remote sensing
Article
Analysis of the Effects of Drought on Vegetation
Cover in a Mediterranean Region through the Use of
SPOT-VGT and TERRA-MODIS Long Time Series
Mehrez Zribi 1, *, Ghofrane Dridi 2 , Rim Amri 1,3 and Zohra Lili-Chabaane 3
1
2
3
*
CESBIO-Centre d’Études Spatiales de la BIOsphère, Université de Toulouse, CNES/CNRS/IRD/UPS,
18 avenue, Edouard Belin, bpi 2801, 31401 Toulouse CEDEX 9, France; [email protected]
CARTEL, Département de géomatique appliquée, Université de Sherbrooke, Sherbrooke, QC J1K 2R1,
Canada; [email protected]
Université de Carthage/INAT, 43, Avenue Charles Nicolle, 1082 Tunis-Mahrajène, Tunisia;
[email protected]
Correspondence: [email protected]; Tel.: +33-5-61558525
Academic Editors: Yuei-An Liou, Chyi-Tyi Lee, Yuriy Kuleshov, ean-Pierre Barriot, Chung-Ru Ho, Xiaofeng Li
and Prasad S. Thenkabail
Received: 20 September 2016; Accepted: 28 November 2016; Published: 2 December 2016
Abstract: The analysis of vegetation dynamics and agricultural production is essential in semi-arid
regions, in particular as a consequence of the frequent occurrence of periods of drought. In this
paper, a multi-temporal series of the Normalized Difference of Vegetation Index (NDVI), derived
from SPOT-VEGETATION (between September 1998 and August 2013) and TERRA-MODIS satellite
data (between September 2000 and August 2013), was used to analyze the vegetation dynamics
over the central region of Tunisia in North Africa, which is characterized by a semi-arid climate.
Products derived from these two satellite sensors are generally found to be coherent. Our analysis of
land use and NDVI anomalies, based on the Vegetation Anomaly Index (VAI), reveals a strong level
of agreement between estimations made with the two satellites, but also some discrepancies related
to the spatial resolution of these two products. The vegetation’s behavior is also analyzed during
years affected by drought through the use of the Windowed Fourier Transform (WFT). Discussions of
the dynamics of annual agricultural areas show that there is a combined effect between climate and
farmers’ behavior, leading to an increase in the prevalence of bare soils during dry years.
Keywords: Vegetation Anomaly Index; North Africa; vegetation dynamics; agriculture
1. Introduction
In semi-arid regions, drought is a frequent phenomenon leading to serious problems in agriculture
and food safety, and increasing attention has been drawn to its high attendant economic and social
costs [1]. Although it is essential to quantify the occurrence of droughts, this is generally found to be
difficult since they are spatially variable and context-dependent [2]. In general, in situ meteorological
data from weather stations is analyzed through the use of drought indices [3,4] such as the Palmer
Drought Severity Index (PDSI; [5]) or the Standardized Precipitation Index (SPI; [6,7]). The PDSI
is based on long-term historical precipitation and mean temperature data. The SPI is based on
precipitation data only. In areas with a high density of weather stations, drought conditions can be
interpolated with only small associated errors. In the case of regions with sparsely distributed weather
stations, statistical techniques such as the Inverse Distance Weighted method or a stochastic model
of Ordinary Kriging can be considered. These methods are not completely sufficient to ensure that
drought quantification is achieved with high accuracy, in areas where limited ground measurements
are made. The problem is more complicated in areas of the globe in which weather stations are
Remote Sens. 2016, 8, 992; doi:10.3390/rs8120992
www.mdpi.com/journal/remotesensing
Remote Sens. 2016, 8, 992
2 of 16
completely absent. For this reason, in the last two decades scientists have proposed various new
approaches for the estimation of drought conditions, based on the use of remotely sensed satellite
data [8–11].
A number of different indices based on data recorded by remote optical sensors have been
developed to quantify drought [12–14]. In fact, remote sensing allows soil and vegetation dynamics
and its variations to be monitored over time [15–17]. Most studies are based on the use of the
Normalized Difference Vegetation Index (NDVI), which has demonstrated good accuracies for the
quantification of green vegetation cover or vegetation abundance [18]. This index is expressed by:
NDVI = (RNIR − RRED)/(RNIR + RRED), where RNIR is the near-infrared (NIR) reflectance and
RRED is the red reflectance. A high NDVI indicates a strong level of photosynthetic activity [19,20].
The Vegetation Condition Index (VCI) has been tested at different sites, revealing its high potential
for the detection and monitoring of drought. Kogan et al. [12], Seiler et al. [21], and Quiring et al. [22]
observed a high correlation between the VCI and agricultural production in different regions of the
globe (Africa, America, Europe). Gitelsen et al. [23] showed that when tested at six different sites in
Kazakhstan, the VCI could explain 76% of the observed variations in crop density. Amri et al. [14]
proposed the use of the notion of the Vegetation Anomaly Index (VAI), derived from SPOT-VGT data,
and observed a high correlation between this product and precipitation levels measured at the site
studied in the present paper. Mu et al. [24] proposed the Drought Severity Index (DSI), combining the
Vegetation Anomaly Index and an evapotranspiration anomaly index, which has shown considerable
potential for the quantification of drought in different climatic regions.
In North Africa, droughts are the cause of major losses in non-irrigated agriculture, and crop yields
are severely reduced during dry years [14]. Droughts are also an important factor in environmental
degradation, because they limit the development of vegetation cover and make the soil more sensitive
to erosion in the case of intense precipitation.
The aim of this paper is to propose an analysis of variations in vegetation cover, in particular
for the case of annual agricultural land use in a semi-arid region of North Africa, in the context of
drought events. It is based on two satellite products having different spatial resolutions, namely
SPOT-VGT and TERRA-MODIS. Section 2 presents the studied site and the database used in this
study. Section 3 compares the behavior of inter-annual NDVI time series, and provides a discussion of
land use mapping and NDVI anomalies from TERRA-MODIS and SPOT-VGT observations. Section 4
describes the specificities of agricultural areas confronted by drought events. Our conclusions are
presented in Section 5.
2. Database
2.1. Studied Site
The studied site is the Kairouan plain, which is situated in central Tunisia (35◦ –35◦ 450 N;
9◦ 300 –10◦ 150 E) (Figure 1) and is characterized by a semi-arid climate [25]. The average annual rainfall
is approximately 300 mm per year, with the rainy seasons lasting from October to May, followed by dry
summers. The rainfall patterns in this semi-arid area are highly variable over time and space. As shown
in the annual precipitation plot of Figure 2, this site is frequently affected by drought events: during the
15-year period of the present study, from 1998 to 2013, we identify five dry years with an accumulated
precipitation of less than 200 mm. The mean daily temperature in Kairouan City ranges between a
minimum of 10.7 ◦ C in January and a maximum of 28.6 ◦ C in August, with a mean value equal to
19.2 ◦ C. The mean annual potential evapotranspiration (Penman) is close to 1600 mm. The landscape
has no relief, and land use is dominated by agriculture, with two main types of vegetation cover:
annual agriculture and olive trees.
Remote Sens. 2016, 8, 992
Remote Sens. 2016, 8, 992
Remote Sens. 2016, 8, 992
3 of 16
3 of 16
3 of 16
Annual
AnnualPrecipitation
Precipitation(mm)
(mm)
Figure 1.
1. Satellite
Satellite view
view of
of the
the Kairouan
Kairouan plain.
plain.
