Previous PageTable Of ContentsNext Page

The use of remotely sensed data to analyse spatial and temporal trends in patchiness within rehabilitated bauxite mines in the Darling Range

A.K. Prananto, Bob Gilkes and Faron C. Mengler

School of Earth and Geographical Sciences, The University of Western Australia,
Mail Delivery Point M087, 35 Stirling Highway, Crawley WA 6009, Australia.
Email: scuba210@tartarus.uwa.edu.au; bob.gilkes@uwa.edu.au; fmengler@agric.uwa.edu.au

Abstract

The assessment of rehabilitation success is time-consuming and costly for bauxite miners because large areas of land (~550 ha per year in W.A.) are involved. In some cases, rehabilitation results in patches of bare or sparsely vegetated soil. This study uses remote sensing imagery to evaluate the growth of vegetation in rehabilitated bauxite mines in the Darling Ranges, W.A. Three investigations were carried out, viz: (1) a comparison of vegetation biomass between rehabilitated areas and nearby native forest; (2) measurement of temporal trends in vegetation growth in rehabilitated areas; and (3) a comparison of pre- and post-mining vegetation growth. This information enables rehabilitation workers to identify patches of rehabilitated areas that may require further remediation. We used RADARSAT, nine years of Normalised Difference Vegetation Index (NDVI) maps (extracted from LANDSAT TM imagery and Quickbird imagery) and aerial photographs to evaluate 22 study sites of ~one ha. We found that four-year-old rehabilitated areas have higher NDVI values (mean NDVI 47.1 4.4) than native forest (mean NDVI 35.1 2.9); and there is no relationship between vegetation index values and time for native forest. There is a positive relationship between vegetation index values and time for non-patchy and patchy rehabilitated areas. For non-patchy rehabilitated areas, the range of NDVI values decreases with time. For patchy rehabilitated areas, the range of NDVI values increases or stays high with time, until the bare patches become vegetated. This difference in behaviour becomes evident after approximately 2 years of growth.

Keywords

Rehabilitation, vegetation growth, rainfall, forest, bauxite, bare soil, ground LAI

Introduction

In rehabilitated mine pits, the presence of bare patches may be caused by several factors, such as shallow soil above bedrock, fire (Jeltch et al., 1997), animal disturbance (Pickett and White, 1985), mismanagement (Bhuju and Ohsawa, 1999) and disease. Some studies have suggested that such diversity in rehabilitation may be desirable as it maintains species diversity (Petraitis et al., 1989; Connell, 1978) but it can also lead to loss of species (Van de Koppel et al., 2002). Continual monitoring is therefore important to assess the success of rehabilitated areas. Monitoring can be time-consuming and costly due to the size of rehabilitated areas. Around 360ha/yr is rehabilitated at Alcoa World Alumina Australia’s (Alcoa) Huntly Bauxite mine near Dwellingup, W.A. Bare patches cannot be mapped on the ground because they are located within areas of dense vegetation. A way to address this problem is by utilising aerial photographs and remotely sensed data (Woral, 1960; Wilson and Sader, 2002; Apan et al., 2002; Bastin et al., 2002). The large area coverage and the ease of image manipulation, storage and retrieval has led to the widespread use of these images for spatial studies of vegetation.

A remotely sensed image is fundamentally the observation of a particular object/s (soil, vegetation) from a distance, such as is acquired from aircraft and Earth observation satellites (e.g. Landsat TM, SPOT, etc.) The spectral information embedded in the images can be manipulated to produce vegetation indices, which may for example be predictive of Leaf Area Index (LAI) for forest ecosystems (Prince, 1991), and crops (Sellers et al., 1992). Amongst the many vegetation indices, the most commonly used is the Normalised Difference Vegetation Index (NDVI) (Rouse et al. 1974) defined as:

Normalised Difference Vegetation Index (NDVI) (Equation. 1)

where: ρ NIR = Reflectance in the Near-Infrared Band
ρ
Red = Reflectance in the Red Band

