). For the indices, MSS ND had the best results,r2of 0.93 (P>0.01) while on the SPOT XS, the NDGR hadthe strongest relationship.

The stepwise regression analysis using data points acrosslandscapes showed that living tree/shrub cover was the mostinfluential factor explaining variability in all cases, in the

first step (Table 6). The MSS ND had the strongest influencewith r2 of 53.6% (P< 0.001). In the second step all MSSbands were affected by dead vegetation, with greater effecton NIR band 3 where dead grass increased r2 from 45.8% to75.97% (P<0.001). Among the indices, living grass was themost influential.

46

Dow

nloa

ded

by [

Uni

vers

ity o

f N

ew M

exic

o] a

t 09:

34 2

2 N

ovem

ber

2014

SPOTXS -MARCH 1988

H - Rocky Hills C - Roodplalns S - Sandstone Hills

P-Plains K-Kalahari Sandveld D - Dry/Fossil ValleysH - Reid sample sites

Figure 3 SPOT multispectral false colour image of March 1988 showing landscape types and ground sample sites

Discussion

Vegetation cover, dead and living tree/shrub, andherbaceous layer had a greater effect on the MS S and SPOTXS bands than soil, partly because images of an exceptionallywet year, 500 mm annual rainfall, were used (BotswanaMeteorological Services, 1988). The inverse relationshipexhibited between vegetation cover and wavebands was dueto the effect of shadow (Fig. 4; Table 4 and 5) (Colwell,1974). Shadow resulted from the predominantly woodyvegetation with dead plant material (Table 1 and Fig. 2).MSS showed slightly greater effect of shadow which ispartly related to the Landsat overpass at lower sun angles,0937 local time, in contrast to SPOT XS overpass at 1037

(Graetz and Gentle, 1982).The differences in sensitivity shown by individual

landscapes on the SPOT XS data (Table 3) can be explainedby the difference in variability of vegetation cover withineach landscape as indicated by the results of ANO VA (Table2). In addition to percent cover, variations in vegetationtype, amount of dead plant material and soil type in eachlandscape were also influential. For instance, the negativerelationship between wavebands and vegetation cover, alsonoted by Vujakovic (1985) and Ringrose et al., (1989), ispartly landscape dependent, and there are variations overwavebands.

Across landscapes heterogeneity in vegetation and soilwas pronounced resulting in generally weak relationships.

47

Dow

nloa

ded

by [

Uni

vers

ity o

f N

ew M

exic

o] a

t 09:

34 2

2 N

ovem

ber

2014

Table 1 Average percent vegetation cover from field measurements: MSS and SPOT XS sample plots

CODE

H

C

S

K

P

D

MSSDeadCover

15

13

13.8

16.5

7.6

6

GrassCover

14

21.9

17.3

13

15.6

6.5

TreeCover

48

33

29.5

16

14.8

6.5

TotalCover

77

68

61

46

38

19

SD

%

25.119.2

16

11.7

17.7

11.4

SPOTDeadCover

9.714.1

12.2

16

6.5

3.7

XSGrassCover

6.2

18

13.5

12.7

94.5

TreeCover

68.7

34

29.8

24

26.5

13

TotalCover

85

66

56

53

42

21

SD

%

15.5

23.3

24.6

17

30.4

10.4

H: Rocky Hills; C: Floodplains; S: Sandstone Hills; K: Kalahari Sandveld; P: Plains; D: Dry Fossil/Valleys. Deadcover-dead tree and grass canopy; Tree and grass cover-living canopy.* SD: Standard deviation.

Table 2 Analysis of variance in vegetation cover between landscapesand within landscapes using one way ANOVA statistics

NIRGreen RedSPOT XS Wavebands

-A- Rocky Hills -*• Floodplains-•- Sandstone Hills -B- Kalahari Sandveld-e- Plains -H- Dry/Fossil Valleys

Green

-¿r Rocky Hills

••-Sandstone Hills

-e-Plains

Red NIR band 3 NIR band 4MSS Wavebands

-*- Floodplains

-B- Kalahari Sandveld

-X- Dry Fossil/Valleys

Averagecover (%)

SPOT XS

54.11

MSS

52.38

Variance

485.56

453.96

Mean Squarebetweengroups

1975.54

(Df=5)