Figure
Figure
1. Satellite
view of
the Kairouan
plain.
500
500
400
400
300
300
200
200
100
100
0
0
Year
Year
Figure 2. Annual precipitations over the studied period (1998–2013).
Figure 2. Annual precipitations over the studied period (1998–2013).
Figure 2. Annual precipitations over the studied period (1998–2013).
2.2. Satellite Data
2.2. Satellite Data
2.2. Satellite Data
(a) SPOT-VGT
(a) SPOT-VGT
(a) SPOT-VGT
The 10-day synthesis (S10) products derived from SPOT-VGT data are available at full resolution
The 10-day synthesis (S10) products derived from SPOT-VGT data are available at full resolution
The and
10-day
synthesis
(S10) products
derived
SPOT-VGT
data are
at full
(1 km),
include
the 10-day
NDVI data
used from
in the
present study.
For available
these products,
(1 km), and include the 10-day NDVI data used in the present study. For these products,
resolution
(1
km),
and
include
the
10-day
NDVI
data
used
in
the
present
study.
For
these
products,
top-of-atmosphere corrections were applied using the SMAC algorithm, which corrects for molecular
top-of-atmosphere corrections were applied using the SMAC algorithm, which corrects for molecular
top-of-atmosphere
corrections
were applied
the SMAC
algorithm,effects.
which The
corrects
for molecular
and aerosol scattering,
water vapor,
ozone,using
and other
gas absorption
parameters
taken
and aerosol scattering, water vapor, ozone, and other gas absorption effects. The parameters taken
and
water vapor,
ozone, and
gas absorption
The parameters
takenwater
into
intoaerosol
accountscattering,
in the atmospheric
corrections
areother
the aerosol
optical effects.
depth (AOD),
atmospheric
into account in the atmospheric corrections are the aerosol optical depth (AOD), atmospheric water
account
in
the
atmospheric
corrections
are
the
aerosol
optical
depth
(AOD),
atmospheric
water
vapor,
vapor, and ozone, and a Digital Elevation Model used for atmospheric pressure estimation. The water
vapor, and ozone, and a Digital Elevation Model used for atmospheric pressure estimation. The water
and
ozone,
and a Digital
Elevation
used for atmospheric
pressure
estimation.
The
waterwith
vapora
vapor
parameter
is derived
fromModel
data provided
by Météo-France
once
every six
hours,
vapor parameter is derived from data provided by Météo-France once every six hours, with a
1.5° × 1.5° grid cell resolution. Although the AOD is currently retrieved from B0 data (blue spectral
1.5° × 1.5° grid cell resolution. Although the AOD is currently retrieved from B0 data (blue spectral
Remote Sens. 2016, 8, 992
4 of 16
parameter is derived from data provided by Météo-France once every six hours, with a 1.5◦ × 1.5◦ grid
cell resolution. Although the AOD is currently retrieved from B0 data (blue spectral band of SPOT-VGT,
0.43–0.47 µm), combined with the NDVI, prior to 2001 it was a static dataset, which varied as a function
of latitude only. Different systematic errors (misregistration of the different channels, calibration of the
linear array detectors for each spectral band) are corrected in the final P product, which is re-sampled
to a plate carree geographic projection. The S10 products are available at http://free.vgt.vito.be. In the
present study, we make use of this product for the period between September 1998 and August 2013.
(b) TERRA-MODIS
In this study we used 250-m spatial resolution, 16-day composites of MODIS NDVI data
(MOD13Q1, collection 5). This product is retrieved from atmosphere-corrected, daily bidirectional
surface reflectance observations, using a compositing technique based on product quality assurance
metrics to remove low quality pixels. The 250-m spatial resolution is the finest available from the
MODIS NDVI dataset, and the 16-day composite was selected in order to ensure a high probability of
having the best quality pixel (reduced cloud effects) representing the NDVI within each 16-day period.
Here, we make use of this product in order to analyze the influence of drought, during the period
between September 2000 and August 2013.
2.3. Precipitation
Precipitation estimations were based on measurements recorded by a network of 30 rain gauges,
distributed over the entire site. As our studied site is mainly flat, without mountains, the terrain was
assumed to have no influence on the rainfall’s spatial distribution. The Inverse Distance Weighting
(IDW) interpolation algorithm, which computes values at non-sampled points by estimating the
weighted average of data observed at nearby points, was used to estimate daily precipitation maps [26].
The precipitation time series and NDVI data set are thus available with similar spatial resolutions.
As a consequence of the small number of rain gauges and the assumption of rainfall with no particular
spatial distribution, the difference between precipitation products at 1 km and 250 m resolutions is
very small.
2.4. NDVI Temporal Series
Figure 3 shows the NDVI time series over the last 13 years (2000–2013), for three types of
vegetation: pastures, annual agricultural areas, and olive groves, based on data provided by the
SPOT-VGT and TERRA-MODIS sensors. For each sensor, the NDVI and precipitation data correspond
to an average value, taken over a selected area characterized by homogenous vegetation cover. It is
retrieved using high resolution land use maps and direct observations in the site. Among the three types
of vegetation cover, the NDVI with the lowest dynamic range is observed for olive groves. These are
characterized by a dispersed vegetation cover, with a spacing of approximately 15 m between two
successive trees accompanied by a high percentage of bare soil. The highest temporal dynamics are
observed in the areas characterized by annual agricultural vegetation, where the NDVI can be seen to
vary strongly between wet and dry years. For example, a maximum NDVI equal to 0.6 is observed
during 2010–2011, the wettest year (430 mm precipitation), whereas 0.1 is observed during the period
2000–2001 (176 mm precipitation), the driest hydrological year. In all land use cases, a high level
of agreement is found between the NDVI dynamics observed with SPOT-VGT and TERRA-MODIS.
However, during two different one-year periods (2004–2005 and 2006–2007) some discrepancies were
observed in the case of the annual agriculture vegetation cover, for which the maximum value of NDVI
derived from TERRA-MODIS data is clearly lower. This behavior could be attributed to differences
in the data processing algorithms, in particular for cloudy conditions. It can also be noted that
the precipitation recorded between November and January, which is essential for growth of annual
agriculture, was relatively low (approximately 105 mm) during these two periods, when compared
to that recorded during wet years (170 mm for 2003–2004 and 190 mm for 2005–2006). It is thus
Remote Sens. 2016, 8, 992
5 of 16
likely that the NDVI values retrieved from the TERRA-MODIS data are more suitable for the accurate
determination of the vegetation dynamic.
(a)
(b)
Figure 3. Temporal variations of the spatially averaged NDVI for three types of vegetation cover
(pasture, annual agriculture and olive groves), from September 1998 to September 2013; (a) SPOT-VGT;
(b) MODIS.
3. Analysis of Vegetation Dynamics with SPOT-VGT and TERRA-MODIS NDVI Products
3.1. Temporal Dynamics of Land Use Mapping
In order to accurately analyze the vegetation dynamics corresponding to the different types
of vegetation cover present at the studied site, it is essential to consider annual land use maps.
As a consequence of the lack of high-resolution satellite images, suitable for the retrieval of
land-use maps for all of the studied years between 1998 and 2013, we adopt an approach based
on the use of low spatial resolution SPOT-VGT and TERRA-MODIS NDVI images. For this,
we consider three characteristic classes, labelled: “olive trees and bare soil”, “annual agriculture”
and “pastures”. As shown by the NDVI time series, olive groves are affected by quite small variations
in NDVI, between dry and wet years, with the NDVI remaining very close to that of bare soil
(between 0.15 and 0.2). The similarity of these values
1 is due to the large average spacing between olive
trees (between 15 and 20 m), as well as to the continuous use of soil tilling (up to five times per year),
a practice that is commonly used to increase infiltration and eliminate the growth of grass and weeds
Remote Sens. 2016, 8, 992
6 of 16
following rainfall events. For this reason, we group olive groves and bare soils together in the same
class of land use.