As NDVI is a ratio it is not very sensitive to multiplicative noise (illumination differences, cloud shadows, atmospheric attenuation) present in the multiple bands. In addition, NDVI separates green vegetation from other vegetation by the absorption of red light by chlorophyll in green vegetation, which reflects the near-infrared wavelengths due to scattering caused by internal leaf structure (Tucker, 1979). High NDVI values therefore indicate high leaf biomass and canopy closure (Sader and Winne, 1992). NDVI does not have a large working range being insensitive once canopy closure is complete. This produces large errors when estimating forest biomass with intermediate (0.5 < LAI < 3) to large LAI (LAI ≥ 3) values (Rondeaux et al., 1996). Several studies addressed this problem by incorporating reflectance in the middle-infrared band (Landsat TM band 5; 1.55 – 1.75 μm) into the index (Boyd et al., 2000; Eklundh et al., 2003).

Reflectance at middle infrared (MIR) wavelengths decreases with increasing leaf area as a result of increased absorption of water in the canopy (Boyd et al., 2000). MIR has also been shown to be less affected by atmospheric effects than either the near infrared or red bands (Kaufman and Remer, 1994).

The inclusion of the middle-infrared band into the NDVI therefore should produce stronger relationship between NDVI and biomass for forest with LAI greater than 0.5. NDVI incorporating the middle infrared band is known as the Corrected Normalised Difference Vegetation Index (Nemani et al., 1993). It is defined as:

Corrected Normalised Difference Vegetation Index (NDVIc) (Equation 2.)

where: ρ MIR = Reflectance in the Middle-Infrared Band
ρ
MIRmin = Minimum reflectance value in the Middle-Infrared Band in the data set
ρ
MIRmax = Maximum reflectance value in the Middle-Infrared Band in the data set

NDVIc is more sensitive to differences in biomass than is NDVI for rehabilitated and native forest areas with LAI greater than 0.5 (Rondeaux et al. 1996). There is little information on the use of NDVIc for monitoring temporal and spatial changes in rehabilitated and native forest. No studies have attempted to identify the evolution of bare patches within rehabilitated bauxite-mines using either NDVI or NDVIc. The aims of this study were to use remotely sensed images to: (1) compare vegetation biomass within rehabilitated areas and nearby native forest; (2) analyse temporal changes in vegetation growth within selected rehabilitated areas, in particular rehabilitated areas with patches of bare soil; and (3) compare pre- and post-mining vegetation growth. The outcomes will assist in the early identification of bare patches, and may reduce the cost of monitoring plant growth in rehabilitated areas. This study will also discuss NDVI values derived from Quickbird (0.6m pixel size).

Previous studies have shown radar imagery to be useful for monitoring biomass (Tanaka et al., 1998; Lucas et al., 2000); however, information on its use for vegetation growth analysis is minimal. Radar imagery operates by making use of energy transmitted at microwave frequencies. RADARSAT in particular operates at a single microwave frequency, known as C-band (5.3GHz frequency) and generates one channel of data. The ratio of scattered to incident microwave energy is known as the radar backscatter. The advantages of using radar imagery are its all-weather coverage, ability to penetrate through cloud, high spatial resolution, day/night acquisition and independence of the signal from solar illumination angle. However, radar imagery is very sensitive to surface roughness and speckle noise can be a problem, particularly for relatively small and closely spaced vegetation. In this study we briefly consider RADARSAT imagery.

Methods

Study site description

Our study site is an area around the Huntly bauxite mine located in the Darling Range, approximately 110km south-southwest of Perth, Western Australia between coordinates: 406850E 6399650N (top left) and 420950E 6389200N (bottom right). The Darling Range is characterised by ancient lateritic soil with a coarse-textured, gravely surface horizon over duricrust. Bauxite is duricrust and loam rich in aluminium hydroxides and oxyhydroxides found just beneath the duricrust. The study area has a Mediterranean-type climate, with hot, dry summers and cool, wet winters. Annual rainfall ranges from ~1200mm in the west to ~400mm in the east. The forest is dominated by Eucalyptus marginata (Jarrah) and Conymbia callophyla (Marri). Middle-storey vegetation consists mainly of Banksia grandis, Allocasuarina fraseriana and Persoonia longifolia. Herbs and shrubs from the families of Liliaceae, Leguminosae, Orchidaceae, Apiaceae, Epacridaceae, Asteraceae, Restionaceae and Cyperceae form the undergrowth.

