Abstract

A more complete picture of the timing and patterns of the ENSO signal during the seasonal cycle for the whole of Africa over the three last decades is provided using the normalized difference vegetation index (NDVI). Indeed, NDVI has a higher spatial resolution and is more frequently updated than in situ climate databases, and highlights the impact of ENSO on vegetation dynamics as a combined result of ENSO on rainfall, solar radiation, and temperature.

The month-by-month NDVI–Niño-3.4 correlation patterns evolve as follows. From July to September, negative correlations are observed over the Sahel, the Gulf of Guinea coast, and regions from the northern Democratic Republic of Congo to Ethiopia. However, they are not uniform in space and are moderate (~0.3). Conversely, positive correlations are recorded over the winter rainfall region of South Africa. In October–November, negative correlations over Ethiopia, Sudan, and Uganda strengthen while positive correlations emerge in the Horn of Africa and in the southeast coast of South Africa. By December with the settlement of the ITCZ south of the equator, positive correlations over the Horn of Africa spread southward and westward while negative correlations appear over Mozambique, Zimbabwe, and South Africa. This pattern strengthens and a dipole at 18°S is well established in February–March with reduced (enhanced) greenness during ENSO years south (north) of 18°S. At the same time, at ~2°N negative correlations spread northward. Last, from April to June negative correlations south of 18°S spread to the north (to 10°S) and to the east (to the south of Tanzania).

1. Introduction

El Niño–Southern Oscillation (ENSO) is one of the main modes of natural climate variability at the global and interannual (2–7 yr) scale (de Viron et al. 2013). During ENSO events, the atmospheric circulation and precipitation patterns are strongly disturbed for several months worldwide and more particularly in the tropics (e.g., Dai and Wigley 2000). For that reason, the predictability and prediction of ENSO events have long been considered (Latif et al. 1998; Chen et al. 2004) and forecasts, based either on numerical and statistical models or a combination of the two, are now routinely performed. They are used in early warning systems in different parts of the world and help in mitigating ENSO impacts. Because of Africa’s strong reliance on primary production for its agricultural needs and its many socioeconomic problems such as endemic poverty, low human development index, poor governance, and armed conflicts, it is highly vulnerable to climate variability and change (Thornton et al. 2006).

The ENSO impacts on the African climate and environment have been the subject of numerous studies. Impacts on climate have been assessed primarily by an analysis of rainfall and temperature data while impacts on the environment have been assessed primarily by an analysis of remotely sensed data of vegetation photosynthetic activity. However, most of these analyses have been performed at seasonal to annual time scales (Williams and Hanan 2011; Kogan 2000) or have focused on the impact of particular ENSO events [e.g., the 1997/98 event in Anyamba et al. (2001, 2002); Verdin et al. 1999] or on specific regions within Africa (e.g., Martiny et al. 2009; Brown et al. 2010) and rarely on the continent as a whole (e.g., Nicholson and Kim 1997). Therefore, neither the precise timing of the signal of ENSO nor the patterns (e.g., propagations, dipoles) have been fully documented. In addition, while the signal of ENSO in the summer rainfall areas and semiarid tropics (Sahel, southern Africa, Horn of Africa) is regularly investigated (e.g., Segele et al. 2008; Fauchereau et al. 2009; Martiny et al. 2009; Brown et al. 2010; Crétat et al. 2012; Fontaine et al. 2011; Mohino et al. 2011, among many others), the humid to subhumid regions and those experiencing winter rainfall at the northwest and southwest tips of the continent (the Mediterranean fringes of Morocco and Algeria and the Atlantic fringe of South Africa, respectively) have drawn less attention (see Knippertz et al. 2003; Malhi and Wright 2004; Balas et al. 2007; Philippon et al. 2011). However, several of these regions are densely populated and are net exporters of high-quality agricultural products. This calls for a better knowledge of their climate variability, particularly as it relates to ENSO. Furthermore, most of the studies performed on the African climate and ENSO teleconnections consider long time periods (usually 50 years from the 1950s). However, these teleconnections are subject to a strong decadal variability, and for many regions in Africa, they have become much more intense since the end of the 1970s (e.g., Janicot et al. 1996; Richard et al. 2000; Knippertz et al. 2003; Philippon et al. 2011). In parallel, since the end of the 1970s, in situ climate data in Africa have become more and more scarce and less reliable [see Malhi and Wright (2004) for a discussion on the decline in the number of temperature and precipitation stations in humid Africa and its consequence on gridded products]. Unfortunately, rainfall estimates from satellites still do not adequately capture the climatology and variability in rainfall over the continent and its subregions, and they feature important biases in rainfall patterns and quantities (Nicholson et al. 2003; Ali et al. 2005; Dinku et al. 2007; McCollum et al. 2000). High-resolution products are also not available for a sufficiently long period of time to enable a consistent and precise mapping of the ENSO signal on local climate conditions and environment.

Thus, the primary objective of this study is to provide a more complete and updated picture of the timing and patterns of ENSO signal in the whole of Africa over the three last decades. With this aim in mind we analyze the normalized difference vegetation index (NDVI), which provides an indirect estimate of the vegetation photosynthetic activity (Tucker 1979), rather than rainfall and temperature to circumvent the issue of in situ data scarcity and reliability over the last decades, and to document the ENSO signal at fine space scales. Indeed, the frequent and regular update of the NDVI data, their high spatial and temporal resolutions, and their coverage of the whole of Africa are among the advantages of using NDVI over rainfall. Moreover, NDVI is a good indicator of the climatic conditions. NDVI variability in the African summer rainfall semiarid environments is closely related to rainfall and soil moisture availability (Camberlin et al. 2007; Martiny et al. 2006; Nicholson et al. 1990, among others). In both the African winter rainfall semiarid environments and the subhumid to humid environments, NDVI variability and its climatic controls have been far less studied as compared to the summer rainfall semiarid environments or the Amazonian humid environments. It has been shown that the variability in photosynthetic activity in the primary forests of Amazonia is primarily controlled by light variability (Nemani et al. 2003) especially during the dry season (Huete et al. 2006; Xiao et al. 2006), and that the ENSO impact on these type of forests is through the modulation of cloudiness and illumination (Graham et al. 2003; Pau et al. 2010). While over equatorial regions NDVI data are constrained by saturation issues (i.e., above 0.7 NDVI is no longer an indicator of growth during the wettest months) and biases related to water vapor content and cloudiness, these aspects, especially a possible ENSO signal, have not yet been explored for Africa.

We also consider NDVI at the monthly time step in order to follow the spatial patterns and evolution during the seasonal cycle of NDVI anomalies associated with ENSO. This time step is of greater interest than the seasonal time step for the scientific community working on vegetation phenology because it enables one to point out the ENSO signal at different vegetation phenological stages. A secondary underlying objective is also to reconsider the climatic variables through which ENSO impacts NDVI. To that end, the NDVI relationships with rainfall are reassessed. In particular, comparing the rainfall–NDVI relationship intensity to the ENSO–NDVI relationship is informative. For the humid environments, few analyses are performed considering the links between NDVI, rainfall, and cloud cover.

The paper is organized into five sections. Section 2 presents the different databases used, namely Global Inventory Modeling and Mapping Studies (GIMMS) NDVI, Global Precipitation Climatology Centre (GPCC) rainfall, International Satellite Cloud Climatology Project (ISCCP) total cloud cover, and the Niño-3.4 index. Section 3 gives a rapid overview of the methods and approach used. Results are presented in section 4 for the three main regions detected according to the ENSO signal timing: the summer rainfall regions, the winter rainfall regions, and the equinoctial rainfall regions. For each region and at the pixel scale, we assess 1) the link between the timing of the ENSO and the pattern in NDVI response through correlations, 2) asymmetries in the NDVI response between the warm and cold ENSO events, and 3) NDVI sensitivity to rainfall, especially in winter rainfall regions and humid and subhumid environments where that sensitivity has been less explored as compared to the semiarid environments. To complement these results at the pixel scale, we also present for six regional NDVI indices the rainfall and total cloud cover seasonal cycle anomalies that develop during the warm and cold ENSO events. Finally in section 5, we synthesize and discuss the findings in the context of the African continent as a whole.

2. Data

a. NDVI

For our study purposes we have worked with the longest available NDVI data, collected by the Advanced Very High Resolution Radiometer (AVHRR) sensor onboard the National Oceanic and Atmospheric Administration (NOAA) satellites. These data were obtained from the Famine Early Warning Systems Network (FEWS NET) Africa data portal (http://earlywarning.usgs.gov/fews/africa/index.php). This portal provides 10-day composite NDVI images of Africa at 8-km spatial resolution from July 1981 to the present, processed by the Global Inventory Monitoring and Modeling Studies group (Tucker et al. 2005) at the National Aeronautical and Space Administration (NASA). NDVI is calculated from the near-infrared (NIR) and red (VIS) top-of-atmosphere reflectances, using the following algorithm: NDVI = (NIR − VIS)/(NIR + VIS). Values for vegetated land generally range from about 0.1 to 0.8, with values lower than 0.15 indicating sparse vegetation and values greater than 0.6 indicating dense vegetation. GIMMS NDVI data have been corrected for 1) stratospheric aerosols due to volcanic eruptions during April 1982–December 1984 and June 1991–December 1993, 2) artifacts due to satellite drift, which is especially important in tropical regions, and 3) subpixel cloud contamination (Pinzón et al. 2005). The same desert calibration has been applied for all the sensors (NOAA-7 to NOAA-17). No correction has been applied for atmospheric effects due to tropospheric aerosols, water vapor, Rayleigh scattering, or stratospheric ozone.

For our purposes we selected the period July 1981–June 2008 and upscaled the data to 16-km resolution and a monthly time step by simple spatial and temporal averages. Following Martiny et al. (2006), pixels with NDVI mean values <0.12, corresponding to bare soils and desert areas, were excluded from the study. These areas are the Sahara, the Namibian coast, northern Kenya, northern Somalia, and northeastern Ethiopia, which are all sparsely inhabited, hyperarid regions.

Figures 1a and 1b present the months of mean green-up and dormancy onsets of vegetation. Following Philippon et al. (2007), the green-up (dormancy) onset is defined as the month when NDVI extends up (down) to the annual mean level. This simple approach proposed for the Sahelian semiarid environment is nonetheless adequate for the whole of Africa since even the equatorial evergreen vegetation, despite high NDVI values, has a marked phenological rhythm (Gond et al. 1997). Moreover, at the monthly time step our method leads to results that are consistent with those obtained in other studies using more complex methods (Brown and de Beurs 2008; Vrieling et al. 2011; Zhang et al. 2003). Where bimodal regimes exist, only the green-up and dormancy onsets that respectively follow and precede the lowest mean monthly NDVI value have been retained. This artificially increases the vegetative season duration but there are years and regions for which the short dry season is sometimes suppressed (e.g., as in 1997/98 in East Africa; Anyamba et al. 2002).

Fig. 1.

(a) Month of the vegetative season onset (average over 27 yr) defined as the month when NDVI extends up the annual mean level. (b) Month of the vegetative season end, defined as the month when NDVI extends down the annual mean level.

Fig. 1.

(a) Month of the vegetative season onset (average over 27 yr) defined as the month when NDVI extends up the annual mean level. (b) Month of the vegetative season end, defined as the month when NDVI extends down the annual mean level.

