Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 1
Copyright Ó 2010, Paper 14-003; 59367 words, 10 Figures, 0 Animations, 3 Tables.
http://EarthInteractions.org
Intraseasonal Variability of
Satellite-Derived Rainfall and
Vegetation over Southern Africa
Hector Chikoore and Mark R. Jury*,1
University of Zululand, KwaDlangezwa, South Africa
Received 31 March 2008; accepted 24 January 2010
ABSTRACT: Satellite-derived rainfall and vegetation [normalized difference
vegetation index (NDVI)] data for the period 1981–2000 are used to reveal
spatial and temporal interrelationships via principal component analysis. The
unimodal seasonal cycle peaks in austral summer over the Zambezi Valley and
interior plateau. Spectral analysis indicates major cycles of intraseasonal rainfall events at ;20 and 40 days in response to atmospheric wave forcing that is
partially phase locked to the annual cycle. The spatial loadings exhibit a ‘‘center
of action’’ along the eastern edge of the Kalahari (6238E), extending from the
western Zambezi toward central South Africa, spanning 208 of latitude. The
loading pattern is consistent with tropical–temperate troughs and associated
northwest (NW) cloud bands. Twenty-day rainfall oscillations occur over the
Agulhas Current and Angola, hence NW cloud bands are a slower terrestrial
harmonic of the faster modes at either end. NDVI also exhibits intraseasonal
cycles operating near 40 days with spatial loadings collocated with the slow
rainfall mode at a 10–20-day lag. It is postulated that an earlier rainfall event and
subsequent ‘‘greening’’ results in a moisture flux that promotes the next rainfall
event. This is reflected in a composite analysis of low-level velocity potential.
The vegetation draws airflow toward itself in a self-sustaining way.
KEYWORDS: Intraseasonal rainfall variability; Satellite vegetation patterns; Southern Africa
1
Additional affiliation: University of Puerto Rico, Mayaguez, Puerto Rico.
* Corresponding author address: Mark Jury, Physics Department, University of Puerto Rico,
P.O. Box 9015, Mayaguez, PR 00681.
E-mail address: [email protected]
DOI: 10.1175/2010EI267.1
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 2
1. Introduction
While there has been significant progress in understanding climate variability
and predictability in many regions of Africa, the role of vegetation in the African
climate system is less established. Whereas vegetation growth and distribution are
largely determined by climate (Woodward 1987; Wang 2004), vegetation and landuse characteristics can feed back on the climate (Zeng and Neelin 2000; Wang
2004).
The impact of vegetation variability has received increased attention since
Charney (Charney 1975) pioneered research on vegetation–climate dynamics in a
study of Sahel drought. Research relating African vegetation to climate was initially based on satellite-derived, large-scale vegetation response patterns to rainfall
(e.g., Justice et al. 1986; Malo and Nicholson 1990; Davenport and Nicholson
1993). Several studies have observed or modeled effects of an altered land surface, including changes in available moisture, changes in surface roughness, and
changes in aerosol properties of the overlying atmosphere. Some modeling work
has focused on vegetation response patterns to climate change (e.g., Drew 2004).
Advances in numerical modeling have not only included simulating vegetation
response to climate variability but also atmospheric response to vegetation change
(Wang 2004).
The normalized difference vegetation index (NDVI) is a robust indicator for
vegetation condition and has been used in many studies as a measure of vegetation.
Research has shown that rainfall and vegetation are best related in semiarid regions
(e.g., Davenport and Nicholson 1993; Zeng et al. 2002) because of high evaporative losses during dry spells.
Gondwe and Jury (Gondwe and Jury 1997) investigated the sensitivity of vegetation to the summer climate over southern Africa using relationships between
NDVI and satellite outgoing longwave radiation (OLR) as a measure of convective
rainfall. The study found an inverse relationship between NDVI and OLR. The
NDVI reaches a minimum over southern Africa in September and rises sharply
thereafter but is less variable during the austral summer.
Davenport and Nicholson (Davenport and Nicholson 1993) studied the relationship between rainfall and NDVI for East Africa and determined that the NDVI–
rainfall relationship is strongest over the semiarid bushland/thicket or shrubland
zones, but the wetter woodlands exhibit little correlation. Gondwe and Jury (Gondwe
and Jury 1997) also found that the Zambezi Valley is less sensitive to climate impacts than the southern plateau. A point of saturation is reached after which vegetation conditions (NDVI) do not improve even with increasing rainfall (Davenport
and Nicholson 1993).
Anyamba et al. (Anyamba et al. 2002) studied El Niño and La Niña events via
NDVI anomalies over East and southern Africa during the period 1997–2000. They
detected a distinct contrast in vegetation anomalies during the warm 1997/98 and
cold 1999/2000 ENSO events. Their study determined an inverse relationship
between sea surface temperatures (SSTs) of the central Pacific and NDVI anomalies over southern Africa. A similar inverse relationship between southern African
rainfall and Pacific SSTs has long been established. NDVI anomalies were used
in Anyamba et al. (Anyamba et al. 2002) to show that the drought of 1997/98
over southern Africa was not as severe as previous droughts associated with warm
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 3
ENSO events. Anyamba and Eastman (Anyamba and Eastman 1996) determined
that vegetation variability over southern Africa is largely modulated by the ENSO
phenomenon at time scales ranging from 4 to 7 years.
Satellite NDVI data has only been available since the 1980s and hence previous
studies used a limited set of data, for example, 1982–85 in Davenport and Nicholson
(Davenport and Nicholson 1993), 1982–87 in Farrar et al. (Farrar et al. 1994), 1983–
88 in Richard and Poccard (Richard and Poccard 1998), and 1997–2000 in Anyamba
et al. (Anyamba et al. 2002). Most studies over southern Africa have used monthly
NDVI data (e.g., Anyamba et al. 2002; Gondwe and Jury 1997; Jury et al. 1997b),
limiting the ability to discern responses at intraseasonal time scales.
It is known that vegetation responds to rainfall via soil moisture—an integrator
(Malo and Nicholson 1990). Soil moisture exhibits a longer memory than most
atmospheric variables (as do SSTs) and thus has potential to influence atmospheric
variability and predictability (Entin et al. 1999). In the short term, besides providing moisture for evaporation, near-surface soil moisture controls the partitioning
of available energy at the ground surface into sensible and latent heat exchange with
the boundary layer. Over longer periods, soil moisture also modulates droughts and
floods (Pan et al. 1999). Thus, an understanding of the distribution and linkages of
soil moisture to evapotranspiration is essential to predict upward feedbacks of land
surface processes on weather and climate.
Yet, intraseasonal oscillations of vegetation and interactions with intraseasonal
rainfall have not received much attention. This is mainly due to the use of monthly
data resolution that is too coarse for intraseasonal analysis. This study therefore
pioneers research on intraseasonal variability of vegetation over southern Africa
using 10-day NDVI data and advances the understanding of associated processes
and interactions.
2. Background
The vegetation of southern Africa (south of 158S) is dominated by grassland
savanna (Jury et al. 1997b), but forests and woodlands occur in the east with dry
and thorny savannas in the central desert. In the west, the soil is nearly bare.