Remotely sensed data sources

Aerial photographs of the study site (years 2001, 2002 and 2003) were provided by the Department of Land Information (DLI) with a spatial resolution of 0.4m. They were used for locating bare patches and rock outcrops, and were related to multispectral remotely sensed images and field observations. Three types of remotely sensed data were used. Six years (1994, 1996, 1998, 2000, 2002 and 2003) of cloud-free multivariate satellite images from Landsat TM (provided by DLI) with spatial resolution of 25m were used to derive two vegetation indices. Landsat TM band 3 and 4 were used to produce NDVI; and bands 3, 4 and 5 were used to derive NDVIc. NDVI values derived from Quickbird data with a spatial resolution of 0.6m provided images with more detail. One “Radarsat” radar image with a spatial resolution of 20m was also analysed (provided by DLI). Additionally, several GIS coverages were used for site information, including: the age of rehabilitation; boundaries; location of stockpiles; type of vegetation; and topography. ERMapper 6.4 and ESRI’s ArcGIS 8.3 were used to perform various manipulations. Study sites for detailed analysis were selected and subjected to two types of investigation.

Comparison of rehabilitated and forest areas

The criteria for selecting paired sites for comparison were that the two sites had the same landscape features (e.g. altitude, topographic situation, similar aspects, degree of slope, geology) and no outcrops. Rehabilitated sites commenced in year 1991 were chosen as there were different management practices used prior to this year. Polygons for three pairs of sites were produced, resulting in 6 selected study sites. These were in several pairs: a.) Adjacent to each other (R1SWA/F1SWA)*; b.) Further apart (R2SEF/F2SEF)*; and c.) Randomly chosen (R3NER/F3NER)*.

* A label beginning with R means rehabilitated area and F means forest area.

Temporal analyses of rehabilitated areas

Temporal analyses were divided into: 1.) non-patchy and 2.) patchy rehabilitated areas. The size of the rehabilitated area selected had to be greater than one hectare with rehabilitation commenced and completed after the year 1990. A rehabilitated area was classified, as “patchy” if one or more patch of bare soil was visible in aerial photographs and persisted over several years.

1.) Non-patchy rehabilitated areas (N-RA): rehabilitated areas completed in years 1994, 1997 and 1999 were analysed for each year that satellite data were available until year 2003. Three replicate sites were identified for each individual rehabilitation year, giving 9 study sites labelled as: R94-1, R94-2, R94-3, R97-1, R97-2, R97-3, R99-1, R99-2 and R99-3.

2.) Patchy rehabilitated areas (P-RA): initially, 12 rehabilitated areas with bare patches visible in aerial photographs were selected and labelled “patchy-area 1 to 12”. After further consideration of the definition of a “patchy” rehabilitated area, the number of sites was reduced to 7 and these were patchy-area 2, 4, 6, 8, 9, 10 and 11. Patch size, shape and amount of plant growth were calculated for individual bare patches for each year of data.

Data and image processing

To analyse rehabilitated and forest areas, GIS coverages (roads, rehabilitation boundaries) were overlaid over aerial photographs and remotely sensed images. Aspect and degree of slope maps were created from DLI topographic data (5 metre contours) using ESRI’s Spatial Analyst tool. The amount of precipitation recorded immediately prior to the day each image was obtained was determined to assist with interpretation as will be discussed later. ERMapper 6.4 was used to derive vegetation indices from Landsat TM data. For the six Landsat TM images, band 5 minimum and maximum values were individually recorded for NDVIc calculations. NDVI was calculated according to Rouse et al. (1974) (Equation 1.) and NDVIc according to Nemani et al. (1993) (Equation 2.). Individual vegetation index values were multiplied by 100 and saved as unsigned 8 bits, giving a range of 0 to 70 (instead of the normal index range, from -1 to 1). Images were converted to ESRI GRID. The Spatial Analyst “zonal function” was then used to analyse the six different years of NDVI and NDVIc images for the selected study sites. Output parameters: area size, minimum, maximum, range, mean, standard deviation, sum, variety, majority, minority and median were produced. Statistical analyses were carried out on these data. Image interpretations were evaluated on several groundtruthing expeditions.