Green-up (Fig. 1a) starts in March north of the equator and shifts northward to the Sahara margins in August [at an average rate of ~0.05 km per day according to Zhang et al. (2005)]. South of the equator, green-up is much more uniform with large areas showing an onset in November or December without any clear north–south propagation. The Congo basin, Kenya, southern Uganda, and northern Tanzania equatorial regions are subject to bimodal regimes, and experience a main onset of green-up by October–November. Lastly, the two winter rainfall regions located at the northwest and southwest tips of Africa (i.e., the Mediterranean margins of Morocco and Algeria and the Atlantic margin of South Africa, respectively) experience a green-up start in December and June, respectively. These results are consistent with those obtained by Zhang et al. (2004, 2005) using Moderate Resolution Imaging Spectroradiometer (MODIS) NDVI. The dormancy onset (Fig. 1b) pattern is much more homogeneous than the green-up onset pattern. Dormancy onset occurs in October–November in tropical Africa north of the equator and in the winter rainfall region of South Africa. Dormancy onset occurs in May–June south of the equator, in East Africa as well as in the winter rainfall region of Northern Africa. This makes vegetative seasons lasting 1) between 8 and 4 months from the Guinean Gulf coast to the Sahel, 2) between 7 and 5 months from 10°S to South Africa, 3) around 7–8 months in East Africa, and 4) 9 months and upward in the equatorial region. In the rest of the study area, those pixels/months that did not fall within the vegetative season (i.e., between the green-up and dormancy months) were excluded from the analysis. This ensured that we interpreted signals in NDVI that are truly related to vegetation greenness and not to soil reflectance values, for example, as may happen in semiarid environments.

b. Rainfall

Rainfall data used in this study originate from the Global Precipitation Climatology Center (http://gpcc.dwd.de/), which provides gauge-based gridded monthly precipitation datasets for the global land surface, in different spatial resolutions. We selected the GPCC full data reanalysis product (Rudolf and Schneider 2005), which compiles the most comprehensive global collection of in situ monthly precipitation data from 1901 to 2009. We chose a 0.5° × 0. 5° spatial resolution and extracted grid points covering Africa for the period January 1981–December 2008. Note that over that period and area, the number of stations has gradually decreased from ~3642 in January 1981 to ~1150 in December 2005 and ~483 in December 2007 (Fig. 2b). If the latter decrease is not attributed solely to the delay in the delivery and processing of data at GPCC, then the African rain gauge network has suffered strong deteriorations since the 1990s. There are fewer and fewer operating stations, while data of those that are operating are not fed systematically into the international system. In addition, the quality of the data is often poor with gaps of missing data and inefficient quality controls (UNECA 2011). This leads to a poor spatial distribution of data with large regions (i.e., Angola, Democratic Republic of Congo, Nigeria; Fig. 2a) undocumented over the last decade.

Fig. 2.

(a) Map of GPCC grid points documented by rain gauges only in January 1981 (crosses) and both in January 1981 and December 2004 (circle). (b) Evolution from January 1981 to December 2007 of the number of rain gauges used in the GPCC database over Africa.

Fig. 2.

(a) Map of GPCC grid points documented by rain gauges only in January 1981 (crosses) and both in January 1981 and December 2004 (circle). (b) Evolution from January 1981 to December 2007 of the number of rain gauges used in the GPCC database over Africa.

c. Cloud cover

The cloud cover data downloaded from the Koninklijk Nederlands Meteorologisch Instituut (KNMI) climate explorer (http://climexp.knmi.nl) originates from the widely used ISCCP D2 dataset. We have worked with the total cloud amount (%), which is provided globally on an equal-area map grid with 280-km spatial resolution and a monthly time step over the period from July 1983 to June 2006. The “D” dataset improves over the previous one (“C”) in terms of radiance calibration, cloud detection, and radiative modeling. Moreover the spatial and temporal resolutions have been increased and the temporal coverage extended, and additional products are also available (Rossow et Schiffer 1999).

d. Niño-3.4 index

Finally, the Niño-3.4 (N3.4) sea surface temperature (SST) index was downloaded from the Climate Prediction Center database (http://www.cpc.ncep.noaa.gov/data/indices/). It documents the sea surface temperature over the area 5°N–5°S, 170°–120°W and has been computed from the Extended Reconstructed SST (ERSST) V3b database (Smith et al. 2008). We selected the period January 1981–December 2008 and a monthly time step. We upscaled data to a monthly overlapping, 3-month time step. Our analyses focus on 4 of the 6 phases of the ENSO seasonal cycle (Larkin and Harrison 2002), which is presented in Fig. 3. These are the ONSET from ~May0 to July0, the PEAK from ~July0 to December0, the DECAY from ~January+1 to April+1, and lastly the POST phase from May+1. On average, the lowest T values (below 27°C; Fig. 3, large gray dots) are recorded during the PEAK and DECAY phases, which are also characterized by a large year to year variability (Fig. 3, spread between the small open gray circles). Note as well that any significant trend is observed in the time series for each of these four phases. Other ENSO “representative” indices such as the multivariate ENSO index (MEI) also exist and recently distinct types of ENSO variability (e.g., eastern Pacific and central Pacific), having somewhat different timings and impacts, have also been highlighted (Kao and Yu 2009; Newman et al. 2011; Ren and Jin 2011). However, the objective here is to provide a general picture of the relationship with ENSO for Africa. There is no justification for considering specific types of ENSO, which may result in a slightly different NDVI signal in a given region only.

Fig. 3.

Seasonal cycle (full large gray dots and dashed line) and scatterplot over 1981–2007 (small opened gray circles) of 3-month NIÑO-3.4 data; black thin line: trend lines (none is significant at the 95% level). The onset, peak, and decay phases start in MJJ, ASO, and JFM respectively.

Fig. 3.

Seasonal cycle (full large gray dots and dashed line) and scatterplot over 1981–2007 (small opened gray circles) of 3-month NIÑO-3.4 data; black thin line: trend lines (none is significant at the 95% level). The onset, peak, and decay phases start in MJJ, ASO, and JFM respectively.

3. Methods and approach

To assess precisely and understand the signal of ENSO on NDVI in the whole of Africa during the last three decades, we have conducted two sets of analyses. In the first set, we have computed correlations between monthly overlapping, 3-month SST values over the Niño-3.4 region and monthly NDVI values with 1-month lag. For example, the April–June N3.4 value is correlated with July NDVI, the May–July N3.4 value is correlated with August NDVI and so on. The 1-month lag enables the time response of vegetation to climatic perturbations to be taken into account. The 1-month time step enables the evolution of the correlation patterns during the vegetation phenological cycle (green-up, peak, senescence …) to be followed precisely. This is an improvement over preceding studies, which have usually worked with a seasonal time step (e.g., Williams and Hanan 2011) that can mask rapid month-to-month changes in the correlation patterns and dilute signals by aggregating months that do not show concordant relationships with ENSO. Results are presented as monthly maps in Fig. 4 where correlations that are not significant at the 90% level are masked out (dark gray shades) and in four tables documenting southern, western, eastern, and central Africa (see Fig. 7, left top panel for their location). For each month, the total number of pixels analyzed and the percentage of pixels significantly correlated with Niño-3.4 are shown. Also, the limits of these four large regions have been chosen according to several criteria: they are coherent in terms of 1) mean climate [e.g., the western Africa (WAF) region is influenced by the West African monsoon], 2) rainfall interannual variability [see Poccard et al. (2000), who extracted four dominant modes of rainfall variability across Africa, with our WAF, southern Africa (SAF), and eastern Africa (EAF) regions matching well with three of these modes], 3) seasonality of the ENSO–rainfall teleconnection (see Camberlin et al. 2001; our four regions roughly correspond to their regions 1, 2, 3, and 5) and 4) NDVI–rainfall relationships (results in section 4 and Fig. 6). Note that these regions are roughly equal in terms of number of pixels analyzed. These tables are a valuable summary of the different regional ENSO signals in vegetation and rainfall.

Fig. 4.

July to June correlation maps between 3-month Niño-3.4 and monthly NDVI with 1-month lag (i.e., April–June Niño-3.4 correlated to July NDVI). Pixels/months that do not fall within the vegetative season are in light gray. Pixels for which the correlation is not significant at the 90% level are in dark gray. The significance level at 90% (95%) equals 0.32 (0.38).

Fig. 4.

July to June correlation maps between 3-month Niño-3.4 and monthly NDVI with 1-month lag (i.e., April–June Niño-3.4 correlated to July NDVI). Pixels/months that do not fall within the vegetative season are in light gray. Pixels for which the correlation is not significant at the 90% level are in dark gray. The significance level at 90% (95%) equals 0.32 (0.38).

To document possible asymmetries between the cold and warm events signals, and to assess the level of NDVI anomalies (expressed in % of departure from the mean), we have also conducted composite analyses. Three samples of years were created according to N3.4 anomalies (note that N3.4 values follow a normal distribution). The “warm” sample contains years/trimesters when the N3.4 anomaly is above or equal to 0.5 std. The “cold” sample contains years/trimesters when the N3.4 anomaly is below or equal to −0.5 std. The “normal” sample contains years when the N3.4 anomaly is between −0.5 and 0.5 std. Differences in NDVI monthly field between the warm and normal sample as well as the cold and normal sample are presented as monthly maps in Figs. 5a and 5b, respectively. As for the correlations maps, differences that are not significant at the 90% level according to a Student’s t test are masked out. It is instructive to compare the results of the correlation and composite analyses. Indeed, for those pixels that are significantly correlated to ENSO, the composite maps show whether the correlation signal comes primarily from the ENSO cold or warm phases or from both. Those pixels that are not significantly correlated to ENSO but that show large deviations from normal in the composite analysis indicate that the relationship with ENSO is not linear and is strongly asymmetrical (i.e., only the cold or warm ENSO events relate to disturbances in the local climate and subsequently to vegetation photosynthetic activity). It will be seen in the results section that the Guinean region of western Africa well exemplifies this.

Fig. 5a.

July to June composite maps of NDVI anomalies (in % of the 1981–2007 mean) during warm ENSO events. Pixels/months that do not fall within the vegetative season are in light gray. Pixels for which the composite does not pass the significance test at 90% are in dark gray.

Fig. 5a.

July to June composite maps of NDVI anomalies (in % of the 1981–2007 mean) during warm ENSO events. Pixels/months that do not fall within the vegetative season are in light gray. Pixels for which the composite does not pass the significance test at 90% are in dark gray.

Fig. 5b.

As in (a), but for cold ENSO events.

Fig. 5b.

As in (a), but for cold ENSO events.

In a second set of analyses, to understand the origin of the ENSO signal detected in NDVI, we have reassessed the sensitivity of NDVI to rainfall. It is presented in Fig. 6 as correlation maps between monthly overlapping 3-month rainfall values and monthly NDVI values with 1-month lag (i.e., the April–June rainfall amount is correlated with July NDVI, the May–July rainfall amount is correlated with August NDVI, and so on). Given the difference of spatial resolution between NDVI (~16 km) and rainfall (~50 km) data, several pixels of NDVI fall within the same pixel covered by the rainfall grid. This can enhance the signal particularly when vegetation is homogeneous across the rainfall pixel. Note that the 1-month lag is the lag for which the correlations between vegetation and rainfall mean seasonal cycles are the highest for most of the pixels regardless of vegetation type (not shown). This is coherent with findings by Martiny et al. (2006) and Klein and Röhrig (2006), who show for several semiarid and semihumid regions in western, eastern, and southern Africa that the vegetation response usually lags behind rainfall by 1 to 1.5 months. Similarly, the 3-month cumulative rainfall amount has already been found to be best correlated with NDVI in semiarid environments in particular (Nicholson et al. 1990; Klein and Röhrig 2006). Indeed, vegetation usually does not respond directly to rainfall but to soil moisture, which is related to rainfall accumulated over several months (Malo and Nicholson 1990). Such NDVI–rainfall correlation maps for the whole of Africa and over the whole seasonal cycle have never been produced before and provide interesting new insights as compared to maps produced in previous studies at the annual time scale (e.g., Camberlin et al. 2007) or regional spatial scale (e.g., Brown et al. 2010), particularly with regard to the evolution of vegetation sensitivity to rainfall during the phenological cycle. These maps, when related to the NDVI–ENSO correlation maps in Fig. 4 that are built in the same way, indicate the potential role of rainfall in transmitting the ENSO signal to vegetation. In addition, Tables 14 give for each month the percentage of pixels where the NDVI–rainfall correlation is significant at the 90% level (column 3) and the percentage of pixels where the NDVI–rainfall correlation is significant amongst those pixels where the NDVI–ENSO correlation is significant (column 4). To complement and synthesize these pixel-scale analyses, composite analyses have also been performed for six regional indexes chosen for their representativeness of the vegetative and climatic context and of the NDVI–ENSO relationship. These indexes document 1) a semiarid area experiencing a bimodal rainfall cycle in Tanzania, 2) a semiarid summer rainfall area in Botswana, 3) a semiarid winter rainfall area in the western Cape region of South Africa, 4) a subhumid region in the Guinean domain of western Africa, and two humid regions in 5) Gabon and 6) the Democratic Republic of Congo (Fig. 7, left top panel). Generally speaking the three last humid regions usually attract little attention. Figure 7 presents the six indexes’ NDVI and rainfall mean seasonal cycles, based on all years (thick full line) and for composite ENSO years (thin dashed and dotted dashed lines, for the cold and warm event samples respectively). In addition, for the three humid regions where rainfall is not as strong a limiting factor as for the semiarid regions, the total cloud cover mean and composite seasonal cycles are given to explore the potential impact of ENSO through light availability. Note that we have worked with the residuals of the total cloud cover with the effect of the synchronous rainfall removed. These residuals have been obtained from the linear regression of the rainfall mean seasonal cycle on the cloud cover mean seasonal cycle. The reason for working with residuals is that cloud cover variations are partly expressed in rainfall. Thus the residuals might sign the potential effect of nonprecipitating clouds on vegetation as they also act as a barrier to the light. These regional-scale analyses of NDVI, rainfall, and cloud cover seasonal cycles are instructive for they show clearly the shifts and amplitude changes in the seasonal cycles, the asymmetries in the response to ENSO warm and cold events, and the synchronicity (in time and/or intensity) between the responses of the three variables.

Fig. 6.

July to June correlation maps between 3-month GPCC rainfall amounts and monthly NDVI with 1-month lag. Pixels/months that do not fall within the vegetative season are in light gray. Pixels for which the correlation is not significant at the 90% level are in dark gray. The significance level at 90% (95%) equals 0.32 (0.38). The 3-month isohyets are superimposed as black contours.

Fig. 6.

July to June correlation maps between 3-month GPCC rainfall amounts and monthly NDVI with 1-month lag. Pixels/months that do not fall within the vegetative season are in light gray. Pixels for which the correlation is not significant at the 90% level are in dark gray. The significance level at 90% (95%) equals 0.32 (0.38). The 3-month isohyets are superimposed as black contours.

Table 1.

Southern Africa (and southwestern South Africa from July to October). 1) Number of pixels of NDVI studied (bold indicates the largest and the smallest number of pixels); 2) percentage of pixels significantly correlated to Niño-3.4 at the 90% level (the number in parentheses and italics is the median correlation); 3) percentage of pixels significantly correlated to rainfall at the 90% level; and 4) percentage of pixels significantly correlated to rainfall among those significantly correlated to Niño-3.4. Note that the significance level at 90% (95%) equals 0.32 (0.38).

Southern Africa (and southwestern South Africa from July to October). 1) Number of pixels of NDVI studied (bold indicates the largest and the smallest number of pixels); 2) percentage of pixels significantly correlated to Niño-3.4 at the 90% level (the number in parentheses and italics is the median correlation); 3) percentage of pixels significantly correlated to rainfall at the 90% level; and 4) percentage of pixels significantly correlated to rainfall among those significantly correlated to Niño-3.4. Note that the significance level at 90% (95%) equals 0.32 (0.38).
Southern Africa (and southwestern South Africa from July to October). 1) Number of pixels of NDVI studied (bold indicates the largest and the smallest number of pixels); 2) percentage of pixels significantly correlated to Niño-3.4 at the 90% level (the number in parentheses and italics is the median correlation); 3) percentage of pixels significantly correlated to rainfall at the 90% level; and 4) percentage of pixels significantly correlated to rainfall among those significantly correlated to Niño-3.4. Note that the significance level at 90% (95%) equals 0.32 (0.38).
Table 2.

As in Table 1, but for western Africa (and northwestern Africa from January to March).

As in Table 1, but for western Africa (and northwestern Africa from January to March).
As in Table 1, but for western Africa (and northwestern Africa from January to March).
Table 3.

As in Table 1, but for East Africa (and north of East Africa in June–October).

As in Table 1, but for East Africa (and north of East Africa in June–October).
As in Table 1, but for East Africa (and north of East Africa in June–October).
Table 4.

As in Table 1, but for central Africa.

As in Table 1, but for central Africa.
As in Table 1, but for central Africa.
Fig. 7.

Location of the six regional indices and four large regions studied. Also shown are mean (full line) and composite (dashed line for cold and dashed dotted line for warm ENSO events) seasonal cycles of NDVI (unitless), rainfall (mm), and the residues of total cloud cover (%) for the six regional indices. Stars (circles) denote month for which the cold (warm) composite is significantly different from the mean (at the 90% level).

Fig. 7.

Location of the six regional indices and four large regions studied. Also shown are mean (full line) and composite (dashed line for cold and dashed dotted line for warm ENSO events) seasonal cycles of NDVI (unitless), rainfall (mm), and the residues of total cloud cover (%) for the six regional indices. Stars (circles) denote month for which the cold (warm) composite is significantly different from the mean (at the 90% level).

4. ENSO signal in vegetation photosynthetic activity

Figure 4 suggests that Africa can be divided into three main regions according to the sign and timing of the correlations between NDVI and N3.4. These are the summer rainfall regions, the winter rainfall regions and the equinoctial rainfall regions. Results for each of these three regions are presented in sections 4a4c, respectively.

a. Summer rainfall regions

The three summer rainfall regions of Africa are the southern African region (35°–10°S), the western African region (5°–16°N, 20°W–20°E), and the northeastern African region (4°–16°N, 20°–40°E).

1) Southern Africa

In southern Africa, where the vegetative growing season spans from approximately December to May (Figs. 1a,b), the significant correlations with N3.4 (Fig. 4) are mainly negative (shades of yellow), indicating that vegetation photosynthetic activity tends to be dampened during the warm phases of ENSO and to increase during the cold ones. The negative correlations appear as early as December (Fig. 4) at the borders of South Africa, Zimbabwe, and Mozambique. By March, these correlations have intensified and spread over most of the region south of 18°S (the Namibia/Angola border area) while north of 18°S positive correlations are recorded (actually they appear as early as January) so that ~41% of the pixels are significantly correlated to ENSO (Table 1, column 2). According to the vegetation map of the Global Land Cover 2000 project [which has provided for the year 2000 a harmonized land cover database over the whole globe making use of a dataset of 14 months of preprocessed daily global data acquired by the VEGETATION instrument on board the Satellite Pour l’Observation de la Terre (SPOT) 4 satellite (Mayaux et al. 2004); not shown] the transition from shrub/grass dominant vegetation (in the south) to tree dominant vegetation (in the north) is located near 18°S. From April to June the negative correlations gradually shift to the north and the east reaching Angola, Zambia, and Malawi, and then the north of Mozambique and the south of Tanzania. Lastly, it is noteworthy that in northern Namibia and southeast South Africa, NDVI correlations switch from being positive at the beginning of the vegetative season (November–December) to being negative at the core (February–April), a signal potentially interesting for forecasting purposes. This is coherent with findings by Richard et al. (2002), who have shown that preseason (October) rainfall anomalies in South Africa tend to be out of phase with rainfall anomalies during the rainy season (December–April), which is a pattern and mode of variability significantly related to ENSO.

The instructiveness of the composite maps (Figs. 5a,b) about the linearity of the relationship to ENSO and the symmetry of the cold and warm event impacts in southern Africa is obvious. The timing, patterns, and intensity of the warm and cold event impacts are clearly different. During cold events (Fig. 5b), vegetation photosynthesis is significantly enhanced at the beginning of the vegetative season (January) in Botswana and at the end of the vegetative season (April–May) in Namibia and north of 20°S. During warm events (Fig. 5a), vegetation photosynthesis increases in November in the southeast part of South Africa, then decreases substantially in South Africa, Botswana, and Namibia for ~4 months in a row from February to April.

Looking at Table 1 (columns 3 and 4) and Fig. 6, it is evident that the impact of ENSO on NDVI in southern Africa is mainly through the effect of rainfall. Vegetation sensitivity to rainfall is very high over the region: from November to May (Table 1, column 3), 60% to 75% of the pixels have significant 1-month lag correlations between monthly NDVI and 3-month rainfall amounts. Moreover, most of the correlation values are above 0.6 or 0.8 (Fig. 6). The concordance between some isohyets and the patterns of significant correlations (e.g., the isohyets 60 and 210 mm in January, and 60 and 360 mm in February and May) is also remarkable. This is consistent with previous findings performed at the annual time step (Malo and Nicholson 1990; Martiny et al. 2006; Camberlin et al. 2007) showing that the relationship between NDVI and rainfall in Africa is strong and linear for those areas with an annual rainfall between 200 and 1000 mm and with open grasslands and rainfed crops as dominant vegetation cover. Note as well in Fig. 6 the gradual increase in the sensitivity of vegetation to rainfall (i.e., higher correlations—0.55 on average in January as against 0.61 in March—which are also more coherent in space) during the vegetative season, which is an effect of the lagged response of vegetation to rainfall and of the intraseasonal persistence of vegetation anomalies (Philippon et al. 2007). Table 1 (column 4) further indicates that between 75% and 90% of the pixels significantly correlated to ENSO are also significantly correlated to rainfall accumulated during the previous trimester. Lastly, looking at the example of Botswana in Fig. 7, the amplitude of the NDVI seasonal cycle is strongly decreased during warm ENSO events in accordance with the rainfall seasonal cycle. The increase in the amplitude of the NDVI seasonal cycle is less pronounced during the cold events although rainfall amounts seem to increase noticeably in January–February. Two hypotheses can be evoked here to explain this apparent weak response of NDVI to rainfall excesses recorded during cold ENSO events. First, the rainfall anomalies can be due to few very intense rain events that are not particularly effective for vegetative growth since much of the water is lost through runoff (this could also explain why the rainfall anomalies do not pass the significance test). Second, vegetation is on average very close to its optimum and cannot increase its photosynthetic productivity or be any denser in cover and biomass than it already is.

These results agree with and complement the results from several previous studies that have explored the impact of ENSO on vegetation and rainfall in southern Africa. The north–south dipole by 18°S in the NDVI–ENSO correlation pattern is evident in the studies by Anyamba and Eastman (1996) and Anyamba et al. (2001), who have mapped the evolution of NDVI anomalies during the ENSO years 1986–89 and 1997–98. It is also coherent with composite maps of Williams and Hanan (2011) for the season December–February (DJF), where the limit around ~18°S is clearer during cold than warm ENSO events. However, these authors have worked with net photosynthesis modeled by the Simple Biosphere Model version 3, and not with observed NDVI. Similarly, Brown et al. (2010) obtained coherent patterns of negative correlations between the multivariate ENSO index and NDVI cumulated over March–May at the border between South Africa, Zimbabwe, and Mozambique, and in northern Namibia/southern Angola. Over South Africa, these authors obtain either a nonsignificant or positive correlation between March–May NDVI and MEI. It is obvious from our results that ENSO has its strongest negative impact in South Africa in January–March and not April–May when the vegetative season has come to an end in part of the country (see the May map; Fig. 4) and when the ENSO has entered into its post phase. This points to the importance of considering a monthly time step to accurately follow the spatial evolution of the ENSO signal during the phenological cycle. With regard to the ENSO impact on rainfall in southern Africa and the mechanisms involved, when analyzing both observations and atmospheric general circulation model (AGCM) outputs, Richard et al. (2000) and Mason (2001) note that convection and rainfall over southern Africa (the southwest Indian Ocean) are decreased (increased) during warm ENSO years and the subtropical high pressure belt is weakened, leading to reduced moisture fluxes toward the continent. Working at the intraseasonal time scale, Fauchereau et al. (2009) and Pohl et al. (2009) show that the ENSO–rainfall teleconnection at the seasonal scale arises from a modulation of the frequency of the tropical–temperate troughs (TTT), which are the dominant rain-bearing systems.

2) Western Africa

Western Africa (5°–17°N, 20°W–20°E) records its main growing season from ~May to November (Figs. 1a,b). In comparison with southern Africa, correlations between ENSO and vegetation photosynthetic activity are weaker—according to Table 2, column 2, barely 10% of the pixels are significantly correlated with ENSO—and less spatially coherent. The most consistent signals are observed in July–August over Senegal and Mali and along the Gulf of Guinea coast with negative correlations (Fig. 4). This is consistent with Propastin et al. (2010) and Williams and Hanan (2011), who found West African vegetation to be less influenced by ENSO warm events than the southern Africa vegetation. With regard to the Sahel, the composite maps show that only the warm events (Fig. 5a) have a significant signal, in July–August mainly over the western (Senegal) and eastern (Chad) parts of the Sahelian band. Brown et al. (2010) note that ENSO events more strongly impact the start of the vegetative season (delayed during warm events) than its core (see their Fig. 7). This weak sensitivity of vegetation to ENSO over the Sahel could be explained by both a weak sensitivity of vegetation to rainfall and a weak sensitivity of climate (rainfall especially) to ENSO. First, on average NDVI–rainfall correlations barely exceed 0.55 (compared to 0.61 for southern Africa). Nicholson et al. (1990) and Martiny et al. (2006) computed the rain use efficiency (RUE; expressed as the NDVI to rainfall ratio) for small regions in western, southern, and eastern Africa and noted that the Sahel had the smallest RUE. They attribute it to the shortness and intensity of its rainy season. Second, during the Sahelian vegetative season ENSO is either in its onset or post phases (and not in its peak or decay phases as is the case for South Africa; Fig. 3). Moreover, the teleconnection between ENSO and the West African monsoon is affected by the strong decadal variations of the climate background state. Whereas the teleconnection was high in the 1970s and 1980s (Janicot et al. 2001), it has decreased over the two last decades in favor of the Mediterranean basin (Rodríguez-Fonseca et al. 2011; Fontaine et al. 2011). This could explain the weak impact we found in our study.

The composite maps for the Guinean region (Figs. 5a,b) provide evidence for a stronger signal of ENSO than that suggested by the correlation analyses alone. Indeed, during cold events (Fig. 5b) large positive anomalies of NDVI are recorded from June to September over West Africa south of 10°N up to northern Gabon. As a consequence, in terms of the NDVI seasonal cycle (Fig. 7 and the example of Guinea), the cold events are associated with a unimodal cycle (the July–August vegetation photosynthesis decrease is suppressed). With regard to the parameters involved in the NDVI–N3.4 relationship over the Guinean region, it is obvious that even if weak positive correlations between NDVI and rainfall accompanies the northward shift of the ITCZ and associated West African monsoon (April–June; Fig. 6), vegetation sensitivity to rainfall accumulated the previous trimester is generally low over the region. Moreover, few pixels are correlated to both ENSO and rainfall, suggesting that the ENSO signal in vegetation might not be through rainfall. Figure 7 brings interesting insights. First, the positive and significant anomalies of NDVI from July to September associated with cold events are synchronous with positive and significant anomalies of rainfall. As opposed to the semiarid environments, the delay in the vegetation response to rainfall during these months and over that region appears small. Enhanced rainfall might allow vegetation to reach its potential photosynthetic level, which is of a magnitude close to that observed in June or October. Second, the large positive anomaly in NDVI in August is also associated with a negative anomaly for cloud cover. That additional supply of light could contribute to the increase in vegetation photosynthesis. It is not incompatible for that region and for the months investigated to have higher rainfall amounts associated with a reduced cloud cover. Indeed, the July–August little dry season usually experiences a uniform cover of nonprecipitating stratus clouds (Knippertz et al. 2011) that brings less light and less rainfall than the broken cover of vertically developed cumulus clouds of the intertropical convergence zone (ITCZ).

3) Northern part of eastern Africa

The northern part of eastern Africa encompasses South Sudan, western and central Ethiopia, Uganda, and western Kenya. This region under the double influence of the West African monsoon and the East African equinoctial rainy seasons has a vegetative season spanning from approximately May to November (Figs. 1a,b). The most consistent correlations with N3.4 are observed at the end of the season (i.e., August–October; Fig. 4) and are negative. That pattern of negative correlations fits very well with the area of intensive cultivation (GLC2000 map, not shown). It seems to be triggered by the both cold and warm events but whereas the largest positive anomalies of NDVI are observed in October during cold events (Fig. 5b), the largest negative anomalies of NDVI are recorded in August–September during the warm events (Fig. 5a). This suggests ENSO-induced changes in the phase of the seasonal cycle. A last interesting signal seen only in the cold event composite maps is the positive anomalies of NDVI over western Ethiopia from June to August.

This signal of ENSO in NDVI is mainly through the effect of rainfall. As seen in Table 3, column 4, between 45% and 47% of the pixels significantly correlated with N3.4 in September–October are also significantly correlated with rainfall accumulated over the previous trimester. In addition, the ENSO–NDVI correlation patterns for September–October fit well with the rainfall–NDVI correlation patterns shown in Fig. 6, falling within the 60–360-mm isohyets. In their study, Williams and Hanan (2011) did not find any ENSO impact on net photosynthesis in the northern part of eastern Africa probably because of the seasonal time scale considered [June–August (JJA) and September–November (SON)], which merges months with different, or opposite, ENSO signals. But the negative impact of ENSO on summer rainfall in Uganda and northwest Kenya has been documented by numerous studies such as those of Ogallo (1988), Camberlin (1995), Phillips and McIntyre (2000), and Indeje et al. (2000) while the negative impact of ENSO on summer rainfall in Ethiopia has been documented by Gissila et al. (2004).

b. Winter rainfall regions

The two semiarid, winter rainfall regions of Africa are located in western South Africa (primarily along the Atlantic coast) and northwestern Africa (along the Atlantic and Mediterranean coasts).

1) Western South Africa

In western South Africa, the vegetative growing season spans the period from July to October (Figs. 1a,b), the Southern Hemisphere winter. Little is known about the impact of ENSO on vegetation over this region as most of the studies devoted to southern African NDVI variability have focused on the summer rainfall region. First, as opposed to the summer rainfall region, correlations for this winter rainfall region (Fig. 4) are positive, meaning that enhanced (diminished) vegetation photosynthetic activity is expected during warm (cold) ENSO events. However, looking at the composite maps (Figs. 5a,b), NDVI anomalies pass the significance test during cold events only. Second, correlations with N3.4 are weaker and are significant mainly in August (Fig. 4), with 24% only of the pixels analyzed significantly correlated to N3.4 (Table 1, column2). This could be attributed first to the type of vegetation itself. Shrubs dominate the Fynbos and Succulent Karoo biomes, which are endemic to the region. Shrubs are known to exhibit a lower amplitude in NDVI values between seasons (see the western Cape index NDVI seasonal cycle in Fig. 7, with an amplitude of 0.05 only). They are also known to be less sensitive to rainfall variations. In the study by Camberlin et al. (2007) there is no significant correlation between annual NDVI and annual rainfall. At the monthly time step, correlation values between June–October NDVI and rainfall accumulated the previous trimester (Fig. 6) are on average around 0.56 [against 0.7 for the summer rainfall region; cf. section 4a(1)], and are significant for 50%–65% of the pixels (Table 1, column3). The Succulent Karoo and the Fynbos biomes respond to rainfall in different ways. Vegetation of the Succulent Karoo biome is the most sensitive (Fox et al. 2005). Its growth starts with the first significant rains at the end of the summer (April–May), continues throughout the winter, and then drops sharply in spring (October) as rainfall declines (Esler and Rundel 1999). Conversely, vegetation within the Fynbos biome exhibits a bimodal pattern with NDVI peaks in August and November (Hoare and Frost 2009) that do not match well with the seasonal cycle of rainfall, which varies from a winter regime in the west to a nonseasonal and equinoctial regime in the eastern part of the biome (Rouault and Richard 2003). Moreover, the relatively long summer drought period appears not to be a limiting factor to photosynthetic activity (Stock and Allsopp 1992). These points explain the somewhat lower NDVI sensitivity to rainfall over this region as compared to the summer rainfall area of southern Africa.

As for the semiarid, summer rainfall environments, the signal of ENSO is through rainfall as suggested by Table 1 (column 4). The 42%–65% of pixels that show a significant correlation with ENSO in July–September also show a significant correlation to rainfall. As shown in the rainfall seasonal cycle of the western Cape in Fig. 7, 1) most of the precipitation in the region falls from May to August, and 2) June–July monthly amounts are strongly lowered during cold ENSO events. This is consistent with the findings by Philippon et al. (2011), who have recently highlighted a positive correlation between ENSO and the May–July (MJJ) rainfall amount over the region (r ~0.5 over the period 1979–99). During ENSO years the rain-bearing systems (extratropical troughs mainly) in MJJ are more frequent and deeper and are shifted toward the north, thus carrying more rainfall over the region.

2) Northwestern Africa

In northwestern Africa the vegetative season spans the period from January to May (Figs. 1a,b), the winter of the Northern Hemisphere. As for the western South African region, correlations with N3.4 are positive but weak and significant in January and over the Mediterranean-facing region of Morocco and Algeria only (Fig. 4 and Table 2, column 2, with barely 10% of the pixels showing significant correlations with N3.4). According to the composite analyses, the ENSO signal in NDVI in Algeria is related to the warm events mainly but very few pixels pass the significance test and the NDVI positive anomalies recorded from January to March are very weak (~15% above the mean; Fig. 5a). As for the other semiarid environments studied, rainfall variations are the main driver of NDVI variability over the region. From January to May (Table 2, column 3) ~40%–50% of the pixels analyzed show significant correlations between monthly NDVI and rainfall accumulated over the previous trimester (October–April), with correlation values that rise above 0.7 (Fig. 6).

As for western South Africa, little attention has been paid to vegetation and rainfall variability in this region. Mariotti et al. (2002) have analyzed the composite pattern in seasonal rainfall according to ENSO warm and cold events over the region. They have observed that the relationship shifts from positive to negative between autumn and spring. Thus, the positive correlation we observe between ENSO and NDVI in winter could result from a combination of the vegetation sensitivity to autumn rainfall added to persistence in NDVI anomalies.

c. Equinoctial rainfall regions

The equatorial regions of Africa, namely East Africa (10°N–10°S, 32°–45°E) and central Africa (~12°S–5°N, 8°–32°E), have equinoctial rainfall regimes, that is, two rainy seasons coinciding with the northward and then the southward passage of the ITCZ, which produces rain in October–December (the short rains) and March–May (the long rains) in East Africa (Tanzania index in Fig. 7), and in September–November and March–May in central Africa (Gabon and DRC indexes in Fig. 7; Nicholson and Grist 2003; Balas et al. 2007; Samba and Nganga 2012).

1) East Africa

The signal of ENSO in the East African vegetation photosynthetic activity is asymmetrical between the two rainy seasons. During the short rains, positive correlations between NDVI and N3.4 (Fig. 4) suggest that warm (cold) ENSO events are associated with positive (negative) NDVI anomalies. Significant correlations emerge in October over eastern Ethiopia/southern Somalia first [note as well the west–east dipole by 40°E due to the persistence of NDVI anomalies that have emerged in August to the west; cf. section 4a(3)]. They next spread southward to northern and eastern Kenya by November, and then to northern Tanzania by January (Fig. 4). By that time, more than 40% of the pixels studied display significant correlations with N3.4 (Table 3, column 2). Significant correlations first start to fade away over Tanzania in February, then in Kenya and southern Somalia in March. When looking at the composite maps, the asymmetry of the cold and warm ENSO events signal is obvious. While warm events are associated with significant large positive NDVI anomalies from October to February (Fig. 5a), only a few pixels in December show significant negative anomalies during cold events (Fig. 5b). This asymmetry clearly appears on the Tanzania index as well (Fig. 7). The NDVI seasonal cycle amplitude is particularly enhanced during warm events with a pronounced peak in January. These results complete and agree with previous studies on NDVI performed at the seasonal scale for the East African region. For example, the NDVI anomaly pattern associated with the 1997/98 warm ENSO event shown by Linthicum et al. (1999) and Anyamba et al. (2002) featured strong positive anomalies from December to March in Kenya and the southern parts of Somalia, Ethiopia, and Sudan.

This marked response of vegetation to ENSO in general and to warm events in particular, during and shortly after the short rains, arises from both a high sensitivity of vegetation to rainfall and a high sensitivity of rainfall to ENSO. Table 3 shows in column 3 that from November to January, monthly NDVI is significantly correlated to the rainfall accumulated over the previous trimester for more than 60% of the pixels, while column 4 shows that from November to February more than 70% of the pixels of NDVI that are significantly correlated to ENSO are also significantly correlated to rainfall. The NDVI–rainfall correlations often rise above 0.7 (Fig. 6). An interesting point highlighted as well in column 3 of Table 3 is that the vegetation sensitivity to rainfall does not decline in the dry period stretching between the short rains and the long rains. On the contrary, in Kenya, the highest correlations (>0.8) are observed between November–January (NDJ) rains and February NDVI (Fig. 6). This can be related to the vegetation time response to rainfall, which is about one month (Davenport and Nicholson 1993; Martiny et al. 2006), but also to the shortness of the January–February dry season, which can also be relatively wet, as was the case in 1998, and thus sustain high vegetation photosynthetic activity between the two rainy seasons (Linthicum et al. 1999; Anyamba et al. 2001).

The teleconnection between equatorial East African (Kenya, Uganda, Tanzania) short rains and ENSO has been documented by many authors, including Indeje et al. (2000) and Mutai and Ward (2000). The teleconnection operates through a modification of the Walker cell circulation over the Indian Ocean. In warm ENSO years, the low-level (high level) winds blowing to the east (west) along the equator are weakened or turn westward (eastward), bringing moisture to the region and driving convective rains. Mapande and Reason (2005) focused on western Tanzania, a region at the transition between the opposite-signed ENSO-impacted eastern and southern African regions, and reported that above-normal October–April rainfall was associated with warm ENSO events.

The correlation patterns of NDVI-N3.4 during the long rains [March–May (MAM); Fig. 4] are radically different from those that occur during the short rains. Few pixels have significant correlations with N3.4 (between 10% and 18% according to Table 3, column2). From March to April correlations switch very quickly from largely positive to somewhat negative (except in southern Tanzania). By May, the negative correlations have vanished. In June, sparse (and low) positive correlations appear in northwestern Kenya, Somalia, and eastern Ethiopia. This weak impact of ENSO on vegetation during the long rains as compared to the short rains first arises from a weaker impact of ENSO on climate and rainfall in particular. Camberlin and Philippon (2002) found a weak negative correlation between Niño-3 and the March–April rainfall amount, switching to positive for May rainfall. This is consistent with our NDVI–N3.4 correlation maps, but on the whole ENSO only explains a very small part of the long rains variability, as pictured as well in the Tanzania rainfall seasonal cycle in Fig. 7. Second, the vegetation sensitivity to rainfall variability is also lower than during the short rains: between 30% and 40% of the pixels have a significant 1-month lag correlation between monthly NDVI and 3-month rainfall amounts (against 50% to 66% during the short rains and the January–February dry season; Table 3, column3). Correlation values are also lower as seen in Fig. 6 (0.5–0.6 vs 0.7–0.8 in November–February). This is consistent with Martiny et al. (2006), who noticed a higher RUE during the short rains than during the long rains.

2) Central Africa

Vegetation response to climate and ENSO in central Africa has been less studied and is less well known as compared to the semiarid regions. This lack of studies and interest for the region comes primarily from data issues: 1) NDVI saturates during the greenest months and is also perturbed by cloudiness, 2) the large rainfall amounts do not seem to be a limiting factor to vegetation growth, and 3) long-term climate data in the region are scarce. Nonetheless, two interesting signals in the NDVI–N3.4 correlation patterns are observed: one at the beginning of the driest of the two rainy seasons (i.e., from February to March) and one during one of the two dry seasons (i.e., from July to September); see the mean seasonal cycles in Fig. 7. In February–March positive correlations stretch from 15°S to approximately the equator (Fig. 4), a signal that comes mainly from warm events with significant positive anomalies of NDVI to the west of Angola, the Democratic Republic of Congo (DRC), Congo, and Gabon (Fig. 5a). This is in line with the results obtained by Anyamba et al. (2001) for the 1997/98 ENSO event and Los et al. (2001), where positive anomalies of NDVI are obvious over the whole of the Congo basin in March (see their Fig. 2) during warm ENSO events. From March to April–May, a band of weak negative correlations shifts from the equator to 7°N, a signal associated with cold events mainly (Fig. 5b). It suggests that ENSO might modulate the northward propagation of the ITCZ during the boreal spring with propagation sped up during cold events. In July–September, which are relatively dry months over the region, sparse negative correlations are observed at the equatorial Atlantic coast (Gabon mainly) and to the east of the DRC (Fig. 4). These correlations seem respectively in line with the negative correlations observed in Guinea [see section 4a(2)] and to the north of East Africa [see section 4a(3)], and seem mainly triggered by warm events (negative anomalies of NDVI in Fig. 5a).

To further document these two signals, the NDVI, rainfall, and residual cloud cover mean seasonal cycles during warm and cold events for the Gabon and DRC are presented in Fig. 7. First, as opposed to the semiarid environments (i.e., Tanzania, Botswana) where a 1- to 2-month lag is observed, there is barely a lag between NDVI and rainfall mean seasonal cycles: the two NDVI maxima (minima) are concurrent with the two rainfall maxima (minima), that is, MAM and SON (JJA and DJF). Thus considering a 1-month lag and 3-month accumulated rainfall to explore the NDVI–rainfall relationships for that region might not be as suitable as for the semiarid regions (very few pixels display significant NDVI–rainfall correlations in Fig. 6 and Table 4, column 3). Second, it is noteworthy that the highest NDVI peak is recorded during the driest of the two rainy seasons (MAM), which appears also as the one with the lowest residual cloud cover for Gabon. This agrees with results by Gond et al. (2013), who have mapped vegetation in central Africa by classifying seasonal profiles of the MODIS enhanced vegetation index (EVI). For the forest classes, the highest EVI peak coincides with the driest of the two rainy seasons. In agreement with the correlation and composite maps, large deviations from the mean are observed in NDVI during warm and cold events in July–August for both Gabon and the DRC (Figs. 5a,b). Although they do not pass the significance test, scatterplots for August (not shown) reveal that during cold events NDVI values are systematically equal to (1984, 1999) or above (1983, 1988, 1987, 1998) the mean. Interestingly, significant deviations from the mean are also recorded in rainfall in August with higher amounts during cold events. This additional water supply could help vegetation to sustain a high photosynthetic activity during the dry season (roughly 50 mm month−1). Note that unlike for Guinea there is not any clear and significant signal in the cloud cover residuals for Gabon. For the DRC the deviation is positive during cold events, which does not support the hypothesis of more active vegetation due to an additional supply of light. However, as illustrated in Fig. 2a, very few rain gauge data have been fed into the GPCC database over the DRC for the last decade. Thus the quality of rainfall and cloud cover residuals composite cycles is questionable.

Climate variability in central Africa and how it relates to ENSO has been the subject of very few studies. Whereas ENSO appears as the primary driver of temperature variations in all the tropical rain forest regions (South America, Southeast Asia, and central Africa), and of precipitation in South America and Southeast Asia, Malhi and Wright (2004) could not find any significant relationship between ENSO and precipitation in central Africa. However, it should be noted that these authors computed interannual correlations for all months together, assuming the absence of any seasonality in the ENSO–rainfall relationships. The central Africa rainfall–ENSO relationships have also been documented in studies by Nicholson and Kim (1997), Poccard et al. (2000), and Camberlin et al. (2001), and more specifically in the studies by Balas et al. (2007) and Misra (2010). Camberlin et al. (2001) and Balas et al. (2007) show that negative rainfall anomalies in October–December (OND) and/or DJF over Gabon and Congo are associated with warm events in the equatorial Pacific while Misra (2010) shows that positive anomalies of rainfall in DJF are recorded over the eastern parts of the DRC during warm events in association with anomalous upper-level divergent winds over the western Indian Ocean. None of these studies considers the July–September dry season, which nonetheless seems critical for photosynthetic activity of vegetation.

5. Discussion and summary

The aim of this study was to provide a precise and updated picture of the timing and patterns of the ENSO signal during the seasonal cycle in the whole of Africa and over the last three decades. We used the normalized difference vegetation index rather than climate data for that purpose. This was because NDVI has a higher spatial resolution and is more frequently updated than in situ climate databases. Moreover, by using the longest available NDVI database (i.e., the one built from the NOAA AVHRR missions) it is possible to take into account the interannual variability. Therefore, using NDVI brings out signals that are potentially stronger, or occur at a higher spatial resolution, than when using short-term remotely sensed or long-term in situ climate products, respectively. However, one of the drawbacks of using NDVI is that signals are often difficult to explain for the humid environments (e.g., central Africa) where NDVI saturates and is biased due to cloudiness, and vegetation responses to climate variations are complex and not well understood. They are also difficult to interpret for cultivated areas under irrigation where water supply may induce variations in photosynthetic activity that are likely to obscure the natural response to climate.

We have mapped first the month-by-month evolution of the 1-month lag correlations between the Niño-3.4 index (3-month values) and NDVI (monthly values) over the whole of Africa (Fig. 4). Although patterns are not necessarily the same from one event to the next (Myneni et al. 1996; Kogan 2000; Lyon and Mason 2007), the use of a monthly time step adds important new insights about the timing and patterns of ENSO signal to the findings of previous studies, which are based largely on annual or seasonal time scales and on a regional spatial scale. Starting from July (the beginning of the peak phase of ENSO) and going to June (the post phase of ENSO), the teleconnection patterns over the whole of Africa evolve as follows. From July to September, negative correlations between NDVI and N3.4 are observed north of the equator (including the Sahel, the Gulf of Guinea coast, and regions from the northern Democratic Republic of Congo to Ethiopia) but they are not uniform in space and are moderate (~0.3). Conversely, positive correlations are recorded over the winter rainfall region of South Africa. In the period from October to November, negative correlations over Ethiopia/Sudan/Uganda disappear while positive correlations emerge in the Horn of Africa (so that a west–east dipole is observed in October), and in the southeast coast of South Africa. By December, the end of the peak phase of ENSO, with the settlement of the ITCZ south of the equator, positive correlations over the Horn of Africa spread southward and westward while negative correlations appear over Mozambique, Zimbabwe, and South Africa. This pattern strengthens and a dipole at 18°S is well established in February–March (the decay phase of ENSO), with reduced photosynthetic activity during ENSO years south of 18°S and enhanced activity north of 18°S. In the meantime, by ~2°N negative correlations spread northward. Lastly, at the end of the ENSO decay phase and the beginning of its post phase (April–June) a northward (to 10°S) and eastward (to the south of Tanzania) spread of the negative correlations south of 18°S occurs. Thus, the NDVI–N3.4 teleconnections feature regional-scale dipoles and propagative patterns both in space (from one region to another one) and time (from one month to another). These patterns have not been identified with such accuracy in previous studies. They suggest that ENSO influences the location of the ITCZ (its southward and northward but also eastward and westward displacements) and the convection within it during most of the seasonal cycle.

Although correlation analyses at the monthly time step point out interesting patterns, composite analyses performed at the pixel scale (Figs. 5a,b) or the regional scale (Fig. 7) highlight the strong asymmetry of the impact of the cold and warm ENSO events for numerous regions. For instance, in Gabon, southern and eastern Africa vegetation photosynthetic activity is modulated during warm ENSO events mainly, with an activity dampened in Gabon in July–August and southern Africa from December to May, and increased in eastern Africa from October to February. But the clearest example is for the Guinean domain in western Africa, which records strong positive anomalies of NDVI in July–August during cold ENSO events, a signal barely perceptible in correlation analyses. Moreover, the value of the complementary regional-scale approach (i.e., regional indices; Fig. 7) is particularly obvious. Signals that were not that noticeable at the pixel scale (both in the correlation and composite analyses) are much more evident at the regional scale, which is known to enhance the signal-to-noise ratio.

Figure 8 summarizes our findings. When the four large African regions studied are classified according to the strength of the ENSO signal in NDVI, it appears that the regions with the largest ENSO signal both in terms of spatial extent (i.e., the number of pixels having a significant correlation with N3.4) and correlation levels are southern Africa and eastern Africa. In southern Africa ENSO impacts NDVI of the southern Africa winter rainfall region in JAS and in the southeast region in November–December (Fig. 8; class 4), and of the summer rainfall region south of 18°S from December to May (Fig. 8; class 2). However, for the latter region it must be remembered that its teleconnection with ENSO is subject to a strong decadal variability, so the ENSO signal could evolve in the next decades. In eastern Africa, ENSO impacts NDVI of the northwestern region under the influence of the West African monsoon from August to October (Fig. 8; class 2) and of the October–December short rains and January–February little dry season (Fig. 8; class 4). It is interesting to note that all these regions are semiarid but not all of them have their rainy seasons phased with the peak of ENSO. For all these regions and months the NDVI–N3.4 correlations get stronger during the progress of the vegetative season (Tables 1 and 3), in agreement with the similar increase of vegetation sensitivity to rainfall (Fig. 8, bottom panels and thick lines). Clearly, the impact of ENSO on vegetation there is through rainfall. Indeed, pixels for which the NDVI–N3.4 correlation is significant display higher correlations with rainfall than pixels for which the NDVI–N3.4 correlation is not necessarily significant (Fig. 8; compare the thick and thin lines; e.g., class 4, January, and eastern Africa with correlation values of 0.7 vs 0.6). Lastly, vegetation sensitivity to ENSO and to rainfall in the winter rainfall region of South Africa is comparable to that in the summer rainfall region (Table 1, August vs March, and Fig. 8, August class 4 vs March class 2). Generally speaking, more attention should be given to the winter rainfall regions of Africa since they hold biodiversity hotspots, are densely inhabited, and sustain an intense agricultural export industry, and because of their location at the transition between the midlatitudes and the tropics, which is an area that is expected to be strongly affected by climate change (Mc Kellar et al. 2007; Giorgi and Lionello 2008).

Fig. 8.

(top) Month-by-month and region-by-region number of pixels showing a significant NDVI–Niño-3.4 correlation (correlation is negative for class 2 and positive for class 4) and a significant positive NDVI–rainfall correlation (note that the significant negative NDVI–rainfall correlations are very few and thus have not been considered). SAF indicates southern Africa, EAF East Africa, WAF West Africa, and CAF central Africa. (bottom) Solid lines indicate median NDVI–rainfall correlations month by month for the four regions for (left) class 2 and (right) class 4. As a benchmark, the thin lines display the median NDVI–rainfall correlations for pixels showing a significant and positive NDVI–rainfall correlation but for which the NDVI–Niño-3.4 correlation is not necessarily significant. Red indicates SAF, green EAF, blue WAF, and black CAF. The significance level considered for all the correlations is set at 90%.

Fig. 8.

(top) Month-by-month and region-by-region number of pixels showing a significant NDVI–Niño-3.4 correlation (correlation is negative for class 2 and positive for class 4) and a significant positive NDVI–rainfall correlation (note that the significant negative NDVI–rainfall correlations are very few and thus have not been considered). SAF indicates southern Africa, EAF East Africa, WAF West Africa, and CAF central Africa. (bottom) Solid lines indicate median NDVI–rainfall correlations month by month for the four regions for (left) class 2 and (right) class 4. As a benchmark, the thin lines display the median NDVI–rainfall correlations for pixels showing a significant and positive NDVI–rainfall correlation but for which the NDVI–Niño-3.4 correlation is not necessarily significant. Red indicates SAF, green EAF, blue WAF, and black CAF. The significance level considered for all the correlations is set at 90%.

Conversely, the less impacted regions are western and central Africa. With regard to the Sahel and northwestern Africa semiarid regions, the relatively weak impact of ENSO on NDVI illustrates first the weak impact of ENSO on climate and for the Sahel over rainfall in particular over the last two decades, a fact pointed out by Fontaine et al. (2011). As for southern Africa, the Sahel–ENSO teleconnection is subject to strong decadal variations. Although vegetation in the Sahel and northwestern Africa is somewhat less sensitive to rainfall than vegetation in eastern and southern Africa, the main driver of its variability is nonetheless rainfall. The western Africa region where vegetation appears to be the most affected by ENSO is the Guinean wet region (Fig. 8, class 2, July–August). According to the composite analyses, the July–August vegetation photosynthesis is significantly enhanced during cold events. Interestingly, this response is associated with significant and synchronous increases of precipitation and decreases of cloud cover, two features that have never been shown before. Although the quality of the NOAA AVHRR NDVI data is questionable for that region given the extensive and shallow cover of nonprecipitating clouds that affect it in July and August, these results call for a further study of the rainfall–cloud cover–NDVI relationships to understand the respective contribution of rainfall and light to vegetation photosynthesis in that humid environment. Indeed, it is not clear what role exactly rainfall plays on vegetation in this region given that we have always considered in our correlation analyses rainfall accumulated over 3 months and a 1-month lag with NDVI, which is not strictly valid for every region of Africa. Moreover, we have considered the total cloud cover only. An analysis of cloud types would corroborate our hypothesis of a uniform cover of low clouds replaced by a more broken cover of vertically developed clouds during cold ENSO events enabling a larger illumination of vegetation. We have found very few ENSO-related signals in NDVI for central Africa, and the lack of studies related to vegetation and climate variability for this region is a handicap to the interpretation and evaluation of the robustness of our results. The main signals concern the months of February–March and July–August (NDVI low). In February–March the photosynthetic activity is enhanced during warm ENSO events. In July–August the photosynthetic activity is decreased during warm ENSO events and increased during cold ones. However, as compared to the Guinean region of western Africa, the associated signals in rainfall and cloud cover are not very coherent and significant. That lack of response to ENSO might be related to different factors: 1) the low sensitivity of climate and rainfall in particular to ENSO, 2) the low sensitivity of vegetation to rainfall, and 3) the quality of rainfall and vegetation data. From the climate side, it must be stressed that unfortunately the spatial coherence of rainfall in central Africa is lower than in most other parts of Africa and its climate variability seems to be largely independent of ENSO, at least its rainfall variability. From the vegetation side, its sensitivity to rainfall is not well known and has not yet been properly evaluated, nor its sensitivity to light or temperature. The time response of vegetation to rainfall might be longer or shorter than the one opted for in that study (i.e., 3 months accumulated rainfall and a 1-month lag). Moreover, the variability of the length and intensity of the dry season, which has not been explored in this study, might play a much more dominant role in the photosynthetic activity of vegetation than that of rainfall accumulated during the rainy season. It must be also noticed that great disparities exist among the different tree species in terms of their sensitivity and time response to climate variability, and that can weaken the signal. For instance, Couralet et al. (2010) noticed from a survey of semideciduous, overstorey trees in the Mayombe forest of the DRC that while the growth of some species is mainly dependent on the early rainy season rains, the growth of other species is impacted by rains at the end of the rainy season. With regard to data quality, in situ rainfall data available in the region are very scarce and irregularly distributed as illustrated by Fig. 2a. Hence, rainfall information for most of the grid points is determined by the interpolation of data from a neighboring grid point rather than from the average values derived from rain gauges. Therefore, the accuracy of the rainfall and residual cloud cover composite cycles for the Gabon and DRC indexes in particular (Fig. 7) is questionable and could explain the weak match with the NDVI composite cycles. One way to circumvent that problem would be to use rainfall estimates derived from satellites. Unfortunately, these estimates, especially those based on infrared data, are still unreliable for the region. For example, the multisatellite rain product from the Global Precipitation Climatology Project (GPCP) overestimates by a factor of 2 the estimates derived from rain gauges from the Global Precipitation Climatology Center (GPCC) (Mc Collum et al. 2000). The NOAA AVHRR NDVI data themselves over the region are compounded by the persistence of clouds, which, despite the 10-day “filtering,” contaminate the data and make the vegetation signal uncertain. It would also be worth checking some of these results using other NDVI databases such as the Moderate Resolution Imaging Spectroradiometer (MODIS) as well as another vegetation index such as the enhanced vegetation index (EVI). For instance, Huete et al. (2002) have evaluated the performance of the MODIS NDVI and EVI products over a wide range of biomes (from the semiarid grassland of Arizona to the tropical broadleaf forest of Brazil) and compared it to the products derived from NOAA AVHRR. They note that when compared to NOAA AVHRR NDVI, MODIS NDVI has a greater seasonal dynamic range over the wet environments and during the wet growing season. They attribute the differences to the atmosphere water vapor content that strongly affects the AVHRR near-infrared band and decreases NDVI values. The authors note as well that when compared to MODIS NDVI, MODIS EVI does not become as easily saturated when viewing rain forests. However, MODIS data have been acquired since April 2000 only, which limits somewhat analyses of the interannual variability. Still, there are interesting (and probably real) vegetation signals over central Africa that deserve to be further studied using databases different from the ones we have considered in this study and additional parameters. For instance, the effect of temperatures has not been explored, and the ENSO signal could pass through this parameter for some months and regions, as exemplified by the NDVI–mean temperature correlation pattern in January–February, which features a dipole by 18°S very close to the one featured in the NDVI–ENSO correlation pattern.

REFERENCES

REFERENCES
Ali
,
A.
,
A.
Amani
,
A.
Diedhiou
, and
T.
Lebel
,
2005
:
Rainfall estimation in the Sahel. Part II: Evaluation of rain gauge networks in the CILSS countries and objective intercomparison of rainfall products
.
J. Appl. Meteor.
,
44
,
1707
1722
.
Anyamba
,
A.
, and
J. R.
Eastman
,
1996
:
Interannual variability of NDVI over Africa and its relation to El Niño/Southern Oscillation
.
Int. J. Remote Sens.
,
17
,
2533
2548
.
Anyamba
,
A.
,
C. J.
Tucker
, and
J. R.
Eastman
,
2001
:
NDVI anomaly patterns over Africa during the 1997/98 ENSO warm event
.
Int. J. Remote Sens.
,
22
,
1847
1859
.
Anyamba
,
A.
,
C. J.
Tucker
, and
R.
Mahonney
,
2002
:
From El Niño to La Niña: Vegetation response patterns over eastern and southern Africa during the 1997–2000 period
.
J. Climate
,
15
,
3096
3103
.
Balas
,
N.
,
S. E.
Nicholson
, and
D.
Klotter
,
2007
:
The relationship of rainfall variability in west central Africa to sea-surface temperature fluctuations
.
Int. J. Climatol.
,
27
,
1335
1349
.
Brown
,
M. E.
, and
K.
de Beurs
,
2008
:
Evaluation of multi-sensor semi-arid crop season parameters based on NDVI and rainfall
.
Remote Sens. Environ.
,
112
,
2261
2271
.
Brown
,
M. E.
,
K.
de Beurs
, and
A.
Vrieling
,
2010
:
The response of African land surface phenology to large scale climate oscillations
.
Remote Sens. Environ.
,
114
,
2286
2296
.
Camberlin
,
P.
,
1995
:
June–September rainfall in north-eastern Africa and atmospheric signals over the tropics: A zonal perspective
.
Int. J. Climatol.
,
15
,
773
783
.
Camberlin
,
P.
, and
N.
Philippon
,
2002
:
The East African March–May rainy season: Associated atmospheric dynamics and predictability over the 1968–97 period
.
J. Climate
,
15
,
1002
1019
.
Camberlin
,
P.
,
S.
Janicot
, and
I.
Poccard
,
2001
:
Seasonality and atmospheric dynamics of the teleconnection between African rainfall and tropical sea-surface temperature: Atlantic vs ENSO
.
Int. J. Climatol.
,
21
,
973
1005
.
Camberlin
,
P.
,
N.
Martiny
,
N.
Philippon
, and
Y.
Richard
,
2007
:
Determinants of the interannual relationships between remote sensed photosynthetic activity and rainfall in tropical Africa
.
Remote Sens. Environ.
,
106
,
199
216
.
Chen
,
D.
,
M. A.
Cane
,
A.
Kaplan
,
S. E.
Zebiak
, and
D.
Huang
,
2004
:
Predictability of El Niño over the past 148 years
.
Nature
,
428
,
733
736
.
Couralet
,
C.
,
F. J.
Sterck
,
U.
Sass-Klaassen
,
J.
Van Acker
, and
H.
Beeckman
,
2010
:
Species-specific growth responses to climate variations in understory trees of a central African rain forest
.
Biotropica
,
42
,
503
511
.
Crétat
,
J.
,
Y.
Richard
,
B.
Pohl
,
M.
Rouault
,
C.
Reason
, and
N.
Fauchereau
,
2012
:
Recurrent daily rainfall patterns over South Africa and associated dynamics during the core of the austral summer
.
Int. J. Climatol.
, 32, 261–273,
doi:10.1002/joc.2266
.
Dai
,
A.
, and
T. M. L.
Wigley
,
2000
:
Global patterns of ENSO-induced precipitation
.
Geophys. Res. Lett.
,
27
,
1283
1286
.
Davenport
,
M. L.
, and
S.
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
.
de Viron
,
O.
,
J. O.
Dickey
, and
M.
Ghil
,
2013
:
Global modes of climate variability
.
Geophys. Res. Lett.
,
40
,
1832
1837
,
doi:10.1002/grl.50386
.
Dinku
,
T.
,
P.
Ceccato
,
E.
Grover-Kopec
,
M.
Lemma
,
S. J.
Connor
, and
C. F.
Ropelewski
,
2007
:
Validation of satellite rainfall products over East Africa’s complex topography
.
Int. J. Remote Sens.
,
28
,
1503
1526
.
Esler
,
K. J.
, and
P. W.
Rundel
,
1999
:
Comparative patterns of phenology and growth form diversity in two winter rainfall deserts: The Succulent Karoo and Mojave Desert ecosystems
.
Plant Ecol.
,
142
,
97
104
.
Fauchereau
,
N.
,
B.
Pohl
,
C. J. C.
Reason
,
M.
Rouault
, and
Y.
Richard
,
2009
:
Recurrent daily OLR patterns in the southern Africa/southwest Indian Ocean region, implications for South African rainfall and teleconnections
.
Climate Dyn.
,
32
,
575
591
.
Fontaine
,
B.
,
M.
Gaetani
,
A.
Ullmann
, and
P.
Roucou
,
2011
:
Time evolution of observed July–September sea surface temperature–Sahel climate teleconnection with removed quasi-global effect (1900–2008)
.
J. Geophys. Res.
,
116
,
D04105
, doi:10.1029/2010JD014843.
Fox
,
S. C.
,
M. T.
Hoffman
, and
D.
Hoare
,
2005
:
The phenological pattern of vegetation in Namaqualand, South Africa and its climatic correlates using NOAA-AVHRR NDVI data
.
S. Afr. Geogr. J.
,
87
,
85
94
.
Giorgi
,
F.
, and
P.
Lionello
,
2008
:
Climate change projections for the Mediterranean region
.
Global Planet. Change
,
63
,
90
104
.
Gissila
,
T.
,
E.
Black
,
D. I. F.
Grimes
, and
J. M.
Slingo
,
2004
:
Seasonal forecasting of the Ethiopian summer rains
.
Int. J. Climatol.
,
24
,
1345
1358
.
Gond
,
V.
,
J.
Fontès
, and
P.
Loudjani
,
1997
:
African biomes dynamics by remote sensing temporal analysis
.
C. R. Acad. Sci.
,
320
,
179
188
.
Gond
,
V.
, and
Coauthors
,
2013
:
Vegetation structure and greenness in central Africa from Modis multi-temporal data
.
Philos. Trans. Roy. Soc. London
,
B368
,
doi:10.1098/rstb.2012.0309
.
Graham
,
E. A.
,
S. S.
Mulkey
,
K.
Kitajima
,
N. G.
Philipps
, and
S. J.
Wright
,
2003
:
Cloud cover limits net CO2 uptake and growth of a rainforest tree during tropical rainy seasons
.
Proc. Natl. Acad. Sci. USA
,
100
,
572
576
.
Hoare
,
D.
, and
P.
Frost
,
2009
:
Phenological description of natural vegetation in southern Africa using remotely-sensed vegetation data
.
Appl. Veg. Sci.
, 7, 19–28,
doi:10.1111/j.1654-109X.2004.tb00591.x
.
Huete
,
A. R.
,
K.
Didan
,
T.
Miura
,
E. P.
Rodriguez
,
X.
Gao
, and
L. G.
Ferreira
,
2002
:
Overview of the radiometric and biophysical performance of the MODIS vegetation indices
.
Remote Sens. Environ.
,
83
,
195
213
.
Huete
,
A. R.
, and
Coauthors
,
2006
:
Amazon rainforests green-up with sunlight in dry season
.
Geophys. Res. Lett.
,
33
,
L06405
,
doi:10.1029/2005GL025583
.
Indeje
,
M.
,
F. H. M.
Semazzi
, and
L. J.
Ogallo
,
2000
:
ENSO signals in East African rainfall seasons
.
Int. J. Climatol.
,
20
,
19
46
.
Janicot
,
S.
,
V.
Moron
, and
B.
Fontaine
,
1996
:
Sahel droughts and ENSO dynamics
.
Geophys. Res. Lett.
,
23
,
515
518
.
Janicot
,
S.
,
S.
Trzaska
, and
I.
Poccard
,
2001
:
Summer Sahel–ENSO teleconnection and decadal time scale SST variations
.
Climate Dyn.
,
18
,
303
320
.
Kao
,
H.-Y.
, and
J.-Y.
Yu
,
2009
:
Contrasting eastern-Pacific and central-Pacific types of ENSO
.
J. Climate
,
22
,
615
632
.
Klein
,
D.
, and
J.
Röhrig
,
2006
: How does vegetation respond to rainfall variability in a semi-humid West African in comparison to a semiarid East African environment? Proc. Second Workshop of the EARSeL SIG on Remote Sensing of Land Use and Land Cover: Application and Development, Bonn, Germany, European Association of Remote Sensing Laboratories, 149–156.
Knippertz
,
P.
,
U.
Ulbrich
,
F.
Marques
, and
J.
Corte-Real
,
2003
:
Decadal changes in the link between El Niño and springtime North Atlantic oscillation and European–North African rainfall
.
Int. J. Climatol.
,
23
,
1293
1311
,
doi:10.1002/joc.944
.
Knippertz
,
P.
,
A. H.
Fink
,
R.
Schuster
,
J.
Trentmann
,
J. M.
Schrage
, and
C.
Yorke
,
2011
:
Ultra-low clouds over the southern West African monsoon region
.
Geophys. Res. Lett.
,
38
,
L21808
,
doi:10.1029/2011GL049278
.
Kogan
,
F. N.
,
2000
:
Satellite-observed sensitivity of world land ecosystems to El Niño/La Niña
.
Remote Sens. Environ.
,
74
,
445
462
.
Larkin
,
N. K.
, and
D. E.
Harrison
,
2002
:
ENSO warm, (El Niño) and cold, (La Niña) event life cycles: Ocean surface anomaly patterns, their symmetries, asymmetries, and implications
.
J. Climate
,
15
,
1118
1140
.
Latif
,
M.
,
D.
Anderson
,
T.
Barnett
,
M.
Cane
,
R.
Kleeman
,
A.
Leetmaa
,
J.
O’Brien
,
A.
Rosati
, and
E.
Schneider
,
1998
:
A review of the predictability and prediction of ENSO
.
J. Geophys. Res.
,
103
(
C7
),
14 375
14 393
.
Linthicum
,
K. J.
,
A.
Anyamba
,
C. J.
Tucker
,
P. W.
Kelley
,
M. F.
Myers
, and
C. J.
Peters
,
1999
:
Climate and satellite indicators to forecast Rift Valley fever epidemics in Kenya
.
Science
,
285
,
397
400
.
Los
,
S. O.
,
S. J.
Collatz
,
L.
Bounoua
,
P. J.
Sellers
, and
C. J.
Tucker
,
2001
:
Global interannual variations in sea surface temperature and land surface vegetation, air temperature, and precipitation
.
J. Climate
,
14
,
1535
1549
.
Lyon
,
B.
, and
S. J.
Mason
,
2007
:
The 1997–98 summer rainfall season in southern Africa. Part I: Observations
.
J. Climate
,
20
,
5134
5148
.
Malhi
,
Y.
, and
J.
Wright
,
2004
:
Spatial patterns and recent trends in the climate of tropical rainforest regions
.
Philos. Trans. Roy. Soc. London
,
B359
,
311
329
.
Malo
,
A. R.
, and
N.
Nicholson
,
1990
:
A study of rainfall and vegetation dynamics in the African Sahel using normalized difference vegetation index
.
J. Arid Environ.
,
19
,
1
24
.
Mapande
,
A. T.
, and
C. J. C.
Reason
,
2005
:
Interannual rainfall variability over western Tanzania
.
Int. J. Climatol.
, 25, 1355–1368,
doi:10.1002/joc.1193
.
Mariotti
,
A.
,
N.
Zeng
, and
K. M.
Lau
,
2002
:
Euro-Mediterranean rainfall and ENSO—A seasonally varying relationship
.
Geophys. Res. Lett.
,
29
, doi:10.1029/2001GL014248.
Martiny
