Estimating vegetative productivity gradients around watering points

Geografisk Tidsskrift, Danish Journal of Geography 103(1) 1

DANISH

JOUR

NAL OF GEOGRA

PHY

2003

Since the drought of the early 1970s and a prolonged pe-riod of below-average rainfall in the Sahel (Hulme, 2001),is has often been discussed whether or not the region isundergoing 'desertification'/'land degradation' (e.g.Thomas (1993), Matheson & Ringrose (1994)). Ferlo innorthern Senegal is part of the Sahelian region. In Ferloland degradation is claimed to start around wateringpoints, due to high grazing pressure (Prins, 1997).

This paper hypothesizes that high grazing pressurearound watering points causes land degradation. To testthis hypothesis, coarse resolution NOAA AVHRR satel-lite images are examined for systematic variation in theintegrated vegetation index, iNDVI around 44 boreholesin the Ferlo over 3 years. Using coarse resolution satelliteimages facilitates estimation of iNDVI over large areas,including many boreholes and over several growing sea-sons. By analysing many gradients, gradients can be sta-tistically linked to possible causal factors. Also, errors as-sociated with extrapolating findings spatially and tempo-rally from one borehole in one year are avoided. Samplinga large number of boreholes additionally led to a roughclassification system for identifying boreholes, correlated

by the type of gradient found, suitable for more detailedstudies of possible causal mechanisms.

Lange's (1969) study of grazing gradients (or pios-phere patterns) is recognized as the first of its kind, al-though it was suggested already in 1932 that wateringpoints used by grazing animals could be used to determinevegetation growth (Osborn et al. (1932) in Lange (1969)).However, the use of grazing gradients to assess how graz-ing influences vegetation did not become common beforethe late 1970s (e.g. Graetz & Ludwig, 1978; Fatchen &Lange, 1979). Since then, the grazing gradient method hasbeen developed and applied mainly in Australia (e.g. An-drew & Lange, 1986b; Pearson et al., 1990). From the late1980s, the method has often been combined with remotelysensed data in erosion forecasting and assessment of landdegradation at large scale (e.g. Andrew, 1988; Hanan etal., 1991; Chewings et al., 1992; Bastin et al., 1993a;Bastin et al., 1993b ). Satellite imagery has increasinglybeen used in combination with models of animal move-ment to survey grazing impacts (e.g. Pickup & Chewings,1988; Pickup, 1994), and the grazing gradient method hasremained a popular tool for rangeland assessment (e.g.

Abstract This paper tests the hypothesis that high dry season grazing pres-

sures around watering points in rangelands in northern Senegal has

a negative effect on vegetative productivity. Time series of NOAA

AVHRR satellite images for three rainy seasons are used to estimate

integrated NDVI (iNDVI) at a resolution of 1 km2, and assess the

dependence of iNDVI on distance to the nearest dry season water

source. The observed 'iNDVI gradients' around watering points dif-

fer greatly between watering points and through time, leading to a

rejection of the hypothesis. Possible explanations are discussed,

and it is argued that iNDVI gradients are not well-suited as indica-

tors of land degradation.

KeywordsLand degradation, grazing gradients, NOAA AVHRR NDVI, remote

sensing, Senegal.

Morten Lind, Kjeld Rasmussen & Hanne Adriansen: Institute of Ge-

ography, University of Copenhagen, Øster Voldgade 10, DK-1350

Copenhagen K, Denmark.

E-mail:[email protected] (Morten Lind), [email protected] (Kjeld Ras-

mussen) & [email protected] (Hanne Adriansen)

Alioune Ka: Centre de Suivi Ecologique, Rue Leon Gontran Damas,

Dakar - Fann, Senegal.

E-mail: [email protected]

Geografisk Tidsskrift

Danish Journal of Geography 103(1): 1-15, 2003

Estimating vegetative productivity gradients aroundwatering points in the rangelands of NorthernSenegal based on NOAA AVHRR data

Morten Lind, Kjeld Rasmussen, Hanne Adriansen & Alioune Ka

2 Geografisk Tidsskrift, Danish Journal of Geography 103(1)

Bastin et al., 1998). Most of these studies using satellite re-mote sensing have been carried out in Australian range-lands. Hanan et al. (1991), however, studied the range-lands of northern Senegal, the subject of the present paper.

The use of grazing gradients and satellite imagery fordegradation assessment has been refined by Pickup &Chewings (1994). They classify grazing gradients as 'nor-mal', 'inverse' and 'composite'. Normal gradients show anincrease in vegetation cover with increasing distancefrom water. Inverse gradients involve decreasing vegeta-tion cover with increasing distance from water. Finally, acomposite gradient combines the two. Pickup & Chew-ings (1994) further distinguish temporary from perma-nent gradients, where permanent refers to gradients thatpersist after a good rain. Permanent gradients are consid-ered indicators of degradation.

A grazing gradient (or a piosphere pattern) can be de-fined as:

'...patterns reflecting the concentricity of stocking pres-sure around water.' (Andrew & Lange, 1986a, p.395),

while for the purpose of studying land degradation Pickupand Chewings (1994) defined a grazing gradient as:

'... spatial patterns in soil or vegetation characteristics re-sulting from grazing activities and which are symptomaticof land degradation.' (Pickup & Chewings, 1994, p.598).

This, however, requires a definition of land degradation.We will apply the definition put forth by Williams &Balling (1996):

'Reduction of biological productivity of dryland ecosys-tems, including rangeland pastures and rainfed and irri-gated croplands, as a result of an acceleration of certainnatural processes' (Williams & Balling, 1996, p.17).

This definition may be criticized for not capturing otherprocesses often considered to characterise land degrada-tion in the Sahel environment, such as changes in speciescomposition, e.g. reduction of woody cover or invasion ofundesirable species, and loss of biodiversity. Williams &Balling's definition has, however, the advantage of beinguniversal and applicable at large scale.

