1. Introduction
Africa is highly vulnerable to climate change, evident from the reliance on seasonal precipitation for agriculture, water supply, and energy generation over the majority of sub-Saharan Africa (Basher and Briceño 2006; Meadows 2006). The region exhibits relatively low adaptive capacity (Kusangaya et al. 2014), as the El Niño drought event of 2015/16 amply demonstrates (Archer et al. 2017; Baudoin et al. 2017). Therefore, future changes in rainfall over this region need to be identified and understood, such that stakeholders, from civil society to policymakers, can make informed decisions about future adaptation planning in key sectors (Collins et al. 2012; Knutti et al. 2010).
We focus here on southern Africa, where the rainfall climate (see Figs. S1 and S2 in the online supplemental material) is broadly characterized by a local summer wet season in austral summer, a pronounced rainfall gradient from the dry southwest to the humid north and northeast, and a peak wet season rainfall maximum extending from the continent into the adjacent southwest Indian Ocean, referred to as the south Indian Ocean convergence zone (SIOCZ; e.g., Lazenby et al. 2016), a preferred locus for tropical–temperate weather systems. For this region the multimodel mean projections from the CMIP3 and CMIP5 model ensembles show a degree of consistency in the dominant signals of change with drier austral summer conditions projected over the southwest of the region as well as a northeasterly shift in the SIOCZ and a later onset of the rainy seasons with indications of early cessation of rainfall (Niang et al. 2014; Shongwe et al. 2009).
For most of the tropics, at least at regional scales relevant to decision-making, considerable uncertainty is evident across the ensemble of global and regional models currently available in both sign and magnitude of future precipitation projections (Rowell 2012; Knutti and Sedláček 2013; McSweeney and Jones 2013). Central and southern Africa are no exception (see section 3) and there is a clear need for improved understanding due in part to a complex climatological setting in which regional climate drivers as well as remote influences affect the region (Christensen et al. 2013; IPCC 2007; Kusangaya et al. 2014). This presents challenges to climate adaptation policy, and the persistence of uncertainty in climate projections has led to the development of “decision-making under climate uncertainty” approaches in adaptation (e.g., Hallegatte et al. 2012). These approaches appreciate the cascade of uncertainty typically associated with climate risk assessment, resulting from climate projections, downscaling techniques, impact modeling, and so on, and emphasize appropriate methods for managing risk under such conditions including, for example, robust or low regrets options or adaptive management (e.g., Dessai et al. 2009).
There is therefore considerable interest in improving our understanding of physical mechanisms driving the particular patterns of projected model changes, so that we may determine and potentially improve the robustness and credibility of projections. The physical processes driving projected change are numerous and may vary among models. Projection uncertainty is a result of varying processes found within models including, among others, parameterization schemes, climate sensitivity, and regional patterns of SST changes.
Based on analyses of projections, various mechanisms have been proposed that link increased global temperatures and precipitation changes, notably the thermodynamic “wet get wetter” process of rainfall change (Held and Soden 2000, 2006; Allen et al. 2010; Christensen et al. 2013; Chou and Neelin 2004; Meehl et al. 2007; Chou et al. 2009; Seager et al. 2010). This operates at the largest scales (e.g., zonal means) and involves the increase in global specific humidity in a warmer atmosphere leading to an increase (decrease) in precipitation in the regions of mean moisture convergence (divergence). In the tropics this is likely to be offset by the weakening of the mean tropical overturning circulation associated with a reduction in convective mass flux in regions of present-day high ascent (Vecchi et al. 2006; Chadwick et al. 2013; Christensen et al. 2013; Ma and Xie 2013; DiNezio et al. 2013).
Chadwick et al. (2013, hereafter C13), however, proposed that the wet-get-wetter mechanism alone does not adequately explain the global pattern of multimodel mean (MMM) projected rainfall change. The spatial correlation of future precipitation change ΔP and mean precipitation P globally is low, such that at regional and seasonal scales in the tropics other processes dominate. These processes are substantially related to changes in the spatial location of moisture convergence and hence convection. These include dynamic effects of regional gradients in near-surface temperature change over oceans, that is, the “warmer get wetter” process (Xie et al. 2010), land–sea temperature contrasts (Dong et al. 2009; Byrne and O’Gorman 2013), land surface processes (Pitman 2003), aerosol direct, indirect, and semidirect effects (Huang et al. 2007; Lohmann and Feichter 2005; Ackerman et al. 2000; Hansen et al. 1997), and changes in circulation (Shepherd 2014). The thermodynamic balance of the “upped ante” (Neelin et al. 2003) mechanisms additionally contributes, leading to results such as the so-called modified warmer-get-wetter mechanism (Huang et al. 2013). The modified warmer-get-wetter mechanism is described as the combination of SST changes (the warmer-get-wetter effect) modified by background climatological moisture and SSTs due to the nonlinear relationship between tropical convection and SSTs.