Miombo woodlands characterize the savanna regions but the largest of the savannas is the Kalahari, which is composed of grasslands and small groups of trees
covering most of Botswana, Namibia, and part of South Africa. The trees and
shrubs of the arid savannas in southern Africa are largely deciduous. The vegetation biomes of southern Africa have been detailed by Rutherford and Westfall
(Rutherford and Westfall 1986).
Spatial gradients exist between the dense vegetation cover to the northeast and
the arid southwestern desert (Gondwe and Jury 1997). Satellite-derived vegetation
images show a distinct boundary between the ‘‘green’’ northeast and the ‘‘brown’’
western areas (Figure 1). Similar gradients exist in the central Great Plains of the
United States (Wang et al. 2001) and in the Sahel from the Guinea coast to the
Sahara Desert (Zeng et al. 2002). The rainfall of southern Africa also exhibits
west–east gradients from less than 50 mm of annual rainfall to more than 900 mm
in the east (Bartman et al. 2003).
Rainfall is a key climatic element over southern Africa—characterized by high
spatial and temporal variability occurring at several time scales. More than 80% of
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 4
Figure 1. Summer mean NDVI over southern Africa, with place names used in the
text.
the region’s annual rainfall is received during the austral summer from December
to February (Buckle 1996; Bartman et al. 2003). The rainfall is quite convective
and related to the proximity of the intertropical convergence zone (ITCZ). Within
this seasonal cycle, there is an alternating sequence of erratic wet and dry spells at
near monthly intervals (Makarau 1995; Tennant and Hewitson 2002) with critical
implications for rain-fed agriculture. About 100 mm of rainfall are received during
a wet spell over many days (Jury and Nkosi 2000), while dry spells are frequently
established over large areas with more homogeneity than wet spells (Vogel 2000).
During a dry spell, an intense midtropospheric anticyclone is often established with
a center over Botswana causing subsidence and restricting rainfall over the subcontinent (Buckle 1996).
Intraseasonal oscillations (ISOs) refer to cycles operating at time scales shorter
than the annual cycle. A well-known tropical fluctuation is the Madden–Julian
oscillation (MJO), which has eastward-moving areas of zonal winds, moisture
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 5
convergence, and surface pressure in the tropical zone that bring rain every
30–60 days. At times, the MJO slows and weakens, especially over the tropical
Atlantic. Hence incoming wave energy is somewhat disorganized over Africa.
Several studies have found major cycles of intraseasonal rainfall over southern
Africa operating between 20 and 60 days (Pohl et al. 2007). Makarau (Makarau
1995) found five main wet spells and two main dry spells during the summer, each
lasting 5–10 days. Rainfall cycles of 10–25 days and 40–50 days were found using
spectral analysis. Wet spells were related to troughs extending from Angola in the
northwest to the Mozambique Channel in the southeast. Similarly, Levey and Jury
(Levey and Jury 1996) found a combination of 20–30- and 40–60-day oscillations
in rainfall over southern Africa. The study found the MJO prevalent in wet years,
while the 20–30-day ISO seemed more dominant in dry years. About one-third of
the rainfall oscillations were found to arise locally from stationary modes linked to
tropical–temperate troughs, with two fast ISOs ‘‘inside’’ a slower MJO. Another
one-third of ISOs propagate eastward in concert with a ‘‘right hook’’ pattern in the
subtropical jet stream. Similar synoptic features were described earlier by Tyson
(Tyson 1981), Lyons (Lyons 1991), and Jury et al. (Jury et al. 1996). Wet spells
have also been associated with weakening and bifurcation of upper westerlies south
of Africa (Jury and Levey 1997). Anyamba (Anyamba 1992) detected a 40–50-day
oscillation and a more dominant 20–35-day oscillation in OLR anomalies over the
tropical western Indian Ocean, with eastward propagation linked to the tropical MJO.
Many studies have described interannual rainfall variability over southern
Africa and its remote forcing, but intraseasonal characteristics have received less
attention. The objective of this paper is to establish the intraseasonal spatial distribution and temporal variability of rainfall and vegetation modes over southern
Africa using gridded station–satellite merged [Climate Prediction Center (CPC)
Merged Analysis of Precipitation (CMAP)] rainfall and NDVI data. Seasonal
characteristics are described in less detail and system propagation will not be
analyzed here. The interaction between the ISO patterns of vegetation and rainfall
over the subcontinent is investigated. We hypothesize that, in addition to a lagged
vegetation response to rainfall, there is ‘‘biotic memory’’ imparted by moisture
fluxes that follow vegetation growth. These influence the subsequent rainfall
event, helping the northwest cloud bands to find ‘‘anchor points’’ over southern
Africa.
3. Data and methods
The National Centers for Environmental Prediction (NCEP) provide CMAP
5-day-averaged (pentad) precipitation data at 2.58 3 2.58 resolution for the period
1981–2000. The CMAP precipitation data are used in this study to analyze spatial
and temporal patterns of variability over southern Africa and the adjacent oceans.
The 5-day pentad data are produced by merging rain gauge observations with
satellite precipitation estimates inferred from infrared radiation (IR), OLR, Microwave Sounding Unit (MSU), and Special Sensor Microwave Imager (SSM/I).
Over oceanic regions, the satellite infrared and microwave data are merged with
gauge observations from some islands. The CMAP precipitation data are thus continuous in time and space. For comparison with the NDVI data, we average the
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 6
pentad data into 10-day blocks. Earlier studies on rainfall variability over southern
Africa have mainly used station data over terrestrial domains (e.g., Makarau 1995;
Mulenga 1998; Unganai and Mason 2001; Tennant and Hewitson 2002). Tazalika
and Jury (Tazalika and Jury 2008) studied seasonal and intraseasonal rainfall
characteristics of tropical central Africa using CMAP rainfall data, as done here.
Detailed spatial coverage of the presence and condition of vegetation is provided
by satellite-derived biophysical indicators such as the NDVI, leaf area index,
fraction of vegetation cover, fraction of absorbed photosynthetically active radiation, and directional albedos. NDVI is often preferred because it seems to best
represent vegetation responses to climatic effects particularly in semiarid regions
(Davenport and Nicholson 1993). NDVI is a measure of vegetation reflectance in
the visible band that indicates ‘‘color.’’ High reflectances are linked to green and
vibrant vegetation while lower reflectances refer to brown vegetation or bare surfaces. The NDVI has proven a robust indicator of biomass production, photosynthetic activity, evapotranspiration, rainfall, and crop yield in semiarid regions
(Gondwe and Jury 1997). It is less variable in humid forest regions where high
rainfall induces NDVI ‘‘saturation’’ (Brunsell 2006). Davenport and Nicholson
(Davenport and Nicholson 1993) determined that NDVI reflects soil moisture better
than rainfall. Other uses for NDVI include land-cover classification, crop and drought
monitoring, and desertification studies (Tucker and Sellers 1986; Gutmann 1990).
The NDVI data (1982–2000) are acquired by the Advanced Very High Resolution Radiometer (AVHRR) aboard the National Oceanic and Atmospheric
Administration (NOAA) polar-orbiting environmental satellites. The data are extracted from the U.S. Geological Survey (USGS)/African Data Dissemination
Service (ADDS) Web sites. The NDVI data have been detailed by Tucker et al.