If this definition is accepted, the measurement of landdegradation will be based on indicators of biological pro-ductivity. Satellite remote sensing, and particularly vege-tation estimates from NOAA AVHRR data, can be used to

estimate 'net primary productivity' (NPP) through thetime-integral (iNDVI) of the 'normalized difference vege-tation index' (NDVI). The relationship has been furtherdeveloped, calibrated and validated for this particular re-gion by Rasmussen (1998). Other methods for assessingNPP using NOAAAVHRR data have been developed, seePrince & Goward (1995). However, here the spatial vari-ation of vegetative productivity within a relatively smalland homogeneous region is the main focus, and thusiNDVI will suffice as an indicator of cumulative vegeta-tive growth. Using standard GIS routines, both directionaland average gradients of iNDVI around each wateringpoint can be determined.

The study area: Climate, soils, vegetation, and pro-duction systems

The present study deals with the rangelands of the Ferlo,northern Senegal, see Figure 1.

The climate of the Ferlo is characterized by a mono-modal rainfall pattern with a short, relatively well-definedrainy season and a long dry season. The largest precipita-tion occurs in July, August and September. Hot, dry peri-ods occur both before and after the rainy season in April-June and October-November. December-March is thecool, dry period. The north-south rainfall gradient is ap-proximately 1.1 mm km-1, and mean annual rainfall forthe period 1986-1996 was approximately 200 mm in thenorth and 400 mm in the south of the study area. In Figure2, the rainfall for a number of stations in the Ferlo is givenfor 1990, 1991 and 1992.

The Ferlo can be divided into two main zones, definedby the dominating soils: To the west, sandy soils prevailand the topography is dominated by ancient longitudinaldunes, created in drier periods associated with the iceages. To the east, soils are finer, harder and reddish, con-taining iron compounds. The transition between the twozones is gradual. The Ferlo is dissected by a dry, fossil val-ley system, most pronounced in the east. The distributionof vegetation is controlled by these soil and geomorpho-logical characteristics: To the north-west, the vegetationcan be described as bushed grassland, with a woody coverseldom exceeding 5 % and more typically around 2 %. Tothe south the woody cover increases to 5 - 20 %. To theeast the vegetation is dominated by bushes and low trees,and woody cover is in the order of 10 - 30 %. Woody coveris greatest in the fossil valleys where water is more read-


ily available. The herbaceous layer is dominated by annu-als, particularly in the north, and includes species such asSchoenfeldia gracilis, Cenchrus biflorus and Zorniaglochidiata. The relative abundance of these specieschanges each year and from place to place, varying withrainfall - especially in the beginning of the rainy season -and with grazing pressure (Hiernaux, 2000).

Before boreholes were dug in the Ferlo, the area wasused as rainy season pasture only, since surface water wasnot available in the dry season. In the 1950s, the Frenchcolonial administration began installing boreholesequipped with motor pumps in Ferlo, which meant that thearea could be used on a permanent basis (Ba, 1986). Thepossibility of staying in the area during the dry seasonmeant that local peoples became semi-sedentary, settlingaround the boreholes. A new mobility arose called 'micro-nomadism', where livestock are moved daily, though onlywithin the area around a borehole, a 'pastoral unit' (Bar-rell, 1982).

Today, mobility patterns vary with availability of re-sources, the animals herded, and the importance of agri-culture in a household's production strategy. Touré (1990)divides pastoral mobility in Ferlo into three categories:Daily, seasonal and occasional migrations. Daily migra-tion within the pastoral unit, the above-mentioned micro-nomadism, is most widespread among pastoralists. Sea-sonal migration implies returning regularly to the same ar-eas. Areas for this type of migration are defined at bothmedium and large scale. Finally, occasional migration isunpredictable, often coinciding with pump-breakdown,destruction of pasture by bush fire and outbreak of dis-ease. Occasional migrations can be either medium or largescale depending of the instigating event.

Herds consist of cattle, sheep and goats, though there isa tendency to specialise in sheep production among certaingroups of pastoralists. These specialised sheep herdersmaintain a high mobility, migrating as far as Gambia insearch of appropriate pastures (Adriansen, 2002).

Figure 1: The position of Ferlo in Senegal. Thiessen polygons are shown together with positions of boreholes and rainfall stations.


Water is supplied by temporary ponds, boreholes andantennas (secondary water outlets piped from boreholes).Ponds can be used free of charge, while a fee per head oflivestock is common for boreholes and antennas. Bore-holes are usually closed until the temporary ponds start todry out (Alissoutin, 1997). Therefore, in the rainy seasonand early dry season, when water is widely available,herds tend to be much more spread out than late in the dryseason, when livestock depends entirely on well-water.Another factor influencing the spatial distribution of live-stock and grazing pressures is the location of camps. Per-manent camps are often located near temporary ponds,while camps of migrating pastoralists are often found onthe best pastures.

The spatio-temporal distribution of grazing pressure isof central importance for the present study. If it is re-vealed, as suggested by Hiernaux (2000), that dry seasongrazing pressure has limited or no importance for vegeta-tive productivity, while rainy season grazing pressure ex-erts significant control, the hypothesis tested in this paperwill be compromised. According to Hiernaux vegetativeproductivity should be lowest in the vicinity of temporaryponds and pastoralist camps, which are often located 5 to10 km from the wells. However, the effects of grazingpressures at different times of the year on long-term pro-ductivity trends are complex and location-specific anddepend on vegetative species composition, herd composi-tion and the timing of grazing.

Agricultural encroachment is taking place in thesouthern and western parts of the Ferlo, and conflicts over

access to land are therefore commonplace. This increasesthe pressure on the remaining rangelands of the Ferlo(Freudenberger & Freudenberger, 1993).