Results

A comparison of rehabilitated and forest areas

Initially, the vegetation index values for native forest were higher than for rehabilitated areas (<3-years-old) but after about 5 years of growth, both NDVI and NDVIc values for the corresponding rehabilitated areas exceeded those for native forests and were not patchy (Table 1. and Figure 1.)

Table 1. Temporal analysis of NDVI and NDVIc mean values from year 1994 to 2003, for topographically matched, non-patchy rehabilitated areas (rehabilitated areas operation completed in year 1991) and native forest that are located: a.) adjacent to one another; b.) further apart; and c.) randomly located

 

Rehabilitated forest

Native forest

 

Year

NDVI

NDVIc

NDVI

NDVIc

mean

std. dev.

mean

std. dev.

mean

std. dev.

mean

std. dev.

a.

1994

31.15

7.00

27.90

6.52

33.25

4.69

30.25

4.30

 

1996

42.70

6.47

35.05

6.15

35.35

4.21

28.45

3.84

 

1998

40.00

5.06

37.30

4.50

32.85

2.55

30.75

2.45

 

2000

42.50

5.36

34.50

5.41

28.90

2.43

21.90

2.17

 

2002

38.90

2.55

32.75

3.08

23.00

3.03

17.90

2.26

 

2003

56.75

3.67

48.80

3.71

40.05

2.67

32.65

2.24

b.

1994

34.03

4.18

28.88

3.58

40.44

2.78

35.38

2.62

 

1996

42.34

2.91

32.22

2.62

42.34

2.62

31.84

2.56

 

1998

52.00

2.92

47.66

2.82

38.97

4.76

34.97

4.33

 

2000

49.91

2.04

39.25

2.17

36.09

2.85

25.81

2.59

 

2002

44.44

1.50

34.88

1.27

34.00

2.60

25.00

2.26

 

2003

61.72

1.74

49.72

1.57

48.69

3.61

37.47

3.15

c.

1994

41.94

8.67

35.59

8.00

38.44

2.01

33.72

2.01

 

1996

53.50

7.49

41.94

7.36

44.25

2.72

34.17

2.30

 

1998

56.32

5.83

50.29

5.82

43.06

2.91

38.92

2.76

 

2000

53.09

4.23

40.35

4.23

39.97

1.85

30.03

1.59

 

2002

45.03

4.06

33.59

3.83

31.58

2.45

23.67

2.04

 

2003

62.15

3.57

46.68

3.44

51.31

1.94

39.89

1.78

a.

b.

Figure 1. Relationships between mean vegetation index values and time (a.)NDVI and (b.)NDVIc for non-patchy rehabilitated areas (♦), patchy rehabilitated areas (■), (both commenced in year 1994) and native forest (▲).

Temporal analyses of rehabilitated areas

Trends in NDVI and NDVIc values with time were investigated. The sensitivity of NDVIc was similar to that of NDVI for this type of vegetation. For native forest, there is no trend in vegetation index values over time (Table 2.). There was a systematic trend for increasing index with time for both non-patchy and patchy rehabilitated areas (Table 3.). The trends with time for non-patchy rehabilitated areas and patchy rehabilitated areas were the same although values were different (Figure 1.). A difference between non-patchy and patchy rehabilitated areas relates to differences in both minimum values and maximum range values and this applies for both NDVI and NDVIc (Table 3.). For a non-patchy rehabilitated area the range value decreases with time (both minimum and maximum values increase, and converge). For a patchy rehabilitated area, the range value is high or increases with time, because minimum values which are equal to or close to zero persist whereas the maximum values increase.