Generally, the processes operating over the ocean are better understood than those over land but in all cases differing representation of these processes across models is likely to drive projection uncertainty. Most previous analyses have focused on the MMM projected change quantities; however, Rowell et al. (2015) highlighted the importance of understanding the mechanisms of change within individual models and concluded that further investigation should be aimed at developing expert judgment of process-based mechanisms and their reliability of projections (Rowell et al. 2015).
In this context, the aims of this paper are 1) to determine the mechanisms of projected regional precipitation changes by decomposition into thermodynamic and dynamic components; 2) to quantify the contribution to total ensemble projection uncertainty, as represented by intermodel spread, associated with these mechanisms; and 3) to identify possible causes of uncertainty, and to draw inferences regarding the robustness and credibility of projected changes. The mechanism of change decomposition method of C13 is used to identify causes of precipitation change (aim 1) and associated uncertainty (aim 2) (reported in section 3a) and causes of uncertainty are inferred (aim 3) through analysis of the intermodel spread (reported in section 3b). This paper focuses on projected regional precipitation changes over southern Africa (SA) and the adjacent southwest Indian Ocean (SWIO) sector (0°–30°S, 10°E–80°E), where wet season rainfall is dominated by the SIOCZ (Cook 2000; Lazenby et al. 2016).
2. Data and methods
a. Data
We use output from simulations of the twentieth century and the twenty-first century [under the representative concentration pathway 8.5 (RCP8.5) emissions scenario] from 20 models [those used in Kent et al. (2015, Table 1 therein)] from the World Climate Research Programme (WCRP) CMIP5 multimodel dataset, which provide results for the most recent Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC) (Meehl et al. 2007; Taylor et al. 2012) (see Table 1). We consider both the MMM of the 20 chosen CMIP5 models and the spread of model projections across the 20-model ensemble, to address aims 1 and 2 and aim 3, respectively. Lazenby et al. (2016) present the biases in the MMM of the larger full CMIP5 ensemble and we note that this is not significantly different from that of the MMM of the 20 CMIP5 models used here (see Figs. S1 and S2). Both MMMs represent well the dominant circulation features and the spatial structure of rainfall and its annual cycle. However, there is an overall positive precipitation bias compared to observations over the main climatological rainfall features over the Indian Ocean and central continent in austral summer, although not in the summer rainfall season over the Sahel zone (~15°N).
CMIP5 model list of the 20 models used including modeling center, institute ID, and atmospheric resolution.
Monthly data were extracted for key diagnostic variables to understand potential physical processes linked to precipitation changes over the SA–SWIO sector. The period of analysis for projected climate changes is 2071–2100 (in the RCP8.5 experiment) minus the historical period 1971–2000. Only the first ensemble member was utilized in creating the MMM. All model data were interpolated to a common grid of 1.5° × 1.5° to ensure uniformity.
b. The “mechanism of change” decomposition methodology
To determine the main contributors to both the MMM precipitation change (see section 3a) and associated uncertainty (i.e., intermodel spread; section 3b) over the study domain, the C13 decomposition of change methodology was applied to data from each model for each calendar month. We then focus specifically on the early and main SA–SWIO wet seasons October–December (OND) and December–February (DJF), respectively.
In this formulation ∆q and ∆M* represent the thermodynamic and dynamic components of change, respectively. The C13 method has the advantage that the dynamical term ∆M* can be further separated into 
Note that 
∆qcc, the Clausius–Clapeyron change in surface q for the change in mean 2-m temperature at each location (expected under fixed relative humidity), expressed as ∆qcc = (es2/es1)q − q, where es is the saturation vapor pressure and q is historical boundary layer specific humidity [the ratio expresses future (es2) by historical (es1) saturation vapor pressure], and
∆qrh, the residual of ∆q − ∆qcc, associated with changes in near-surface relative humidity.
ΔP is the change in precipitation expressed per degree global warming,
ΔPT − M*∆qcc is the thermodynamic change due to Clausius–Clapeyron-driven increases in specific humidity,
ΔPRH = M*∆qrh is the change due to near surface relative humidity changes,
ΔPWeak = q
ΔPShift = q
ΔPCross = ΔqΔM* is the cross component of precipitation change, associated with interactions between the other components.
c. Analysis of uncertainty
To quantify model uncertainty in projected change we assess the spread in the precipitation change components over the 20-model member ensemble (section 3b). The dominant spatial pattern of variation in ∆PShift between models is identified through EOF analysis, applied to the cross-model ∆PShift fields (standardized) for two separate domains within our study region: the southern African continental land region (10°–30°S, 10°–40°E) and the Indian Ocean (0°–30°S, 40°–80°E) region, (see section 3c). Potential drivers of this uncertainty are then assessed through correlations of the ∆PShift EOF component scores with diagnostic fields (SST and circulation indices) across the model ensemble and through composite analysis of diagnostic fields from models, sampling the upper and lower 25% of models from the EOF scores (see section 3c).