1027.06

(Df=5)

Mean Squarewithin groups

218.07

(Df=26)

212.81

(Df=10)

F-ratio

9.06

4.826

Figure 4 SPOT XS and Landsat MSS 1988 wet season reflectancecurves of landscape types using mean values.

p>0.01

The NIR bands had the highest variability; 445.44 in SPOTXS, and 233.46 and 234 in MSS NIR bands 3 and 4,respectively (Table 4). Vegetation cover from field plotswas variable; at the SPOT XS level, an average of 54.11%with a variance of 485.86 was recorded for all plots, and52.38% with a variance of 453.96 at the MSS level (Table 1and 2). The moderate relationships with indices indicatedthe limitations of vegetation indices in minimising variationsacross landscape types, a factor noted by Huete (1988),Pickup et al, ( 1993) and Nicholson and Farrar, (1994).

As in Vujakovic (1985), strong relationships wereobserved with the RED band across landscapes. Ringroseand Matheson (1987) also noted the variable reflectance inthe NIR band over the rangelands of Botswana, and concludedthat the RED band was the most suitable for vegetationcover studies. Despite the high coefficient of determinationin the SPOT XS RED band established in this study (Table 4and 5), the reflectance curves in Figure 4 show low separationof landscapes in terms of levels of cover in this band. TheRED band has been noted by Pickup et al, (1993) tounderestimate cover in woodland areas and overestimatecover in areas of bright soil background in central Australia.

The stronger relationships recorded between landscapes

48

Dow

nloa

ded

by [

Uni

vers

ity o

f N

ew M

exic

o] a

t 09:

34 2

2 N

ovem

ber

2014

Table 3 Linear regression analysis at individual landscape: Vegetationcover and SPOT XS bands and indices

Table 4 Linear regression analysis across landscape types: Relationshipbetween percent vegetation cover with SPOT XS and LandsatMSS bands and indices

Landscape R2 Intercept Slope P-value

error Variable R2 Standarderror

Intercept Slope Variance

HC

S

KPDREDHC

S

KP

D

NIRH

C

SK

P

DNDVI

H

cS

KPD

NDGRH

C

SK

P

D

H-Rocky

0.05

0.93

0.14

0.370.001

0.88

0.01

0.91

0.79

0.37

0.55

0.67

0.91

0.02

0.450.13

0.15

0.93

0.17

0.20.63

0.340.870.95

0.020.69

0.61

0.35

0.69

0.99

Hills

4.8

3.94

22.47

4.7

15.88 .

3.32

4.914.39

10.92

4.71

10.61

5.46

1.49

- 14.34

17.895.49

14.68

2.45

4.51

12.9514.66

37.75.591.93

4.347.98

15.18

4.75

8.87

0.71

74.38

288.09

133.45

75.37

54.82

-173.43

87.32187.94

160.78

70.51

105.21

-255.57

77.06

82.65

197.53-0.03

26.94

-108.16

52.6-8.22

-134.29

37.7-47.73

-171.62

36.08

-385.14

-156.59

-115.49

-90.69

-1227.7

1.75-36.74

-10.9

-2.78

-1.8919.81

-0.52

-26.39

-14.96

-2.25

-8.41

24.24

0.51

-0.82

-6.732.77

0.77

5.09

55

119.21378

36.22181.37510.37

44.36

397.15

209.36

165.83

141.28

1350.00

C-Floodplains S-Sandstone

0.75

0.01

0.36

0.36

0.94

0.1

0.89

0.01

0.01

0.36

0.1

0.2

0.05

0.76

0.10.61

0.44

0.05

0.57

0.360.05

0.380.010.05

0.49

0.05

0.05

0.37

0.050.01

Hills

SPOT XSXS1 0.43 17.3 130.9 -10.8 297.5

XS2 0.71 12.2 ' 112.2 -8.2 148.4

XS3 0.14* 21.1 89.3 -1.7 445.4

NDVI 0.56 15.1 -35.1 175.4 227.1NDGR 0.67 13.1 -145.4 195.1 171.4

Df=31

MSSMSS 0.6 14.3 155.6 -8.2 204.4

MSS2 0.75 11.5 120.5 -6.4 131.9

MSS3 0.55 15.3 151.5 -5.3 233.5

MSS4 0.55 15.3 175 -5.1 234.3

NDVI 0.59 14.6 -138.4 228.4 214.5

ND 0.62 13.9 -8.61 208.9 195.5

NDGR 0.67 13.11 -250.1 275.9 171.9

Df=15

P>0.001 *:P>0.05 XS1-XS3: SPOT XS wavebands

MSS1-MSS4: Landsat MSS wavebands.