Table 2. Correlation coefficient (r, positive) for the linear relationships between vegetation indices and time, for non-patchy rehabilitated areas, patchy rehabilitated areas and native forest.

site ID

Rehabilition
start year

NDVI (r)

NDVIc (r)

 

Non-Patchy rehabilitated sites

94-1

1994

0.82

0.62

94-2

1994

0.79

0.59

94-3

1994

0.83

0.62

 

Patchy rehabilitated sites

patchy-area 2

1995

0.94

0.85

patchy-area 4

1994

0.86

0.67

patchy-area 6

1991

0.77

0.47

patchy-area 8

1992

0.79

0.42

patchy-area 9

1992

0.78

0.45

patchy-area 10

1995

0.93

0.82

patchy-area 11

1996

0.96

0.93

 

Forest

F1SWA

native

0.14

0.11

F2SEF

native

0.00

0.07

F3NER

native

0.10

0.02

Pre- and post- mining vegetation comparison

The amount of vegetation existing pre- and post- mining for individual sites was estimated using mean vegetation index values. All rehabilitated areas reached the pre-mining condition (index values) after only 4 years of growth. For some rehabilitated areas, after 6 years of growth, the average vegetation index values exceed those for pre-mining conditions (Figure 2.).

Figure 2. Temporal trends of mean NDVI and NDVIc (1994, 1996, 1998, 2000, 2002 and 2003) values before and after mining activity, for three rehabilitated sites. Rehabilitation took place in 1999.

Table 3. Summary NDVI values for one example of a patchy- and non-patchy rehabilitated area. Rehabilitation of both areas commenced in 1994. Spectral data were collected in years 1994, 1996, 1998, 2000, 2002 and 2003.

Year

Patchy rehabilitated area

Non-patchy rehabilitated area

Minimum

Maximum

Range

Minimum

Maximum

Range

1994

0

23

23

0

8

8

1996

0

49

49

0

64

64

1998

3

58

55

11

65

54

2000

9

55

46

27

62

35

2002

5

52

47

33

55

22

2003

14

65

51

55

69

14

Discussion

NDVI and NDVIc were higher for rehabilitated areas than for forest by 4 to 6 years after rehabilitation (Table 2. and Figure 2.). This is consistent with qualitative observations from aerial photographs. Greater plant growth might be a consequence of ripping of the soil during site preparation, which maximises access of vegetation to water, including the extra water stored following the clearing of vegetation during mining. Rehabilitated areas receive large applications of seeds (5kg of native seed/ha) and are fertilised, which would also enhance plant growth. The growth of vegetation depends on rainfall and during the early years, changes in NDVI and NDVIc values over the two-year period between NDVI observations are strongly dependant on the amount of rainfall in the two years before the image was taken. Vegetation index values should normally increase with time until a constant NDVI value is achieved after about four years. An apparent relative reduction in NDVI was observed for year 2002, for every rehabilitation site. This drop could be due to disease, fire or drought in the two years before the image was captured, but this did not occur in this instance. Rainfall occurring immediately (days) before the image was recorded could be responsible for this effect, as it will increase NDVI values and this may not have occurred in 2002 (Davenport and Nicholson, 1993; Wang et al.2003). A week prior to the 1998 image and 2003 images, 43.2 mm and 44.9mm of rain was recorded so that soil and vegetation was wet, whereas for other years only about 8 mm of rain fell in the week before image capture. This effect however was not observed for native forest where vegetation index values were unchanged presumably due to the wet soil being obscured by vegetation.

Both vegetation indices can be used to monitor vegetation growth within rehabilitated areas, including sites with and without patches of bare soil. However, there is a stronger relationship between NDVI values and time than NDVIc with time for this particular type of rehabilitation. In addition, a study at Dwellingup, W.A, showed that NDVI related closely to the ground LAI (R2 = 0.7, Wallace, 1996). Nonetheless, both indices are adequate for recognising patches of bare soil, although not informing on the size of patches. This is because each pixel in NDVI and NDVIc images provides an index value representing average vegetation density, which is proportional to the absorption of photosynthetically active radiation (Sellers, 1985). Therefore, for a uniformly growing rehabilitated area, the index values in every pixel, should increase uniformly with time following the mean vegetation index. This trend was observed for non-patchy rehabilitated areas.