Data and methods

NOAA AVHRR satellite images

The choice of data sourceSeveral sources of satellite data are available for assessingvegetation state and productivity. We have chosen NOAAAVHRR data, which are widely available at low cost. Inthe Australian literature on grazing gradients, Landsatdata are widely applied, and Bastin et al. (1995) argue thatLandsat MSS is far superior to NOAA AVHRR for suchapplications. Their conclusion has the following basis:

• The low spatial resolution of NOAA AVHRR compli-cates the process of relating vegetation state to featuressuch as fences and well-defined landscape elements.

• NOAA's low radiometric resolution (one visible andone near-infrared band only) does not allow use of thespectral indices, a widely applied technique for Land-sat data.

Nonetheless, in the present context we will argue that useof NOAA AVHRR data is the only realistic option:

• Since seasonal vegetative productivity is the focus,rather than the vegetative state for one or a few dates,only data-sources allowing an estimation of time-inte-grated vegetative productivity are relevant. Due to theshort rainy season where the probability of cloud coveris high, satellite/sensor systems with a low repetivity,such as Landsat MSS and TM, are unsuitable. Veryfew, if any, cloud-free scenes are likely to be availablefor any one rainy season in a given location. In con-trast, at least 20 NOAA AVHRR scenes are generallyused for assessing cumulative vegetative growth.

• Methods for estimating vegetative productivity fromtime series of NOAAAVHRR data are well-developedand have been validated for the study region (Ras-mussen, 1998).

• Spatial variability in vegetative productivity at sub-km2 scale does certainly exist, but the mobile andhighly flexible grazing system makes this variabilityless important to a land degradation perspective.

Processing NOAA AVHRR data

0

100

200

300

400

500

mm

/yea

r

PodorGuede

FanayeDagana

NdioumAere Lao

SaldeThilogne

Keur MomYang Yang

MatamOgo

KanelLinguere

DahraRanerou

BarkedjiSemme

BakelThiel

graph left to right = north to south

1990 1991 1992

Figure 2: Rainfall in Ferlo 1990, 91 and 92. Rainfall for the sta-tions shown in Figure 1.


AVHRR HRPT data for 1990, 1991 and 1992 are avail-able from the ground receiving station at Centre de SuiviEcologique, Dakar, Senegal. Data have undergone a stan-dard processing sequence at the Institute of Geography,University of Copenhagen, leading to a comprehensiveand consistent series of images. Radiometric calibrationfollows the NOAA NESDIS guidelines using the methodsuggested by Rao & Chen (1995) and Rao & Chen (1996)for channel 1 and 2. For the thermal channels 3, 4 and 5in-flight calibration data embedded in the HRPT datastream are applied.

The Normalized Difference Vegetation Index, NDVIis calculated as:

where NIR is the calibrated reflectance factor of the nearinfrared channel (channel 2) and RED is the calibrated re-flectance factor of the visible channel (channel 1).

Cloud masking is performed in two steps. The firststep is an automatic masking performed through decision-tree classification, and the second step is a manual mask-ing for line drop-outs, cloud edges, haze or mist, dust andother disturbing features not detected by the automatic al-gorithm. In addition, automatic masking is applied toavoid scan angles greater than 42 degrees.

Images are geometrically rectified by combining auto-matic methods, applying orbital parameters, and manualmatching between the image and a vector file representa-tion of the Senegal River, the Gambia River and the coast-line from Mauritania to Guinea Bissau. Accuracy is esti-mated to be in the order of 1 km.

Following the standard processing sequence a numberof NDVI images (usually 3-5 images) are combined as 10-day composites using the Maximum Value Compositingmethod (MVC) (Holben, 1986; Cihlar et al., 1994). Thistechnique depresses atmospheric effects on NDVI, in-cluding the effects of varying water vapour and aerosolcontents of the atmosphere. The compositing procedure isapplied for the growing season, which for the Ferlo regionnormally lasts from late June to early October. The time-integral of NDVI, iNDVI, is obtained as a sum of all 10-day composites for the growing season. iNDVI is knownto be strongly correlated with NPP as NDVI is related tothe fraction of photo-synthetically active radiation ab-sorbed by vegetation, fAPAR. fAPAR is a key input pa-rameter in the Production Efficiency Model (PEM) used

by several authors, e.g. Kumar & Montieth (1981), Béguéet al. (1991), Prince & Goward (1995).

To obtain absolute values of NPP AVHRR data can beused as an input to a light-absorption model like the PEM.This would be an advantage when temporal comparisonsare in focus and when areas of continental scale are beingstudied. It is arguable that data should be converted intoNPP rather than using the 'raw' iNDVI-data, as demon-strated by Rasmussen (1998). While use of models such asPEM, as well as calibration with field data, may be morewell-founded theoretically, using iNDVI to estimate cu-mulative vegetative productivity is believed to be a robustapproach, demanding no additional inputs. In particular,iNDVI is useful for studying relatively small, homoge-neous areas, and when spatial variability, rather than ac-curate identification of temporal trends, is the focus.

Delimitation of pastoral units

A maximum area of 707 km2 around each borehole is con-sidered in the iNDVI analysis. This corresponds to a 15km radius corresponding to the maximum cattle mobilityaround the borehole (Hanan et al., 1991). Pastoral unitshave been delimited using the Thiessen Polygon method,which in most cases leads to limits corresponding rela-tively well to the administrative boundaries. This methodallocates each 1 km2 pixel to the polygon associated withthe closest well. The shape of polygons ultimately de-pends on the spatial distribution of boreholes, and sincethese are not evenly distributed some polygons haverather asymmetrical shapes, see Figure 1. Where bore-holes are situated close to one another (inside approxi-mately 30 km) other factors than distance might influencethe choice of water supply for livestock. Therefore, aThiessen Polygon is merely an approximation of 'realworld' dry-season grazing borders related to livestockmobility around the borehole.