m x 20 m pixel size for SPOT XS, a factor noted in Markhamand Townshend (1981). Other factors such as imaging timeand date, wavelengths of the bands, sensor view angle andatmospheric effect also had a role on the two sensors (Crippen,1988). The SPOT XS scene was taken at 19° left while MSSis acquired at nadir view. Epiphano and Huete (1995) havedemonstrated the effect of sensor geometry on the NDVI.The MSS NIR band 3 wavelength, 700-800 nanometres,corresponds to the peak of the reflectance of NIR by greenvegetation which explains, in part, the better resultsdemonstrated in this band and the corresponding NDVI incontrast to band 4 and SPOT XS band 3 and their respectiveNDVls (Tucker, 1979).

ConclusionK-Kalahari Sandveld P-Plains D-Dry/Fossil Valleys

resulted from lower variability compared to across landscapetypes (Table 4 and 5). Variance in vegetation cover betweenlandscapes was reduced from 453 to 379.39 for MSS andfrom 485.86 to 387.22 in SPOT XS data. These results showthat relationships improve at a general level betweenlandscapes, which is comparable to studies based on coarsespatial resolution NOAA data performed at a regional scale(Prince and Tucker, 1986).

In general Landsat MSS produced better correlationsthan SPOT XS in this study which is partly due to the coarsespatial resolution of MSS data, 80 m x 80 m compared to 20

The study has shown that stratification of landscape atdifferent scales can yield different degrees of relationshipsbetween vegetation cover and spectral data. Variation invegetation cover within each of the six landscape types inthe study area was found to be significantly different fromvariation across landscape types combined. Individuallandscapes exhibited different relationships with SPOT XSbands and indices. Overall the SPOT XS RED and theNDGR performed better with moderate to strong relationshipsin four (Floodplains, Sandstone Hills, Plains and Dry/FossilValleys) out of the six landscape types. Relationships atindividual landscape level need further investigation withadequate ground data. High variability across landscapes,

49

Dow

nloa

ded

by [

Uni

vers

ity o

f N

ew M

exic

o] a

t 09:

34 2

2 N

ovem

ber

2014

Table 5 Linear regression analysis between landscape types: Percentvegetation cover and SPOT XS and Landsat MSS bands andindices using mean values

Variable

SPOT XS

XS1XS2XS3NDVI

NDGR

Df=4

MSS

MSS1

MSS2

MSS3

MSS4

NDVI

NDNDGR

Df=4

R2

0.69

0.85

0.82

0.66

0.72

0.82

0.78

0.81

0.79

0.86

0.93

0.80

Standard

error

13.43

9.35

10.11

14.16

12.72

1011.2

10.2

10.8

8.96.4

10.6

Intercept

142.1

112.4

117.5

-39.7

-170.7

156.7

111.1

161.7

190.2

-56.8

-28.4

-332.6

Slope

-12.2

-8.1

-5.8

187.7

220.3

-8-5.7

-5.9

-5.8

277.6

277.7

347.4

Variance

180.40

87.40

102.30

200.6

161.9

100.03

125.22

104.18

115.72

80.91

41.54

113.36

P-value

0.05

0.01

0.02

0.10.05

0.02

0.02

0.02

0.02

0.01

0.01

0.02

resulted in weak to moderate relationships with best resultsin the RED band, r2 = 0.75 and 0.71 ( P>0.001) for MSS andSPOT XS, respectively. Most studies in Botswana havebeen carried out across landscape types. Stronger relationshipswere obtained using average vegetation cover betweenlandscapes, indicating that better estimates of cover occur atgeneralised scales, producing relative than absolute covervalues. The RED band maintained the strongest relationshipof r2 = 0.85 (P>0.01) for SPOT XS, while the NDVI indexbased on the RED and NIR band 3, had the best results withi2= 0.93 (P>0.01) for landsat MSS.