,
N.
,
P.
Camberlin
,
Y.
Richard
, and
N.
Philippon
,
2006
:
Compared regimes of NDVI and rainfall in semi-arid regions of Africa
.
Int. J. Remote Sens.
,
27
,
5201
5223
.
Martiny
,
N.
,
N.
Philippon
,
Y.
Richard
,
P.
Camberlin
, and
C. J. C.
Reason
,
2009
:
Predictability of NDVI in semi-arid African regions
.
Theor. Appl. Climatol.
, 100, 467–484.
Mason
,
S. J.
,
2001
:
El Niño, climate change, and southern African climate
.
Environmetrics
,
12
,
327
345
.
Mayaux
,
P.
,
E.
Bartholomé
,
S.
Fritz
, and
A.
Beward
,
2004
:
A new land-cover map of Africa for the year 2000
.
J. Biogeogr.
,
31
,
861
877
.
McCollum
,
J. R.
,
A.
Gruber
, and
M. B.
Ba
,
2000
:
Discrepancy between gauges and satellite estimates of rainfall in equatorial Africa
.
J. Appl. Meteor.
,
39
,
666
679
.
McKellar
,
N. C.
,
B. C.
Hewitson
, and
M. A.
Tadross
,
2007
:
Namaqualand’s climate: Recent historical scenarios
.
J. Arid Environ.
,
70
,
604
614
.
Misra
,
V.
,
2010
:
Interaction of interannual and diurnal variations over equatorial Africa
.
J. Geophys. Res.
,
115
,
D01111
,
doi:10.1029/2009JD012512
.
Mohino
,
E.
,
S.
Janicot
, and
J.
Bader
,
2011
:
Sahel rainfall and decadal to multi-decadal sea surface temperature variability
.
Climate Dyn.
,
37
,
419
440
.
Mutai
,
C. C.
, and
N. M.
Ward
,
2000
:
East African rainfall and the tropical circulation/convection on intraseasonal to interannual time scales
.
J. Climate
,
13
,
3915
3939
.
Myneni
,
R. B.
,
S. O.
Los
, and
C. J.
Tucker
,
1996
:
Satellite-based identification of linked vegetation index and sea surface temperature anomaly areas from 1982-1990 for Africa, Australia and South America
.
Geophys. Res. Lett.
,
23
,
729
732
.
Nemani
,
R. R.
,
C. D.
Keeling
,
H.
Hashimoto
,
W. M.
Jolly
,
S. C.
Piper
,
C. J.
Tucker
,
R. B.
Myneni
, and
S. W.
Running
,
2003
:
Climate-driven increases in global terrestrial net primary production from 1982 to 1999
.
Science
,
300
,
1560
1563
.
Newman
,
M.
,
S. I.
Shin
, and
M. A.
Alexander
,
2011
:
Natural variation in ENSO flavors
.
Geophys. Res. Lett.
,
38
,
L14705
,
doi:10.1029/2011GL047658
.
Nicholson
,
S. E.
, and
J.
Kim
,
1997
:
The relationship of the El Niño–Southern Oscillation to the African rainfall
.
Int. J. Climatol.
,
17
,
117
135
.
Nicholson
,
S. E.
, and
J. P.
Grist
,
2003
:
The seasonal evolution of the atmospheric circulation over West Africa and equatorial Africa
.
J. Climate
,
16
,
1013
1030
.
Nicholson
,
S. E.
,
M. L.
Davenport
, and
A. R.
Malo
,
1990
:
A comparison of the vegetation response to rainfall in the Sahel and East Africa, using normalized difference vegetation index from NOAA AVHRR
.
Climatic Change
,
17
,
209
241
.
Nicholson
,
S. E.
, and
Coauthors
,
2003
:
Validation of TRMM and other rainfall estimates with a high-density gauge dataset for West Africa. Part II: Validation of TRMM rainfall products
.
J. Appl. Meteor.
,
42
,
1355
1368
.
Ogallo
,
L. J.
,
1988
:
Relationships between seasonal rainfall in East Africa and the Southern Oscillation
.
J. Climatol.
,
8
,
31
43
.
Pau
,
S.
,
G. S.
Okin
, and
T. W.
Gillepsie
,
2010
:
Asynchronous response of tropical forest leaf phenology to seasonal and El Niño–driven drought
.
PLoS One
,
5
,
e11325
, doi:10.1371/journal.pone.0011325.
Philippon
,
N.
,
L.
Jarlan
,
N.
Martiny
,
P.
Camberlin
, and
E.
Mougin
,
2007
:
Characterization of the interannual and intraseasonal variability of West African vegetation between 1982 and 2002 by means of NOAA AVHRR NDVI data
.
J. Climate
,
20
,
1202
1218
.
Philippon
,
N.
,
M.
Rouault
,
Y.
Richard
, and
A.
Favre
,
2011
:
The influence of ENSO on South African winter rainfall
.
Int. J. Climatol.
, 32, 2333–2347,
doi:10.1002/joc.3403
.
Phillips
,
J.
, and
B.
McIntyre
,
2000
:
ENSO and interannual rainfall variability in Uganda: Implications for agricultural management
.
Int. J. Climatol.
,
20
,
171
182
.
Pinzón
,
J.
,
M. E.
Brown
, and
C. J.
Tucker
,
2005
: EMD correction of orbital drift artifacts in satellite stream data. Hilbert–Huang Transform: Introduction and Applications, N. Huang, Ed., World Scientific, 167–186.
Poccard
,
I.
,
S.
Janicot
, and
P.
Camberlin
,
2000
:
Comparison of rainfall structures between NCEP/NCAR reanalyses and observed data over tropical Africa
.
Climate Dyn.
,
16
,
897
915
.
Pohl
,
B.
,
N.
Fauchereau
,
Y.
Richard
,
M.
Rouault
, and
C. J. C.
Reason
,
2009
:
Interactions between synoptic, intraseasonal and interannual convective variability over southern Africa
.
Climate Dyn.
,
33
,
1033
1050
.
Propastin
,
P.
,
L.
Fosto
, and
M.
Kappas
,
2010
:
Assessment of vegetation vulnerability to ENSO warm events over Africa
.
Int. J. Appl. Earth Obs. Geoinf.
, 12 (Suppl. 1), S83–S89,
doi:10.1016/j.jag.2009.10.007
.
Ren
,
H.-L.
, and
F.-F.
Jin
,
2011
:
Niño indices for two types of ENSO
.
Geophys. Res. Lett.
,
38
,
L04704
,
doi:10.1029/2010GL046031
.
Richard
,
Y.
,
S.
Trzaska
,
P.
Roucou
, and
M.
Rouault
,
2000
:
Modification of the southern African rainfall variability/ENSO relationship since the late 1960s
.
Climate Dyn.
,
16
,
883
895
.
Richard
,
Y.
,
P.
Camberlin
,
N.
Fauchereau
, and
H.
Mulenga
,
2002
:
Cohérence intra-saisonnière de la variabilité pluviométrique en Afrique du Sud
.
Espace Geogr.
,
2
,
63
72
.
Rodríguez-Fonseca
,
B.
, and
Coauthors
,
2011
:
Interannual and decadal SST-forced responses of the West African monsoon
.
Atmos. Sci. Lett.
,
12
,
67
74
.
Rossow
,
W. B.
, and
R. A.
Schiffer
,
1999
:
Advances in understanding clouds from ISCCP
.
Bull. Amer. Meteor. Soc.
,
80
,
2261
2287
.
Rouault
,
M.
, and
Y.
Richard
,
2003
:
Intensity and spatial extension of drought in South Africa at different time scales
.
Water SA
,
29
,
489
500
.
Rudolf
,
B.
, and
U.
Schneider
,
2005
: Calculation of gridded precipitation data for the global land-surface using in situ gauge-based observations. Proc. Second Workshop of the International Precipitation Working Group, Monterey, CA, EUMETSAT, 231–247.
Samba
,
G.
, and
D.
Nganga
,
2012
:
Rainfall variability in Congo-Brazzaville: 1932–2007
.
Int. J. Climatol.
,
32
,
854
873
.
Segele
,
Z. T.
,
P. J.
Lamb
, and
L. M.
Leslie
,
2008
:
Large-scale atmospheric circulation and global sea surface temperature associations with Horn of Africa June–September rainfall
.
Int. J. Climatol.
,
29
,
1075
1100
.
Smith
,
T. M.
,
R. W.
Reynolds
,
T. C.
Peterson
, and
J.
Lawrimore
,
2008
:
Improvements to NOAA’s historical merged land–ocean surface temperature analysis (1880–2006)
.
J. Climate
,
21
,
2283
2296
.
Stock
,
W. D.
, and
N.
Allsopp
,
1992
: Functional perspective of ecosystems. The Ecology of Fynbos: Nutrients, Fire and Diversity, R. M. Cowling, Ed., Oxford University Press, 241–259.
Thornton
,
P. K.
, and
Coauthors
,
2006
: Mapping climate vulnerability and poverty in Africa. Report to the Department for International Development, 171 pp. [Available from ILRI, P.O. Box 30709, Nairobi 00100, Kenya.]
Tucker
,
C. J.
,
1979
:
Red and photographic infrared linear combinations for monitoring vegetation
.
Remote Sens. Environ.
,
8
,
127
150
.
Tucker
,
C. J.
,
J. E.
Pinzón
,
M. E.
Brown
,
D.
Slayback
,
E. W.
Pak
,
R.
Mahoney
,
E.
Vermote
, and
N.
El Saleous
,
2005
:
An extended AVHRR 8-km NDVI data set compatible with MODIS and SPOT vegetation NDVI data
.
Int. J. Remote Sens.
,
26
,
4485
4498
.
UNECA
,
2011
: Assessment of Africa’s climatic records and recording networks including strategic for rescuing climatic data. United Nations Economic Commission for Africa, African Climate Policy Centre Rep., 25 pp.
Verdin
,
J.
,
C.
Funk
,
R.
Klaver
, and
D.
Roberts
,
1999
:
Exploring the correlation between southern African NDVI and Pacific sea surface temperatures: Results for the 1998 maize growing season
.
Int. J. Remote Sens.
,
20
,
2117
2124
.
Vrieling
,
A.
,
K. M.
de Beurs
, and
M. E.
Brown
,
2011
:
Phenological characterization and variability of African farming systems with NDVI time series
.
Climatic Change
, 109, 455–477,
doi:10.1007/s10584-011-0049-1
.
Williams
,
C. A.
, and
N. P.
Hanan
,
2011
:
ENSO and IOD teleconnections for African ecosystems: Evidence of destructive interference between climate oscillations
.
Biogeosciences
,
8
,
27
40
.
Xiao
,
X.
,
S.
Hagen
,
Q.
Zhang
,
M.
Keller
, and
B.
Moore
,
2006
:
Detecting leaf phenology of seasonally moist tropical forests in South America with multi-temporal MODIS images
.
Remote Sens. Environ.
,
103
,
465
473
.
Zhang
,
P.
,
B.
Anderson
,
M.
Barlow
,
B.
Tan
, and
R. B.
Myneni
,
2004
:
Climate-related vegetation characteristics derived from Moderate Resolution Imaging Spectroradiometer (MODIS) leaf area index and normalized difference vegetation index
.
J. Geophys. Res.
,
109
,
D20105
,
doi:10.1029/2004JD004720
.
Zhang
,
X.
,
M. A.
Friedl
,
C. B.
Schaaf
,
A. H.
Strahler
,
J. C. F.
Hodges
,
F.
Gao
,
B. C.
Reed
, and
A.
Huete
,
2003
:
Monitoring vegetation phenology using MODIS
.
Remote Sens. Environ.
,
84
,
471
475
.
Zhang
,
X.
,
M. A.
Friedl
,
C. B.
Schaaf
,
A. H.
Strahler
, and
Z.
Liu
,
2005
:
Monitoring the response of vegetation phenology to precipitation in Africa by coupling MODIS and TRMM instruments
.
J. Geophys. Res.
,
110
,
D12103
,
doi:10.1029/2004JD005263
.