(Tucker et al. 2005).
NDVI is a ratio of vegetation reflectance at two wavelengths—one band in the
visible region (0.58–.68 mm) and another in the near infrared (NIR; 0.725–
1.1 mm). Vegetation reflectance maximizes at about 0.8 mm, making the NIR
channel more sensitive to vegetation color (Tucker 1979; Tucker et al. 1991; Jury
et al. 1997b). Since the NDVI is a measurement of the intercepted or absorbed
photosynthetically active radiation (Myneni et al. 1995), a sum of the NDVI results
in a quantity highly related to net primary production (Tucker et al. 2005). Vegetation reflects a greater proportion of near-infrared light than visible (red) light,
but bare soils or little vegetation will have similar reflectances in both bands.
Mathematically, NDVI is a nonlinear function ranging from 21 to 11, but typical
values on the Earth’s surface vary between 20.1 and 0.7 with negative values
indicating water bodies, snow surfaces, and clouds (Wang et al. 2003). NDVI
values below 0.3 indicate scant vegetation and water stress conditions (Jury et al.
1997a), while values above 0.5 reflect dense and vibrant vegetation. The summer
mean vegetation pattern over southern Africa is shown in Figure 1.
3.1. Quality control and subsampling
The ADDS NDVI are obtained at 10-day or dekadal time resolution to enable
sufficient overpasses of the satellite to produce cloud-free images based on the
maximum value at each pixel. NDVI values are often reduced in the presence of
water vapor, clouds, snow, and aerosols. In regions of agricultural activity, types of
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 7
crops, soil types, and irrigation may influence observed changes in NDVI (Gondwe
and Jury 1997). The soil background also affects NDVI magnitudes with darkcolored soils such as vertisols and fluvisols increasing NDVI, while light-colored
soils reduce it (Nicholson and Farrar 1994). Atmospheric interference tends to
reduce NDVI through cloud contamination, Rayleigh scattering, and absorption by
atmospheric constituents such as ozone and aerosols. The ADDS data are corrected
for these effects.
NDVI data are available at a high spatial resolution of 8 km 3 8 km. For the
purposes of this climate-oriented study, the data are reduced to a 2.58 3 2.58 grid
through a process of subsampling by averaging with neighboring pixels. NDVI
values near a coast or lake are usually low because of spatial averaging with water
reflectance (Davenport and Nicholson 1993) so we interpolate using values from
adjacent inland pixels. Here the ADDS NDVI data are extracted in the latitudes of
108–358S.
The data may be viewed as a ‘‘stack’’ of NDVI images over time. Extracting a
value for each geographic point in the stack creates a time series. The data are
standardized for further analysis with respect to their historical mean and standard
deviation. This procedure assists intercomparisons.
Wavelet analysis is used to both filter and analyze the time series of standardized
CMAP precipitation and NDVI data. Wavelet analysis is widely used in the geophysical sciences as an alternative method to spectral analysis techniques to examine
how the power spectrum changes with time (Torrence and Compo 1998). While
Fourier expansion consists of breaking up a signal into sine waves of various frequencies, it has no time resolution. Wavelet analysis transforms a one-dimensional
time series (or frequency spectrum) into a two-dimensional image depicting evolution of frequencies with time (Lau and Weng 1995).
Both the satellite CMAP rainfall and NDVI data are prefiltered using wavelet
analysis. For rainfall at pentad resolution, cycles between 10–35 days and 20–70
days are retained as suggested by earlier studies (Levey and Jury 1996; Makarau
1995). We refer to these as ‘‘fast’’ and ‘‘slow’’ modes, respectively. The time resolution of the NDVI data is 10 days so only 20–70-day cycles can be resolved. We
investigate how the spectral energy changes over the study period through modulus
analysis. This resolves the wavelet power at each time step for periods allowed by
the filter.
Principal component analysis (PCA) or empirical orthogonal function (EOF)
analysis is used to analyze the space–time variance of pentad CMAP precipitation
and dekad NDVI data. PCA is a useful statistical method that is widely used in
climate research as a method for data reduction and factor analysis. In this study,
PCA is used to express the spatiotemporal variability of rainfall and vegetation as a
number of principal components. Each principal component is a linear combination of the original variables with eigenvalues that consists of a spatial pattern and a
time series.
PCA is applied as a data reduction or structure detection method that describes
the fundamental modes of variability. The first principal component (PC1) represents the greatest variance in the data, PC2 the second greatest variance, and so on.
Seasonal and intraseasonal cycles of rainfall and vegetation variability are described by temporal scores. It is accepted that PCA is a statistical technique that
requires careful interpretation to attach physical significance.
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 8
Rotation is often applied to isolate spatial and temporal structure in the modes
(Mestas-Nuñez 2000). This additional step is available for extraction of principal
components that maximizes the variance (varimax) and enhances the spatial
loadings of the dominant modes (Jury 1999). This is applied in the case of the fast
intraseasonal modes, but for the annual cycle and slow modes we find that rotation
is unnecessary—we analyze the first unrotated mode instead.
The number of PCs that represent significant variance is determined by scree
test, which identifies the contribution of the eigenvalue to the total variance. A
number of studies have used PCA to study rainfall over southern Africa (e.g.,
Mulenga 1998; Jury 1999; Unganai and Mason 2001). Here, PCA is performed on
satellite CMAP precipitation and NDVI data using consecutive maps over a 20-yr
period (1981–2000). All loadings are made positive so that the time scores reflect
wet or green conditions when departures are positive.
In this study, composite analysis is used to detect common features for wet spells
and ‘‘green events,’’ and also to contrast wet and dry summers during the period
1981–2000. Although useful, it is understood that the results will depend on the
choice of samples included in the composite, for example, a mixing of events of
different phase, propagation, and amplitude contributed by weather systems that
may be both tropical and extratropical.
Intraseasonal composites are analyzed for selected events with large-amplitude
changes in NDVI (.2s) over the ‘‘Kalahari center of action’’ (188–308S, 208–278E).
These are referred to as green events. We compute area-averaged CMAP rainfall, the vertical gradient of specific humidity, layer-averaged dewpoint depression, latent heat flux, and low-level velocity potential for the same group of cases
and area using NCEP reanalysis data obtained from the NOAA Operational
Model Archive Distribution System (NOMADS) Web site. Intraseasonal composites are used here to qualitatively describe the upward impact of vegetation on
the boundary layer. Compositing is potentially problematic because events in the
composite may be having intraseasonal cycles with different speeds, different
maxima, and different amplitudes. We consider the spread of individual cases in
the composite and find reasonable coherence for green events in the center of
action. Coherence diminishes with time away from the dekad of peak vegetation
change, as described later.
NDVI data are analyzed for seasonal and intraseasonal fluctuations with a focus
on the summer rainfall season. Cross correlation is performed to study relationships between NDVI and rainfall at various leads and lags, particularly for green
events to reveal the strength of relationships. A simultaneous or lagged response
(Table 1) was shown in 10 out of 17 events. The remaining events appear externally
forced or locally uncoupled. These cases are excluded from the composite group.
4. Results
4.1. Seasonal cycles
Standardized, unfiltered, and unrotated data are used to reveal the seasonal cycle.