Acknowledgements

I am indebted to Giles D'Souza for his assistance in theearly stages of the analysis of the data. Geoff Pickup, SharonNicholson and Mogodisheng Sekhwela made critical reviewsof the earlier drafts. Field data and images were funded bythe Swedish Agency for Research Cooperation (SAREC)and data processing and analysis benefitted from theAustralian Centre for International Agricultural Research(ACIAR) grant.

XS1-XS3: SPOTXS wavebands; MSS1-MSS4: Landsat MSS wavebands

Table 6 Results of Stepwise regression analysis across landscape types

Predictor:(Pred.)

VariableBands

Living tree/shrub canopy-Living herb layer-

Dead tree/shrub canopy-

DeadSoili

Step

Pred.

Dependent variable:

XS1XS2

XS3MSS1

MSS2

MSS3

MSS4

LT

LT

LT

LT

LT

LT

LT

Dependent variable:

XS-NDVI

XS-NDGRMSS-NDVI

MSS-ND

MSS-NDGR

LT

LT

LTLT

LT

1 herb layer-

:olour-

1

R2

wavebands.

28.948

31.942.4

49.6

45.8

40.4

RMSE

1.16

1.67

2.311.64

2.17

2.28

2.56

vegetation indices.

36.4

44.4

39.6

53.6

46.4

0.08

0.07

0.06

0.06

0.05

LTLH

DT

DH

SC

Step 2

Pred.

SCLH

DT

DH

DH

DH

DH

LH

LH

LH

LH

LH

R2

46.2

74.948.8

58.4

73.8

75,9

66.9

79.174.4

70.9

69.1*

62.7*

RMSE

1.03

1.192.04

1.45

1.62

1.6

1.97

0.45

0.050.04

0.050.04

Step 3

Pred.

DT

DH

SCLH

SC

SC

DT

SC

R2

81.558.9

75.9*

83.6*

82.2

83.4*

77.6

73.1*

RMSE

1.03

1.86

1.141.34

1.4

0.04

0.05

0.04

P<0.001 *:P<0.05

XS1-XS3: SPOT XS wavebands; MSS1-MSS4: Landsat MSS wavebands.

RMSE-root mean square error

50

Dow

nloa

ded

by [

Uni

vers

ity o

f N

ew M

exic

o] a

t 09:

34 2

2 N

ovem

ber

2014

References

Bhalotra, Y. P. R. 1985. Rainfall maps of Botswana. Department ofMeteorological Services, Ministry of Works and Communication,Botswana Government.

Botswana Meteorological Services 1988. Gaborone, Botswana.

Choudhury, B. J. and Tucker, C. J. 1987. Satellite observed season andinter-annual variations of vegetation over the Kalahari, The GreatVictoria Desert and the Great Sandy Desert, 1979-1984. Remotesens. Environ. 23:233-241.

Colwell, J. E. 1974. Vegetation canopy reflectance. Remote Sens.Environ. 3:175-183.

Curran, P. J. and Williamson, H. D. 1986. The sample size required tocharacterise ground and remotely sensed data to known levels oferror. Rem. Sens. Environ. 20:31 -41.

Epiphanio, J. C. N. and Huete A. R. 1985. Dependence of NDVI andS AVI on sun/sensor geometry and its effect on FAPAR relationshipsin alfalfa. Remote Sens. Environ. 51 :351 -360.

Government of Botswana 1990. Land systems map of the Republic ofBotswana, 1:2000 000. Soil Mapping and Advisory Services Project,AG-.DP/BOT/011. Ministry of Agriculture. Gaborone.

Graetz, R. D. and Gentle, M. R. 1982. The relationship betweenreflectance in Landsat wavebands and the composition of anAustralian semi-arid shrub rangeland. Photogramm. Eng. and Rem.Sens. 48:1721-1730.

G.R.M. International 1987. Re-establishment of a national rangeresource assessment and monitoring programm. Final Report.Ministry of Agriculture, Republic of Botswana, Gaborone. 224p.

Huete, A. R. 1988. A soil adjusted vegetation index (SAVI). Remotesens, of Environ. 25:295-309.

Jacobberger, P. A., and Hooper, D. M. 1991. Geomorphology andreflectance patterns of vegetation-covered dunes at the Tsodilohills, north-west Botswana. Int. J. remote sensing 12(11):2321 -2342.