The existence of large differences in minimum and maximum index values for pixels of the vegetation index maps indicates the presence of a bare patch(s) (Table 4.), which correspond to the near-minimum values. For a rehabilitated area up to 2 years old, a vegetation index map containing a minimum value of ~ zero is normal. With time vegetation covers bare areas so the minimum value should increase and the range of index values should decrease. This trend was observed for non-patchy rehabilitated sites. For patchy sites, persistent bare patches maintain a minimum values at or close to zero. This increases range (or maintains consistent highly range), until bare patches become vegetated. This trend becomes noticeable particularly after 2 years of rehabilitation growth. Analysis of NDVI and NDVIc values for rehabilitated areas from the time of rehabilitation onward can indicate the possibility of bare patches, allowing early remediation to be carried out, before site access is restricted. For all patchy rehabilitated areas, NDVI values from Quickbird for year 2004 showed similar results to Landsat TM values. Areas with bare patches had minimum values of ~zero. Quickbird has a much smaller pixel size (0.6m) versus Landsat TM (25m) allowing smaller bare patches to be detected. Groundtruthing confirmed this interpretation and helped us determine the causes of bare patches. Factors included human intervention (insufficient fertiliser, lack of seeds, tracks), plant disease, death of plants and rehabilitation areas located on low points or pit sumps, where seasonal waterlogging killed vegetation. Further testing is needed to determine the various causes of individual bare patches.

Figure 3. Part of Huntly study site displayed as (a.) aerial photograph and (b.) NDVI map (Landsat TM) showing a non-patchy rehabilitated area, a patchy rehabilitated area and native forest (Scale 1: 50000).

Figure 4. (a.) NDVI, (b.) NDVIc, (c.) RADARSAT and (d.) GIS map for year 1995. (Scale 1:50000). Legend for GIS coverages: road = grey (line) year of rehabilitation colours: 1989=yellow; 1990=pink; 1991=brown; 1992=green; 1993=blue; 1994=grey; and 1995=red.

From the RADARSAT image, young rehabilitated areas up to two years old can be distinguished from the older rehabilitated and native forest (Figure 4.). Bare patches within rehabilitated areas that are visible from the aerial photographs and vegetation index maps cannot be recognised from radar imagery. Radarsat is incapable of detecting small bare patches (particularly with size less than 5000 m2) because surrounding trees cause too much speckle noise. This lack of sensitivity and many cloud-free days in Western Australia makes radar imagery redundant for these applications.

Conclusions

NDVI may be a more sensitive vegetation index than NDVIc for monitoring vegetation on rehabilitated bauxite mines. For recognising bare patches, both vegetation indices are adequate. The minimum, maximum and range values of indices computed by ESRI Spatial Analyst help delineate bare patches and allow managers to monitor the changes in the bare patch itself (e.g. in size, shape, vegetation growth) over time and to provide timely remediation. The high spatial resolution (0.6m) of Quickbird data will enable the relationship between NDVI and ground LAI, to be developed at convenient spatial scale. Further investigation of NDVI derived from Quickbird is therefore highly recommended so as to provide a robust tool for evaluating rehabilitation performance.

Acknowledgements

We acknowledge the support of the DLI, Western Australia for providing the aerial photography and Landsat TM imagery. Wesfarmers CSBP limited provided initial NDVI data at the commencement of this project. We also thank Alcoa, particularly Dr. Ian Colquhoun for allowing us to publish this research.