Derivation of iNDVI profiles

In the iNDVI gradient analysis eight directions are con-sidered, see Figure 3. Using a spreadsheet software pack-age the iNDVI data were sorted and divided into eightblocks according to the eight directions. Within eachblock mean iNDVI was calculated for each kilometre in-terval from the borehole until a maximum of 15 km or theborder of the Thiessen Polygon was reached. Information

NDVI = NIR - RED NIR + RED


on distance and direction was obtained from suitable geo-graphic information system tools.

In order to test to what extent the profiles can be de-scribed by a simple model, a second order polynomial fitwas found for the omnidirectional profile of each pastoralunit. The second order polynomial was chosen for itsmathematical simplicity combined with its ability to ap-proximate the observed iNDVI profiles.

Classification of iNDVI profiles

A classification of omnidirectional profiles for all pas-toral units and all three years was carried out based on adecision tree-classification model.

Step 1: Determining the profile rangeAll profiles were chosen as starting one kilometre distantfrom the borehole to eliminate the effects of the 'sacrificearea' dominating the nearest circumference of the bore-hole.

In order to eliminate the effect of the N-S rainfall gra-dient the end-points of the omnidirectional gradients wereset based on the weighted representation of individual di-rectional profiles. Omnidirectional profiles were "cut off"

if they were based on an insufficient number or an unevendistribution of directional profiles. In theory, as few as 3individual directions would generate an acceptable omni-directional profile if the individual directions representedwere well distributed. However, because of the eraticshape of polygons no cases with less than four directionswere found.

The majority of omnidirectional profiles producedfrom step 1 had a range between 10 and 15 km. Profilesshorter that 7 km were excluded from the classification.

Step 2: Categorizing profilesFour categories of omnidirectional profiles were consid-ered: (1) normal gradients, (2) non-gradients, (3) inversegradients and (4) composite gradients.

In order to separate non-gradients from gradients, basicdescriptive statistics have been computed for the omnidi-rectional profiles. A lower limit of 0.1 of the iNDVI range(maximum iNDVI minus minimum iNDVI divided by themean iNDVI) was used to separate gradients (above 0.1)from non-gradients (below 0.1). In this way, inter-annualdifferences in the absolute level of iNDVI attributableto varying rainfall are excluded from analysis. A rangecorrection is applied, dividing the iNDVI range by meaniNDVI, thereby accounting for different ranges (profilelengths) of omnidirectional gradients in the data set.

In accordance with the afore-mentioned definition byPickup & Chewings (1994), normal gradients are charac-terized by a steady growth in iNDVI with increasing dis-tance from the borehole, whereas inverse gradients arecharacterized by a falling iNDVI.

Composite gradients are cases where a sudden changein the iNDVI profile leads to a consistent and clear in-crease or decrease for two km or more. By clear we mean,once again, that the distance-corrected range divided bythe mean on one side of a given minimum or maximumexceeds 0.1. Changes occurring over shorter ranges than 2km have been interpreted as noise in the iNDVI profile.

Results

Omni-directional gradientsOmni-directional iNDVI-gradients for 44 boreholes for1990, 1991 and 1992 are shown in Figure 4. It can be seenimmediately that the pattern is quite complex. Normal, in-verse and composite gradients are observed, as well ascases of non-gradients.

Figure 3: Integrated NDVI for Tessekre. The Pastoral unit is con-fined by the Thiessen Polygon and divided into eight sectors ac-cording to direction.


Amali

Distance from borehole (km)

iND

VI

10

11

12

13

14

15

16

17

18

19

20

21

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Atch Bali


iND

VI

5

6

7

8

9

10

11

12

13

14

15

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Belil Bogal


iND

VI

2

3

4

5

6

7

8

9

10

11

12

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Bisnabe Bokidive


iND

VI

6

7

8

9

10

11

12

13

14

15

16

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Bombode


iND

VI

4

5

6

7

8

9

10

11

12

13

14

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Boudy Sakho


iND

VI

9

10

11

12

13

14

15

16

17

18

19

20

21

22

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Civiabe Oriental


iND

VI

10

11

12

13

14

15

16

17

18

19

20

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Dar Salam


iND

VI

8

9

10

11

12

13

14

15

16

17

18

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Darou Naime


iND

VI

12

13

14

15

16

17

18

19

20

21

22

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Dendoudy


iND

VI

6

7

8

9

10

11

12

13

14

15

16

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Diagle


iND

VI

7

8

9

10

11

12

13

14

15

16

17

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Djagaly Thianor


iND

VI

9

10

11

12

13

14

15

16

17

18

19

20

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Dodji Oulof


iND

VI

11

12

13

14

15

16

17

18

19

20

21

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Fedia


iND

VI

5

6

7

8

9

10

11

12

13

14

15

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Ganine Erogne


iND

VI

6

7

8

9

10

11

12

13

14

15

16

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Gaoudou Goti


iND

VI

456789

101112131415161718

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Gueye Kadar


iND

VI

5

6

7

8

9

10

11

12

13

14

15

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

1990

1991

1992


Kara Vendou


iND

VI

4

5

6

7

8

9

10

11

12

13

14

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Kodiolele


iND

VI

2

3

4

5

6

7

8

9

10

11

12

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Labgar


iND

VI

89

101112131415161718192021222324

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Linde Peulh


iND

VI

15

16

17

18

19

20

21

22

23

24

25

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Lougre Thioly Diallo


iND

VI

6

7

8

9

10

11

12

13

14

15

16

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Loumbi Sandarabe


iND

VI

4

5

6

7

8

9

10

11

12

13

14

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Loumboul Samba Abdou


iND

VI

6

7

8

9

10

11

12

13

14

15

16

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Mbar Touabab


iND

VI

6

7

8

9

10

11

12

13

14

15

16

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Mbem Mbem


iND

VI

15

16

17

18

19

20

21

22

23

24

25

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Mbiddi


iND

VI

5

6

7

8

9

10

11

12

13

14

15

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Namarel Pelel Daba


iND

VI

5

6

7

8

9

10

11

12

13

14

15

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Niassante


iND

VI

6

7

8

9

10

11

12

13

14

15

16

17

18

19

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Ourourbe Daka (Bano)


