Volume 122, Issue 11 p. 5654-5676
Research Article
Free Access

Sahel precipitation and regional teleconnections with the Indian Ocean

Ellen L. E. Dyer

Corresponding Author

Ellen L. E. Dyer

Department of Physics, University of Toronto, Toronto, Ontario, Canada

Correspondence to: E. L. E. Dyer,

[email protected]

Search for more papers by this author
Dylan B. A. Jones

Dylan B. A. Jones

Department of Physics, University of Toronto, Toronto, Ontario, Canada

Search for more papers by this author
Ryan Li

Ryan Li

Department of Physics, University of Toronto, Toronto, Ontario, Canada

Search for more papers by this author
Hiromitsu Sawaoka

Hiromitsu Sawaoka

Department of Physics, University of Toronto, Toronto, Ontario, Canada

Search for more papers by this author
Lawrence Mudryk

Lawrence Mudryk

Department of Physics, University of Toronto, Toronto, Ontario, Canada

Search for more papers by this author
First published: 16 May 2017
Citations: 10

Abstract

The drought in the Sahel in the 1980s has been associated with Indian Ocean warming, although the Sahel has experienced a recovery in precipitation since the 1990s, despite continued warming in the Indian Ocean. Using the Community Earth System Model (CESM), we examined the linkages between the pattern of Indian Ocean warming and changes in atmospheric circulation over the Indian Ocean and North Africa to determine how they impact Sahel precipitation. The influence of the Indian Ocean on Sahel precipitation was investigated using a series of sea surface temperature (SST) sensitivity experiments. We identified two mechanisms by which the Indian Ocean can alter Sahel precipitation. The first mechanism is associated with perturbations in SSTs on the equator that alter Sahel precipitation by modulating the Asian monsoon circulation and driving changes in descent in North Africa. The second mechanism is associated with SST perturbations that cover more of the basin and alter the overturning circulation between the Indian and Atlantic Oceans. These two mechanisms result in different precipitation responses in the Sahel: the first induces an increase in precipitation as a result of warming in the Indian Ocean, whereas the second produces a decrease in Sahel precipitation in response to warming. Our results suggest that obtaining robust projections of precipitation in the Sahel will require reliably capturing the scale and spatial patterns of Indian Ocean warming.

Key Points

  • Sahel rainfall response to Indian Ocean SSTs is sensitive to the pattern of SST changes
  • Large-scale circulation over the Indian Ocean impacts local circulation over the Sahel
  • Indian Ocean warming increases Sahel rainfall through changes in the Asian monsoon

1 Introduction

Sahel precipitation is a complex issue from both physical and socioeconomic standpoints. In the Sahel, rainfall can affect the ability to produce crops, as rain-fed agriculture is widespread. A change in the duration of the rain season, or variability in precipitation extremes, can have a severe impact on local populations. The Sahel experienced a drying trend starting in the 1950s that culminated in severe drought in the 1980s [Dai et al., 2004; Held et al., 2005]. This was one of the first environmental crises to invoke an international response in aid and development initiatives and also from environmental research [Batterburry and Warren, 2001]. The Sahel remains an important region for climate research, focusing on the interaction between the Sahel and the large-scale climate and how this may affect future precipitation.

As noted by Lu and Delworth [2005], studies of Sahel precipitation can be split into two broad groups: regional and local. Regional studies have looked at the influence of the surrounding ocean basins and atmospheric circulation changes on Sahel precipitation [Giannini et al., 2003; Held et al., 2005; Joly et al., 2007; Hagos and Cook, 2008; Lu, 2009; Giannini et al., 2013]. Local studies, on the other hand, have focused on the impact of local phenomena such as the African Easterly Jet (AEJ) and the Saharan Heat Low (SHL) [Cook, 1999; Chou and Neelin, 2003; Sealy et al., 2003; Zhang et al., 2008; Biasutti et al., 2009; Nie et al., 2010; Pomposi et al., 2015]. Regionally, changes in the gradients in sea surface temperatures (SSTs) in the Atlantic, associated with changes in atmospheric greenhouse gases and anthropogenic aerosols, have been shown to have influenced twentieth century Sahel precipitation by shifting the location of the Intertropical Convergence Zone [Held et al., 2005]. It has also been suggested that warming in the Indian Ocean could have had a stabilizing effect on the atmosphere over the Atlantic and African sectors, leading to less favorable conditions for convection and precipitation between the 1960s and 1980s [Bader and Latif, 2003; Giannini et al., 2008]. Giannini et al. [2013] have argued that the increased precipitation over the Sahel since the 1990s is due to the recent rapid warming of the Atlantic Ocean, which supplied enhanced moisture flux from the Atlantic to the Sahel, resulting in an increase in rainfall intensity despite the stabilizing influence of the warm Indian Ocean.

Local circulation features such as the AEJ have been shown to have a strong impact on Sahel precipitation. The AEJ provides an environment that allows for the formation of African Easterly Waves (AEWs), which modulate much of the convection, in particular, mesoscale convective systems (MCSs) that are responsible for the majority of the precipitation in the Sahel [Sealy et al., 2003; Mohr and Thorncroft, 2006]. While the modulating effect of the AEJ on Sahel precipitation is clear, the positive correlation between the AEJ and Sahel precipitation amount is not conclusive as it can also be associated with divergence at lower levels, leading to a decrease in precipitation [Cook, 1999]. To the north of the AEJ, the SHL is an area of shallow dry convection that is accompanied by an anticyclonic circulation in the midtroposphere. The West African monsoon has been shown to be a “mixed-type” monsoon due to the interaction of the region of deep convection over the rain band and the shallow dry convection to its north [Nie et al., 2010]. This interaction may modulate precipitation amounts and the location of the rain band.