References

Apan A.A, Raine S.R and Paterson M.S (2002) Mapping and analysis of changes in the riparian landscape structure of the Lockyer Valley catchment, Queensland, Australia. Landsc. & Urb. Plng. 59, 43-57

Bastin G.N, Ludwig J.A, Eager R.W, Chewings V.H and Liedloff A.C (2002) Indicators of landscape function: comparing patchiness metrics using remotely-sensed data from rangelands. Ecological Indicators 1, 247-260

Bhuju D.R and Ohsawa M (1999) Species dynamics and colonization pattern in an abandoned isolated planted forest in an urban landscape. Ecol. Res. 14, 139-153

Boyd D.S, Wicks T.E and Curran P.J (2000) Use of middle infrared radiation to estimate the leaf area index of boreal forest. Tree Physiology 20, 755-760

Connell J.H (1978) Diversity in tropical rain forests and coral reefs. Science 199, 1302 - 1310

Davenport M.L and Nicholson S.E (1990) On the relation between rainfall and the Normalised Difference Vegetation Index for diverse vegetation types in East Africa. Int. J. of Rem. Sensg. 14, 2369-2389

Eklundh L, Hall K, Eriksson H, Ardo J and Pilesjo P (2003) Investigating the use of Landsat thematic mapper data for estimation of forest leaf area index in southern Sweden. Can. J. Rem. Sensg. 29, 349-362

Kaufman Y.J and Remer L.A (1994) Detection of forest using mid-IR reflectance: An application for aerosol studies. IEEE Transactions on Geoscience and Remote Sensing 32, May 1994

Lucas R.M, Cronin N, Milne A.K, Witte C, Denhan R, Dong Y (2000) Carbon losses from land cover change in Australia: A role for SAR. IEEE 2525-2527

Nemani R, Pierce L, Running S and Band L (1993) Forest ecosystem processes at the watershed scale: sensitivity to remotely-sensed leaf area index estimates. Int. J. of Rem. Sensg. 14, 2519-2534

Petraitis P.S, Latham R.E and Niesenbaum R.A (1989) The maintenance of species diversity by disturbance. Quat Rev. Biol. 64, 393-418

Pickett S.T.A and White P.S (Eds.) (1985) The Ecology of Natural Disturbance and Patch Dynamics. Academic Press, Orlando

Rouse J.W.J, Haas R.H, Schell J.A and Deering D.W (1974) Monitoring vegetation systems in the Great Plains with ERTS. In Proceedings of the 3rd Earth Resources Technology Satellite-1 Symposium, 10 – 14 Dec. 1973, Washington D.C. Edited by Freden S.C, Marcanti E.P and Becker M.A. NASA SP-351, Scientific and Technical Information Office, NASA, Washington D.C., 309-317

Sader S.A and Winne J.C (1992) RGB-NDVI colour composites for visualizing forest change dynamics. Int. J. of Rem. Sensg. 13, 3055-3067

Sellers P.J (1985) Canopy reflectance, photosynthesis and transpiration. Int. J. Rem. Sensg. 6, 1335-1372.

Sellers P.J, Berry J.A, Collatz G.J, Field C.B and Hall F.G (1992) Canopy reflectance, photosynthesis and transpiration III. A reanalysis using improved leaf models and a new canopy integration scheme. Int. J. of Rem. Sensg. 42, 187- 216

Tanaka S, Sugimura T, Senoo T and Nishi T (1998) Forest stand structure analysis in Mt. Fuji area with ops and SAR data. Adv. Space Res 22, 685-688

Tucker C. J (1979) Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment 8, 127-150

Van de Koppel J, Rietkerk M, Van Langevelde F, Kumar L, Klausmeier C.A, Fryxell J.M, Hearne J.W, Van Andel J, Ridder N.D, Skidmore A, Stoosnijder L and Prins H.H.T (2002) Spatial heterogeneity and irreversible vegetation change in semiarid grazing system. The American Naturalist 159, 209-218

Wallace J (1996) A report to Alcoa: Relationships between Satellite Image Data and Leaf Area Index in Jarrah Forest. CSIRO Mathematical and Information Sciences.

Wang J, Rich P.M and Price K.P (2003) Temporal responses of NDVI to precipitation and temperature in the central Great Plains, USA. Int. J. of Rem Sensg. 24, 2345-2364

Wilson E.H and Sader S.A (2002) Detection of forest harvest type using multiple dates of Landsat TM imagery. Remote Sensing of Environment 80, 385-396

Woral G.A (1960) Patchiness in vegetation in the northern Sudan. Journal of Ecology 48, 107-117

Previous PageTop Of PageNext Page