iND

VI

7

8

9

10

11

12

13

14

15

16

17

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Ranerou


iND

VI

8

9

10

11

12

13

14

15

16

17

18

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Revane


iND

VI

6

7

8

9

10

11

12

13

14

15

16

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Samaly


iND

VI

10

11

12

13

14

15

16

17

18

19

20

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Sare Liou


iND

VI

2

3

4

5

6

7

8

9

10

11

12

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Kambe


iND

VI

910111213141516171819202122232425

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15


Directional profilesIn some cases, directional gradients around a certainborehole are very different from each other. This may bedue to local discontinuities in the landscape, e.g. the edgeof the Senegal River valley or the Fossil Valley system ofthe Ferlo. In other cases, such as the one illustrated in Fig-ure 5, the differences are not stable from one year to thenext and are likely the result of spatial rainfall variability.

Fitting of quadratic functions and classification of omni-directional profilesIn Figure 4 quadratic functions have been fitted to the om-nidirectional profiles. Subsequently, the profiles havebeen classified, using the procedure described above. Thedistributions of the four main types, mentioned above, and5 sub-types of composite gradients for the three yearsstudied are shown in Figure 6 a, b and c. From Figs 4 & 6the following patterns emerge:

• Overall, inverse gradients are most common in thenorth-central Ferlo in 1990 and 1991.

• Inverse gradients are most pronounced (steeper gradi-ent) in cases of high values of iNDVI, associated withhigh vegetative productivity and above-average rain-fall.

• The clearest cases of normal gradients are found at thenorthern edge of the Ferlo, in the ferrigenuous, easternFerlo, and, in 1992, in the southern part.

• Composite gradients with decreasing followed by in-creasing iNDVI are particularly widespread in the east-central part of the study area.

Correlation between yearsIt is noteworthy that gradients may change from normal toinverse between years for the same borehole. The correla-tions between years have been calculated and the resultsare shown in Figure 7, a, b, c and Figure 8. It appears thatpositive correlations dominate with median values of cor-relation coefficients in the order of 0.6 - 0.7. However,negative correlation is not unusual.

Tessekre


iND

VI

8

9

10

11

12

13

14

15

16

17

18

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Thiargny


iND

VI

12

13

14

15

16

17

18

19

20

21

22

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Thiel


iND

VI

12

13

14

15

16

17

18

19

20

21

22

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Tialambol


iND

VI

8

9

10

11

12

13

14

15

16

17

18

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Tiekinguel (Petel Bi)


iND

VI

2

3

4

5

6

7

8

9

10

11

12

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Velingara


iND

VI

10

11

12

13

14

15

16

17

18

19

20

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Yare Lao


iND

VI

5

6

7

8

9

10

11

12

13

14

15

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Younoufere

Distance from borehole (km)iN

DV

I

8

9

10

11

12

13

14

15

16

17

18

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Tatki


iND

VI

6

7

8

9

10

11

12

13

14

15

16

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Figure 4: Omnidirectional iNDVI profiles for 44 boreholes. See Figure 1 for position of boreholes.


Discussion and conclusion

The complex pattern observed negates the simple hy-pothesis that high grazing pressure around boreholes gen-erally causes the development of a normal iNDVI gradi-ent indicating land degradation. The considerable varia-tion in gradient type observed within the study area, inter-annually and as well as spatially, may be related to varia-tions in soil type, landscape and vegetation types. Thougha full analysis of these factors requires more data thanpresently available some possible explanations of the ob-served gradients are presented:

Limiting factors of production: Phosphorous and waterInverse gradients are most pronounced in years with highrainfall (and therefore high iNDVI), see Figure 4. Vegeta-tive productivity in high rainfall years may therefore belimited by available phosphorous of which manure is aprimary source (Turner, 1998a and b). A high concentra-tion of manure in the vicinity of the wells would explainthe strong inverse gradient. In low rainfall years, waterwill be the limiting factor and the inverse gradient willtherefore be weakened or eliminated.

Presence of unpalatable speciesClose to wells, where high concentrations of livestock arefound in the dry season, unpalatable species may some-

times attain dominance. These may contribute signifi-cantly to the NDVI signal. Cassia obtusifolia is an exam-ple in the south, while Calotropis procera (eaten but notpreferred by the livestock) is widespread in the northernparts. These may give rise to composite (uneven) gradi-ents. This has also been observed in Australia by Pickup& Chewings (1994).

The significance of long term rainy season grazing pres-sureIn the rainy season, the distribution of grazing pressure isbelieved to be controlled mainly by the location of tem-porary ponds and the location of pastoralist camps, thetwo of which often coincide. Frequently camps are foundat distances of 5-10 km from boreholes. Hence, the result-ing rainy season grazing pressure is presumed to be simi-lar year after year. Certain cases of composite gradients,where iNDVI reaches a minimum at some distance fromthe well (e.g. in Thiel in the southern part of Ferlo) may beexplained in this way, see Figure 4. Here the observedminimum in iNDVI is found 7 to 8 km from the well,which corresponds well with the location of a number oftemporary ponds and pastoralist camps.

Problems associated with the methodThe methodology suffers from a number of problems:• It is assumed that iNDVI provides a good estimate of

7

9

11

13

15

17

19

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15


N-E E S-E Omni-dir

S S-W W N-W N

B

7

9

11

13

15

17

19

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Distance from borehole (km)S S-W W N-W N

N-E E S-E Omni-dir

A

Figure 5: Directional iNDVI profiles for the Namarel Pelel Daba borehole reflecting two situations. In 1990 (Figure 5A) large differencesbetween individual directions can be observed, whereas differences are much smaller in the lower rainfall situation of 1991 (Figure 5B).


cumulative vegetative growth. This has been demon-strated to be a potentially problematic assumptionwhen vegetation cover is low and soil colour thereforehas a significant impact (Huete, 1988). This is particu-larly important in dry years.