Several studies have examined the linkages between regional and local influences on Sahel precipitation. Chung and Ramanathan [2006] suggested that a weakening of the Asian monsoon, driven by changes in SST gradients in the Indian Ocean, could have led to an increase in Sahel precipitation as a result of a decrease in upper level easterly outflow from Asia toward northern Africa. Flaounas et al. [2012] showed that the onset of the West African monsoon is linked to the onset of the Indian monsoon through the influence of westward propagating Rossby waves from Asia, building on the mechanism for the connection between the Asian monsoon and Northern Africa first proposed by Rodwell and Hoskins [1996]. Hameed and Riemer [2012] found a strong correlation between variations in Sahel precipitation between July and September and surface level pressure differences between the Azores high and the low-pressure region extending from South Asia across North Africa. Chen [2005] suggested a direct connection between the Asian monsoon and the local atmospheric circulation over the Sahel. He argued that the descent over North Africa, linked to ascent in the Asian monsoon region and midtropospheric radiative cooling over the Sahara, coupled with the low-level ascent due to the SHL, lead to the midtropospheric divergence which strengthens the Saharan High and, consequently, the AEJ. Lu and Delworth [2005] have argued that the increase in Sahel precipitation since the 1990s is due to a change in warming in the Indian Ocean, which drove a latitudinal shift in the AEJ and the exit region of the Tropical Easterly Jet (TEJ), resulting in a shift in the rain belt over the Sahel. Similarly, Hagos and Cook [2008] suggested that the partial recovery of Sahel precipitation in the 1990s may be the result of a shift in Indian Ocean SSTs which caused increased divergence over the Atlantic rather than over the Sahel.

Here we revisit the connection between regional and local effects, focusing on the response of Sahel precipitation to variations in SSTs in the Indian Ocean. Using the NCAR Community Earth System Model (CESM), we investigate the regional and local circulation responses to change in gradients in Indian Ocean SSTs and how these circulation changes in turn influence precipitation in the Sahel. We emphasize that our objective here is not to explain the drought and recovery of rainfall in the Sahel. We are interested in understanding the linkage between the pattern of warming in the Indian Ocean and Sahel precipitation changes. We begin in section 2 with a description of the methodology and data sets employed in the analysis, along with an evaluation of the model simulation and a discussion of the various perturbation experiments that are conducted. In section 5 we present the results of the experiments. Discussions and conclusion are presented in section 6.

2 Data and Methods

2.1 CESM and Data Products

We employ the NCAR CESM version 1.0.4 with active and coupled atmosphere (CAM 5.1) [Neale et al., 2012] and land (CLM 4.0) [Oleson et al., 2010] models with a data ocean using the merged Hadley Centre for Climate Prediction and Research Global sea-Ice coverage and Sea Surface Temperatures (HadISST) and National Ocean and Atmospheric Administration (NOAA) Optimal Interpolation analyses [Hurrell et al., 2008]. The model is run at a horizontal resolution of 1.9° × 2.5° with estimated twentieth century forcing. Results from an unperturbed, baseline simulation are compared with the European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis, ERA-Interim [Dee et al., 2011] to assess the representation of important circulation features and precipitation patterns in CESM. Although ERA-Interim has a much shorter time series than its predecessor (ERA-40), it has a much better hydrological cycle, representation of clouds, and important circulation features such as the AEJ [Trenberth et al., 2011; Dee et al., 2011]. Precipitation time series from the CESM baseline simulation are also compared to the Climate Research Unit (CRU) TS 3.0 precipitation data set, which uses a network of weather stations [Harris et al., 2014].

The Sahel has one rainy season from June to September, during which rain rates can reach up to 5 mm/d in August. Monthly climatologies (for 1980–2000) for Sahel precipitation for CESM, CRU, and ERA-Interim are shown in Figure 1a. There is good agreement between CESM and the data products, especially CRU, during the rainy season. CESM also reproduces the twentieth century drought well, relative to the CRU time series as shown in Figure 1b. Projections of Sahel precipitation do not agree across fully coupled climate models [Held et al., 2005; Biasutti and Giannini, 2006] and that is why we use the Atmospheric Model Intercomparison Project (AMIP) protocol with a data ocean and twentieth century SSTs for our analysis. For example, Kamga et al. [2005] evaluated an earlier version of the coupled NCAR model, climate system model (CSM) version 1.3 over West Africa and found that the model underestimated precipitation rates during the wet season in West Africa, the Sahel, and the Guinea region and did not capture the persistent drying trend in the Sahel between the 1950s and 1980s.

Details are in the caption following the image
(a) Climatology of Sahel precipitation (mm/d) from CESM, CRU, and ERA-Interim for 1980–2000. (b) Rainy season (June–September) average precipitation (mm/d) from CESM and CRU for 1940–2010, smoothed with a 3 year running average. Area averaged for the Sahel is 10–20°N, and 15°W–15°E. The CESM precipitation is from the baseline AMIP configuration run. Error bars in Figure 1a show 1 standard deviation of the monthly mean.

Precipitation in the Sahel peaks in August and September, in the later part of the West African monsoon rainy season as the rain band moves north from the coast. Consequently, we focus here on the months of August and September for our analysis. Shown in Figure 2 are the jet streams that appear over Africa during August and September from both CESM and ERA-Interim. Near the surface (925 hPa), there is strong inflow from the Atlantic and the Guinea coast in the form of the Westerly African Jet. This jet has a large influence on the West African Monsoon. At this same level over the Indian Ocean, the Somali Jet, associated with the transport of moisture into the Asian Monsoon, is present. In the midtroposphere, the AEJ extends from the central Sahel out over the Atlantic Ocean. This is accompanied by the Saharan anticyclone circulation to the north of the Sahel. South of the equator, there is another easterly flow, the AEJ-S, which is much weaker than the northern AEJ. The AEJ has been shown to play a role in the strength and location of Sahel precipitation [Sealy et al., 2003; Mohr and Thorncroft, 2006]. In the midtroposphere (700 hPa) there is also an anticyclonic circulation over the Arabian peninsula. In the upper troposphere (175 hPa), the TEJ extends across the Indian Ocean from the southern exit of the Asian Monsoon flow. The core is at 10°N and can reach wind speeds of over 25 m/s and 15 m/s over the Sahel. The TEJ extends across Africa and is an important feature linking the Asian monsoon and North African regions. The TEJ is stronger in CESM than in ERA-Interim and extends farther west. Conversely, the AEJ-N is weaker in CESM than it is in ERA-Interim. In the lower troposphere Atlantic inflow is stronger in CESM than ERA Interim.

Details are in the caption following the image
Comparison of CESM and ERA-Interim winds averaged for August and September from 1980 to 2000. Filled contours represent the zonal component of wind. CESM winds are from the baseline AMIP configuration run and were interpolated onto ERA-Interim pressure levels for comparison.

To better understand the model simulation of the circulation connecting Africa and the surrounding regions, it is useful to look at the velocity potential, which is used as a proxy for large-scale ascent and descent. In Figure 3 the divergent component of the wind from CESM and ERA-Interim is superimposed on velocity potential contours, where positive velocity potential indicates convergence and negative velocity potential indicates divergence. As can be seen in Figure 3, there is good agreement between CESM and ERA-Interim, suggesting that this version of CESM, in an AMIP configuration, does capture well the large-scale vertical motions in the region. In the upper troposphere (175 hPa), there is strong divergence over the western equatorial Pacific, near southeast Asia, and convergence in the Southern Hemisphere over the Atlantic. This pattern represents the combined influences of the Asian monsoon and Hadley and Walker circulations [Tanaka et al., 2004]. In the lower troposphere (at 700 hPa), the dipole is shifted and Africa is now characterized by strong divergence, especially in the west and north, with convergence in the equatorial western Pacific, which reflects the Walker circulation. Near the surface (at 925 hPa), there is a north-south dipole across the equator with a gradient that is especially noticeable in the Indian Ocean. This is indicative of the strong inflow from the Indian Ocean into the Asian monsoon region which leads to the strong ascent over the region and divergence aloft. At this level near the surface (925 hPa), there is also convergence over North Africa, which is strongest over the Sahel, with inflow from the Atlantic.

Details are in the caption following the image
Same as in Figure 2, except for the velocity potential (m2/s). Contours show the velocity potential (zero contour of velocity potential is the solid magenta contour) and vectors represent the divergent wind component.

As a preliminary step toward understanding the impact of regional atmospheric circulation changes on Sahel precipitation, we examine the circulation changes associated with particularly wet and dry years in the baseline CESM simulation. Shown in Figure 4 is the difference in circulation between wet and dry composites near the surface (Figure 4a) and in the lower troposphere (Figure 4b). Wet and dry composite years were determined by selecting years that were wetter or dryer than 1 standard deviation of the average precipitation within the Sahel region, during the rainy season, without detrending.

Details are in the caption following the image
Difference between lower and midtropospheric winds at (a) 912 hPa and (b) 690 hPa in wet and dry composites from CESM baseline run from 1950 to 2000 for August–September (m/s). All major features are significant to a 95% significance level. Composites are not detrended.

During wet years, there is increased inflow from the Atlantic along the Guinea coast. There is increased moisture flux from the Atlantic Ocean at 10°N into West Africa at this level. Near the surface, there is anomalous cyclonic flow over North Africa, with southerly flow across the Sahara, and easterly flow across the Arabian Sea that weakens the Somali Jet. The anomalous easterly flow over the Arabian Sea is also present in the midtroposphere (Figure 4b) and is accompanied by anomalous easterly winds at 20°N in the eastern Sahel. This northward shift of the eastern core of the AEJ during wet years has been previously highlighted by Nicholson [2013]. At this level, there are anomalous westerly winds (deceleration in the easterly winds) over West Africa, which indicate a change in the position and strength of the AEJ. This anomalous westerly flow is associated with increased moisture flux into West Africa, although between 20 and 30°N there is increased easterly moisture flux across the Sahel which extends into the Atlantic (not shown). These differences will provide a useful context for interpreting the results of the SST perturbation experiments discussed below.

2.2 Perturbation Experiments

To assess the potential impact of changes in Indian Ocean SSTs on Sahel precipitation, it is useful to understand the patterns of variability of precipitation in the Sahel and SSTs in the Indian Ocean. Figure 5 shows the first three empirical orthogonal functions (EOFs) for African precipitation during August–September using the available years 1950–2000 from the baseline CESM simulation. The first EOF resembles the rain belt pattern and accounts for the bulk of the variability (31.5%). The second EOF pattern accounts for 17.6% of the variability and resembles the West African monsoon precipitation, and extends into Central Africa. The third EOF is a dipole centered on 10°N, the southern part of which includes coastal West African and Central African precipitation. All of these EOF patterns feature a dipole signal along the Atlantic coast of the western Guinea region. The variability in precipitation captured by these EOFs reflects the local dynamical interactions between the Sahel and the other regions in Western Africa.

Details are in the caption following the image
First three EOFs for African precipitation (25°S–35°N, 20°W–55°E) using the detrended CESM control from 1950 to 2000 for the August–September average.

The first three EOFs for Indian Ocean SSTs during the same period are shown in Figure 6. An examination of the correlation between the EOFs of Indian Ocean SSTs and Sahel precipitation was not done; these EOFs are examined purely as an indication of patterns of natural variability in both fields. The bulk of the SST variability represents warming throughout the most of the basin, with greater warming in the western and central Indian Ocean. The second and third EOFs resemble the variations in SSTs due to the seasonal cycle and the Indian Ocean dipole, respectively. We use these patterns of variations as a guide to construct recognizable perturbations to Indian Ocean SSTs for our experiments. Studies of the effect of the variations in Indian Ocean SSTs have previously been done, but the patterns of warming follow the complex warming of the twentieth century with temperature changes localized in different sections of the Indian Ocean basin [Lu and Delworth, 2005; Chung and Ramanathan, 2006; Lu, 2009].

Details are in the caption following the image
First three EOFs for Indian Ocean SSTs (30°S–30°N, 30°E–110°E) using the detrended CESM control from 1950 to 2000 for the August–September average.

The experiments conducted here attempt to better identify the precipitation response in the Sahel to different patterns of SST changes in the Indian Ocean. The imposed SST perturbations are simplified and amplified versions of different patterns captured by the EOFs that we take as an indication of existing patterns of variability and change. These include the Indian Ocean dipole, the seasonal change in SST gradients, an equatorial Indian Ocean warming, and total basin warming. By perturbing SSTs, we alter convergence and divergence over the Indian Ocean and can then examine the corresponding change over the Sahel. As these experiments are done in an AMIP configuration and with SST perturbations that are constant, we do not capture all the possible feedbacks that should be associated with a change in SST, but this allows us to isolate a direct response in the atmospheric circulation to an SST perturbation.

All of the SST perturbations used for the experiments presented here are shown in Figure 7, with supporting information in Table 1. For example, the LON_PG experiment imposes a warming with a longitudinal gradient (between 50°E and 95°E) consisting of positive SST anomalies in the western Indian ocean, decreasing to zero on the eastern edge of the domain. This pattern is similar to the warming pattern projected by the Coupled Model Intercomparison Project (CMIP5) models [Zheng et al., 2013]. The LON_NG experiment, in contrast, reflects an amplification of the seasonal cycle in Indian Ocean SSTs, with colder temperatures in the western Indian Ocean, off the coast of East Africa. The LAT_NG experiment imposed a negative SST anomaly that varied from −3 K to 0 K from 10°S to 10°N, whereas the LAT_PG experiment imposed a positive anomaly that decreased from +3 K to 0 K over the same latitude range. As discussed above, some of the experiments were designed to be simplified representations of the patterns of variability in Indian Ocean SSTs, but others were chosen to match experiments in the literature. The TP_-1K experiment, for example, is similar to an experiment done by Bader and Latif [2003]. We have conducted different variations of this experiment with different degrees of warming and cooling (e.g., TP_1K, TP_3K, and TP_-3K). The results from these experiments will be discussed in section 5.

Details are in the caption following the image
SST perturbations applied in the CESM SST sensitivity experiments.
Table 1. CESM SST Perturbation Experiments
Experiment Run Type SST Perturbation
LON_NG long run −3 K:0 K longitudinal gradient perturbation 10°S:10°N 50:95°E, (coolest at 50°E)
LON_PG long run 3 K:0 K longitudinal gradient perturbation 10°S:10°N 50:95°E (warmest at 50°E)
LAT_NG long run −3 K:0 K latitudinal gradient perturbation 10°S:15°N 50:100°E (coolest at 10°S)
LAT_PG long run 3 K:0 K latitudinal gradient perturbation 10°S:15°N 50:100°E, (warmest at 10°S)
CP_1K long run 1 K perturbation 10°S:10°N 50:95°E, central Indian Ocean
TP_1K short runs 1 K perturbation, total basin
TP_3K short runs 3 K perturbation, total basin
TP_-1K short runs −1 K perturbation, total basin
TP_-3K short runs −3 K perturbation, total basin
EP_3K long run 3 K perturbation 10°S:20°N 50:95°E, central patch

Two types of perturbation experiments were carried out: long perturbation runs (LR) and short perturbation runs (SR). The LR simulations were run from 1940 to 2000 with a sustained Indian Ocean SST perturbation throughout all months and years. The SR runs are an ensemble of 1 year simulations that were conducted for computational expediency. These simulations were run for every even year from 1980 to 2000, inclusive, with SST perturbation applied during all months and using baseline conditions at the start of the simulation. Starting these runs in January allows time for the model to spin up before the rainy season in the Sahel. LR experiments include CP_1K, EP_3K, LON_NG, LON_PG, LAT_NG, and LAT_PG. SR runs include TP_1K, TP_3K, TP_-1K, and TP_-3K. As the LR experiments have SST perturbations that were fixed for all months and all years of the simulation, one of the LR experiments (LON_NG) was repeated as an SR simulation to see if there was any significant bias between the LR and SR configurations. Also, to test whether having the SST perturbations applied throughout the year introduced a bias in the precipitation response, we also repeated the LON_NG experiment as an SR run, but with the SST perturbation applied only from June to September. We found that the results of these LON_NG sensitivity tests all showed similar responses by the beginning of the rainy season and, in particular, by the August–September period that we examine in this analysis. For the rest of the discussion in this study we focus only on the SR and LR experiments sampled for years that overlap for both sets of experiments (every even year between 1980 and 2000). In the design of these perturbation experiments, we aim to alter the mean state of Sahel precipitation rather than changing the variability and therefore examine the average climatological response to the perturbations.

3 Results

The precipitation changes in the Sahel that result from the SST perturbations (Figure 7) are greatest in the months of August and September, which is consistent with results from past studies [Bader and Latif, 2003; Sealy et al., 2003; Nicholson, 2013] and also with these months being the most variable part of the rainy season, as previously noted. We find that the imposed Indian Ocean SST perturbations do not alter the onset of the rainy season, therefore, for the results presented here we will focus only on the months of August and September. The precipitation changes for the various experiments are shown in Figure 8. Some of the precipitation changes are localized to the Sahel rain band while others show a broader pattern of changes that extend to Central Africa, with isolated changes in the western sector of the Sahel. We find that in some experiments the precipitation response is of the same sign as the SST perturbations, whereas in others the response is opposite in sign. We refer to these as Group 1 (Figures 8a–8e) and Group 2 (Figures 8f–8j), respectively.

Details are in the caption following the image
(a–j) Difference (experiment – baseline) in mean (August–September) precipitation due to the Indian Ocean SST perturbations in CESM. (k) The CESM AMIP configuration baseline precipitation. Units are mm/d and stippling in Figures 8a–8j indicates 95% significance.

In the Group 1 experiments, the change in precipitation is localized to the Sahel and extends from the western Sahel to the Horn of Africa. There is also a dipole pattern with the change in Sahel precipitation opposite in sign to the change in precipitation in western Central Africa. The LON_NG experiment (Figure 8a) is a good example of the precipitation response in this group. In contrast, the Group 2 experiments, of which TP_1K (Figure 8h) is a good example, are characterized by broader precipitation changes of the same sign in western Africa. Another interesting difference between the two groups is the response of East African precipitation. In Group 1, the change in precipitation over the Horn of Africa is the same sign as over the western Sahel, whereas in Group 2 the precipitation response over the Horn of Africa is opposite to that over the western Sahel. The pattern of the precipitation response in Group 1 closely resembles the pattern of EOF 1 (Figure 4) for West African precipitation, whereas the pattern of the response in Group 2 more closely resembles EOF 2.

To understand the connections between Indian Ocean SST perturbations and the Sahel precipitation response, we examine the changes in circulation in the midtroposphere across the region, which are shown in Figure 9. One feature which is apparent in all of the Group 1 experiments is the presence of an anomalous cyclonic or anticyclonic circulation across South Asia and northeastern Africa. For experiments in Group 1 with an increase (decrease) in Sahel precipitation, there is an anomalous anticyclonic (cyclonic) circulation between 20°N and 30°N. Associated with this anomalous anticyclonic (cyclonic) flow is an anomalous easterly (westerly) flow across the central and eastern Sahel at 20°N. In all Group 1 experiments there is anomalous westerly flow from the Atlantic around 25°N over the western Sahel. However, for the experiments with a decrease in Sahel precipitation (Figures 9a and 9c) this westerly flow becomes northerly over the central Sahara, and connects to anomalous easterly flow to the Atlantic Ocean around 10°N, forming a center of anomalous anticyclonic flow at 15°N on the Atlantic coast of the western Sahel. For experiments with reduced precipitation, the anomalous westerly flow in the central Sahel at 20°N, in the AEJ acceleration region, and the anomalous easterly flow at 10°N suggest a shifted and weaker AEJ. For the experiments with increased precipitation (Figures 9b, 9d, and 9e), the circulation pattern, which consists of stronger easterly winds in the central Sahel at 20°N and anomalous westerly winds over the coast of West Africa, between the equator and 10°N, is similar to the circulation pattern shown in Figure 4b for the composite of wet years minus dry years.

Details are in the caption following the image
Difference (experiment − baseline) in August–September averaged winds (m/s) at 690 hPa in CESM.

In the Group 2 experiments, the circulation response is also dominated by anomalous cyclonic or anticyclonic flow in the midtroposphere, but the circulation pattern is shifted westward relative to that in Group 1 and is centered over northeastern Africa and the Arabian Peninsula, with no consistent change occurring over the Asian subcontinent. In the Group 2 experiments with reduced precipitation (Figures 9f, 9h, and 9j) the circulation is cyclonic, whereas for those with increased precipitation (Figures 9g and 9i) the circulation is anticyclonic. Similar to Group 1, the Group 2 experiments with reduced precipitation have anomalous westerly flow in the central Sahel at 20°N, whereas those with increased precipitation have enhanced easterly flow in this region. However, there is little resemblance between the large-scale circulation changes in Group 2 and the composite differences shown in Figure 4b.

The large-scale changes in the midtropospheric horizontal flow suggest that there must be significant changes in the vertical motions regionally and locally in the Sahel. The main precipitation response across the Sahel is between 10° and 20°N (see Figure 8), consequently in Figure 10, we have plotted the change in vertical motion at 15°N, from the eastern Atlantic Ocean to the western Pacific Ocean. Across the whole region, the background state in August and September consists of ascent throughout much of the troposphere. In the Group 1 experiments the imposed SST perturbations drive changes in the vertical motions with a dipole pattern, with changes in the western part of the domain opposite in sign to those in the east. For example, in the Group 1 experiments with enhanced Sahel precipitation (Figures 10b, 10d, and 10e), there is weakened ascent in the east (vertical velocities become less negative), extending from the Arabian Sea to the western Pacific, whereas there is enhanced ascent in the west, across western Africa. In the Group 2 experiments, the circulation response over the western and eastern parts of the domain, western Africa and the western Pacific Ocean, respectively, is similar and opposite to that over the central region (the Arabian Sea and the Indian Ocean). In the Group 2 experiments with basinwide heating in the Indian Ocean, such as in TP_3K (Figure 10j), there is enhanced ascent over the Arabian Sea with reduced ascent over West Africa and the western Pacific. In all Group 2 experiments, there is a change in sign in the circulation response over western and central Africa in the midtroposphere; enhanced ascent extends down to about 500 hPa, below which there is weakened ascent. In Group 1 experiments, the change in vertical motion in the eastern Sahel is uniform throughout the troposphere at this latitude.

Details are in the caption following the image
Difference (experiment − baseline) in mean (August–September) vertical motion along 15°N in CESM. The black contour lines show the baseline vertical velocity, with solid lines for descent (positive values) and dashed lines for ascent (negative values). Color contours are the change in vertical velocity for each experiment relative to the baseline. Units for the color contours are 10−2 Pa/s. Results are sampled every even year between 1980 and 2000.

The meridional structure in the circulation, averaged across the western Sahel, is shown in Figure 11. The background state is characterized by strong ascent between 10° and 15°N (corresponding to the region of ascent shown in Figure 10) and descent around 5°S and 30°N. Between 20° and 25°N there is a shallow region of ascent which is capped at 500 hPa, above which there is descent. There is northerly flow between 10°N and 25°N below 600 hPa. This circulation is the shallow meridional circulation (SMC) described by Zhang et al. [2008]. This connects the Sahara and the Sahel rain band as air from the dry northern Sahara is transported into the region of strong ascent.

Details are in the caption following the image
Mean August–September meridional overturning in the baseline CESM run, averaged from 15°E to 15°W. Arrows indicate the local circulation. The color contours indicate the vertical velocity, with positive values for descent and negative values for ascent. Vertical velocity contours are scaled by 20 hPa/s, whereas vertical velocity vectors are scaled by 200 hPa/s.

The changes in the meridional circulation in the Group 1 experiments are shown in Figure 12, and for Group 2 in Figure 13. In all experiments with reduced Sahel precipitation (Figures 12a, 12d, 13a, 13c, and 13e), there is anomalous descent over the western Sahel, with anomalous northerly flow in the midtroposphere. In contrast, in the experiments with increased precipitation (Figures 12b, 12c, 12e, 13b, and 13d) there is anomalous ascent across the western Sahara with anomalous southerly flow in the middle troposphere. A key difference between the Group 1 and Group 2 experiments is that the anomalous northerly or southerly flow in the Group 1 experiments is confined poleward of 10°–15°N, whereas for the Group 2 experiments the anomalous meridional flow extends from the equator to 40°N and is strongest near the equator. This northerly (southerly) meridional flow is accompanied by an increase in ascent (descent) below 600 hPa, which for Group 1 is north of 20°N, over the SHL region. For Group 2 the anomaly in lower tropospheric ascent (descent) covers a more broad area between 10° and 25°N, and was clearly visible in Figures 10f–10j. The anomalous ascent in the Group 1 experiments is fairly uniform poleward of 10°N and opposite in sign to the response in vertical motions equatorward of 10°N. In contrast, the anomalous ascent in the Group 2 experiments extends across much of the region from 0° to 40°N.

Details are in the caption following the image
Differences (experiment − baseline) in mean (August–September) meridional overturning for Group 1 experiments, averaged from 15°E to 15°W in CESM. Arrows indicate the change in local circulation. The color contours are the change in vertical velocity for each experiment and the white contour lines show the baseline vertical velocity with upward motion indicated by the dashed contours. Vertical velocities were scaled by 200 Pa/s.
Details are in the caption following the image
Differences (experiment − baseline) in mean (August–September) meridional overturning for Group 2 experiments, averaged from 15°E to 15°W in CESM. Arrows indicate the change in local circulation. The color contours are the change in vertical velocity for each experiment and the white contour lines show the baseline vertical velocity with upward motion indicated by the dashed contours. Vertical velocities were scaled by 200 Pa/s.

Anomalous northerly flow in the midtroposphere between 20° and 30°N reflects increased divergent flow linked with enhanced descent above and ascent below about 600 hPa. This divergent flow, as described by Chen [2005], influences the AEJ and is due to descent over North Africa associated with the Asian monsoon, and ascent in the lower troposphere driven by the SHL. A good example of this interaction is Figure 12a. There is increased ascent below 600 hPa between 20°N and 25°N, which is part of the SHL region, and increased descent above 600 hPa, accompanied by an increase in the shallow meridional overturning circulation, which was shown in Figure 11. The correlation between the northerly flow between 700 and 500 hPa (averaged over the western Sahel) and the zonal wind midtropospheric wind is shown in Figure 14. We find that the northerly flow is negatively correlated with the easterly flow at 20°N over northern Africa. This region of negative correlation coincides with the 600 hPa divergent center identified by Chen [2005] as playing an important role in the maintenance of the Saharan high and thus the AEJ. Comparison of Figures 9 and 12 shows that the experiments with weaker easterly flow do indeed have anomalous northerly flow. An increase in this northerly flow enhances the advection of dry air into the rain band region and reduces convection. This can be seen clearly in the LAT_NG experiment (Figure 12d), which has anomalous ascent below and descent above 600 hPa, between 20° and 25°N, that is accompanied by weakened ascent in the rain band, near 15°N.

Details are in the caption following the image
Correlation between meridional winds (15°W–15°E, 10–20°N) at 690 hPa and regional zonal winds at 690 hPa from the CESM baseline run (August–September). Contours indicate a 95% significance.

As discussed above, the meridional circulation responses in the Group 1 experiments are localized mainly in northern Africa, poleward of 10°–15°N, whereas in the Group 2 experiments they extend to the equator. To better understand the different circulation response between the Groups 1 and 2 experiments, it is helpful to look at the upper tropospheric velocity potential. Following Tanaka et al. [2004], we have decomposed the velocity potential into the contributions from the Walker circulation and the monsoon circulation, dominated by the Asian monsoon in this domain. As can be seen in Figure 15, the Walker circulation is associated with outflow from the western Pacific (near Indonesia) and convergence over the eastern Atlantic and eastern Pacific. In contrast, the Asian monsoon circulation reflects divergent flow from South Asia and convergence over the Atlantic Ocean.

Details are in the caption following the image
Decomposition of 168 hPa divergent winds (vectors, m/s) and velocity potential (color contours, m2/s) in the CESM AMIP configuration baseline into the Walker and monsoon components. The monsoon component is shown for August. Red boxes represent regions used for indices calculated in Table 2. From left to right the regions are categorized as Atlantic-West Africa, Indian Ocean, Asian monsoon, and Warm Pool.

Although the Walker circulation is a time-averaged pattern and the monsoon is a pattern that changes throughout the year, both patterns shown in Figure 15 represent a time-averaged climatology. We show that the responses in the two groups are each sensitive to a change in one of these two patterns. The changes in the velocity potential for LON_PG and TP_1K, which are representative of the velocity potential changes in the Groups 1 and 2 experiments, respectively, are shown in Figure 16. In the Group 1 experiments, the changes in velocity potential are characterized by a strengthening or weakening of the Asian monsoon dipole pattern. For example, in the LON_PG experiment shown in Figure 16a, in which there is increased precipitation in the Sahel, there is anomalous divergence over the western Indian Ocean (where SSTs were increased) and anomalous convergence in the Asian monsoon region, weakening the overturning circulation associated with the Asian monsoon. This results in reduced descent over North Africa and a reduction in dry advection into the rain belt by the SMC. In contrast, in the Group 2 experiments, the changes in velocity potential have a zonal orientation, in which anomalous divergence (convergence) over the Indian Ocean drives anomalous convergence (divergence) over the western Pacific (near the Philippines) and over Central America and the eastern Pacific. The Group 2 response resembles more closely a perturbation of the Walker circulation. As shown in Figure 16b, in the TP_1K warming experiment, there is anomalous divergence over the Indian Ocean that drives anomalous descent over the western Pacific, North America, and the Atlantic. In this experiment, West Africa is characterized by anomalous descent, whereas East Africa is characterized by anomalous ascent, reflecting the changes in the divergent field over the Atlantic and Indian Oceans. This is consistent with the reduced precipitation in the western Sahel and the increased precipitation over the horn of Africa obtained in this experiment (see Figure 8h). As discussed above, this dipole structure in the precipitation response across the Sahel is a key feature of the Group 2 experiments, and Figure 16 suggests that it is linked to the changes in the zonal overturning circulation in these experiments.

Details are in the caption following the image
Difference (experiment − baseline) in August-September averaged divergent wind (vectors, m/s) over velocity potential (color contours, m2/s) at 168 hPa for a (a) Group 2 and (b) Group 1 experiment. Results are sampled every even year between 1980 and 2000.

To better illustrate the differences in the response between the two groups of experiments shown in Figure 16, we summarize in Table 2 the change in upper tropospheric velocity potential in four key regions for all experiments. The regions are shown in Figure 15a and all except the Asian monsoon lie along the equator and represent the convergence center over the Gulf of Guinea, the tropical Indian Ocean, and the western Pacific warm pool. For Group 1 experiments, a positive (negative) change in the Indian Ocean is always associated with a positive (negative) change in the Atlantic-West African region. Conversely, in Group 2 experiments a positive (negative) change in the Indian Ocean is associated with a negative (positive) change over the Atlantic-West African region and the warm pool, indicating a shift in the overturning associated with the Walker overturning shown in Figure 15. In Group 1, the relationship resembles the monsoon overturning in Figure 15, with a change over the Warm Pool that is the same sign as the change over the Asian monsoon region and opposite to that over both the Indian Ocean and Atlantic-West African regions. In Group 2, the changes over the Asian monsoon and Warm Pool regions are of opposite sign, indicating that the changes over these two regions are distinct and support the assertion that two different mechanisms connect Indian Ocean SST patterns in the two groups of experiments.