Note that all components of projected change in this analysis are normalized by the mean global surface temperature change ΔTglobal of each individual model, and therefore all quantities are expressed as per degree of global warming. This removes the uncertainty due to intermodel spread in climate sensitivity and makes results scalable to the magnitude of warming (assuming quasi-linearity of regional precipitation change with warming) and therefore tractable when applied to predefined warming levels for policymakers. Removing the effect on ∆P of uncertainty in model ∆Tglobal related to model climate sensitivity (i.e., the high correlation between normalized and absolute precipitation changes; see Fig. 1) indicates a minor influence on our results.
Relation of absolute projected precipitation ∆Pabs change vs normalized precipitation change ΔP for the 20 CMIP5 models for the period 2071–2100 minus 1971–2000 under the RCP8.5 scenario for the seasons (left) OND and (right) DJF. Quantities are area averaged over the study domain (0°–30°S, 10°–80°E).
Citation: Journal of Climate 31, 12; 10.1175/JCLI-D-17-0311.1
3. Results and discussion
a. Changes in multimodel mean precipitation over southern Africa and the adjacent Indian Ocean
First, we consider the annual cycle of zonally averaged MMM precipitation changes over our study domain for land and ocean regions (Figs. 2a and 2b, respectively). The most pronounced feature of projected changes in rainfall over both land and ocean is an opposing dipole of future wetter and drier conditions, oriented to the north and south, respectively, of the climatological axis of maximum rainfall. However, the drying–wetting dipole is not obvious in January–February (Fig. 2a), particularly over land. This ∆P wetter–drier dipole structure effectively straddles the ITCZ with wetting (feature A in Figs. 2a,b) and drying (feature B in Figs. 2a,b) located to the north and south, respectively, of the ITCZ axis. This indicates an effective northward shift in the ITCZ. This dipole pattern, however, has a strong seasonal cycle peaking in austral spring (September–December) over both land and ocean. The OND season is often overlooked in studies of SA–SWIO climate in favor of the main rainy season, DJF, but here we note a remarkably strong rainfall change signal in OND.
Time–latitude plots of projected MMM changes in precipitation, and related mechanisms of change components, derived from the 20 CMIP5 models for the future period 2071–2100 minus 1971–2000 under the RCP8.5 scenario. Values are averaged over longitude bands indicative of (left) the SA continent (10°–40°E; land only) and (right) the Indian Ocean (40°–80°E; ocean only). (a),(b) ∆P (shaded) overlaid with historical climatological P (1971–2000). (c),(d) As in (a),(b), but for ∆PShift (shaded). (e),(f) As in (a),(b), but for ∆Ptwrh (shaded). (g),(h) ∆T overlaid with historical climatological T. (Units of ΔP and associated components are in mm day−1 K−1 global warming and P in mm day−1. Units of ΔT and T are in kelvin. See text for explanation of features marked with A–D.)
Citation: Journal of Climate 31, 12; 10.1175/JCLI-D-17-0311.1
The spatial patterns of MMM precipitation change over our study region for the key seasons OND and DJF (Figs. 3 and 4 ) clearly illustrate this dominant zonally oriented pattern of wetter (drier) conditions located to the north (south) of the mean ITCZ at ~10°S, apparently connecting precipitation changes over the southern African continent with those in the equatorial and southwest Indian Ocean. The continental drying is most pronounced in OND (Fig. 2a, feature B) centered at ~15°S over the southern African continent. This may indicate a later start and reduced length of the growing season rainfall with serious implications on the agricultural sector. In addition to the wetting–drying dipole there is evidence of additional wetting over 25°–30°S over eastern South Africa and across into the southwest Indian Ocean, especially in DJF.
(a) Projected MMM changes in precipitation, and (b)–(g) related mechanisms of change components, derived from the 20 CMIP5 models for the future period 2071–2100 minus 1971–2000 under the RCP8.5 scenario, for the OND season over SA and the SWIO region. The ∆Ptwrh is the sum of ∆PT + ∆PWeak + ∆PRH. Black contours overlaid represent climatological precipitation for the specific season contoured from 0 to 10 mm day−1 in intervals of 2 mm day−1. Units are in absolute change per kelvin global warming. (h) Projected 20 CMIP5 MMM temperature changes for the same time period and region (K).