• Likewise, iNDVI generally provides a better estimateof vegetative productivity where the crown cover of thewoody vegetation is low (Rasmussen, 1998). In thesouthern and eastern parts of the study area, crowncover may in cases exceed 20 %, and the productivityassociated with the woody vegetation may exceed theproductivity of the herbaceous vegetation. Also, brows-ing is very important here, especially in the dry season.Thus, iNDVI is less useful in these areas. However, noalternative method for estimating vegetative productiv-ity at the appropriate scale is presently available.

• Further, it is assumed that grazing in the rainy seasondoes not limit iNDVI significantly. If it did, iNDVIwould not be useful as an indicator of degradation.However, this assumption may be ill-founded in lowrainfall years. Note that this does not apply to long-term effects of rainy season grazing pressure, as dis-cussed later.

• The number of pixels used to calculate iNDVI in-creases with the distance from the well. Thus, values ofiNDVI close to the borehole can be "noisy" and shouldbe interpreted with caution. One source of noise is thegeometrical (co-) registration of the series of NOAAAVHRR images used. Even a minor error of less thanone pixel can generate substantial noise close to theborehole, whereas this effect will be small at greaterdistances.

• It is assumed that boreholes have been well function-ing during the entire study. However, occasional pumpfailure is not unusual, which might help explain someof the variation found between different years for thesame borehole.

Conceptual problems associated with the notion of landdegradationIt should be kept in mind that there are many definitionsof land degradation that emphasize other aspects than justvegetative productivity. From a local pastoralist's or aconservationist's point of view the issues of importanceare quite different (Rasmussen, 1999). The pastoralist willmost often be interested in the quality of grazing seenfrom the perspective of the livestock, and in this case thevegetative productivity or the total quantity of biomassmay be of little interest if unpalatable species are com-

mon. The protein content of fodder is often claimed tolimit livestock production, rather than the energetic value.For a conservationist, land degradation is often related toa loss of bio-diversity, which is not closely related to over-all vegetative productivity. The interpretations in this pa-per strictly concern the definition of land degradationbased on vegetative productivity.

As discussed, the basic argument relating vegetativeproductivity gradients to degradation is that areas withlower productivity (around boreholes) should be seen asfoci of land degradation as it is defined here. Should theabsence of this relation, as indicated by the dominance ofinverse gradients, lead one to interpret absence of degra-dation? Depending on the actual causes of the observedvegetative productivity gradients, different lines of rea-soning will apply:

If inverse gradients are mainly due to manuring effect,the nutrients causing increased vegetative productivity

#³

#³

#³"G

"G

%

"G"G

%

"G

0 50 100 Kilometers

N

"G

%

$

$

"G

$

$

#³

"G

"G

%

#³

"G

"G

%$

"G

$

%

"G

"G

%

"G

#³

$

"G

"G

$%

$

"G

#³

#³

"G

#³

Categories:Normal gradientNon-gradientInverse gradient

"G Composite (decrease-increase)#³ Composite (increase-decrease)$ Composite (decrease-flat or flat-decrease)

Composite (increase-flat or flat-increase)% Double composites

A: 1990

B: 1991

C: 1992

Figure 6: Categories of omnidirectional iNDVI profiles for thethree years 1990, 91 and 92. Note, polygons with no categorizationmeans that too many clouds have been present.


around wells come as a cost to areas further from the well,which as a consequence experience a lower productivity.If this explanation holds, grazing gradients should not beused as degradation indicators at all, but rather as indica-tors of spatial redistribution of nutrients. Clarification ofthis question will require further study of the causes ofvegetative productivity gradients.

If the main cause of the spatial variations in vegetativeproductivity is rainy season grazing pressure, productiv-ity gradients radiating from wells become irrelevant,since distance from a well is not controlling grazing pres-sure. In that case vegetative productivity should, ofcourse, be related to the rainy season grazing pressure, es-timated on a km2-basis.

In summary, the simple hypothesis that vegetativeproductivity will generally be reduced around dry seasonwatering points can be rejected. However, it will not bepossible to interpret this as a proof that degradation, as the

term is understood here, is not taking place. Rather, thegradients observed should be interpreted as indicators ofthe factors controlling vegetative productivity. The greatinter-annual variation shows that rainfall certainly playsan important role in some years, while landscape hetero-geneity, spatial variations in nutrient availability and long-term rainy season grazing pressures are likely to be im-portant in years of good rainfall. These alternative hy-potheses demand validation, and this work is in progress.Finally, the great differences observed among the 44 wellsand 3 rainy seasons studied illustrate that great careshould be taken when interpreting and upscaling/extrapo-lating results from studies of single wells and single years.

Concluding remarksAll in all, the results of the analysis of NOAA AVHRRdata presented, as well as the interpretations indicatedabove, point to the need for a better understanding of theinterplay between climate, vegetation and grazing. Multi-ple factors influence these processes, and findings arelikely to apply only to certain combinations of interrelatedphysical, biological and human factors. Assessment ofland degradation is not straightforward; especially in thelight of the 'dis-equilibrium' paradigm for dryland ecosys-tems (for example described by Behnke et al. (1993)) it isnecessary to suggest operational definitions of land degra-dation and find measurable indicators. The use of iNDVIgradients as a simple indicator of degradation is not gen-erally suggested.