Table 2. Averages of Upper Tropospheric Velocity Potential Differences (Baseline Experiment) in CESM SST Perturbation Experiments in Regions Indicative of the Walker and Monsoon Components of the Circulationa
Experiment Warm Pool Indian Ocean Atlantic-West Africa Asian Monsoon
Group 1
LON_NG −3.9 3.5 3.3 −4.3
LAT_NG −5.0 5.0 4.3 −5.5
LAT_PG 5.9 −9.0 −3.1 3.9
LON_PG 4.6 −4.9 −3.4 6.0
CP_1K 5.7 −9.9 −3.3 3.2
Group 2
TP_1K 3.3 −5.5 0.2 −0.6
TP_3K 1.8 −13.5 5.3 −6.2
EP_3K 6.2 −18.2 0.9 −2.9
TP_-1K −2.6 8.9 −1.6 1.2
TP_-3K −5.9 23.8 −4.3 11.3
  • a Values are scaled by 106 and units are m2/s. Regions used for the indices here are shown in Figure 15a.

4 Discussion and Conclusions

We have examined the impact of perturbations in Indian Ocean SSTs on Sahel precipitation. In all of the SST perturbation experiments considered, the greatest change in precipitation in the rainy season occurred in August and September, effectively altering the second half of the season. The experiments can be categorized into two groups based on the sign and spatial extent of the precipitation change in the Sahel. In the Group 1 experiments, an increase (decrease) in Indian Ocean SSTs produced an increase (decrease) in precipitation across the Sahel rain band. In contrast, in the Group 2 experiments, an increase (decrease) in Indian Ocean SSTs produced a decrease (increase) in precipitation in the Western Sahel; this decrease (increase) extends into Central Africa and is opposite to the change in precipitation produced in Eastern Africa. The key difference between the imposed SSTs perturbations in the experiments is that, in Group 1, the SST changes are located near the equator (10°S–15°N), whereas in Group 2, they have a larger spatial extent and in all cases extend further north than those of Group 1. We found that the two groups of experiments are associated with distinct changes in the large-scale overturning circulation in the troposphere indicating two distinct mechanisms. In Group 1, the mechanism by which a change in Indian Ocean SSTs alter Sahel precipitation is by altering the circulation between the Indian Ocean and the Asian monsoon region. The change in Asian monsoon overturning modulates descent over North Africa, producing a change in precipitation along the Sahel rain band. In Group 2, the mechanism by which a change in Indian Ocean SSTs alter Sahel precipitation is by altering the overturning between the Indian Ocean and the Atlantic and Pacific Oceans, resulting in a perturbation to the Walker circulation.

Although the perturbation patterns in Indian Ocean SSTs considered here are simplified representations of the actual pattern of variability in Indian Ocean SSTs, our results are consistent with previous studies. The relatively large change in August and September precipitation has been previously shown by Bader and Latif [2003] and is consistent with these months being the most variable of the rainy season [Sealy et al., 2003; Nicholson, 2013]. The TP_-1K and TP_1K experiments are similar to those done by Bader and Latif [2003] and produced similar results, in which a decrease in Indian Ocean SSTs resulted in an increase in Sahel precipitation. Furthermore, the Bader and Latif [2003] precipitation response exhibited a dipole structure, with changes in precipitation between the western and eastern Sahel that are opposite in sign and in agreement with the precipitation changes in our Group 2 experiments shown in Figure 8. As noted in section 5, the longitudinal dipole structure in the precipitation response across the Sahel reflects the changes in zonal overturning circulation in the Atlantic and Indian Oceans.

The pattern of the SST perturbations in the LON_PG Group 1 experiment is similar to the warming pattern projected by the CMIP5 models [Zheng et al., 2013]. We found that an increase in SSTs in the Group 1 experiments produced an increase in precipitation across the Sahel rain band. This is in contrast to Lu and Delworth [2005] and Bader and Latif [2003] who found that basinwide warming in the Indian Ocean caused a decrease in precipitation. The SST changes imposed by Bader and Latif [2003] and Lu and Delworth [2005] more closely resemble the Group 2 experiments. Our Groups 1 and 2 results suggest that the different responses obtained by Lu and Delworth [2005] and Bader and Latif [2003] reflect the sensitivity of the Sahel precipitation response to the spatial distribution of the Indian Ocean warming. Indeed, our Group 1 results, which highlight the possible impact of changes in the Asian monsoon on Sahel precipitation, are consistent with those of Chung and Ramanathan [2006] who showed that a change in the seasonal SST gradient in the Indian Ocean weakened the Asian monsoon and resulted in a slight increase in Sahel precipitation. The nature of the experiments used in this study is such that SSTs are being used as a way to force heating changes in the lower troposphere. Another way to impose such changes is by directly perturbing the heat flux into the atmospheric model. In two companion experiments (not shown), we altered sensible heat fluxes to the atmosphere in the Asian monsoon region and over the Arabian Peninsula. Both of these experiments effectively modified the zonal heating gradient between the Asian monsoon and North Africa. The responses were also classified as Group 1, and the Asian monsoon perturbation showed responses over Africa in both circulation and precipitation that were remarkably similar to those of LON_NG and LAT_NG, highlighting the connection between Group 1 responses and the Asian monsoon.

Changes in the midtropospheric zonal flow over Africa produced in this study are consistent with changes that have been obtained in previous studies [Sealy et al., 2003; Lu and Delworth, 2005; Hagos and Cook, 2008]. In years with historically dry anomalies in August and September, the AEJ has been shown to move equatorward, making conditions less suitable for convective and stratiform precipitation [Sealy et al., 2003]. In the Group 1 experiments, the equatorward shift of easterly winds west of 5°W is associated with a decrease in Sahel precipitation. We have also shown that there is a large change in the AEJ acceleration region, or the eastern core, for these experiments. An increase in easterly flow at 20°N is linked with an increase in Sahel precipitation, in agreement with Nicholson [2013]. The change in horizontal winds for Group 2 shows a broad cyclonic or anticyclonic structure which encompasses most of North Africa. This looks quite different from the change in the first group, consistently altering the AEJ flow across the Sahel between 10°N and 20°N. Hagos and Cook [2008] state that a change in the AEJ can also alter the easterly moisture flux, suggesting that the AEJ modulates not only dynamic conditions for precipitation, but also moisture advection. An increase in easterly flow at 10°N and west of 0°, or the western core, could cause an increase in local moisture flux from the western Sahel to the Atlantic. This would be consistent with the subsequent Sahel precipitation decrease in experiments LON_NG, LAT_NG, and TP_1K with this increase in easterly flow in the western core. Indeed the change in TP_1K along with LON_NG and LAT_NG do have an increase in easterly moisture flux into the Atlantic at 10°N (not shown).

Cook [1999] has argued that the thermally direct secondary circulation over the western Sahel, which we refer to as the local SMC maintains the AEJ flow, while Zhang et al. [2008] suggested that this circulation can bring dry air into the lower midtroposphere and inhibit deep convection. Our results agree with both studies, as the increase in the strength of the SMC is associated with an increase in easterly flow at 10°N and also a decrease in precipitation in the rain band. This result is distinct from previous studies which have shown the strength of the SHL to be positively correlated with Sahel precipitation strength [Biasutti et al., 2009; Pomposi et al., 2015]. Here the strength of the low is not explicitly examined, rather it is the strength of the outflow in the midtroposphere into the Sahel to the south that is found to consistently change along with Sahel precipitation. Our results show that enhanced descent over the Sahara intensifies the northerly flow in the SMC as it meets the shallow ascent caused by the thermal low over the Sahara. Therefore, it is not only the change in large-scale ascent over the rain band but also variations in the dry air intrusions which can alter convection and precipitation over the Sahel.

Chung and Ramanathan [2006] and Hagos and Cook [2008] have suggested that a change in the spatial distribution of the Indian Ocean warming could have contributed to the recovery of Sahel precipitation in the 1990s. As mentioned above, Chung and Ramanathan [2006] argued that changes to Indian Ocean SSTs weakened the Asian monsoon and the associated monsoon overturning circulation. Although the climatological experiments conducted here cannot be used to explain the drought and rainfall recovery in the Sahel, the results of Chung and Ramanathan [2006] and Hagos and Cook [2008] are consistent with a shift from a Group 2 warming pattern to a Group 1 pattern. Using a regional model, Hagos and Cook [2008] argued that Indian Ocean SST changes resulted in a shift in midtropospheric divergence from the Sahel to the Atlantic Ocean, leading to a recovery of Sahel precipitation. While we do not explicitly examine the midtropospheric divergence fields here, we do see that Indian Ocean SST changes can alter ascent and descent regions over the Sahel (Figures 12 and 13), causing an amplification or shift in the midtropospheric divergence. We have also found that certain warming experiments, generally those which have a warming concentrated on the equator or south of the equator, can cause an increase in divergence in the midtroposphere over the Atlantic (not shown). This highlights the importance of the specific pattern of SST perturbation, and why isolating the teleconnection between the Sahel and specific changes in the Indian Ocean will be important to understand future projections of precipitation in the Sahel. The use of a global model like CESM enables us to capture the influence of the SSTs on the large-scale Walker and Asian monsoon circulations and how changes in those large-scale circulations impact the local circulation over the Sahel.

The spatial pattern, extent, and magnitude of the SST perturbations all have an effect on the resultant Sahel precipitation response. Changing basinwide warming will have a different effect compared to changing heating gradients and equatorial warming. Some companion studies (not shown) were done to test the effect of heating and cooling in various other locations in the basin including the southern Indian Ocean. Changing SSTs in the southern Indian Ocean changes the gradient in SSTs throughout the basin, and these perturbations resulted in a precipitation response that was classified as Group 1 even though the extent and magnitude of these changes were larger than those shown here in the Group 1 experiments. This highlights that the fact that the pattern and location of SST changes can influence the degree to which a monsoon type or a Walker overturning type response is triggered. In Held et al. [2005], models which correctly simulated the twentieth century Sahel drought were perturbed by a 2 K increase in global ocean SSTs. Two out of the three models predicted a decrease in Sahel precipitation, while the third predicted increased precipitation in Africa. Held et al. [2005] pointed out that it is not only gradients but also uniform warming, which may drive a Sahel precipitation response. Different climate models have different land surface feedbacks and also have different cloud responses, which Held et al. [2005] argued has a substantial effect on how sensitive Sahel precipitation is to ocean warming in different models. We would extend this argument to add that differences between models in the sensitivity of the Asian monsoon to SST changes, for example, could in part determine whether Sahel precipitation increases or decreases given the same future change in Indian Ocean SSTs.

Given how difficult it has been to attribute Sahel rainfall changes during the twentieth century, making robust projections of future rainfall will be challenging. The Indian Ocean has been warming since the 1950s, particularly the central Indian Ocean [Bader and Latif, 2003]. Vecchi and Soden [2007] and Zheng et al. [2013] have suggested that future SST changes in the Indian Ocean may be more closely related to the Indian Ocean dipole. But Li et al. [2016] have suggested that the pattern of warming in the Indian Ocean is uncertain due to biases in the CMIP5 models. Furthermore, changes in the strength of the Asian monsoon have also been projected by climate models [Fan et al., 2012]. Our results indicate that to achieve a realistic projection of Sahel precipitation, the scale, and spatial patterns of Indian Ocean SST changes must be well accounted for. This emphasizes the need to better understand the large-scale and local circulation response to changes in the pattern of tropical ocean warming. Here we have isolated the Indian Ocean, but changes in the Atlantic and Pacific Oceans will also impact Sahel precipitation through changes in the both the Walker and Asian monsoon circulations. Additional studies using transient SST experiments across all three ocean basins are needed to investigate the mechanisms identified here, which may be altered by long-term changes in tropical SSTs.

Acknowledgments

This work was supported by the Natural Science and Engineering Research Council (NSERC) of Canada. We thank Paul Kushner, Kimberly Strong, and Alessandra Giannini for helpful comments. The CESM project is supported by the National Science Foundation and the Office of Science (BER) of the U.S. Department of Energy. Computations were performed on the TCS supercomputer at the SciNet HPC Consortium. SciNet is funded by the Canada Foundation for Innovation under the auspices of Compute Canada, the Government of Ontario, Ontario Research Fund–Research Excellence, and the University of Toronto. ERA-Interim reanalysis data from ECMWF is available at http://apps.ecmwf.int/datasets/. The CRU TS 3.0 rainfall data set is available for download at https://crudata.uea.ac.uk/cru/data/hrg/.