Kgathi, D. L., Sekhwela, M.B.M., Tietema, T., Mpotokwane, M. A.1994. Biomass in Botswana, in D. O. Hall and Y. S. Mao (eds),Biomass energy and coal in Africa. Zed Books Ltd in associationwith African Energy Policy Research Network, London, Gaborone.67pp.

Kleinbaum, D. G., Küpper, L. L. and Müller, K. E.1988. Appliedregression analysis and other multivariable methods. Second edition.Duxbury Press, Belmont. 717p.

Markham, B. L. and Barker, I. L. 1987. Radiometrie properties of USProcessed Landsat MSS data, Remote Sens. Environ. 22:39-7'1.

Nellis, M. D. and Bussing C. E. 1989. Desertification and sustainabledevelopment issues in Botswana. In J. W. Frazler, B. J. Epstein andF. A. Schoolmaster ill, Papers and Proceedings of AppliedGeography Conferences, Vol.12. 99-106.

Nicholson, S. E. and Farrar, T. J. 1994. The influence of soil type onthe relationships between NDVI, rainfall, and soil moisture insemiarid Botswana. I. NDVI response to rainfall. Remote sensingEnviron. 50:107-120.

Pickup, G., Chewings, V. H. and Nelson, D. J. 1993. Estimatingchanges in vegetation cover over time in arid rangelands usingLandsat MSS data. Remote Sens. Environ. 43:243-263.

Price, J. C. 1987. Calibration of satellite radiometers and the comparisonof vegetation indices. Rem. Sens. Environ. 24:15-27.

Price, J. C. 1988. An update on visible and near infrared calibration ofsatellite instruments. Rem. Sens. Environ. 24:419-422.

Prince, S. D. and Tucker, C. J. 1986. Satellite remote sensing ofrangelands in Botswana. II. NOAA AVHRR and herbaceousvegetation. Int. J. remote sensing 7:11:1555-1570.

Richardson, A. J. and Everitt, J. H. 1992. Using Spectral VegetationIndices to Estimate Rangelands Productivity. Geocarto International7:1 :63-69.

Ringrose, S. and Matheson, W. 1987. Spectral assessment of indicatorsof range degradation in the Botswana hard veld environment. Rem.Sens. Environ. 23:379-396.

Ringrose, S., Matheson, W., Mogotsi, B., Tempest, F. 1989. Thedarkening effect in drought affected savanna woodlandsenvironments relative to soil reflectance in Landsat and SPOTwavebands. Remote sens, environ. 30:1 -19.

Sekhwela, M. B. M. 1995. Environmental Impact of woody biomassutilisation in Botswana. The case of fuelwood. Final report. AfricanEnergy Policy Research Network (AFREPREN). Biomass researchGroup. National Institute of Developmental Research andDocumentation (NIR). University of Botswana, Gaborone. 77p.

Timberlake, J. 1980. Vegetation map of south east Botswana. Divisionof Land Utilisation, Department of Agricultural Field Services.Ministry of Agriculture. Gaborone, 83p.

Tucker, C. J. 1979. Red and photographic infrared linear combinationfor monitoring vegetation. Remote Sens. Environ. 8: 127- 150

Tucker, C. J., Vanpraet, C. L., Sharman, M. J., and Van Inttersum, G.1985. Satellite remote sensing of total herbaceous biomassproduction in the Senegalese Sahel: 1980-1984. Remote sensingEnviron!!:233-249.

Tucker, C. J., Justice, C. O. and Prince, S. D. 1986. Monitoring thegrasslands of the Sahel 1984-1985. Int.]. remote sensingl:U:\51l-1581.

Tueller, P. T. 1987. Remote sensing applications in arid environment.Rem. Sens. Environ. 23:143-154.

Van Heist, M. and Kooiman, A. 1992. Modelling fuelwood availabilitywith GIS: A case study in Botswana. ITC Journal 1992-3.

Vujakovic, P. 1985. Evaluation of Landsat MSS data for terrain analysisand rangeland management in the lower Okavango Delta region,Ngamiland, Botswana. PhD Thesis (unpublished). University ofLondon, 322 pp.

51

Dow

nloa

ded

by [

Uni

vers

ity o

f N

ew M

exic

o] a

t 09:

34 2

2 N

ovem

ber

2014