A linear mixing theory developed by Settle et al. [27] is considered, in which it is assumed that
the reflectance (respectively NDVI) of a mixed pixel is given by the sum of the mean reflectance
(respectively NDVI) values of the different land cover classes within the pixel, weighted by their
respective fractional covers. The first step in this approach involves the identification of typical NDVI
profiles, representative of each of the above land cover classes in the disaggregation methodology.
For this step, we consider information related to the class composition of each pixel, retrieved from
a high-resolution land-cover map. Disaggregation techniques are designed to estimate the annual
proportion (between 0 and 1) of specific classes occurring within each pixel, by considering the NDVI
time series acquired for each year. The result is a set of three fractional images, each corresponding
to one specific class of land-cover. While this information describes the composition of each pixel,
it does not provide any indication as to how the classes are spatially distributed within each pixel.
Figure 4 illustrates the results of the three land use mapping using SPOT-VGT and MODIS data for
the wettest (2010–2011) and driest (2000–2001) years. A qualitative comparison of the SPOT-VGT
and MODIS land use maps shows that they are in good agreement. Approximately the same areas
are retrieved, corresponding to a strong presence of “annual agriculture” or “pasture” or “olive
groves and bare soil”. During dry years, a clear decrease is observed in the presence of annual
agriculture. In order to verify the accuracy of the land use maps produced from low-resolution satellite
images, we compared high-resolution SPOT-HRV maps of annual agricultural areas with the proposed
low-resolution SPOT-VGT and TERRA-MODIS NDVI maps over a period of eight years, for which
high-resolution satellite data exists. Table 1 shows the SPOT-HRV and LANDSAT high-resolution
images used to produce the high-resolution land-use maps, for the eight agricultural seasons.
Table 1. High resolution optical data (SPOT, LANDSAT) acquisitions.
Agricultural Year
Date of Acquisition
Satellites
1999/2000
5 January 2000
13 March 2000
9 July 200
SPOT 2
2005/2006
16 November 2005
26 March 2006
22 July 2006
SPOT 2
2006/2007
21 December 2006
25 March 2007
27 July 2007
SPOT 2/LANDSAT
2008/2009
21 December 2008
14 April 2009
20 May 2009
17 July 2009
SPOT 5
2009/2010
8 October 2009
30 December 2009
28 March 2010
24 June 2010
SPOT 5
2010/2011
24 December 2010
29 January 2011
17 March 2011
03 July 2011
SPOT 5
2011/2012
6 November 2011
13 January 2012
31 March 2012
6 July 2012
SPOT 5
2012/2013
26 December 2012
21 January 2013
19 March 2013
11 July 2013
SPOT 5
Remote Sens. 2016, 8, 992
7 of 16
Figure 5 shows the high-resolution land use map of the studied site for the 2012–2013
agricultural season. An approach making use of decision tree learning was applied, based on the use
Remote Sens. 2016, 8, 992
7 of 16
of four multi-temporal data acquisitions for each year [25]. This allowed 11 land use classes to be
identified: rivers, dams, sebkhas, reliefs, urban areas, dry olive, summer and winter vegetables,
cereals,
bare
soils, and
tree crops. Our land
annual
derived
from
and MODIS
Figure
5 shows
the high-resolution
useagriculture
map of theclass
studied
site for
theSPOT-VGT
2012–2013 agricultural
maps includes
several
classesuse
derived
from tree
the SPOT-HRV
use maps,
as use
cereals
and
season.
An approach
making
of decision
learning wasland
applied,
based such
on the
of four
summer
and
winter
vegetables,
which
were
merged
in
order
to
compare
highand
low-resolution
multi-temporal data acquisitions for each year [25]. This allowed 11 land use classes to be identified:
estimations.
The comparison
was made
the SPOT-HRV
scene
data only,
and bare
other
rivers,
dams, sebkhas,
reliefs, urban
areas,according
dry olive, to
summer
and winter
vegetables,
cereals,
regions
the crops.
Kairouan
plain
wereagriculture
masked in class
the SPOT-VGT
and SPOT-VGT
MODIS landand
useMODIS
maps tomaps
allow
soils,
andoftree
Our
annual
derived from
comparisons
to classes
be made
correctly.
Asthe
shown
in Figure
6, use
in the
casesuch
of the
estimated
area of
includes
several
derived
from
SPOT-HRV
land
maps,
astotal
cereals
and summer
annual
agriculture
land
use,were
a better
agreement
between
and TERRA-MODIS
and
winter
vegetables,
which
merged
in orderistofound
compare
high-SPOT-HRV
and low-resolution
estimations.
data
than between
SPOT-HRV
and
data,scene
with data
correlation
coefficients
equal of
to the
0.89
The
comparison
wasthe
made
according
to SPOT-VGT
the SPOT-HRV
only, and
other regions
and 0.59,plain
respectively.
Thisinresult
could be and
explained
differences
in allow
spatial
and temporal
Kairouan
were masked
the SPOT-VGT
MODISby
land
use maps to
comparisons
to
As a result
of theinrelatively
fields
(generally
between
1 and
beresolutions.
made correctly.
As shown
Figure 6, small
in thesize
caseofofthe
theagricultural
total estimated
area
of annual
agriculture
5 ha),
theaTERRA-MODIS
outputs
are between
found to SPOT-HRV
be in better agreement
with the high-resolution
reality
land
use,
better agreement
is found
and TERRA-MODIS
data than between
of SPOT-HRV
land use at the
site. data, with correlation coefficients equal to 0.89 and 0.59, respectively.
the
andstudy
SPOT-VGT
This result could be explained by differences in spatial and temporal resolutions. As a result of the
3.2. Discussion
NDVI
Anomalies
with TERRA-MODIS
SPOT-VGT
Data
relatively
small ofsize
of the
agricultural
fields (generallyand
between
1 and
5 ha), the TERRA-MODIS
outputs
are
found
to
be
in
better
agreement
with
the
high-resolution
reality
of
at the provides
study site.a
In this section, we propose to use an anomaly index applied inland
[14],use
which
quantitative illustration of vegetation stress and the influence of drought on the vegetation cover.
3.2. Discussion of NDVI Anomalies with TERRA-MODIS and SPOT-VGT Data
This index is based on statistics derived from the NDVI time series, and is referred to as the
In this section,
we propose
to use an
anomaly
“Vegetation
Anomaly
Index” (VAI),
written
as: index applied in [14], which provides a quantitative
illustration of vegetation stress and the influence of drought on the vegetation cover. This index is
NDVIi − and
NDVI
i mean
based on statistics derived from the VAI
NDVI=time series,
is referred
to as the “Vegetation Anomaly
(1)
i
σi
Index” (VAI), written as:
NDV Ii − ( NDV Ii )mean
VAI
(1)
i =a given month i, (NDVI )
where NDVIi is the NDVI estimate
for
is the mean value of the NDVI
σi
i mean
(
)
σi
during
month
i, derived
from the
described
15Ii years
NDVI
series,
where
NDV
Ii is the
NDVI estimate
forpreviously
a given month
i, ( NDV
meantime
value
of theand
NDVI
)mean isofthe
corresponds
the standard
deviation
of described
the NDVI15
values
for series,
monthand
i over
the same 15
during
month i,toderived
from the
previously
yearsestimated
of NDVI time
σi corresponds
period. deviation of the NDVI values estimated for month i over the same 15 year period.
toyear
the standard
(a)
(b)
Figure 4. Cont.
Remote Sens. 2016, 8, 992
Remote Sens. 2016, 8, 992
8 of 16
8 of 16
(c)
(d)
(e)
(f)
(g)
(h)
Figure4.4.(a,b)
(a,b)“Annual
“Annualagriculture”
agriculture”maps
mapsderived
derivedfrom
fromSPOT-VGT
SPOT-VGTobservations
observationsfor
forthe
thewettest
wettest
Figure
(2010–2011)
and
driest
year
(2000–2001);
(c,d)
“olive
trees
and
bare
soil”
maps
derived
from
(2010–2011) and driest year (2000–2001); (c,d) “olive trees and bare soil” maps derived from SPOT-VGT
SPOT-VGT
observations
for
the
wettest
(2010–2011)
and
driest
year
(2000–2001);
(e,f)
“Annual
observations for the wettest (2010–2011) and driest year (2000–2001); (e,f) “Annual agriculture” maps
agriculture”
derived from
TERRA-MODIS
for the wettest
(2010–2011)
and driest
derived
from maps
TERRA-MODIS
observations
for theobservations
wettest (2010–2011)
and driest
year (2000–2001);
year
(2000–2001);
(g,h)
“olive
trees
and
bare
soil”
maps
derived
from
TERRA-MODIS
observations
(g,h) “olive trees and bare soil” maps derived from TERRA-MODIS observations for the wettest
for the wettest
and driest year (2000–2001).
(2010–2011)
and (2010–2011)
driest year (2000–2001).
Remote
Sens.
2016,
8, 992
Remote
Sens.
2016,
8, 992
Remote Sens. 2016, 8, 992
of 16
9 of916
9 of 16
Figure
5. High
resolution
land
useuse
map
derived
from
SPOT/HRV
images
in 2012–2013.
Figure
High
resolution
land
map
derived
from
SPOT/HRV
images
2012–2013.
Figure
5.5.High
resolution
land
use map
derived
from
SPOT/HRV
images
inin
2012–2013.
Annual Agriculture Area (ha)
Annual Agriculture Area (ha)
100000
100000
90000
90000
80000
80000
70000
70000
60000
60000
50000
50000
40000
40000
30000
30000
20000
20000
10000
10000
0
0
SPOT HR
SPOT HR
SPOT-VGT
SPOT-VGT
MODIS
MODIS
Figure
6. Variations
in the
surface
area
of annual
agriculture
over
a period
of eight
years.
Figure
6. Variations
in the
surface
area
of annual
agriculture
over
a period
of eight
years.
Figure 6. Variations in the surface area of annual agriculture over a period of eight years.
Figure
7 illustrates
thethe
mean
level
of of
thisthis
index
forfor
thethe
studied
site,
using
data
from
both
Figure
illustrates
mean
level
index
studied
site,
using
data
from
both
Figure
7 7illustrates
the mean
level
of this
index
for the
studied
site,
using
data
from
both
SPOT-VGT
andand
TERRA-MODIS
forfor
thethe
period
of of
interest,
between
2000
andand
2013.
TheThe
VAI
is
SPOT-VGT
TERRA-MODIS
period
interest,
between
2000
2013.
VAI
SPOT-VGT and TERRA-MODIS for the period of interest, between 2000 and 2013. The VAI isis
computed
forforthe
hydrological
years
beginning ininSeptember,
September,
over
the
studied
region.
We retrieve
first
computed
the
hydrological
years
beginning
over
the
studied
region.
We
first
computed for the hydrological years beginning in September, over the studied region. We first
retrieve
approximately
the same
behavior
for these
products
and
positive
negativelevels
levelsfor
approximately
the same
VAI VAI
behavior
for these
twotwo
products
and
positive
orornegative
retrieve
approximately
the same
VAI behavior
for these
two
products
and
positive
or negative
levels
forapproximately
approximatelythe
the
same
periods.
However,
various
discrepancies
are encountered
due
to
the
same
periods.
However,
various
discrepancies
are
encountered
due
to
the
for approximately the same periods. However, various discrepancies are encountered duedifferent
to the
different
spatial
resolutions
of
these
two
sensors.
The
VAI
index
can
be
seen
to
have
a
positive
trend
spatial resolutions
of theseoftwo
sensors.
The VAI
can be
to have
a positive
trend
for
different
spatial resolutions
these
two sensors.
The index
VAI index
canseen
be seen
to have
a positive
trend
forboth
bothsatellite
satelliteproducts,
products,which
which
can
be
firstly
explained
byslightly
a slightly
positive
trend
in autumn
can
be
firstly
explained
by
a
positive
trend
in
autumn
rainfall
for both satellite products, which can be firstly explained by a slightly positive trend in autumn
rainfall
(between
November
and January)
observed
13 years
of the studied
In
(between
November
and January)
observed
duringduring
the 13the
years
the studied
period.period.
In
addition,
rainfall
(between
November
and January)
observed
during
the 13ofyears
of the studied
period.
In
addition,
a
small
increase
in
the
surface
of
irrigated
areas
was
observed
during
the
last
10
years
of
a small increase
in the surface
irrigated
areas wasareas
observed
during theduring
last 10the
years
the
studied
addition,
a small increase
in theofsurface
of irrigated
was observed
lastof10
years
of
theperiod,
studied
period,
maycontributed
also have contributed
to trend
this positive
[28].confirms
This outcome
which
maywhich
also
have
this positive
Thistrend
outcome
the trend
the studied
period,
which
may also havetocontributed
to this[28].
positive
trend [28].
This outcome
confirms
thein
trend
reported
inof
a global study
of drought
by on
[24,29],
based on the interpretation
theDSI
reported
a global
study
by [24,29],
based
the interpretation
the NDVI of
and
confirms
the trend
reported
in adrought
global study
of drought
by [24,29],
based on the of
interpretation
of the
NDVI
and DSI
(Drought
Severity
Index)
indices,
showing
that the
studied
(and North
Africa is
(Drought
Severity
Index)
indices,
showing
that
the studied
(andregion
North
in general)
NDVI
and DSI
(Drought
Severity
Index)
indices,
showing
thatregion
the studied
regionAfrica
(and North
Africa
in general)
is wetter.
becoming
wetter.
We find
a more significant
trend
for the SPOT-VGT
time
series,
with
becoming
We
find
a
more
significant
trend
for
the
SPOT-VGT
time
series,
with
a
slope
equal
in general) is becoming wetter. We find a more significant trend for the SPOT-VGT time series, with
a slope
equal
to
0.13,
whereas
the
TERRA-MODIS
data
leads
to
a
slope
of
0.06.
This
difference
is
due
0.13,equal
whereas
the whereas
TERRA-MODIS
data leads todata
a slope
of to
0.06.
This of
difference
is difference
due mainly
the
a to
slope
to 0.13,
the TERRA-MODIS
leads
a slope
0.06. This
is to
due
mainly to the discrepancy between the annual maximum values of NDVI determined for these two
mainly to the discrepancy between the annual maximum values of NDVI determined for these two
products, during the one-year periods between 2004–2005 and 2006–2007 (see Figure 3).
products, during the one-year periods between 2004–2005 and 2006–2007 (see Figure 3).
Remote Sens. 2016, 8, 992
10 of 16
discrepancy between the annual maximum values of NDVI determined for these two products, during
the one-year
periods between 2004–2005 and 2006–2007 (see Figure 3).
Remote Sens. 2016, 8, 992
10 of 16
September
2
VAI SPOT-VGT
1.5
October
1
0.5
November
0
December
-0.5
-1
y = 0.13 x - 0.87
January
-1.5
-2
February
March
Hydrological year
(a)
September
2
VAI TERRA-MODIS
1.5
October
1
November
0.5
0
December
-0.5
-1
y = 0.0671x - 0.4698
January
-1.5
February
-2
March
Hydrological year
(b)
Figure 7. Temporal dynamics of the VAI over the Kairouan plain: (a) using SPOT-VGT data; (b) using
Figure 7. Temporal dynamics of the VAI over the Kairouan plain: (a) using SPOT-VGT data;
TERRA-MODIS data (2000–2013).
(b) using TERRA-MODIS data (2000–2013).
Figure 8 compares the mean annual VAI indices for the two products, for the 13 years considered
in
this
The twothe
products
are found
to be
in very
good
and an
linear
Figurestudy.
8 compares
mean annual
VAI
indices
for
the agreement,
two products,
forapproximately
the 13 years considered
relationship
be products
established
between
twoinindexes:
in this
study. Thecan
two
are
found the
to be
very good agreement, and an approximately linear
relationship can be established between
VAI the two =indexes:
α VAI
SPOT −VGT
MODIS
+β
(2)
Table 2 lists the linear regression
parameters
for each
VAISPOT
αVAI MODIS
+ βmonth of the year, showing that
−VGT =retrieved
for most months α is close to 1, and α is close to zero.
(2)
Table 2 lists the linear regression parameters retrieved for each month of the year, showing that
for most months α is close to 1, and α is close to zero.
Remote Sens. 2016, 8, 992
11 of 16
Remote Sens. 2016, 8, 992
11 of 16
2
mean annual VAI from SPOT-VGT
1.5
-2
1
0.5
0
-1.5
-1
-0.5
0
0.5
1
1.5
2
-0.5
-1
-1.5
-2
mean annual VAI from TERRA-MODIS
Figure 8. Relationship between the VAI indices estimated annually from TERRA-MODIS and
Figure 8. Relationship between the VAI indices estimated annually from TERRA-MODIS and
SPOT-VGT
data.
SPOT-VGT
data.
Table
2. Parameters derived
linear
regression
between
VAISPOT-VGT
and
VAIMODIS and
over the
Table
2. Parameters
derivedfrom
from
linear
regression
between
VAI
VAIstudied
SPOT-VGT
MODIS over the
site, site,
for the
period
betweenbetween
2000 and2000
2013. and 2013.
studied
for
the period
Month
September
Month
October
September
November
December
October
January
November
February
December
March
January
April
February
May
June
March
July
April
August
α
0.78 α
1.05
0.880.78
0.961.05
1.020.88
0.99
0.96
1.08
1.02
1.07
0.870.99
1.041.08
1.071.07
0.96
β
0.0075 β
0.01
0.03 0.0075
0.06 0.01
0.02 0.03
−0.02
0.06
−0.05
0.02
0.0028
0.1 −0.02
0.066 −0.05
0.07 0.0028
0.06
May
0.87
0.1
June
1.04
0.066
4. Dynamics of the Area used forJuly
Annual Agriculture
1.07
0.07
0.96
0.06
In the previous sections we August
propose a method for
the quantification
of drought, through the use
of satellite time series. The area used for annual agriculture is certainly the most sensitive to drought
events, as of
shown
Figure
3. Itfor
is essential
gain a clear understanding of the influence of drought
4. Dynamics
the in
Area
Used
AnnualtoAgriculture
in this area, in order to improve its agricultural productivity. In this section, following the analysis of
the
temporal
variations
observed
in the NDVI
data, we
the behavior ofof
thedrought,
NDVI signals
in the use
In the
previous
sections
we propose
a method
fordiscuss
the quantification
through
the spectral
domainThe
through
the use
the windowed
Fourier
transform,
of satellite
time series.
area used
forofannual
agriculture
is certainly
the thus
mostallowing
sensitivethe
to drought
specificities
of in
extremely
to be analyzed.
events,
as shown
Figure dry
3. Ityears
is essential
to gain a clear understanding of the influence of drought in
this area,
in order
to Vegetation
improve Using
its agricultural
In this section, following the analysis of the
4.1. Analysis
of the
a Windowedproductivity.
Fourier Transform
temporal variations observed in the NDVI data, we discuss the behavior of the NDVI signals in the
The Fourier Transform (FT) can be used to analyze the frequency content of a signal in the time
spectral domain through the use of the windowed Fourier transform, thus allowing the specificities of
domain, by decomposing the original signal into a set of superimposed sine and cosine basis
extremely dry years to be analyzed.
4.1. Analysis of the Vegetation Using a Windowed Fourier Transform
The Fourier Transform (FT) can be used to analyze the frequency content of a signal in the time
domain, by decomposing the original signal into a set of superimposed sine and cosine basis functions.
By definition, these basis functions are not local—i.e., they have an infinite span and are globally
Remote Sens. 2016, 8, 992
12 of 16
uniform over time. Thus, the FT does not represent any temporal variability that may be present in the
signal, and cannot provide any information about the time of occurrence of any specific frequential
event. For such applications, the Windowed Fourier Transform (WFT) allows frequency components
present in a time series to be localized in the time domain. This technique applies a fixed-width moving
window over the data, and can be used to identify specific frequency components corresponding to
Remote Sens. 2016, 8, 992
12 of 16
dry years.
The
short-time
Fourier transform
the NDVI
signal,
X(t) is
defined
as:infinite span and are
functions.
By definition,
these basis of
functions
are not
local—i.e.,
they
have an
globally uniform over time. Thus, theZFT does not represent any temporal variability that may be
∞
present in the signal, and cannot provide any information
about the time of occurrence of any specific
X (t, f ) =
x (t1 )h∗ (t1 − t) e−i2π f t1 dt1
(3)
frequential event. For such applications,−the
∞ Windowed Fourier Transform (WFT) allows frequency
components present in a time series to be localized in the time domain. This technique applies a
where the
functionmoving
h(t) is window
a Hamming
and
multiplied
the signal X(t)
fixed-width
over window
the data, centred
and can at
be time
used t to
identify
specificbyfrequency
components
corresponding
to
dry
years.
prior to the Fourier transformation. The window function allows the signal to be viewed close to the
Fourier
transformprovides
of the NDVI
signal,estimation
X(t) is defined
time t only,The
andshort-time
the Fourier
transform
a local
of as:
the spectrum centred around
−i 2πft1 to the ordinary Fourier transform and
∞ days.
* Similarly
time t. The width of this window wasX (set
t , f )to
= 90
dt1
(3)
−∞ x(t1 )h (t1 − t )e
spectrum, the spectrogram can be written as:
where the function h(t) is a Hamming window centred at time t and multiplied by the signal X(t)
prior to the Fourier transformation. The window function allows
the signal to be viewed close to the
Sx (t, f ) = | X (t, f )|2
time t only, and the Fourier transform provides a local estimation of the spectrum centred around
time t. The width of this window was set to 90 days. Similarly to the ordinary Fourier transform and
Sx spectrum,
can be expressed
in decibels as:
the spectrogram can be written as:
P S=x (t10log
Sx(t(, t,f )f ))
, f ) = (X
2
(4)
(4)
(5)
be expressed in decibels as:
Figure S9x can
shows
the resulting WFT spectrograms (P values), for the annual agricultural cover
P = 10
log(vegetation
S x (t , f ))
(5)observed
produced from TERRA-MODIS time series. For
this
cover, a clear difference can be
between dryFigure
and wet
seasons,
with
the
spectrogram
revealing
a
greater
intensity
at
high
frequencies
9 shows the resulting WFT spectrograms (P values), for the annual agricultural cover
during wet
years.from
Figure
10a shows the
spectrograms
corresponding
to the
relatively
dry year
from 2001
produced
TERRA-MODIS
time
series. For this
vegetation cover,
a clear
difference
can be
and2010
wet seasons,
spectrogram
revealingHigher
a greaterlevels
intensity
at high power
to 2002,observed
and the between
wet yeardry
from
to 2011,with
as a the
function
of frequency.
of spectral
during wetduring
years. Figure
10a shows
the(more
spectrograms
to the relatively
dry
occur atfrequencies
high frequencies,
the wettest
year
than 4 corresponding
dB). This observation
is confirmed
in
year from 2001 to 2002, and the wet year from 2010 to 2011, as a function of frequency. Higher levels
Figure 10b, which plots the values of the spectrogram at 30 Hz (a high frequency). A similar behavior
of spectral power occur at high frequencies, during the wettest year (more than 4 dB). This
is observed for the NDVI temporal signal, with higher power occurring during wet seasons, and lower
observation is confirmed in Figure 10b, which plots the values of the spectrogram at 30 Hz (a high
power occurring
dry years,
during
which the
the vegetation
cover
are low.
This leads
frequency).inAthe
similar
behavior
is observed
for dynamics
the NDVI of
temporal
signal, with
higher
power
to a relatively
low
intensity
in the temporal
of theinNDVI
signals.
the
occurring
during
wet seasons,
and lower variations
power occurring
the dry
years, Conversely,
during which during
the
dynamics
of the
vegetationresulting
cover arefrom
low. This
leads to ain
relatively
low intensity
in and
the temporal
wet seasons,
NDVI
variations
fluctuations
the vegetation
cover
rainfall events,
variations
the NDVI
signals.
Conversely,
duringinthe
wet
seasons,
NDVI
variations
resulting
from
etc. occur
with a of
much
higher
intensity.
Although
this
section
we
present
the results
obtained
with
fluctuations in the vegetation cover and rainfall events, etc. occur with a much higher intensity.
TERRA-MODIS products only, it is interesting to note that the results obtained with the SPOT-VGT
Although in this section we present the results obtained with TERRA-MODIS products only, it is
products
are verytosimilar.
interesting
note that the results obtained with the SPOT-VGT products are very similar.
Figure 9. WFT spectrogram for annual agriculture cover derived from TERRA-MODIS data.
Figure 9. WFT spectrogram for annual agriculture cover derived from TERRA-MODIS data.
Remote Sens. 2016, 8, 992
13 of 16
Remote Sens. 2016, 8, 992
13 of 16
(a)
(b)
Figure 10. (a) Spectrogram dynamics as a function of frequency, for a dry and a wet year;
Figure 10. (a) Spectrogram dynamics as a function of frequency, for a dry and a wet year;
(b) Spectrogram dynamics at a high frequency (Fr = 30 Hz).
(b) Spectrogram dynamics at a high frequency (Fr = 30 Hz).
4.2. Monitoring of Bare Soils in Agricultural Areas
4.2. Monitoring of Bare Soils in Agricultural Areas
In the cultivated zones analyzed here, the bare soils represented a greater surface area during
drythe
years,
as a result
of theanalyzed
farmers’ tendency
avoid
sowing
in the absence
of rainfall
events,
In
cultivated
zones
here, theto
bare
soils
represented
a greater
surface
areaduring
during dry
between
and January.
aimsowing
of the next
section
is to illustrate
thisevents,
behavior,
years,the
asperiod
a result
of theNovember
farmers’ tendency
to The
avoid
in the
absence
of rainfall
during
by
analyzing
the
NDVI
time
series.
the period between November and January. The aim of the next section is to illustrate this behavior,
In order to analyze the dynamics of bare soils, together with their relationship with drought
by analyzing
the NDVI time series.
events occurring during the studied period, we studied yearly remotely-sensed land use maps,
In order to analyze the dynamics of bare soils, together with their relationship with drought events
recorded between 1998 and 2013, and broke them down into three classes (olive groves and bare soil
occurring
during
the
studied (class
period,
we pastures
studied(class
yearly
land
use maps,
(class 1),
annual
agriculture
2) and
3)).remotely-sensed
In order to evaluate
the dynamics
of recorded
bare
between
1998
and
2013,
and
broke
them
down
into
three
classes
(olive
groves
and
bare
soil
(class 1),
soils selected from class 1, during the course of wet and dry years, it is proposed to accurately
annual
agriculture
(class
2)
and
pastures
(class
3)).
In
order
to
evaluate
the
dynamics
of
bare
determine the surface area covered by the olive groves only, corresponding to that identified during soils
selected
class
during the course
of wet
and dry
years, it is
proposed
accurately
determine
the from
wettest
year1,(2010–2011).
Under these
climatic
conditions,
a very
small to
percentage
of bare
soils the
remain
Even
in natural
can be found
As this the
olive
grove year
surface
area uncultivated.
covered by the
olive
grovesareas,
only, pastures
corresponding
to thateverywhere.
identified during
wettest
class hasUnder
remained
approximately
constant in
recent
years,
any increase
class
1 coverage
a
(2010–2011).
these
climatic conditions,
a very
small
percentage
of in
bare
soils
remain during
uncultivated.
dry
year
can
be
attributed
to
an
increase
in
the
surface
area
of
bare
soil.
Even in natural areas, pastures can be found everywhere. As this olive grove class has remained
Using this hypothesis, Figure 11 illustrates the temporal variation of surface areas covered by
approximately constant in recent years, any increase in class 1 coverage during a dry year can be
bare soils, determined from data recorded by the SPOT-VGT and TERRA-MODIS sensors. These two
attributed to an increase in the surface area of bare soil.
estimations are in good agreement, with the highest surface areas occurring during the driest years.
Using this hypothesis, Figure 11 illustrates the temporal variation of surface areas covered
by bare soils, determined from data recorded by the SPOT-VGT and TERRA-MODIS sensors.
These two estimations are in good agreement, with the highest surface areas occurring during the
driest years. As shown in Figure 11, only the TERRA-MODIS sensor was able to detect a strong
Remote Sens. 2016, 8, 992
14 of 16
As shown
in Figure
Remote
Sens. 2016,
8, 992
11, only the TERRA-MODIS sensor was able to detect a strong increase14inof the
16
surface area of bare soil during the dry years of 2000, 2001 and 2007. This can be explained by
comparing the spatial resolutions of the two sensors: since that of SPOT-VGT is equal to 1 km, the
increase
in the
areabare
of bare
duringathe
dry years
of 2000,
2001than
and 2007.
probability
of surface
retrieving
soilsoil
covering
complete
pixel
is lower
in theThis
casecan
of be
the
explained
by
comparing
the
spatial
resolutions
of
the
two
sensors:
since
that
of
SPOT-VGT
is
equal
to
TERRA-MODIS observations, which have a resolution of 250 m. These results clearly illustrate effects
1that
km, are
the related
probability
of retrieving
bare soil
a completeby
pixel
lower than
in the case
of the
to drought,
but which
maycovering
also be influenced
theisfarmers’
agricultural
practices.
TERRA-MODIS
observations,
which
have
a
resolution
of
250
m.
These
results
clearly
illustrate
effects
In general, farmers do not sow cereals or vegetables whenever there are no significant rainfall events
that
are beginning
related to drought,
but which may
also(end
be influenced
the farmers’
at the
of the agricultural
season
of autumn,bybeginning
of agricultural
winter). Thispractices.
result is
Inconfirmed
general, farmers
do not sow between
cereals orprecipitation
vegetables whenever
there are
no covered
significant
events
at
by the correlation
and the surface
area
byrainfall
bare soil,
during
the
beginning
of
the
agricultural
season
(end
of
autumn,
beginning
of
winter).
This
result
is
confirmed
the period between November and January. The correlations between bare soil surface area and
by
the correlation
between
precipitation
the0.8,
surface
covered by bare
during the period
precipitation
levels
are equal
to 0.36and
and
for area
the SPOT-VGT
and soil,
TERRA-MODIS
data,
between
November
and
January.
correlations
between
bare soil surface
area and
respectively.
For this
reason,
the The
influence
of drought
on vegetation
cover could
notprecipitation
be accurately
levels
are equal
to 0.36
and 0.8, forareas,
the SPOT-VGT
TERRA-MODIS
respectively.
Fortaken
this
identified
in these
agricultural
since the and
farmers’
agriculturaldata,
practices
were not
reason,
the influence of drought on vegetation cover could not be accurately identified in these
into account.
agricultural areas, since the farmers’ agricultural practices were not taken into account.
Precipitation (november-january)
0.40
0.35
SPOT-VGT
MODIS
0
50
0.25
0.20
0.15
100
150
200
Precepitation
Bare soil area
0.30
0.10
0.05
0.00
250
300
Figure
Figure11.
11.Variations
Variationsininbare
baresoil
soilsurface
surfacearea
areafrom
from1998
1998toto2013,
2013,determined
determinedfrom
fromSPOT-VGT
SPOT-VGTand
and
TERRA-MODIS
data.
TERRA-MODIS data.
5.5.Conclusions
Conclusions
This
Thispaper
paperproposes
proposestotointerpret
interpretSPOT-VGT
SPOT-VGTand
andTERRA-MODIS
TERRA-MODISNDVI
NDVItime
timeseries
seriesfor
forthe
the
purposes
of
analyzing
the
behavior
of
vegetation
as
a
consequence
of
drought.
In
the
first
purposes of analyzing the behavior of vegetation as a consequence of drought. In the first step,step,
land
land
mapping
is implemented
using
from
both
of these
satellites.
Three
agricultural
classes
use use
mapping
is implemented
using
datadata
from
both
of these
satellites.
Three
agricultural
classes
are
are
considered:
olive
groves
and
bare
soil,
pastures,
and
annual
agricultural
areas.
The
results
obtained
considered: olive groves and bare soil, pastures, and annual agricultural areas. The results obtained
from
TERRA-MODIS
observations
areare
found
to be
qualitatively
coherent
in the
casecase
of
fromSPOT-VGT
SPOT-VGTand
and
TERRA-MODIS
observations
found
to be
qualitatively
coherent
in the
the
class ofclass
agricultural
land use.land
Theseuse.
results
are validated
through
the use
of high-resolution
of latter
the latter
of agricultural
These
results are
validated
through
the use of
SPOT/HRV
land
use
maps
recorded
over
a
period
of
eight
years.
For
the
total
surface
area
covered
high-resolution SPOT/HRV land use maps recorded over a period of eight years. For the total
surface
by
annual
agricultural
fields,
a
better
agreement
is
found
between
SPOT-HRV
and
TERRA-MODIS
area covered by annual agricultural fields, a better agreement is found between SPOT-HRV and
data
than betweendata
SPOT-HRV
and SPOT-VGT
data,and
withSPOT-VGT
correlation data,
coefficients
equal to 0.89
and 0.59,
TERRA-MODIS
than between
SPOT-HRV
with correlation
coefficients
respectively.
VAI
(Vegetation
Anomaly
Index)
estimations
are
proposed
for
a
period
of
more
equal to 0.89 and 0.59, respectively. VAI (Vegetation Anomaly Index) estimations are proposedthan
for a
13period
years,of
through
the use
SPOT-VGT
NDVI
The VAI NDVI
estimations
are
more than
13 of
years,
throughand
theTERRA-MODIS
use of SPOT-VGT
andproducts.
TERRA-MODIS
products.
found
to be
coherent, despite
differences
in spatial
resolution
between
these two
sensors,between
and a linear
The VAI
estimations
are found
to be coherent,
despite
differences
in spatial
resolution
these
relationship
is
established
on
a
monthly
basis,
between
the
two
sets
of
VAI
estimations.
positive
two sensors, and a linear relationship is established on a monthly basis, between the twoA
sets
of VAI
trend
is observed
for thetrend
VAI isindex
determined
withindex
thesedetermined
two products.
the
case
of annual
estimations.
A positive
observed
for the VAI
with For
these
two
products.
For
agricultural
areas,
a
Windowed
Fourier
Transform
spectrogram
is
computed
from
the
NDVI
time
the case of annual agricultural areas, a Windowed Fourier Transform spectrogram is computed from
series,
to simplify
the analysis
of thethe
temporal
of this product,
to determine
whento
the NDVI
time series,
to simplify
analysisvariations
of the temporal
variationsand
of this
product, and
certain
cyclicwhen
events
occur. cyclic
Drought
yearsoccur.
are qualitatively
characterized
by spectrograms
having
determine
certain
events
Drought years
are qualitatively
characterized
by
low
levels
of
high
frequency
spectral
power.
The
dynamics
of
the
VAI
and
NDVI
characterizing
of
spectrograms having low levels of high frequency spectral power. The dynamics of the VAI and
agricultural
areas are strongly
related toareas
variations
in bare soil
surface
was
NDVI characterizing
of agricultural
are strongly
related
to coverage.
variationsThis
in relationship
bare soil surface
derived from the inter-annual dynamics of land use mapping. Using the two satellites, SPOT-VGT and
Remote Sens. 2016, 8, 992
15 of 16
TERRA-MODIS, we are able to show that an increase in bare soil coverage occurs during dry years.
This outcome is related mainly to the decision, made by most farmers, not to sow new crops during
agricultural years that are preceded by an autumn with low levels of precipitation. This aspect is
clearly revealed by the TERRA-MODIS data, due to its higher spatial resolution and greater sensitivity
for the detection of fields with bare soils.
Acknowledgments: This study was funded by the AMETHYST project (ANR-12-TMED-0006-01, ANR TRNSMED
program). We would like to thank VITO and NASA for kindly providing us with its SPOT-VEGETATION and
TERRA-MODIS NDVI products, and the ISIS program for providing us with SPOT images. We wish to thank the
Tunisian Ministry of Agriculture for providing us with the precipitation data used in this study.
Author Contributions: Mehrez Zribi and Ghofrane Dridi proposed the methodologies used to compare the
SPOT-VGT and TERRA-MODIS products. Rim Amri developed various algorithms for the processing of ground
and satellite data. Zohra Lili Chabaane contributed to data collection. Mehrez Zribi and Ghofrane Dridi wrote
and revised the manuscript.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
Heim, R.R. A review of twentieth-century drought indices used in the United States. Bull. Am. Meteorol. Soc.
2002, 84, 1149–1165.
Quiring, S.M. Developing objective operational definitions for monitoring drought. J. Appl. Meteorol. Climatol.
2009, 48, 1217–1229. [CrossRef]
Gommes, R.; Petrassi, F. Rainfall variability and drought in Sub-Saharan Africa since 1960.
In Agro-Meteorology Series Working; Version 9; Food and Agriculture Organization: Rome, Italy, 1994.
Vicente-Serrano, S.M.; Beguería, S.; López-Moreno, J.I. A Multi-scalar drought index sensitive to global
warming: The standardized precipitation evapotranspiration index—SPEI. J. Clim. 2010, 23, 1696–1718.
[CrossRef]
Palmer, W.C. Meteorologic Drought; Research Paper; U.S. Department of Commerce, Weather Bureau:
Washington, DC, USA, 1965.
McKee, T.B.; Doesken, N.J.; Kleist, J. The relationship of drought frequency and duration to time scales.
In Proceedings of the 8th Conference on Applied Climatology of American Meteorological Society, Anaheim,
CA, USA, 7–22 January 1993.
McKee, T.B.; Doesken, N.J.; Kleist, J. Drought monitoring with multiple time scales. In Proceedings
of the 9th Conference on Applied Climatology of American Meteorological Society, Dallas, TX, USA,
15–20 January 1995.
Kogan, F.N. Application of vegetation index and brightness temperature for drought detection. Adv. Space Res.
1995, 15, 91–100. [CrossRef]
Ji, L.; Peters, A. Assessing vegetation response to drought in the northern Great Plains using vegetation and
drought indices. Remote Sens. Environ. 2003, 87, 85–89. [CrossRef]
Singh, R.P.; Roy, S.; Kogan, F.N. Vegetation and temperature condition indices from NOAA AVHRR data for
drought monitoring over India. Int. J. Remote Sens. 2003, 24, 4393–4402. [CrossRef]
Amri, R.; Zribi, M.; Lili-Chabaane, Z.; Wagner, W.; Hauesner, S. Analysis of ASCAT-C band scatterometer
estimations derived over a semi-arid region. IEEE Trans. Geosic. Remote Sens. 2012, 50, 2630–2638. [CrossRef]
Kogan, F.N. Global drought watch from space. Bull. Am. Meteorol. Soc. 1997, 78, 621–636. [CrossRef]
Gu, Y.; Brown, J.F.; Verdin, J.P.; Wardlow, B. A five-year analysis of MODIS NDVI and NDWI for grassland
drought assessment over the central Great Plains of the United States. Geophys. Res. Lett. 2007, 34. [CrossRef]
Amri, R.; Zribi, M.; Duchemin, B.; Lili-Chabaane, Z.; Gruhier, C.; Chebouni, A. Analysis of vegetation
behavior in a semi-arid region, using SPOT-VEGETATION NDVI data. Remote Sens. 2011, 3, 2568–2590.
[CrossRef]
Paris Anguela, T.; Zribi, M.; Habets, F.; Hasenauer, S.; Loumagne, C. Analysis of surface and root soil
moisture dynamics with ERS scatterometer and the hydrometeorological model SAFRAN-ISBA-MODCOU
at Grand Morin watershed (France). Hydrol. Earth Syst. Sci. 2008, 5, 1903–1926. [CrossRef]
Remote Sens. 2016, 8, 992
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
16 of 16
Albergel, C.; Zakharova, E.; Calvet, J.C.; Zribi, M.; Pardé, M.; Wigneron, J.P.; Novello, N.; Kerr, Y.; Mialon, A.;
Nour-ed-Dine, F. A first assessment of the SMOS data in southwestern France using in situ and airborne soil
moisture estimates: The CAROLS airborne campaign. Remote Sens. Environ. 2011, 115, 2718–2728. [CrossRef]
Zribi, M.; Pardé, M.; Boutin, J.; Fanise, P.; Hauser, D.; Dechambre, M.; Kerr, Y.; Leduc-Leballeur, M.;
Reverdin, G.; Skou, N.; et al. CAROLS: A new airborne l-band radiometer for ocean surface and land
observations. Sensors 2011, 11, 719–742. [CrossRef] [PubMed]
Sellers, P.J. Canopy reflectance, photosynthesis and transpiration. Int. J. Remote Sens. 1985, 6, 1335–1372.
[CrossRef]
Myneni, R.B.; Los, S.O.; Asrar, G. Potential gross primary productivity of terrestrial vegetation from 1982 to
1990. Geophys. Res. Lett. 1995, 22, 2617–2620. [CrossRef]
Propastin, P.; Kappas, M. Modeling net ecosystem exchange for grassland in Central Kazakhstan by
combining remote sensing and field data. Remote Sens. 2009, 1, 159–183. [CrossRef]
Seiler, R.A.; Kogan, F.; Wei, G. Monitoring weather impact and crop yield from NOAA AVHRR data in
Argentina. Adv. Space Res. 2000, 26, 1177–1185. [CrossRef]
Quiring, S.M.; Ganesh, S. Evaluating the utility of the Vegetation Condition Index (VCI) formonitoring
meteorological drought in Texas. Agric. For. Meteorol. 2010, 150, 330–339. [CrossRef]
Gitelson, A.A.; Kogan, F.N.; Zakarin, E.; Spivak, L.; Lebed, L. Using AVHRR data for quantitative estimation
of vegetation conditions: Calibration and validation. Adv. Space Res. 1998, 22, 673–676. [CrossRef]
Fensholt, R.; Proud, S.R. Evaluation of Earth Observation based global long term vegetation
trends-Comparing GIMMS and MODIS global NDVI time series. Remote Sens. Environ. 2012, 119, 131–147.
[CrossRef]
Zribi, M.; Chahbi, A.; Lili, Z.; Duchemin, B.; Baghdadi, N.; Amri, R.; Chehbouni, A. Soil surface moisture
estimation over a semi-arid region using ENVISAT ASAR radar data for soil evaporation evaluation.
Hydrol. Earth Syst. Sci. 2011, 15, 345–358. [CrossRef]
Shepard, D. A tow dimensional interpolation function for regularly spaced data. In Proceedings of
the 23rd ACM National Conference of the Association for Computing Machinery, Princeton, NJ, USA,
27–29 August 1968.
Orvos, P.I.; Homonnai, V.; Várai, A.; Bozóki, Z.; Jánosi, I.M. Trend analysis of a new MODIS drought severity
index with emphasis on the Carpathian Basin. Q. J. Hung. Meteorol. Serv. 2014, 118, 323–333.
Shabou, M.; Zribi, M.; Lili-chabaane, Z.; Amri, R.; Selmi, A. Evolution de l’occupation du sol sur le
site semi-aride de Merguellil (Tunisie centrale), en utilisant des données optiques de hautes résolutions
(SPOT/HRV, Landsat). In Proceedings of the International Workshop: Water Resources in Kairouan Plain:
Natural Constraints and Social Developments, Tunis, Tunisia, 16–17 November 2011.
Settle, J.J.; Drake, N.A. Linear mixing and the estimation of ground cover proportions. Int. J. Remote Sens.
1993, 14, 1159–1177. [CrossRef]
© 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC-BY) license (http://creativecommons.org/licenses/by/4.0/).