The first principal component (PC1) is the only mode meeting the scree test criteria; succeeding modes are noisy and do not provide useful insights. Seasonal
rainfall PC1 accounts for 34.2% of the total variance across the domain of study.
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 9
Table 1. Correlation coefficients for dekadal NDVI and rainfall at different lags
based on events with 2s change of NDVI (green events). Dates refer to the beginning of the rain event, and lags refer to 10-day periods thereafter. Significant
positive coefficients are bold; values <0.2 are omitted for clarity.
Lags
Events with large-amplitude
change of NDVI
3
3
2
1
1
1
1
3
3
3
3
2
3
1
1
2
2
*
Dec 1983*
Nov 1984
Dec 1985*
Nov 1986
Dec 1986
Oct 1987*
Nov 1988*
Feb 1989*
Nov 1991
Nov 1992*
Nov 1993*
Dec 1995
Jan 1997*
Jan 1999
Jan 2000*
Nov 2000*
Dec 2000
0
1
0.25
0.42
0.23
0.69
0.70
0.76
0.68
0.76
0.84
0.63
0.39
0.91
2
3
0.49
0.79
0.85
0.41
0.44
0.95
0.83
0.44
0.93
0.65
0.73
0.60
Included in composite group.
Maximum spatial loadings are observed in the region of the Zambezi River valley
(;158S, 258–508E) extending northwest–southeast from Angola toward Mozambique
(Figure 2a) reflecting the zonal ITCZ. Time scores show the seasonal oscillation
peaking typically during the austral summer from December to February (DJF)
when the ITCZ is at its most southerly position, becoming a minimum from June to
August (JJA; Figure 2b). Time scores reflect above normal rainfall in the 1984/85,
1988/89, 1996/97, and 1999/2000 seasons. Low rainfall associated with El Niño is
evident for the 1982/83 and 1991/92 seasons. Such oscillations appear to depend on
the large-scale zonal flow that advects Indian (Atlantic) Ocean air masses and
weather systems in wet (dry) years. The dominant frequency is at the annual cycle,
but there is some spectral energy with a lower-frequency ‘‘2 years at a time’’
characteristic (e.g., 1981–82, 1988–89, 1993–94, and 1996–97).
The annual cycle for NDVI has a similar pattern, cycle, and variance (34%), so it
is omitted from discussion here. Its relationship with rainfall is discussed later.
4.2. Intraseasonal variability
In this section we analyze the timing and duration of wet and dry spells and the
associated vegetation responses and feedbacks. We study the 20- and 40-day oscillations independently using a selective wavelet band filter applied prior to PCA.
From the results of the PCA for data filtered to retain cycles between 20 and
70 days, only the first PC is retained for discussion according to scree test criteria.
Rainfall PC1 accounts for about 12% of variance across the domain of study. The
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 10
Figure 2. (a) Spatial loading and (b) time score for rainfall (unfiltered, unrotated) PC1
explaining 34% of total variance.
spatial loadings for PC1 maximize in a northwest–southeast axis from the western
Zambezi toward central South Africa (Figure 3a). The loading zone spans about
208 of latitude. Mulenga (Mulenga 1998) determined a similar loading pattern for
midsummer rainfall (December–February). The loading region is where tropical
lows and subtropical troughs interact to produce northwest (NW) cloud bands on
the eastern edge of the Kalahari ;238E (Harrison 1984; Levey and Jury 1996;
Vogel 2000; Tennant and Hewitson 2002). Variations of the NW cloud band thus
determine intraseasonal rainfall variability over the subcontinent and facilitate the
poleward transport of tropical moisture and momentum (D’Abreton and Lindesay
1993; Crimp et al. 1998; Tennant and Hewitson 2002). In drier summers, tropical–
temperate troughs shift east of the subcontinent (;558E) and convection is
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 11
Figure 3. (a) Spatial loading and (b) time scores for intraseasonal rainfall (filtered
20–70 days, unrotated) for PC1.
displaced to the Indian Ocean. The time score amplifies in December–January at
the peak of the annual cycle. Time scores reflect strong pulsing during the late
1980s (Figure 3b) but lower amplitudes in the dry summers of 1982/83 and 1991/92.
The modulus spectrum has maximum energy varying between 30 and 50 days that
may be attributed to the influence of tropical waves such as MJO.
Satellite vegetation NDVI data filtered by wavelet analysis to retain cycles between 20 and 70 days enables a determination of responses to major rainfall events.
The first unrotated vegetation mode PC1 is positively loaded in a northwest–
southeast band in the central interior extending through the western Zambezi toward central South Africa on the eastern edge of the Kalahari (Figure 4a) and is
referred to as the Kalahari center of action. Gondwe and Jury (Gondwe and Jury
1997) found a similar loading pattern over the Kalahari using monthly NDVI over a
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 12
Figure 4. (a) Spatial loading, (b) time scores, and (c) the modulus spectrum for
intraseasonal (20–70 days, filtered) vegetation NDVI (unrotated) for PC1.
Shading in (a) defines the center of action.
shorter time period. The NDVI loading pattern is collocated with the CMAP
rainfall loading (Figure 3a), except for a truncation on the southern flank due to the
terrestrial nature of the data and the contour analysis. This is an important result
and most of the subsequent analysis hinges on this collocation.
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 13
Intraseasonal NDVI PC1 accounts for 12% of variance across the domain and
reflects the increased sensitivity of midrange vegetation to slowly evolving weather
events. The loading pattern is affected by the greener and browner areas to the east
and west being less sensitive to climatic forcing. Time scores for vegetation PC1
exhibit amplification in summer, more so in the wet summers of 1989, 1994, and
2000 (Figure 4b). Consistent pulsing of vegetation during the dry season may be an
artifact of standardized data due to low values in winter and spring. The modulus
spectrum reveals 25–45-day cycles as confined by the filter (Figure 4c). Rainfall
modulus (not shown) displays a similar characteristic, except for a stronger annual
component with adjacent years being disconnected.
High-frequency rainfall PCA modes exhibit oscillations mainly around 20 days.
Spatial loadings are interpreted for rotated modes because they bring out seasonality within the ISO. The percentage of variance explained by each ISO mode is
lower than that explained by the seasonal mode (34.2%) thus establishing the
annual cycle and the associated ITCZ as the dominant factor in rainfall variability
over southern Africa.
Although the first four 20-day modes of variability are significant according to
scree test criteria, we focus attention on the first two fast modes that are aligned
northwest–southeast with the first slow mode. The first mode of 20-day rainfall
oscillations (PC1) exhibits positive spatial loadings over the Agulhas Current region
and the adjacent southeast coastal margins (Figure 5a). This mode of variability
is modulated by westerly waves of the subtropical jet stream. Frontal activity over
the warm Agulhas Current contributes to enhanced convection and rainfall there
(Walker 1990; Jury et al. 1993; Mason 1995). Unlike other modes, seasonality is
limited over the Agulhas region (Figure 5b); oscillations continue through the winter.
The loading pattern for the second mode of 20-day rainfall oscillations (PC2) is
focused on southeastern Angola (Figure 6a), possibly representing a western anchor point for the summer ITCZ where easterly waves and a thermal low interact.