Citation: Journal of Climate 31, 12; 10.1175/JCLI-D-17-0311.1
As in Fig. 3, but for the DJF season.
Citation: Journal of Climate 31, 12; 10.1175/JCLI-D-17-0311.1
The decomposition can aid in understanding the processes driving these changes. The thermodynamic component ∆PT (Figs. 3d and 4d) results in a wetting signal everywhere but whose magnitude is proportional to a rise in temperature through the Clausius–Clapeyron relation (i.e., it represents the wet-get-wetter process and maps substantially onto mean rainfall). It is substantially offset by an equivalent pattern of drying from the weakening of the tropical overturning circulation ∆PWeak (Figs. 3e and 4e), which is inversely proportional to M* and as such maps onto mean P. In addition, over land wetting from ∆PT is offset by drying from ∆PRH (Figs. 3f and 4f), due to reduced relative humidity, presumably resulting from moisture supply not keeping pace with increasing temperature. Together, these terms constitute the thermodynamic residual term ∆Ptwrh (Figs. 2e, 2f, 3g, and 4g), which drives a net wetting signal, peaking over the oceanic climatological ITCZ and humid land regions. Therefore, ∆P is partly composed of a thermodynamic wet-get-wetter process magnifying the mean ITCZ rainband. However, the thermodynamic ∆PRH response also drives up to ~50% of the drying over subtropical land regions during OND (feature C in Fig. 2e) centered on ~20°S. This is broadly coincident with the maxima in ∆T (Fig. 2g, feature D), suggestive of a land–atmosphere positive feedback response in early wet season, which is overcome during the peak DJF wet season. As such over land in OND ∆Ptwrh is an important driver of the spatial pattern of ∆P (r = 0.70, where r is the spatial correlation), substantially associated with ∆PRH.
However, over the study region as a whole, it is clear from Figs. 2–4 and Table 2 that the spatial pattern of ∆P in OND or DJF is not closely related to that of mean precipitation (P) (spatial correlations are not significant at the 95% confidence interval) or to the thermodynamic residual term ∆Ptwrh (r = 0.46 and 0.30 for OND and DJF, respectively), such that the notable dipole features of ∆P are not driven by the wet-get-wetter process. Rather, ∆P most closely matches the dynamical component ∆PShift (r = 0.94 for OND and 0.96 for DJF) such that it is changes in the location of convection that explain the dipole. Note that ∆PShift contributes most of the wetting–drying dipole over land and almost all of it over ocean, especially the drying signal south of the ITCZ. We postulate here that this projected northward shift of the ITCZ over the Indian Ocean may be related to projected changes in the SST structure through a “warmest get wetter” mechanism. The north–south gradient in SST over the Indian Ocean, which is broadly representative of the ∆P dipole [~(5°N–20°S)] in the present day of ~(2–4) K over the austral summer–winter seasons (Fig. 2h), is enhanced in the MMM projections by ~0.5 K (Figs. 2h, 3h, and 4h). This may lead to a northward displacement of convection toward the warmest oceanic waters, similar to that noted for the tropical Pacific (Widlansky et al. 2013).
Spatial correlations of MMM ∆P vs the different mechanisms of change components for both DJF and OND for the SA–SWIO region (0°–30°S, 10°–80°E). (Using the Student’s t test correlation values above 0.44 are deemed significant at the 95% confidence interval, shown in boldface.)
b. Quantifying uncertainty in projected change in precipitation: Ensemble spread
Uncertainty across the multimodel ensemble remains a feature in projections of future precipitation across the tropics and is a major barrier to effective use of climate information in adaptation activities, notwithstanding approaches to “decision making under climate uncertainty” (e.g., Lempert and Collins 2007). Here, we assess the contributions to total ensemble uncertainty from the various mechanisms of change in our decomposition. We assess the robustness of projected changes from the magnitude of intermodel spread for each component as represented in two forms: first as maps of the standard deviation of the ensemble at each grid cell (Figs. 5 and 6) and second as box-and-whisker plots of intermodel spread for each component averaged over specific regions of interest (Fig. 7). We average over areas where the MMM change signal shows mean future wetting or drying to help inform interpretation of the robustness of key signals emerging from MMM commonly used (e.g., Flato et al. 2013; IPCC 2013). Further, we consider land and ocean separately given the differing level of importance for adaptation actions and differing mechanisms driving changes. A number of broad signals emerge.
Changes in precipitation averaged over the regions of wetting–drying from the MMM are more robust over land than ocean for both the future wetting and drying signals (i.e., total uncertainty in ∆P is lower for land compared to ocean; Fig. 7). This difference in the spread of ