You are on page 1of 11

Remote Sensing of Environment 114 (2010) 18451855

Contents lists available at ScienceDirect

Remote Sensing of Environment


j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / r s e

Assessing rangeland degradation using multi temporal satellite images and grazing
pressure surface model in Upper Mustang, Trans Himalaya, Nepal
Keshav Prasad Paudel , Peter Andersen
Department of Geography, University of Bergen, Fosswinkelsgate 6. N-5007, Bergen, Norway

a r t i c l e

i n f o

Article history:
Received 4 November 2009
Received in revised form 17 March 2010
Accepted 21 March 2010
Keywords:
Rangeland
Degradation
Remote sensing
Grazing pressure
Residual trend
Cost surface
Trans Himalaya

a b s t r a c t
This study aims to map and discriminate rangeland degradation from the effects of precipitation variability and
thereby identify the driving forces of degradation in the grazing areas of Ghiling in Upper Mustang, Nepal.
Landsat MSS, TM, ETM, and SPOT images covering the years 19762008 were analyzed. 8 km resolution NOAA
NDVI from 1981to 2006 were used to identify the long term interrelationship between vegetation greenness
and precipitation variability. The use of time series residual of the NDVI/precipitation linear regression to
normalize the precipitation effect on vegetation productivity and identify the long term degradation was
extended at the local scale. A weighted grazing pressure surface model was developed combining information
from satellite images, topography, forage availability and detailed eld work data on points of livestock
concentration, herders' ranking of forage quality and grazing pattern in each pasture unit. The grazing pressure
of a given site was dened as the product of annual net stocking density and the inverse of the total friction
of livestock movement. While annual precipitation was found as the dominant factor for the interannual
vegetation variability, degradation in Upper Mustang was the result of grazing induced change and some
localized natural processes.
2010 Elsevier Inc. All rights reserved.

1. Introduction
Temporal variation in rangeland productivity is a function of
climatic and anthropogenic factors. Other natural factors like vegetation ecosystem, topography and soil/hydrology characteristics can be
assumed as relatively constant within annual to a few decadal time
scales (Di et al., 1994). Among the climatic variables in arid and semi
arid environments, precipitation variability has been found to be the
primary determinant for rangeland vegetation dynamics (Le Hourou
and Hoste, 1977; Le Hourou et al., 1988; Nicholson et al., 1990; NoyMeir, 1973; Wang et al., 2001; Whittaker, 1970). High interannual
precipitation variability in arid/semi arid environments represents an
external disturbance to rangeland ecosystems resulting in high variation in vegetation cover.
Changes in vegetation cover or productivity derived from a time
series based satellite images vegetation index (VI) have been widely
used to map, quantify and analyze vegetation change and rangeland
degradation (Anderson et al., 1993; Deering et al., 1975; Liu et al.,
2004; Lyon et al., 1998; Myneni et al., 1997; Pettorelli et al., 2005;
Pickup et al., 1998; Singh, 1989; Tucker, 1979; Tueller, 1989). However, because of the high interannual precipitation variability in semi/
arid environments and its effect on rangeland production trends, a
description of trends in the vegetation cover only is insufcient to

Corresponding author. Tel.: +47 98683685; fax: +47 55583099.


E-mail address: keshav.paudel@geog.uib.no (K.P. Paudel).
0034-4257/$ see front matter 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.rse.2010.03.011

characterize degradation in rangeland (Pickup et al., 1998). It is


essential to distinguish between degradation and changes due to
uctuation of precipitation. In the landscapes where precipitation is
the only determinant for biomass productivity, vegetation can be
naturally restored following good rainfall/snowfall. Looking at such
reversible changes following precipitation, decreasing vegetation
productivity due to declining precipitation can be categorized as
vegetation variability or what Bai et al. (2008) termed false alarm
rather than degradation. In contrast, a decreasing response of vegetation to precipitation over time indicates an inuence of some long
term localized processes, which reduce the ability of those landscapes
to recover after the changes. Such long term reduction in rangeland
productivity systems implies decreasing ecosystem resilience and has
been dened as rangeland degradation (Pickup, 1996; Pickup et al.,
1998). Spatial variation of grazing is generally considered as a long
term process reducing the resilience of specic geographical units.
In addition, active erosion processes in a land unit may also inuence
the spatial pattern of degradation. Hence, for a better understanding
of the underlying driving forces of rangeland degradation in dry
environments, it is necessary to identify and normalize the effect of
precipitation on the rangeland vegetation trends.
This study aims to discriminate long term degradation from precipitation driven vegetation variability in a Trans Himalayan rangeland. The Trans Himalayan rangelands of Nepal are not just a resource
for livestock grazing, but have also been regarded as a storehouse
for biodiversity including many endangered species. The area also
encompasses the headwaters of the major river systems of Nepal and

1846

K.P. Paudel, P. Andersen / Remote Sensing of Environment 114 (2010) 18451855

the Indian subcontinent. It has been suggested that these high altitude
rangelands are severely threatened by rapid degradation due to overgrazing (NTNC, 2008:78). However, such general concepts advocated by policy makers and development agencies often rely rather on
estimations than on empirical nding. Has the Trans Himalayan
rangeland really degraded towards such a critical level?
To identify and normalize the effect of rainfall variability on vegetation productivity, residual trend method has been used recently
(Archer, 2004; Evans and Geerken, 2004; Herrmann et al., 2005;
Wessels et al., 2007). The residual trend method is based on linear
relation between VI and accumulated precipitation. Residual (VIres), the
difference between observed and predicted VI, theoretically represents
the part of the observed VI which is not explained by precipitation.
Thus, a signicant negative trend in time series residuals represents a
rangeland degradation excluding the precipitation effects. One major
problem in this method is that the trend is measured against a linear
regression of all the data, including degraded, so the non-degraded
relationship with precipitation is, in fact, the average of non-and degraded data. Thus in sites degraded before the time series started,
the observed relationship between vegetation index and precipitation
will underestimate the expected production for a given amount of
precipitation and consequent identied degradation magnitude (Prince
et al., 2007; Wessels et al., 2007). However, the calculated interannual
trends is not affected, so long as VIres is independent of rainfall which
indicates the xed reduction in vegetation production independent
of precipitation due to degradation (Evans and Geerken, 2004; Wessels
et al., 2007).
In dry Trans Himalayan rangelands, where seasonal transhumance
and rotational grazing are practiced, grazing and related phenomena
are concentrated around some point of livestock concentration (PLC)
like encampment sites (sheds), water points, livestock trails and
settlements. In addition to these factors, topography and range quality
also inuence the spatial pattern of livestock movement (Cingolani
et al., 2008; Rder et al., 2007). Rder et al. (2007) extended the
gradient concept (Pickup and Chewings, 1994) to a cost surface model,
incorporating most of the above factors which determine spatial
distribution patterns. However, the cost surface model still lacks the
grazing management variables such as livestock density, grazing
pattern and total grazing days in a paddock etc. Without incorporating
net annual stocking density, as the function of size of subpasture units,
forage availability, livestock population of each animal species and
their respective total grazing days in a year (Holechek et al., 2001;
Wehn, 2009), cost surface model alone cannot represent the grazing
pressure. Similarly, only the function of demand and supply does not
represent the spatial pattern of grazing pressure within subpasture
units. This study also aims to develop a weighted grazing pressure
surface model combining net annual stocking density with friction
surface, for the assessment of grazing induced rangeland degradation.
This study aims at mapping and quantifying the extent of rangeland
degradation and identies the driving forces of degradation at the local
level. Thus it presents a methodology that could be used in support
of decision making at community level rangeland management in
the region. It also provides an example of extending existing GIS/
remote sensing analysis framework, incorporating information from
oral history and eld observation.
2. Study area
The Trans Himalaya region (THR) of Nepal, the rain shadow of the
great Himalaya, lies between the Himalayan range and the Tibetan
plateau. The rangelands of Ghiling, located approximately between
2857 to 2903N and 8347 to 8355 E in Upper Mustang in the
Annapurna Conservation Area, have been selected as a test site for
this study (Fig. 1). It is situated between 3000 and 5300 m above sea
level (asl), has an alpine cold, arid to semi arid climate with mean
annual precipitation of 153 mm. The coefcient of variation of annual

precipitation during 1973 to 2008 amounts to 49%. Most of the precipitation occurs during the monsoon (JunSep) as rainfall and during
winter (DecFeb) as snowfall. Mean maximum/minimum temperatures during 19702008 were recorded as 20.8/11.9 in monsoon, 15.6/
3.5 in post-monsoon (OctNov), 11.4/1.02 in winter and 17.1/4.8
in pre-monsoon (MarMay). The seasonal trends in mean temperature indicate a warming in monsoon (0.017 C y 1) and winter
(0.003 C y 1) and cooling in pre-monsoon ( 0.013 C y 1) and
post-monsoon ( 0.016 C y 1). However, these trends are based on
only two stations' last 38 year data, so caution needs to be taken
for further interpretation. Snow cover lasts for 45 months, from
November to March. Owing to snow cover and very low temperature
during post-monsoon to the rst month of pre-monsoon, the high
altitude areas are characterized by a very short growing season. Oral
history from local people evidenced a decline in snow fall and a
substantial decline in ice/snow cover area in the region. Snow-melt
and glacier fed irrigation canals are important sources of soil moisture
for agriculture and rangeland productivity.
Covering 40% of the total area, rangeland is the major land resource
in the region. The vegetation cover exhibits a clear distinction of
belts across altitude. Areas below 4100 m asl comprise shrubberies
and herbs dominated by Caragana spp., Juniperus spp., Lonicera spp.,
Artemisia spp., Rosa spp., and Stipa spp. between 4100 and 4300 m asl a
mixed belt of alpine grassland with shrub and dwarf shrub occurs,
above 4300 m asl there is an alpine grassland dominated by Kobresia
spp., Carex spp. and Stipa spp. Because of sufcient moisture from snow
melt and mist, grass cover above 4300 m asl generally exceeds 85%,
while below 4100 m asl vegetation is characterized by a windblown
Caragana Gerardiana steppe with a vegetation cover of generally less
than 60%.
Ghiling was selected for this study because it is the village with the
highest number of total livestock units including mountain goats,
which is the main source of livestock income of the Upper Mustang
valley (NTNC-ACAP, 2005). Considering the importance of livestock in
their livelihood, the rangeland degradation has been of great concern.
Lulu cattle1, Jhopa2, mountain goat, horse and mule are the main
livestock types reared in Ghiling (Fig. 1). Apart from livestock from
Ghiling, goats from Tange village graze in Dhowa pasture for three
and half months during winter (see Fig. 2a). Between Ghiling and
Chhusang a common pasture is located, called Ripima. Livestock of
both villages graze in this area during summer. Similarly a common
pasture exists between Ghiling and Tange, called Piri and livestock
from both villages graze there during winter.
Local agro-pastoralists divide the rangeland into several smaller
units, called Ri (pasture), on the basis of spatial variation of vegetation
growth pattern, plant community and use pattern, and landscape unit.
Ghiling village comprises a total of 61 subpasture units (Fig. 2a). Local
people have a clear distinction between winter, summer and intermediate grazing pastures with a strictly regulated movement of livestock according to an agro-pastoral calendar. Grazing in Ghiling, like
in other THR, is based on a seasonal transhumance rotation practice,
as a mix of seasonal deferment, transhumance and rotational stocking
grazing management strategies (Paudel, 2006). Thus there are some
encampment sites in summer and winter pasture areas, which act
as the primary PLC (Fig. 2b). Animals are led to pastures in the
morning and returned to their sheds in the evening. Not all agropastoralists of Ghiling own a shed at pasture for their animals and
bring their herd home in the evening. In addition, all animals are
brought to the village in the evening during 4 to 5 months in the
summer. Being a mountainous and dry region, there are few water
points where livestock activities are concentrated as well. In addition

1
Humpless dwarf cattle (Bos Taurus) found in Mustang district of Nepal (see
Kumiko et al. (2004) for detail).
2
Male cross breeding of male yak and common cow.

K.P. Paudel, P. Andersen / Remote Sensing of Environment 114 (2010) 18451855

1847

Fig. 1. Location of the Upper Mustang; the box outlines the study site Ghiling; the table at the bottom right shows the livestock population of Ghiling village.

animal activities are also concentrated along major livestock trails


(Fig. 2b). Livestock uses these major trails as primary routes of travel
between settlement, shed and pasture. Thus, the distance to these

trails also represents decreasing livestock pressure and associated


activities including shrub and scrub uprooting and re wood collection (Cingolani et al., 2008).

Fig. 2. Pasture units and points of livestock concentrations (a) pasture unit boundary (b) points of livestock concentrations (shed/ encampment site, water point, major livestock trail
and settlement location).

1848

K.P. Paudel, P. Andersen / Remote Sensing of Environment 114 (2010) 18451855

3. Methodology
3.1. Remote sensing data
The rangeland vegetation growth peak period in Upper Mustang
occurs during the rst half of September. However, this period often
corresponds to high cloud covers and thus limits the selection of cloud
free images. All available Landsat-Multi Spectral Scanner (MSS),
Thematic Mapper (TM) and Enhanced Thematic Mapper (ETM+)images in USGS archives were reviewed for the post-monsoon period
and cloud free (b10%) images were selected (Table 1). Because of
the lack of cloud free TM image for post-monsoon 2008, the year
when we conducted major eld work, SPOT 4 (XI), 20 m spatial
resolution image, was used for this year. Though most of the ETM+
data from 2003 showed a cloud cover of less than 10%, these were
not used for the trend analysis due to the ETM + SLC failure in 2003.
However, we used ETM+ images of May 2008 to analyze effectiveness of VIs compared with ground measured vegetation cover and
of October 2008 to asses comparability of SPOT 4 (20 m resolution)
image.
Twenty-six years (July 1981December 2006) Normalized Difference Vegetation Index (NDVI) images at 15 day intervals were
acquired from the global inventory modeling and mapping studies
(GIMMS) AVHRR 8 km, bimonthly (19812006) data set, available
at bftp://ftp.glcf.umiacs.umd.edu/glcf/GIMMS/N (Pinzon et al., 2004;
Tucker et al., 2005).

3.2. Pre-processing
The Landsat ETM+ captured on 31 October 2001, which has good
visibility and is almost cloud free, was rst rectied with 56 ground
control points of topographic map (1:50,000). All other images were
co-registered with this image with at least 38 control points as links.
Using the information provided in the header le and following
the equations and calibrating coefcients of Chander et al. (2009) for
Landsat images and following Soudani et al. (2006) for SPOT image,
all images were converted to at-sensors radiance. Following Soudani
et al. (2006) we used the dark object subtraction (DOS) approach
(Chavez, 1996; Schroeder et al., 2006) to minimize the noise due
to atmospheric effects on satellite radiance and calibrate at-sensor
radiance to scaled surface reectance. After geometric, radiometric
and atmospheric correction of high resolution images, all images
were masked out by study area boundary for further analysis. Only
rangeland area, excluding area covered by the main village and Kali
Gandaki River, was selected for analysis.
The NDVI, which is most commonly used to monitor green vegetation dynamic (Lyon et al., 1998; Shabanov and Myneni, 2001;
Tucker, 1979; Tucker and Sellers, 1986), was computed for each high
resolution images. Very high correlation (0.907, n = 50) was found
between the computed NDVI (May 2008) and visually estimated
percent green vegetation cover in the eld (Fig. 3).

Fig. 3. Regression relationship between vegetation cover (%) estimated on the ground
and ETM+ derived NDVI. Field estimation was conducted during 19 to 28 of May 2008.

3.3. Field data collection


Field work was conducted by the rst author during AprilJuly in
2008 and in April/May 2009. During the eld visits, data have been
collected by means of a detailed household survey of all 59 households of Ghiling village, semi to unstructured interviews with key
persons in the village and group discussions with herders and local
agro-pastoralists. Questions included livestock population, composition and trend, grazing patterns, rotation and management, uprooting
of shrubs and processes of rangeland changes.
During the group discussion, to identify pasture unit boundaries
and major trails of livestock movements, participants were asked to
mark relevant information in the maps, using digital display of Google
Earth and printed topographic maps (1:50,000) as reference. The
herders were also asked to mark and divide pastures in to different
subunit areas, according to their experience, particularly attractive, in
terms of perceived forage quality, for livestock. These subunits were
ranked from 100 as the most attractive to 10 as the least attractive.
The rst author drew the pasture boundaries, subunits and trails using
the drawing tools of Google Earth to record the information provided
during group discussion and the mapping exercise was instantly
veried by participants. Subsequently, these sketches were saved and
exported to a GIS. Later these layers were nalized via verication
with DEM, GPS observation points and a topographic base map.
Location of all water points and livestock shed were recorded with
the help of a GPS. Based on data from household survey and group
discussions, total livestock grazing in each pasture unit and grazing
duration for each livestock type was calculated. All livestock was
converted to sheep units (SU) based on the standard body weight

Table 1
Dates, sources and characteristics of satellite images used in this study.
Date
1976
1990
1999
2000
2001
2006
2008
2008
2008

Nov 16
Nov 10
Oct 10
Sep 26
Oct 31
Nov 22
May 27
Oct 18
Oct 16

Source

Resolution (m)

Path/row

Cell center lat/Long ()

Sun elevation ()

Sun azimuth ()

Landsat MSS
Landsat TM
Landsat ETM+
Landsat ETM+
Landsat ETM+
Landat TM
Landsat ETM+
Landsat ETM+
SPOT 4

60
30
30
30
30
30
30
30
20

153/040
142/040
142/040
142/040
142/040
142/040
142/040
142/040

28.6/83.5
28.9/84.1
28.9/84.1
28.9/84.1
28.9/84.1
28.9/84.0
28.8/83.9
28.9/83.9
28.9/83.7

34.0
37.0
49.3
52.8
42.3
37.3
66.7
46.1
48.0

143.7
145.0
146.9
140.5
151.6
155.9
103.2
148.5
152.4

K.P. Paudel, P. Andersen / Remote Sensing of Environment 114 (2010) 18451855

(Ghimire, 1992; Joshi, 1992; Kumiko et al., 2004; Lise et al., 2006), which
determines: 1 goat/ sheep = 1 SU; 1 horse/mule= 7 SU; 1 Jhopa= 6 SU;
1 Lulu cattle= 5 SU.
3.4. Vegetation change analysis
In order to identify long term change in vegetation production in
the study area, the NDVI from high resolution images were regressed
over time (Fig. 4). We applied a linear regression analysis. Pixels with
a negative slope of the regression thus indicate areas of declining
vegetation cover. Pixels exhibiting marginal decreases (i.e. b5%)
during the 32 year period (19762008) were excluded or considered
as stable. The T-test was computed to assess signicance of negative
slope.
3.5. Precipitation data preparation and identication of precipitation
signal
We collected available 13 meteorological stations (Mustang and
Manang district) daily precipitation data and two nearby stations'
(Thakmarpha and Jomsom) monthly temperature data of period 1970
2008 from the Department of Hydrology and Meteorology (DHM),
the Government of Nepal. There is no meteorological station within
the study area but two precipitation stations are located very close to the
study areaGhami in the northern side (about 1 km) and Samar Gaon
in the southern side (about 2.5 km).
Following Evans and Geerken (2004) we computed the correlation
between various amounts of precipitation with different time range
accumulation and NOAA NDVI in order to identify the relationship
between precipitation and vegetation production in THR of Nepal.
The correlation analysis was performed using pairs of accumulated
precipitation of each station with the pixel value of NOAA NDVI where

1849

the station is located. Stations are well distributed, locating different NOAA NDVI pixels, from south to north altitude ranging from
2384 m asl to 3705 m asl. For the correlation computation, all combinations of precipitation accumulation length ranging from 15 days
to 14 months and lag time ranging up to 3 months (with 5 days
increment) was done for each bimonthly NDVI value. 12 month accumulation precipitation with 15 days lag time was found the best
precipitation accumulation period for the vegetation production of
September to November. We also examined the correlation between
seasonal total precipitation and NOAA NDVI maximum value during
September to November.
The daily precipitation of each station was summed every 365 day
period with 15 days lag time from the captured date of high resolution
images. This accumulated precipitation of all available stations was
used for interpolation using the Inverse Distance Weighted (IDW)
method (Hartkamp et al., 1999; Mitas and Mitasova, 1999). Each
interpolated precipitation layers was then masked with study area
boundary. Using interpolated accumulated precipitation grids, we
calculated linear regression between NDVI (high resolution images)
and accumulated precipitation for each pixel. Time series residual
(NDVIres), i.e. observed NDVIprecipitation predicted NDVI, is considered as non-precipitation-triggered time series vegetation production (Geerken and Ilaiwi, 2004). Computing linear regression between
these time series residual and time, the trend of non-precipitation
triggered vegetation was derived for each pixel. Declining trends
through time present in the residuals then indicates changes in vegetation response other than precipitation effects. Pixels exhibiting
marginal decreases (i.e. b5%) in 19762008 were excluded or considered as stable because they reect only uncertainties possibly
caused by difference of image dates within the season/month and the
image calibration processes. Signicance of the trend was assessed by
the T-test.

Fig. 4. Flowchart illustrating steps to identify driving forces of rangeland degradation.

1850

K.P. Paudel, P. Andersen / Remote Sensing of Environment 114 (2010) 18451855

the forage supply than only the area of pasture unit. Thus, annual
net grazing density (Gden) of each pasture unit was calculated as:

3.6. Developing grazing pressure surface


Distance surface layers for each PLC (shed, water points, livestock
trails and settlement) were prepared calculating straight line distance
from each identied PLC in the study area. All subpasture units do not
have their own sheds and water points (Fig. 2) and during the stay
in a given encampment site livestock graze in a group of pasture units.
A group of pasture units may have multiple sheds. To represent such
grazing pattern, we rst created distance layers for a group of pasture units calculating distance from shed(s), from where livestock
graze in those pasture units. Finally, all subshed distance layers were
merged to create one distance layer for the whole study area. Similarly, subdistance layers of water point were also created for a group
of pasture units according to water points use pattern, and were
merged to prepare a single water point distance layer for the whole
study area. Access to water points during grazing in a given pasture
unit is also determined by geographical barriers in a landscape like
cliffs, and deep gullies. During calculation of distance from water
point, a barrier function was also included. For settlements and
livestock trails, we created distance layers for the whole study area.
Each pixel in distance layers represents the shortest distance from the
given PLC. Four distance cost surface layers were calculated converting each distance surface layers into cost surface using a simple linear
transfer function (Eq. (1)) as:
i

Cdist = 1 + 99 dist = Maxdist

where, Cidist is the distance cost surface of each given PLC (i), that
is shed, water point, livestock trails and settlement, disti is the pixel
value in distance layer of i and Maxidist is the maximum distance value
in distance layer of i. The value of distance cost surface for each PLC
ranges from pixel value 1 to 100, where pixel value 1 represents the
cell where given PLC is located i.e. no effort is needed for livestock
spatial movement and cost value 100 represents the maximum effort
require to reach that pixel.
From the elevation contour with an interval of 20 m, we constructed a digital elevation model (DEM) with 30 30 m resolution
and obtained slope in degree. Slope was considered as a topography
factor determining spatial movement of livestock. Slope values (in
degree) were converted to cost surface (0100) as:
Cslope = 100 slope = 90

where, Cslope is the friction surfaces of topography based on slope,


where 90 slope corresponds to cost of 100 and relatively at surface
corresponds to very low cost.
Based on the herders ranking (10100) of range quality (rattract),
we developed friction surface (Cattract) of attractiveness as:
Cattract = 100 rattract :

All individual cost surfaces were combined by calculating the


average value for each pixel. The resulting surface was dened as total
cost (tcost).
Considering the seasonal transhumance and rotational grazing
practices in THR, crude livestock density in a pasture unit cannot
represent the actual annual livestock pressure. So, we computed
annual net grazing density as a function of total livestock units and
grazing days in a given pasture unit and total forage supply in that
pasture unit. Total forage production in a given pasture unit is determined by the percent grass cover in that site. If we assume the
proportional relation between forage production and percent vegetation cover, a pasture unit having 100% vegetation cover can have equal
forage production to twice as large a pasture unit having 50% vegetation cover. Hence, net vegetation cover area (vcov_a) can represent

Gden = SUgd = Vcov a

where
SUgd x = SU x gd x = 365

vcov a = vcov A pu = 100:

SUgd_x is the annual grazing sheep unit (SU) days for livestock
type x (i.e. cattle, horse/mule, goat, Jhopa etc.); SU_x is the product of
total population of livestock type x in sheep unit in a given site and
gd_x is the total number of days in a year they grazed in that given
pasture unit; SUgd is the sum of annual grazing sheep unit of all
livestock type grazing in a given pasture unit, vcov_a is the net
vegetation cover area; A_pu is the area of given pasture unit and vcov
is the average percent vegetation cover derived from time series NDVI
of high resolution images used in this study. We used regression
equation (Fig. 3) derived from the relation between eld estimated
vegetation cover and NDVI to calculate vcov.
Finally the weighted grazing pressure surface model (GP) was
derived by multiplying the inverse of total cost surface with the net
livestock grazing density as:
GP = Gden 1 = tcost:

Identied degradation patches derived from residual trend method


were carefully examined with reference to grazing pressure surface.
In addition, results were checked against the information from eld
observation and in-depth interview and group discussion.
4. Results
4.1. The relationship between precipitation and vegetation
In Upper Mustang we found a very strong relationship (r = 0.823,
p b 0.001) between each 15 days NOAA NDVI between September and
November (19812006) and the 12 months accumulation precipitation ending 15 days before the date of each NDVI image. The correlation between seasonal precipitation and maximum NOAA NDVI
values in Upper Mustang shows the strong inuence of monsoon, premonsoon and winter precipitation to the vegetation production of
September to November (Table 2). Fig. 5 highlights the relationship
between accumulated precipitation and maximum NOAA NDVI values
between September and November in Ghami station, the observation
station nearest to the study area.
The correlation between accumulated precipitation and NDVI
values derived from high resolution images reveals a strong positive
correlation (r = 0.6 to 0.99) for about 66% of the rangeland area in
Ghiling. However on south facing steep cliffs along Kali Gandaki and
its' tributary Kolang Kolang river, an area of very low vegetation cover
(b10%) and where soils have been eroded due to snowmelt water
and monsoon rainfall, a negative correlation was found (Fig. 6).

Table 2
Correlation coefcients (r) between NOAA NDVI maximum (SeptemberNovember)
and seasonal precipitation.
Total precipitation

NOAA NDVI maximum


(SeptemberNovember)
p b 0.01.

Monsoon

Pre-monsoon

Post-monsoon

Winter

0.818

0.801

0.359

0.741

K.P. Paudel, P. Andersen / Remote Sensing of Environment 114 (2010) 18451855

1851

Fig. 5. Relationship between maximum NOAA NDVI (SeptemberNovember) and accumulated precipitation (12 months 15 days earlier to date of maximum NDVI) in Ghami.

4.2. Long term vegetation change and rangeland degradation


As expected for a semiarid/arid environment, vegetation production uctuates strongly according to interannual precipitation
variability. A linear regression analysis of NDVI derived from the
high resolution images from1976 to 2008 reveals that about 20.2%
area of the total rangeland of Ghiling shows a declining trend of
vegetation production (Fig. 7a). The pattern of declining NDVI is
patchy rather than continuous. The declining patches of vegetation
were found mainly in the southeastern pastures (winter grazing). In
contrast, the high altitude summer grazing pastures in the northwestern belt, which benet from the effect of mist, showed mainly a
positive or stable trend of long term vegetation productivity. Fig. 7b
shows the signicance of negative trend found in study area. As
expected, with areas showing negligible declining trends in NDVI
coincide with insignicant results.
According to the analysis some 2.4% of the study area was found
having worse response to the precipitation during 19762008
(Fig. 8a). Fig. 8b highlights that the trends found in the residuals are

signicant over most of the area. A comparison of temporal trends


in NDVI with the temporal trends of NDVIres reveals that most of the
change in vegetation can be explained by precipitation variability
(Figs. 7a and 8a). The correlation between residual NDVI and precipitation was calculated for each pixel and no relationship found
between precipitation and NDVIres (Fig. 9). Some clusters of pixels
that were identied as stable or improving in NDVI trend were found
to decline in the residual trends. This indicates that although NDVI
in those areas increased over time, their response to precipitation
has been declining over time. The spatial pattern of degradation is
generally sporadic however four clusters of degradation can be identied: a) large cluster at north eastern part at Dhowa pasture b)
patches of narrow linear clusters in the south at northern part of
Kali Gandaki river at Rajmangwa pasture, c) near to Ghiling village
in north side and d) winter grazing pasture in eastern side at Ngaila
and Shere pasture.
4.3. Weighted grazing pressure surface and driving factors of degradation
The weighted grazing pressure surface index ranges from 0.35 very
low to 20.28 very high grazing pressure surface (Fig. 10a). As expected
a high grazing induced degradation was in identied degraded patches
after removal of precipitation effect, the largest degraded cluster at
Dhowa pasture, and cluster near village and at Syangmochen highly
coincide with identied high grazing pressure clusters (Fig. 10b).
However, the linear degraded patches along the Kali Gandaki River and
some of the sporadic degraded patches were found in low grazing
pressure areas indicating other driving factors for degradation than
human pressure or declining precipitation. When comparing degraded
patches with topography, the linear clusters of degradation patches in
the southern belt along the Kali Gandaki river were found to be located
on very steep slopes (slope N 50) (Fig. 11). Field observations suggest
a strong wind effect along the Kali Gandaki River and associated high
wind erosion.
5. Discussion

Fig. 6. Correlation between High resolution NDVI and accumulated precipitation.

Available soil moisture, a key determinant for actual evapotranspiration in dry region is primarily a function of accumulated precipitation over certain periods of time rather than instantaneous
precipitation amount. Farrar et al. (1994) found strong correlation
between available accumulated precipitation and soil moisture in dry
rangelands. Different precipitation accumulation period, ranging from

1852

K.P. Paudel, P. Andersen / Remote Sensing of Environment 114 (2010) 18451855

Fig. 7. Overall negative trend in vegetation greenness based on high resolution NDVI (19762008). (a) Decreasing vegetation production. (b) Signicance of declining trends in the
high resolution NDVI. Trends are termed insignicant for pixels in which p N 0.1.

2 to15 months, have been reported in different regions (Geerken and


Ilaiwi, 2004; Nicholson et al., 1990; Nicholson et al., 1998; Richard
and Poccard, 1998; Wang et al., 2001; Yang et al., 1997). Reported
such variation in relation between NDVI and precipitation in different
regions highlights the importance of geo-climatic condition of the given
region. 12 months summed precipitation with 15 days lag time as the
best precipitation accumulation period for maximum NDVI of September to November was found in upper Mustang. In addition to monsoon
and pre-monsoon rainfall, a strong inuence of winter precipitation
(r = 0.741, p b 0.01) for post-monsoon vegetation production highlights
the importance of snow fall for vegetation production in the region.
Local people also have a very strong perception that snow fall is the
major determinant for rangeland vegetation.
In this study, rangeland degradation is dened as long term vegetation productivity decreases due to grazing and related phenomena

and localized natural processes. Dening degradation as the loss of


productivity incorporates agro-pastoralists' perspectives and concerns
(Stocking and Murnaghan, 2001) together with commonly used biophysical indicators for rangeland degradation (Behnke and Scoones,
1993).
According to some authors, shrub encroachment and consequent
decline of proportion of palatable species, have a great effect on
rangeland productivity (Jeltsch et al., 1997; Todd and Hoffman, 1999).
Local people and herders reported no signicant changes in palatable
and non-palatable species composition. However, the process could
be slow one. The study area is subject to a strictly controlled practice
of shrub and scrub uprooting for ensuring a sustained rewood collection. Such woody shrub uprooting in different pasture units may
be regarded as an indirect control mechanism against bush encroachment. However, more detailed eld measurement and vegetation unit

Fig. 8. Declining trend in the residual NDVI based on high resolution images (19762008) (a) Rangeland degradation after reducing precipitation effects (b) Signicance of declining
trends in the residual NDVI. Trends are termed insignicant for pixels in which p N 0.1.

K.P. Paudel, P. Andersen / Remote Sensing of Environment 114 (2010) 18451855

1853

Fig. 9. The relation between residual NDVI and accumulated precipitation.

mapping (e.g. Cingolani et al., 2004) and change analysis is required


to derive conclusion.
The strong impact of high ux of interannual precipitation
variability on NDVI is conrmed by the strong positive correlation
and linear relation with strong determinant of coefcient between
accumulated precipitation and NDVI. In addition, similar to Wessels
et al. (2007) no relation was found between precipitation and residual
(NDVIres) in the study sites which supports the use of residual trend
method. Similarly, negligible (almost zero) relationship found
between predicted NDVI (from precipitation) and residuals (NDVIres)
also conrms our basic assumption behind the use of a residual trend
method to normalize the precipitation effects from time series NDVI
(Archer, 2004). However, used residual trend method still has some
shortcomings like difculties of detecting degradation that occur
within the rst or two years of time series (Wessels et al., 2007) and
lack of proper accounting of the carryover effects of previous years'
precipitation (Wiegand et al., 2004).

Fig. 11. Degradation patches lying in very steep slope: the red boundary shows the slope
greater than 50.

In grazing pressure model we attempted to incorporate the majority of factors inuencing livestock distribution. This was achieved
by combining the cost surfaces and livestock net annual grazing
density. The livestock net annual grazing density is more robust than
generally used stocking density since the former also incorporates
total grazing days in a particular pasture unit in a year and its' area
weighted by forage availability. The model has still some short
comings: rst, we assume forage being directly proportional to
percent vegetation cover. Second, equal weighting has been given to
the components of cost. However, the effect of distance from PLC
or forage amount may not be equal. The weighted grazing pressure
surface model may furthermore be improved by using different
weight of the components of cost, derived from multiple regression analysis. The agreement and high correlation between identied

Fig. 10. Weighted grazing pressure surface (a) showing clear distinction of grazing pressure clusters (overlaid boundary line shows the pasture unit) (b) degraded rangeland patches
and grazing pressure surface (the dark line shows the high agreement between degraded patches and high grazing pressure).

1854

K.P. Paudel, P. Andersen / Remote Sensing of Environment 114 (2010) 18451855

degradation patches and high grazing pressures areas indicates that


the model is useful to identify and predict grazing induced rangeland
degradation.
The increasing trend of degradation in some patches in steeper
slope in southern belt, despite low grazing pressure, could be attributed to a changing pattern of soil water holding capacity and
actual evapotranspiratoin due to increasing wind erosion processes
and declining snow accumulation in terms of thickness and duration.
The snow accumulation in highland pastures signicantly decreases,
owing to irregularity, declining and shifting pattern (from Dec/Jan
toward Feb/Mar) of snowfall, as commonly perceived by local people.
Due to lack of information regarding the form of precipitation, we
could not verify and quantify such shifting patterns of snowfall and its
impact on vegetation productivity. However, a cooling trend in premonsoon (after winter) season and a warming trend during winter
indirectly conrm the experiences of local people. Furthermore, rapid
drying out and snow melt is another feature of south facing slopes.
Increasing wind effect in dry years, as perceived by local people,
signicantly erodes the top soil surfaces of these steeper slopes. In
active eroding sites, where natural processes of erosion occur, rates of
soil loss can be greater than rates of soil formation even with zero
human inuence (Behnke and Scoones, 1993). Such degraded patches
in very low vegetation cover in southern stiff cliff could be further
extended since monsoon rainfall and snowmelt water exacerbate the
ongoing processes.
6. Conclusions
This research extends the use of NDVI residual trend method to
identify human induced land degradation to a high resolution at local
scale, which was originally developed and used in coarse-resolution at
a regional scale (cf. Bai et al., 2008; Wessels et al., 2007). Precipitation
variability greatly inuences the interannual variation in vegetation
production in the high altitude cold arid region. The nding indicates
that rangeland degradation in Upper Mustang is not only a response
to long and short term variation in rainfall as conceptualized in the
non-equilibrium theory (Ellis and Swift, 1988; Miller, 1997; Scoones,
1994) rather is the effects of precipitation, grazing and localized
natural processes. Through a combination of information from eld
work and oral history with the remote sensing and terrain data we
were able to map high grazing pressure areas where rangeland degradation has apparently been caused by human activities (Fig. 10b).
Increasing evapotranspiration and wind erosion in the southern
steeper areas, as the inuence of strong wind and high radiation,
causes for increasing degradation trend within some patches of the
very low grazing pressure areas.
The extended grazing gradient concept and cost surface model to
the weighted grazing pressure surface model is useful to understand
the spatial dispersion of livestock pressure and to detect areas of high
grazing inuences. The methodology could also be used as a management tool for identifying optimal grazing patterns and be very
effective tool for studying grazing induced rangeland degradation.
Combining the residual trend method and weighted grazing pressure
surface model provides a tool to identify effects of precipitation on
vegetation change and understand the causes of degradation other
than precipitation. Integration of information from qualitative in-depth
interview and group discussions within the analysis framework of GIS/
remote sensing has proven to be extremely valuable for validation and
explanation of driving forces of degradation in different spatial units.
Acknowledgements
The research work was partially funded by the Meltzer Foundation
and a grant from the Research Council of Norway during rst author's
stay in YSSP/IIASA, where some part of analysis was performed.
We are grateful for the support and advice of LUC/IIASA research staff

and for the comments and inputs of Sylvia Prieler. The anonymous
reviewer is thanked for valuable comments and suggestions that
greatly helped to improve the manuscript.

References
Anderson, G. L., Hanson, J. D., & Haas, R. H. (1993). Evaluating Landsat Thematic Mapper
derived vegetation indices for estimating above-ground biomass on semiarid
rangelands. Remote Sensing of Environment, 45, 165175.
Archer, E. R. M. (2004). Beyond the climate versus grazing impasse: Using remote
sensing to investigate the effects of grazing system choice on vegetation cover in
the eastern Karoo. Journal of Arid Environments, 57, 381408.
Bai, Z. G., Dent, D. L., Olsson, L., & Schaepman, M. E. (2008). Global assessment of land
degradation and improvement :1. Identication by remote sensing. Report 2008/01.
Wageningen: ISRICWorld Soil Information.
Behnke, R. H., & Scoones, I. (1993). Rethinking range ecology: Implications for rangeland
management in Africa. In R. H. Behnke, I. Scoones, & C. Kerven (Eds.), Range ecology
at disequilibrium: New models of natural variability and pastoral adaptation in African
savannas (pp. 130). London: Overseas Development Institute.
Chander, G., Markham, B. L., & Helder, D. L. (2009). Summary of current radiometric
calibration coefcients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors. Remote
Sensing of Environment, 113, 893903.
Chavez, P. S. (1996). Image-based atmospheric correctionsRevisited and improved.
Photogrammetric Engineering & Remote Sensing, 62, 10251036.
Cingolani, A. M., Renison, D., Tecco, P. A., Gurvich, D. E., & Cabido, M. (2008). Predicting
cover types in a mountain range with long evolutionary grazing history: A GIS
approach. Journal of Biogeography, 35, 538551.
Cingolani, A. M., Renison, D., Zak, M. R., & Cabido, M. R. (2004). Mapping vegetation in a
heterogeneous mountain rangeland using Landsat data: An alternative method to
dene and classify land-cover units. Remote Sensing of Environment, 92, 8497.
Deering, D. W., Rouse, J. W., Haas, R. H., & Schell, J. A. (1975). Measuring forage
production of grazing units from Landsat MSS data. Proceedings of 10th International
Symposium on Remote Sensing of Environment (pp. 11691178). United States: Mich.
Di, L., Rundquist, D. C., & Han, L. (1994). Modelling relationships between NDVI and
precipitation during vegetative growth cycles. International Journal of Remote Sensing,
15, 21212136.
Ellis, J. E., & Swift, D. M. (1988). Stability of African pastoral ecosystems: Alternate paradigms
and implications for development. Journal of Range Management, 450459.
Evans, J., & Geerken, R. (2004). Discrimination between climate and human-induced
dryland degradation. Journal of Arid Environments, 57, 535554.
Farrar, T. J., Nicholson, S. E., & Lare, A. R. (1994). The inuence of soil type on the
relationships between NDVI, rainfall, and soil moisture in semiarid Botswana. 2.
NDVI response to soil moisture. Remote Sensing of Environment, 50, 121133.
Geerken, R., & Ilaiwi, M. (2004). Assessment of rangeland degradation and development
of a strategy for rehabilitation. Remote Sensing of Environment, 90, 490504.
Ghimire, S. C. (1992). The role of small ruminants. In G. B. Abington (Ed.), Sustainable
livestock production in the mountain agro-ecosystem of Nepal: FAO Animal Production
and Health Paper (FAO) 105. Rome: FAO.
Hartkamp, A. D., De Beurs, K., Stein, A., & White, J. W. (1999). Interpolation techniques for
climate variables. NRG-GIS Series 99-01. Mexico, D.F: CIMMYT.
Herrmann, S. M., Anyamba, A., & Tucker, C. J. (2005). Recent trends in vegetation
dynamics in the African Sahel and their relationship to climate. Global Environmental Change, 15, 394404.
Holechek, J., Pieper, R. D., & Herbel, C. H. (2001). Range management: Principles and
practices. Upper Saddle River, N.J.: Prentice Hall.
Jeltsch, F., Milton, S. J., Dean, W. R. J., & Van Rooyen, N. (1997). Analysing shrub
encroachment in the southern Kalahari: A grid-based modelling approach. Journal
of Applied Ecology, 34, 14971508.
Joshi, B. R. (1992). The role of large ruminants. In G. B. Abington (Ed.), Sustainable
livestock production in the mountain agro-ecosystem of Nepal: FAO Animal Production
and Health Paper (FAO) 105 (pp. 4776). Rome: FAO.
Kumiko, T., Masahiro, S., Shreeram, P. N., Bahadur, S. K., Hari, D. J., Nanda, P. S., Hiroshi,
F., Mariko, T., Takahiro, T., & Hirofumi, H. (2004). Mitochondrial DNA analysis of
Nepalese domestic dwarf cattle Lulu. Animal Science Journal, 75, 103110.
Le Hourou, H. N., Bingham, R. L., & Skerbek, W. (1988). Relationship between the
variability of primary production and the variability of annual precipitation in
world arid lands. Journal of Arid Environments, 15, 118.
Le Hourou, H. N., & Hoste, C. H. (1977). Rangeland production and annual rainfall
relations in the Mediterranean Basin and in the African Sahelo-Sudanian zone.
Journal of Range Management, 30, 181189.
Lise, W., Hess, S., & Purev, B. (2006). Pastureland degradation and poverty among
herders in Mongolia: Data analysis and game estimation. Ecological Economics, 58,
350364.
Liu, Y., Zha, Y., Gao, J., & Ni, S. (2004). Assessment of grassland degradation near Lake Qinghai,
West China, using Landsat TM and in situ reectance spectra data. International Journal
of Remote Sensing, 25, 41774189.
Lyon, J. G., Yuan, D., Lunetta, R. S., & Elvidge, C. D. (1998). A change detection
experiment using vegetation indices. Photogrammetric Engineering and Remote
Sensing, 64, 143150.
Miller, D. J. (1997). New perspectives on range management and pastoralism and their
implications for HKH-Tibetan plateau rangelands. In D. J. Miller, & S. R. Craig (Eds.),
Rangelands and Pastoral Development in HKH (pp. 712).
Mitas, L., & Mitasova, H. (1999). Spatial interpolation. Geographical information systems:
Principles, techniques, management and applications (pp. 481). Wiley.