Time scores indicate that this Angola mode is a summer phenomenon with amplitudes peaking during DJF, becoming quiet in winter (Figure 6b). The fast Angola
ISO exhibits strong oscillations in the late 1980s and the late 1990s but amplitudes are rather weak during the early 1990s. Mulenga (Mulenga 1998) studied
the Angola low and determined that it is a quasi-stationary thermal heat low that
drives moisture convergence over southern Africa.
To check the robustness of our interpretations, the spectral character of the
rainfall oscillations is reanalyzed without band filtering in the areas of maximum
loading. The 40-day oscillation is the major spectral peak (after the annual cycle)
for continental areas, while the 20-day ISO emerges in the marine region.
4.3. Seasonal mean ISO
We compare climatological rainfall and vegetation totals with collocated ISO
scores averaged each pentad or dekad over 20 years. This helps us to understand
whether the ISO is random or seasonally phase locked. The continental 40-day
oscillation of rainfall is ‘‘quiet’’ during the winter when rainfall is nearly absent
(Figure 7a). The onset of summer rainfall is marked by a phase-locked event.
Thereafter the mean ISO flattens until the end of December when two successive
large-amplitude oscillations occur: the first (in January) marks the midsummer
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 14
Figure 5. (a) Spatial loading and (b) associated time scores for intraseasonal rainfall
(filtered 10–35 days, rotated) for PC1.
optimum and the second (in March) ends the season (Figure 7a). Our mean rainfall
oscillation exhibits three large and three smaller wet spells during summer in
agreement with Makarau (Makarau 1995). It is unclear whether the 40-day oscillation is related to the tropical MJO, as our methodology highlights stationary
rather than transient features.
Climatological mean NDVI oscillations are more continuous through the year
(Figure 7b) with a distinct dry season minimum in September–October. The vegetation ISO peaks in early January, before the seasonal optimum. It reverses to be
out of phase in February–March and then peaks again in April (Figure 7b). Our
analysis reveals that intraseasonal oscillations of NDVI over southern Africa have
significant amplitude and that part of the variance is seasonally phase locked.
Biotic memory is thought to play a role in these phase relationships (Wang et al.
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 15
Figure 6. (a) Spatial loading and (b) associated time scores for intraseasonal rainfall
(filtered 10–35 days, rotated) for PC2.
2001). A composite of green and brown years is done and these are compared in
Figure 7c. The mean ISO for green years contains shorter cycles than in brown
years, which are reduced in midseason (Figure 7c). The key green event is in
January; otherwise, the two curves are quite similar.
We analyze the mean ISO to evaluate the relationship between high- and lowfrequency rainfall oscillations in Figure 8a. The slow Kalahari oscillation and fast
Agulhas oscillation (both PC1) are in phase at the beginning and end of summer, in
agreement with studies in nearby regions (Levey and Jury 1996; Jury and Nkosi
2000; Tazalika and Jury 2008). During midsummer, the Agulhas and Kalahari
oscillations are out of phase (Figure 8a). It is then that the fast Angola oscillation
(PC2) ‘‘takes over.’’ We find correspondence for major wet spells between the two
continental modes except in January (Figure 8b). Rains during January that are
critical to interannual variability seem driven by the fast Angola mode in isolation.
It is concluded that the 20-day marine mode that is modulated by westerly Rossby
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 16
Figure 7. (a) Mean ISO for rainfall PC1 and its 20-yr climatology (line with dots), (b)
mean ISO for NDVI PC1 (Kalahari, line with dots) and its 20-yr climatology,
and (c) mean ISO for green and brown years in the study period. The
calendar starts on 1 Jul, such that pentad 73 refers to 25–29 Jun.
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 17
Figure 8. Mean 40-day rainfall mode (Kalahari PC1, blue with dots) compared with
(a) 20-day Agulhas PC1 rainfall mode and (b) 20-day Angola PC2.
waves starts and ends the summer rainy season, while the Angola low interacts with
the NW cloud band during December, February, and April wet spells.
4.4. Vegetation feedback on climate
The collocation of dominant PCA modes for seasonal and intraseasonal rainfall
and vegetation NDVI and the correspondence of (10–20-day lagged) time scores
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 18
Table 2. Correlation coefficients for events with large-amplitude change of NDVI vs
rainfall and discrete rainfall events vs NDVI, all averaged over the center of action
at different lags and leads. Insignificant correlation values are omitted. Values in
boldface are significant above the 90% confidence level.
Lags (2)/Leads (1) (dekads)
NDVI green events
23
22
21
0
11
12
13
0.32
0.92
0.80
0.15
Discrete rainfall events
0.14
0.77
0.60
0.26
suggest that strong coupling exists. The center of action along the eastern edge of
the Kalahari is our focus here. We wish to understand the processes responsible
using parameters that describe the moisture fluxes and circulation. It is known that
vegetation responds to (lags) rainfall at intraseasonal time scales (Davenport and
Nicholson 1993), with factors such as soil condition and land use complicating the
problem (Farrar et al. 1994).
We seek the ‘‘upward’’ impact of vegetation on climate over the eastern Kalahari
following the analysis of Wang et al. (Wang et al. 2001). A composite analysis of
10 summer events with large-amplitude changes (.2s) of NDVI is made. Satellite
CMAP rainfall, NCEP reanalysis latent heat flux, and the divergent part of the
circulation for the composite cases are calculated and averaged over the center of
action to enable comparisons and lag cross correlation. The results are shown in
Table 1. Out of 17 events, 10 show a response, either simultaneous (zero lag) or
delayed by 1 or 2 dekads. The remaining events show little response and are excluded from the composite below.
Significant NDVI–rainfall correlations occur at either 1 or 2 dekads (10–20
days) lag. Events with large-amplitude change in NDVI occur mostly during the
early summer, suggesting that the vegetation–rainfall relationship may be strongest then—prior to NDVI reaching saturation with the additional influence of
weak temperature fluctuations and low winds during the late summer (February–
April).
A correlation analysis is made with respect to the composite green event described above and another for a composite of ‘‘discrete’’ rainfall events for comparison. Discrete rainfall events are defined here as occurring after a dry spell
.30 days with no prior change in vegetation. The results are shown in Table 2 at
lags up to 3 dekads. Events with large-amplitude changes of NDVI lag rainfall,
while discrete rainfall events lead NDVI. The weakest correlations between
rainfall and vegetation in either case occur at zero lag. High correlations are
noted at lag 1 and lag 2 for both NDVI green events and discrete rainfall events.
This result confirms the expected influence of rainfall, for example, leading
vegetation by 10–20 days. Similar results were obtained by Justice et al. (Justice
et al. 1991) for semiarid Niger and Mali and Wang et al. (Wang et al. 2001) over
the central United States.
Although NDVI green events follow a significant rainfall event, some critical
research questions arise:
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 19
1) To what extent does the evapotranspiration flux from vegetation greening after a
rainfall event affect the next wet spell?
2) Does the vegetation help to ‘‘anchor’’ cloud bands regardless of external forcing
by the large-scale circulation and heating anomalies?
As outlined above, NCEP reanalysis latent heat flux and low-level velocity potential in the Kalahari center of action were analyzed for the 10-event composite to
determine how an earlier rainfall event and subsequent greening affect the next
rainfall event.
A composite of NCEP reanalysis latent heat flux (Figure 9a) shows that rainfall
leads latent heat flux by 1 dekad, which in turn leads NDVI by another dekad
(Figure 9b). Surface evaporation after a rainfall event results in an increased latent heat
flux into the lower atmosphere. The vertical flux of moisture connects changes in
rainfall and vegetation. It carries the message from soil moisture to the lower atmosphere. The sequence of events before and after an NDVI green event is shown in
Table 3. The response of the boundary layer is observed in dekads D11, D12, . . . up
to D15. We recognize that composites become unreliable with increasing time
away from the peak of the event (D) given the various rhythms of ISO present.
Radiosonde profiles for a station within the center of action were analyzed. We
found that the dewpoint depression and boundary layer height (decline of specific
humidity, ›q/›z) closely follow the latent heat flux, becoming moist and deep after
the peak of rainfall at D21 (cf. Figure 9a) and vegetation maximum. This supports
the supposition that additional moisture flux into the boundary layer from enhanced
vegetation helps sustain the moisture in the absence of external forcing such as
horizontal moisture advection (U ›q/›x).
Variations of composite low-level velocity potential are in phase with the vegetation (Figure 9b). While it was initially expected that the low-level velocity
potential would reflect the large-scale circulation and correspond with rainfall,
instead it closely follows changes in vegetation. It peaks at D coinciding with the
NDVI maximum suggesting that an increase in vegetation ‘‘draws’’ airflow toward
itself in a self-sustaining way.
Radiosonde profiles for Bloemfontein are analyzed for the 700-hPa dewpoint
depression. We recognize that a single level may not indicate convective potential
and also that Bloemfontein is a single point at the southern edge of the center of
action. However, when compared against NCEP model area-averaged boundary
layer height, agreement is found (Figure 10). As the dewpoint depression decreases
(reflecting moist conditions), the moist boundary layer height deepens (Figure 10).
A turning point occurs at D21, coinciding with the latent heat flux maximum. An
upward trend in moist boundary layer height after a NDVI ‘‘big’’ event is noted.
The rise of latent heat flux one dekad after the rainfall event agrees with fluctuations of boundary layer height and dewpoint depression. The composite moist
boundary layer deepens markedly after a rainfall event at D21 but does not diminish as much after the event (Figure 10). The increase of boundary layer height
is distinct over the western semiarid regions. Changes over the more humid east
and northeast are not as marked. The mean composite boundary layer height increases after the rainfall event and deepens further after the vegetation maximum.
This supports the notion that transpiring vegetation acts to sustain moisture through
the boundary layer in the absence of external forcing.
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 20
Figure 9. Ten-event composite-averaged data over the center of action for (a)
CMAP rainfall (rf) and NCEP latent heat flux (lht) and (b) NDVI (Vg) and
NCEP low-level velocity potential (as labeled).
5. Discussion and conclusions
This study has focused on intraseasonal variability of satellite-derived rainfall
and vegetation over southern Africa. It is observed that the annual cycle is the
dominant mode (34%) of rainfall and vegetation variability over southern Africa,
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 21
Table 3. The sequence of events before and after the NDVI green event.
Time lag/lead
Situation
D23
D22
D21
D
D11
D12
D13
Dry, subsident
Rainfall maximum
Latent heat flux maximum
NDVI maximum
Horizontal convergence
Boundary layer deepening
Boundary layer deepening
associated with the summertime ITCZ across the Zambezi Valley. At intraseasonal
time scales, tropical–temperate troughs and the associated NW cloud band are
dominant.
Intraseasonal oscillations of rainfall over southern Africa exhibit spectral peaks
near 20 and 40 days. The slow mode is oriented northwest–southeast along the
eastern edge of the Kalahari spanning 208 of latitude, while fast intraseasonal
oscillations are focused on the subtropical Agulhas region and tropical Angola
Plateau. The 40-day oscillation ‘‘connects’’ two 20-day ISOs, suggesting they
elongate and join to produce NW cloud bands, with a slow harmonic of the faster
modes at either end. This supports the notion that tropical–temperate troughs form
when a tropical low over Angola is coupled to a temperate westerly wave, with a
right hook circulation that undercuts the poleward flowing tropical air.
Figure 10. Composite analysis of NCEP model boundary layer height (hgt) and inverted Bloemfontein 700-hPa dewpoint depression (DD).
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 22
Climatologically, there are three to six wet spells alternating with dry spells in
the summer rainfall season over southern Africa. Some of the intraseasonal variance is phase locked to the annual cycle, particularly at the onset of the summer
rains, and during major wet spells. The 20-day marine mode becomes a harmonic
of the 40-day oscillation at the start and end of the rainy season, while the Angola
mode becomes a harmonic of the 40-day oscillation in midsummer wet spells. The
phase locking of ISO over the oceans suggests that convection associated with
Rossby waves to the south of Africa progress eastward at preferred times of the
year (cf. Figure 8a). It was previously thought that these waves were random.
We note that our slow and fast ISO modes are displaced away from the main
region of ENSO influence. Correlations between CMAP rainfall and Pacific Niño-3
SSTs reach an optimum value along ;308E. However, the intraseasonal amplitude
peaks ;238E. This displacement means that seasonal predictions based on external SST forcing (Ropelewski and Halpert 1987; Janowiak 1988; Hastenrath et al.
1995; Anyamba et al. 2002; Jury et al. 2004) could go astray if strong internal
variability is present in the form of intraseasonal events (Burroughs 2003).
Processes involved in vegetation–rainfall dynamics over southern Africa were
studied, and we specifically sought the upward vegetation feedback in land–
atmosphere interactions. The regional vegetation exhibits a unimodal cycle reaching
an optimum in February–March and a minimum in September–October. Drought
years have an earlier start to the rainy season and deficient rainfall in December and
January.
The Kalahari ‘‘transition zone’’ is dynamic, being located farther west during
wet summers and advancing eastward in drier years. This zone reflects the dominant influence of NW cloud bands. Their associated rainfall imparts the observed
variations in NDVI and hence primary production. The sharp decrease in the
variance accounted for by lower PC modes suggests a lower amplitude response of
vegetation to fluctuations of rainfall west of 218E and east of 278E, in the latitudes
188–298S.
The annual cycle of vegetation has a spatial loading maximum that collocates
with the rainfall mode in a zonal axis around 158S. Intraseasonal oscillations of
vegetation occur at about 40 days in a northwest–southeast band extending from
the western Zambezi Valley southward to central South Africa along the eastern
edge of the Kalahari. The spatial loadings of vegetation collocate with rainfall
loadings despite differences in the way the data were derived and analyzed. Vegetation ISOs are not as seasonally distinct as their rainfall counterparts. This is
partly because of responses during the dry winter/spring and to the continuous
nature of the data.
Upward feedbacks of vegetation on rainfall-producing weather conditions are
more distinct in semiarid regions (Zeng et al. 2002). Thus, the NW cloud band over
Kalahari is a focus of boundary layer–land surface interactions at the intraseasonal time scale. Our results show that the additional vertical moisture flux
from the vegetation affects the next rainfall event through 1) a deeper boundary
layer and 2) enhanced horizontal convergence (cf. Figure 9b). An increase in
vegetation appears to draw the airflow toward itself, enhancing the low-level
buoyancy and slowing the winds through friction, thereby causing convergence
and uplift. Thus vegetation seems to impact horizontal momentum transfer as
much as vertical moisture flux.
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 23
The NCEP model is dependent on precipitation and temperature to produce
latent heat flux, regardless of whether the land surface is vegetated or just soil. The
lag between the latent heat flux and the boundary layer response is dependent on
hydrological transfers in the model, so the NDVI and radiosonde data were important ‘‘reality checks’’ on this process. The way the NCEP model velocity potential follows NDVI suggests the lags are real and created by evapotranspiration
impacts on the boundary layer. In this way, the land surface provides anchor points
for subsequent weather systems. While our analysis has revealed an interaction
between horizontal and vertical fluxes in the center of action, some uncertainty
remains on just how influential the Kalahari can be on tropical–temperate troughs
and attendant NW cloud bands, which may align according to hemispheric-scale
atmospheric Rossby waves. The Kalahari vegetation fraction is quite low, so an
evaluation of soil moisture feedback would provide more insight.
The main contribution of this work is the examination of vegetation–climate
dynamics in southern Africa at the intraseasonal time scale using 10-day NDVI
data over 20 years. This is in contrast to earlier work using monthly NDVI data
over a decade (e.g., Gondwe and Jury 1997; Richard and Poccard 1998; Anyamba
et al. 2002). Modeling studies by Dirmeyer (Dirmeyer 2001) demonstrate increases
in evapotranspiration, temperature, and humidity over semiarid regions that penetrate vertically through the lowest 30% of the troposphere, producing coherent
changes in precipitation of 10%–15% that depend on model and other factors. Here
our research implicates land surface coupling with the overlying boundary layer in
a meridional axis over central southern Africa, which lies just ‘‘upwind’’ from the
commercially important agricultural area of South Africa. An increased understanding of land–biosphere–atmosphere interactions could enhance predictability
of ISO rainfall cycles through consideration of NDVI data. It is hoped that this
realization may contribute toward better land management in southern Africa.
Acknowledgments. CMAP precipitation data were provided by NCEP–CPC from their
Web site at http://www.cdc.noaa.gov. Satellite NDVI data were extracted from the archives
of the USGS’s ADDS. The South African Weather Service provided radiosonde data for
Bloemfontein. This research project was jointly supported by the Agriculture Research
Council and the National Research Foundation of South Africa at the University of
Zululand, KwaDlangezwa, South Africa, with the help of Anna Kozakiewicz.
References
Anyamba, A., 1992: Some properties of a 20-30 day oscillation in tropical convection. J. Afr.
Meteor. Soc., 1, 1–19.
——, and J. R. Eastman, 1996: Interannual variability of NDVI over Africa and its relationship to
El Niño/Southern Oscillation. Int. J. Remote Sens., 17, 2533–2548.
——, C. J. Tucker, and R. Mahoney, 2002: From El Niño to La Niña: Vegetation response patterns
over east and southern Africa during the 1997–2000 period. J. Climate, 15, 3096–3103.
Bartman, A. G., W. A. Landman, and C. J. de W. Rautenbach, 2003: Recalibration of general
circulation model output to austral summer rainfall over southern Africa. Int. J. Climatol., 23,
1407–1420.
Brunsell, N. A., 2006: Characterization of land-surface precipitation feedback regimes with remote
sensing. Remote Sens. Environ., 100, 200–211.
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 24
Buckle, C., 1996: Weather and Climate in Africa. Longman, 321 pp.
Burroughs, W. J., 2003: Weather Cycles, Real or Imaginary? University Press, 317 pp.
Charney, J. G., 1975: Dynamics of deserts and drought in the Sahel. Quart. J. Roy. Meteor. Soc.,
101, 193–202.
Crimp, S. J., J. R. E. Lutjeharms, and S. J. Mason, 1998: Sensitivity of a tropical–temperate
trough to sea-surface temperature anomalies in the Agulhas retroflection region. Water SA,
24, 93–100.
D’Abreton, P. C., and J. A. Lindesay, 1993: Water vapour transport over southern Africa during wet
and dry early and late summer months. Int. J. Climatol., 13, 237–256.
Davenport, M. L., and S. E. Nicholson, 1993: On the relation between rainfall and the normalized
difference vegetation index for diverse vegetation types in East Africa. Int. J. Remote Sens.,
14, 2369–2389.
Dirmeyer, P. A., 2001: An evaluation of the strength of land–atmosphere coupling. J. Hydrometeor.,
2, 329–344.
Drew, G. H., 2004: Modelling vegetation dynamics and their feedbacks over southern Africa in
response to climate change forcing. Ph.D. thesis, University of Cape Town, 253 pp.
Entin, A., A. Robock, K. Y. Vinnikov, S. Qiu, V. Zabeli, S. Liu, A. Namkhai, and T. Adyasuren,
1999: Evaluation of Global Soil Wetness Project soil moisture simulations. J. Meteor. Soc.
Japan, 77, 183–198.
Farrar, T. J., S. E. Nicholson, and A. R. Lare, 1994: The influence of soil type on the relationships
between NDVI, rainfall, and soil moisture in semiarid Botswana. II. NDVI response to soil
moisture. Remote Sens. Environ., 50, 121–133.
Gondwe, M. P., and M. R. Jury, 1997: Sensitivity of vegetation (NDVI) to climate over southern
Africa: Relationships with summer rainfall and OLR. S. Afr. Geogr. J., 79, 52–60.
Gutmann, G. G., 1990: Towards monitoring drought from space. J. Climate, 3, 282–295.
Harrison, M. S. J., 1984: A generalized classification of South African summer rain-bearing synoptic systems. Int. J. Climatol., 4, 547–560.
Hastenrath, S., L. Greischar, and J. van Heeden, 1995: Prediction of summer rainfall over South
Africa. J. Climate, 8, 1511–1518.
Janowiak, J. E., 1988: An investigation of interannual rainfall variability in Africa. J. Climate, 1,
240–255.
Jury, M. R., 1999: Intra-seasonal convective variability over southern Africa: Principal component
analysis of pentad outgoing-longwave radiation departures 1976–1994. Theor. Appl. Climatol., 62, 133–146.
——, and K. M. Levey, 1997: Vertical structure of the atmosphere during wet spells over southern
Africa. Water SA, 23, 51–55.
——, and S. E. Nkosi, 2000: Easterly flow in the tropical Indian Ocean and climate variability over
south-east Africa. Water SA, 26, 147–152.
——, H. R. Valentine, and J. R. E. Lutjeharms, 1993: Influence of the Agulhas current on summer
rainfall along the southeast coast of South Africa. J. Appl. Meteor., 32, 1282–1287.
——, K. M. Levey, and A. Makarau, 1996: Mechanisms of short term rainfall variability over
southern Africa. WRC Rep. 436, Oceanography Department, University of Cape Town,
50 pp.
——, M. Rouault, S. Weeks, and M. Schormann, 1997a: Atmospheric boundary-layer fluxes and
structure across a land–sea transition zone in south-eastern Africa. Bound.-Layer Meteor., 83,
311–330.
——, S. Weeks, and M. P. Gondwe, 1997b: Satellite-observed vegetation as an indicator of climate
variability over southern Africa. S. Afr. J. Sci., 93, 34–38.
——, W. B. White, and C. J. C. Reason, 2004: Modelling the dominant climate signals around
southern Africa. Climate Dyn., 23, 717–726.
Justice, C. O., B. N. Holben, and M. D. Gwyne, 1986: Monitoring east African vegetation using
AVHRR data. Int. J. Remote Sens., 7, 1453–1474.
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 25
——, T. Eck, D. Tanr, and B. N. Holben, 1991: The effect of water vapor on the normalized
difference vegetation index derived for the Sahelian region from NOAA AVHRR data. Int.
J. Remote Sens., 12, 1165–1187.
Lau, K.-M., and H. Weng, 1995: Climate signal detection using wavelet transform: How to make a
time series sing. Bull. Amer. Meteor. Soc., 76, 2391–2402.
Levey, K. M., and M. R. Jury, 1996: Composite intraseasonal oscillations of convection over
southern Africa. J. Climate, 9, 1910–1920.
Lyons, S. W., 1991: Origins of convective variability over equatorial southern Africa during the
austral summer. J. Climate, 4, 23–39.
Makarau, A., 1995: Intra-seasonal oscillatory modes of the southern Africa summer circulation.
Ph.D. thesis, University of Cape Town, 324 pp.
Malo, A., and S. E. Nicholson, 1990: A study of rainfall and vegetation dynamics in the African
Sahel using Normalized Difference Vegetation Index. J. Arid Environ., 19, 1–24.
Mason, S. J., 1995: Sea-surface temperature–South African rainfall associations, 1910–1989. Int.
J. Climatol., 15, 119–135.
Mestas-Nuñez, A. M., 2000: Orthogonality properties of rotated empirical modes. Int. J. Climatol.,
20, 1509–1516.
Mulenga, H. M., 1998: Southern African climatic anomalies, summer rainfall and the Angola low.
Ph.D. thesis, University of Cape Town, 234 pp.
Myneni, R. B., F. G. Hall, P. J. Sellers, and A. K. Marshak, 1995: The interpretation of spectral
vegetation indexes. IEEE Trans. Geosci. Remote Sens., 33, 481–486.
Nicholson, S. E., and T. J. Farrar, 1994: The influence of soil type on the relationships between
NDVI, rainfall, and soil moisture in semiarid Botswana. I. NDVI response to soil moisture.
Remote Sens. Environ., 50, 107–120.
Pan, Z., R. W. Arritt, W. J. Gutowski Jr., and E. S. Takle, 1999: Soil moisture in a regional climate
model: Simulation and projection. Geophys. Res. Lett., 28, 2947–2950.
Pohl, B., Y. Richard, and N. Fauchereau, 2007: Influence of the Madden–Julian oscillation on
southern Africa summer rainfall. J. Climate, 20, 4227–4242.
Richard, Y., and I. Poccard, 1998: A statistical study of NDVI sensitivity to seasonal and interannual rainfall variations in southern Africa. Int. J. Remote Sens., 19, 2907–2920.
Ropelewski, C. F., and M. S. Halpert, 1987: Global- and regional-scale precipitation patterns
associated with El Niño/Southern Oscillation. Mon. Wea. Rev., 115, 1606–1626.
Rutherford, M. C., and R. H. Westfall, 1986: The biomes of southern Africa—An objective categorization. Mem. Bot. Soc. S. Afr., 54, 1–98.
Tazalika, L., and M. R. Jury, 2008: Intra-seasonal rainfall oscillations over central Africa: Space–
time character and evolution. Theor. Appl. Climatol., 94, 67–80.
Tennant, W. J., and B. C. Hewitson, 2002: Intra-seasonal rainfall characteristics and their importance to the seasonal prediction problem. Int. J. Climatol., 22, 1033–1048.
Torrence, C., and G. P. Compo, 1998: A practical guide to wavelet analysis. Bull. Amer. Meteor.
Soc., 79, 61–78.
Tucker, C. J., 1979: Red and photographic infrared linear combinations for monitoring vegetation.
Remote Sens. Environ., 8, 127–150.
——, and P. J. Sellers, 1986: Satellite remote sensing of primary production. Int. J. Remote Sens., 7,
1395–1416.
——, W. W. Newcomb, S. O. Los, and S. D. Prince, 1991: Mean and inter-year variation of
growing-season normalized difference vegetation index for the Sahel 1981–1989. Int.
J. Remote Sens., 12, 1113–1115.
——, M. E. Brown, J. E. Pinzon, D. A. Slayback, R. Mahoeny, N. E. Saleous, and E. N. Vermote,
2005: An extended AVHRR 8-km NDVI dataset comparable with MODIS and SPOT vegetation NDVI data. Int. J. Remote Sens., 15, 3547–3566.
Tyson, P. D., 1981: Atmospheric circulation variations and the occurrence of extended wet and dry
spells over southern Africa. Int. J. Climatol., 1, 115–130.
Earth Interactions
d
Volume 14 (2010)
d
Paper No. 3
d
Page 26
Unganai, L. S., and S. J. Mason, 2001: Spatial characterization of Zimbabwe summer rainfall
during the period 1920-1996. S. Afr. J. Sci., 97, 425–431.
Vogel, C., 2000: Climate and climatic change: Causes and consequences. The Geography of South
Africa in a Changing World, R. Fox and K. Rowntree, Eds., Oxford University Press,
284–303.
Walker, N. D., 1990: Links between South African summer rainfall and temperature variability of
the Agulhas and Benguela currents systems. J. Geophys. Res., 95, 3297–3319.
Wang, G., 2004: A conceptual modeling study on biosphere–atmosphere interactions for physically
based climate modeling. J. Climate, 17, 2572–2583.
Wang, J., K. P. Price, and P. M. Rich, 2001: Spatial patterns of NDVI in response to precipitation
and temperature in the central Great Plains. Int. J. Remote Sens., 22, 3827–3844.
——, P. M. Rich, and K. P. Price, 2003: Temporal responses of NDVI to precipitation and temperature in the central Great Plains, USA. Int. J. Remote Sens., 24, 2345–2364.
Woodward, F. I., 1987: Climate and Plant Distribution. Cambridge University Press, 174 pp.
Zeng, N., and D. J. Neelin, 2000: The role of vegetation–climate interaction and interannual variability in shaping the African savanna. J. Climate, 13, 2665–2670.
——, K. Hales, and J. D. Neelin, 2002: Nonlinear dynamics in a coupled vegetation–atmosphere
system and implications for desert–forest gradient. J. Climate, 15, 3474–3487.
Earth Interactions is published jointly by the American Meteorological Society, the American Geophysical
Union, and the Association of American Geographers. Permission to use figures, tables, and brief excerpts
from this journal in scientific and educational works is hereby granted provided that the source is acknowledged. Any use of material in this journal that is determined to be ‘‘fair use’’ under Section 107 or that
satisfies the conditions specified in Section 108 of the U.S. Copyright Law (17 USC, as revised by P.IL. 94553) does not require the publishers’ permission. For permission for any other from of copying, contact one of
the copublishing societies.