#

#

#

##

##

#

#

##

## #

##

#

#

0 50 100 Kilometers

#

##

#

#

#

# ##

##

##

#

#

#

#

#

#

#

#

#

#

#

##

##

# #

#

#

##

## #

##

#

#

#

#

#

#

#

Correlation coefficients:-1 - -0.9-0.9 - -0.8-0.8 - -0.6-0.6 - -0.4-0.4 - -0.2-0.2 - 0

# 0 - 0.2# 0.2 - 0.4# 0.4 - 0.6# 0.6 - 0.8# 0.8 - 0.9

# 0.9 - 1

N

A: 1990-1991

B: 1990-1992

C: 1991-1992

Figure 7: Spatial distribution of correlation coefficients for corre-lations performed between omnidirectional profiles for differentyears.

Figure 8: Box and Whisker plot of correlation coefficients obtainedfrom correlations performed between omnidirectional profiles fromdifferent years.

1990-1991 (n=26) 1990-1992 (n=24) 1991-1992 (n=35)

corr

elat

ion

coef

ficie

nt

-1,0

-0,8

-0,6

-0,4

-0,2

0,0

0,2

0,4

0,6

0,8

1,0

MaxMin75%25%Median


Acknowledgments

The work presented has been carried out in the context ofthe collaborative project between CSE and IGUC con-cerning development of methodologies for environmentalmonitoring. The project was funded by Danida. The au-thors wish to thank the staff of CSE, in particular AlmamyWade, Alioune Touré, Ousmane Bocoum, Assize Touréfor their contributions. At IGUC we wish to acknowledgethe contributions of Michael Schultz Rasmussen, IngeSandholt, Thomas Theis Nielsen, Rasmus Fensholt andGorm Dybkjær.

Utilisation des données NOAA AVHRR pour l'esti-mation des gradients de productivité végétale au-tour des points d'eau dans les zones de parcours dunord du Sénégal.Morten Lind, Kjeld Rasmussen, Hanne Adriansen &Alioune Ka

RésuméCet article teste l'hypothèse selon laquelle la forte pres-sion de pâturage, durant la saison sèche autour des pointsd'eau dans les zones de parcours du Nord du Sénégal, pré-sente des effets négatifs sur la productivité de la végéta-tion. L'étude est faite en se basant sur une série temporelled'images NOAAAVHRR couvrant trois saisons des pluiespendant lesquelles le NDVI intégré (iNDVI) est estimé àune échelle de 1 Km2. Ensuite une analyse est faite de ladépendance de cet iNDVI à la distance du plus prochepoint d'eau. Les gradients de iNDVI observés sont appa-rus très variables en fonction des points d'eau et de l'an-née, ce qui rejette l'hypothèse de départ. Les raisons pos-sibles à cette variabilité sont discutées et il est retenu queles gradients de iNDVI ne sont pas de bons indicateurs dela dégradation des terres.

Mots-ClefsDégradation des terres, gradients de pâturage, NOAAAVHRR NDVI, télédétection, Sénégal

References

Adriansen, H. K. (2002): A Fulani without cattle is like awoman without jewellery: a study of pastoralists inFerlo, Senegal. Copenhagen, Geographica HafniensiaA11.

Alissoutin, R. L. (1997): Pond Management in the PodorDepartment, Senegal. International Institute for Envi-ronment and Development (IIED), Drylands Pro-gramme, Issue Paper No 72.

Andrew, M. H. (1988): Grazing Impact in Relation toLivestock Watering points. Trends Ecology and Evolu-tion 3(12): 336-339.

Andrew, M. H. & Lange, R. T. (1986a): Development of aNew Piosphere in Arid Chenopod Shrubland Grazedby Sheep: 1. Changes to the Soil Surface. AustralianJournal of Ecology 11: 395-409.

Andrew, M. H. & Lange, R. T. (1986b): Development ofa New Piosphere in Arid Chenopod Shrubland Grazedby Sheep: 2. Changes to the Vegetation. AustralianJournal of Ecology 11: 411-424.

Ba, C. (1986): Les Peuls du Sénégal, étude geographique.Dakar, Les Nouvelles Editions Africaines.

Barrell, H. (1982): Le Ferlo des forages: Gestion ancienneet actuelle de l'espace pastoral. Dakar, ORSTOM.

Bastin, G. N., Pickup, G. & Pearce, G. (1995): Utility ofAVHRR data for land degradation assessment: a casestudy. International Journal of Remote Sensing 16(4):651-672

Bastin, G. N., Pickup, G., Chewings, V. H. & Pearce, G.(1993a): Land Degradation Assessment in CentralAustralia Using a Grazing Gradient Method. Range-land Journal 15(2): 190-216.

Bastin, G. N., Sparrow, A. D. & Pearce, G. (1993b): Graz-ing Gradients in Central Australian Rangelands:Ground Verification of Remote Sensing-Based Ap-proaches. Rangeland Journal 15(2): 217-233.

Bastin, G. N., Tynan, R. W. & Chewings, V. H. (1998): Im-plementing satellite-based grazing gradient methodsfor rangeland assessment in South Australia. TheRangeland Journal 20(1): 61-76.

Bégué, A., Desprat, J. F., Imbernon, J. & Baret, F. (1991):Radiation use efficiency of pearl millet in the Sahelianzone. Agricultural and Forest Meteorology 56: 93-110.

Behnke, R. H., Scoones, I. & Kerven, C. (1993): RangeEcology at Disequilibrium - New Models of NaturalVariability and Pastoral Adaptation in African Savan-nas. London, Overseas Development Institute & Inter-national Institute for Environment and Development(IIED).

Chewings, V. H. Bastin G. N. & Pickup, G. (1992): Re-mote Sensing Models for Determining Grazing Impactin Arid Rangelands. Pp. 169-179 in: Proceedings of the6th Australasian Remote Sensing Conference.Wellington (New Zealand).


Cihlar, J., Manak, D. & D'Iori, M. (1994): Evaluation ofCompositing Algorithms for AVHRR Data Over Land.IEEE Transactions on Geoscience and Remote Sens-ing 32(2): 427-437.