K.P. Paudel, P. Andersen / Remote Sensing of Environment 114 (2010) 18451855


Myneni, R. B., Keeling, C. D., Tucker, C. J., Asrar, G., & Nemani, R. R. (1997). Increased plant
growth in the northern high latitudes from 1981 to 1991. Nature, 386, 698702.
Nicholson, S. E., Davenport, M. L., & Malo, A. R. (1990). A comparison of the vegetation
response to rainfall in the Sahel and East Africa, using normalized difference
vegetation index from NOAA AVHRR. Climatic Change, 17, 209241.
Nicholson, S. E., Tucker, C. J., & Ba, M. B. (1998). Desertication, drought, and surface
vegetation: An example from the West African Sahel. Bulletin of the American
Meteorological Society, 79, 815829.
Noy-Meir, I. (1973). Desert ecosystems: Environment and producers. Annual Review of
Ecology and Systematics, 4, 2551.
NTNC (2008). Sustainable development plan of Mustang. Kathmandu, Nepal: NTNC/GoN/
UNEP.
NTNC-ACAP (2005). Unpublished household survey dataPopulation and livestock of
Upper Mustang. Unpublished database, NTNC-ACAP-UCO, Lomanthang/UMBCP
Paudel, K.P. (2006). Institutions, environmental entitlements and pastoral management: A case of Nyishang, Trans-Himalaya region, Manang District, Nepal. M. Phil.
thesis, University of Bergen, Norway.
Pettorelli, N., Vik, J., Mysterud, A., Gaillard, J., Tucker, C., & Stenseth, N. (2005). Using the
satellite-derived NDVI to assess ecological responses to environmental change.
Trends in Ecology & Evolution, 20, 503510.
Pickup, G. (1996). Estimating the effects of land degradation and rainfall variation on
productivity in rangelands: An approach using remote sensing and models of
grazing and herbage dynamics. Journal of Applied Ecology, 33, 819832.
Pickup, G., Bastin, G. N., & Chewings, V. H. (1998). Identifying trends in land degradation
in non-equilibrium rangelands. Journal of Applied Ecology, 365377.
Pickup, G., & Chewings, V. H. (1994). A grazing gradient approach to land degradation
assessment in arid areas from remotely-sensed data. International Journal of Remote
Sensing, 15, 597617.
Pinzon, J., Brown, M. E., & Tucker, C. J. (2004). Satellite time series correction of orbital
drift artifacts using empirical mode decomposition. Hilbert-Huang Transform:
Introduction and Applications (pp. 173176).
Prince, S. D., Wessels, K. J., Tucker, C. J., & Nicholson, S. E. (2007). Desertication in the
Sahel: A reinterpretation of a reinterpretation. Global Change Biology, 13, 13081313.
Richard, Y., & Poccard, I. (1998). A statistical study of NDVI sensitivity to seasonal and
interannual rainfall variations in Southern Africa. International Journal of Remote
Sensing, 19, 29072920.
Rder, A., Kuemmerle, T., Hill, J., Papanastasis, V. P., & Tsiourlis, G. M. (2007). Adaptation
of a grazing gradient concept to heterogeneous Mediterranean rangelands using
cost surface modelling. Ecological Modelling, 204, 387398.
Schroeder, T. A., Cohen, W. B., Song, C., Canty, M. J., & Yang, Z. (2006). Radiometric
correction of multi-temporal Landsat data for characterization of early successional
forest patterns in western Oregon. Remote Sensing of Environment, 103, 1626.

1855

Scoones, I. (1994). Living with uncertainty: New directions in pastoral development in


Africa. London: Intermediate Technology.
Shabanov, N. V., & Myneni, R. (2001). Variations in northern vegetation activity inferred
from satellite data of vegetation index during 1981 to 1999.Journal of Geophysical
Research, 106 20,069-020,083.
Singh, A. (1989). Digital change detection techniques using remotely-sensed data.
International Journal of Remote Sensing, 10, 9891004.
Soudani, K., Franois, C., Le Maire, G., Le Dantec, V., & Dufrne, E. (2006). Comparative
analysis of IKONOS, SPOT, and ETM+ data for leaf area index estimation in
temperate coniferous and deciduous forest stands. Remote Sensing of Environment,
102, 161175.
Stocking, M., & Murnaghan, N. (2001). Handbook for the eld assessment of land degradation.
London: Earthscan.
Todd, S. W., & Hoffman, M. T. (1999). A fence-line contrast reveals effects of heavy
grazing on plant diversity and community composition in Namaqualand, South
Africa. Plant Ecology, 142, 169178.
Tucker, C. J. (1979). Red and photographic infrared linear combinations for monitoring
vegetation. Remote Sensing of Environment, 8, 127150.
Tucker, C. J., Pinzon, J. E., Brown, M. E., Slayback, D. A., Pak, E. W., Mahoney, R., Vermote, E. F.,
& Saleous, N. E. (2005). An extended AVHRR 8-km NDVI dataset compatible with
MODIS and SPOT vegetation NDVI data. International Journal of Remote Sensing,
26, 44854498.
Tucker, C. J., & Sellers, P. J. (1986). Satellite remote sensing of primary production.
International Journal of Remote Sensing, 7, 13951416.
Tueller, P. T. (1989). Remote sensing technology for rangeland management applications. Journal of Range Management, 42, 442453.
Wang, J., Price, K. P., & Rich, P. M. (2001). Spatial patterns of NDVI in response to
precipitation and temperature in the central Great Plains. International Journal of
Remote Sensing, 22, 38273844.
Wehn, S. (2009). A map-based method for exploring responses to different levels of
grazing pressure at the landscape scale. Agriculture, Ecosystems, and Environment,
129, 177181.
Wessels, K. J., Prince, S. D., Malherbe, J., Small, J., Frost, P. E., & VanZyl, D. (2007). Can
human-induced land degradation be distinguished from the effects of rainfall
variability? A case study in South Africa. Journal of Arid Environments, 68, 271297.
Whittaker, R. H. (1970). Communities and ecosystems. London: Macmillan.
Wiegand, T., Snyman, H. A., Kellner, K., & Paruelo, J. M. (2004). Do grasslands have a
memory: Modeling phytomass production of a semiarid South African grassland.
Ecosystems, 7, 243258.
Yang, W., Yang, L., & Merchant, J. W. (1997). An assessment of AVHRR/NDVIecoclimatological relations in Nebraska, USA. International Journal of Remote Sensing,
18, 21612180.

You might also like