Fatchen, T. J. & Lange, R. T. (1979): Piosphere Patternand Dynamics in a Chenopod Pasture Grazed by Cat-tle. Pp. 160-169 in: Graetz, R.D. & Howes, K. M. W.(ed.): Studies of the Australian Arid Zone IV. Cheno-pod Shrublands. Melbourne, CSIRO Division of LandResources Management.

Freudenberger, M. S. & Freudenberger, K. S. (1993): Pas-toralism in Peril: Pressures on Grazing land in Sene-gal. IIED Drylands Programme, Pastoral Land TenureSeries 4.

Graetz, R. D. & Ludwig, J. A. (1978): A Method for theAnalysis of piosphere Data Applicable to Range As-sessment. Australian Rangeland Journal 2: 126-136.

Hanan, N. P., Prevost, Y., Diouf, A. & Diallo, O. (1991):Assessment of Desertification Around Deep Wells inthe Sahel using Satellite Imagery. Journal of AppliedEcology 28: 173-186.

Hiernaux, P. (2000): Implications of the 'New RangelandParadigm' for natural resource management. Pp. 113-142 in: Adriansen, H., Reenberg, A. & Nielsen, I.(ed.): The Sahel. SEREIN Occasional papers 11, Insti-tute of Geography, University of Copenhagen.

Holben, B. N. (1986): Characteristics of maximum-valuecomposite images from temporal AVHRR data. Inter-national Journal of Remote Sensing 7(11): 1417-1434.

Huete, A.R. (1988): A Soil-Adjusted Vegetation Index(SAVI): Remote Sensing of Environment 25: 295-309.

Hulme, M. (2001): Climatic perspectives on Saheliandesiccation: 1973-1998. Global EnvironmentalChange 11: 19-29.

Juul, K. (1996): Post Drought Migration and Technologi-cal Innovations among Fulani Herders in Senegal: theTriumph of the Tube! IIED Drylands Programme, Is-sue Paper 64.

Kumar, M. & Montieth, J. L. (1981): Remote Sensing ofcrop growth. Pp. 133-144 in: Smith, M. (ed.): Plantsand Daylight Spectrum. London, Academic Press.

Lange, R. T. (1969): The piosphere: Sheep Track andDung Patterns. Journal of Range Management 22:396-400.

Matheson, W. & Ringrose, S. (1994): Assessment ofdegradation features and their development into thepost-drought period in the West-Central Sahel usingLandsat MSS. Journal of Arid Environments 26(2):181-199.

Pearson, J. T., Sparrow, A. D. & Lange, R. T. (1990): Pro-longed Exposure to Sheep Grazing Reduces the Palata-bility of Australian Saltbush populations. Australianjournal of Ecology 15: 337-344.

Pickup, G. (1994): Modelling Patterns of Defoliation byGrazing Animals in Rangelands. Journal of AppliedEcology 31: 231-246.

Pickup, G. & Chewings, V. H. (1988): Estimating the Dis-tribution of Grazing and Patterns of Cattle Movementin a Large Arid Zone Paddock: An Approach using An-imals Distribution models and Landsat Imagery. Inter-national Journal of Remote Sensing 9(9): 1469-1490.

Pickup, G. & Chewings, V. H. (1994): A Grazing GradientApproach to Land Degradation Assessment in Arid Ar-eas from Remotely-sensed Data. International Journalof Remote Sensing 15(4): 597-617.

Prince, S. D. & Goward, S. N. (1995): Global primaryproduction: a remote sensing approach. Journal of bio-geography 22: 815-835.

Prins, E. (1997): Natural Resources and Resource Utiliza-tion in Ferlo Faunal Reserves - a Baseline MappingReport on Natural Resources and Resource Utilization.Pp. 37-45 in: Reenberg, A., Nielsen, I. & Secher-Mar-cussen, H. (ed.): The Sahel: Natural Resource Man-agement Projects; Energy Provision; Decentralisation,Empowerment and Capacity Building. SEREIN Occa-sional Papers No 5, Institute of Geography, Universityof Copenhagen.

Rao, C. R. N. & Chen, J. (1995): Inter-satellite calibrationlinkages for the visible and near-infrared channels ofthe Advanced Very High Resolution Radiometer on theNOAA-7, -9 and -11 spacecraft. International Journalof Remote Sensing 16: 1931-1942.

Rao, C. R. N. & Chen, J. (1996): Post-Launch Calibrationof the Visible and Near-infrared Channels of the Ad-vanced Very High Resolution Radiometer on theNOAA-14 spacecraft. International Journal of RemoteSensing 14: 2743-2747.

Rasmussen, K. (1999): Land degradation in the Sahel-Su-dan. The conceptual basis. Geografisk Tidsskrift, Dan-ish Journal of Geography (2): 151-159.

Rasmussen, M. S. (1998): Developing simple, opera-tional, consistent NDVI - vegetation models by apply-ing environmental and climatic information. Part I. As-sessment of Net Primary Production. InternationalJournal of Remote Sensing 19: 97-117.

Thomas, D. S. G. (1993): Sandstorm in a teacup: under-standing desertification. Geographical Journal 159:318-331.


Touré, O. (1990): Where herders don't herd anymore: -Ex-perience from the Ferlo, northern Senegal. IIED Dry-lands Programme, Issue Paper No 22.

Turner, M. D. (1998a): Long-term effects of daily grazingorbits on nutrient availability in Sahelian West Africa:1. Gradients in the chemical composition of rangelandsoils and vegetation. Journal of Biogeography 25(4):669-682.

Turner, M. D. (1998b): Long-term effects of daily grazingorbits on nutrient availability in Sahelian West Africa:Effects of a phosphorus gradient on spatial patterns ofannual grassland production. Journal of Biogeography25(4): 683-694.

Williams, M. A. J. & Balling, R. R. Jr. (1996): Interac-tions of Desertification and Climate. London, EdwardArnold.

Documents

Estimating vegetative productivity gradients around watering points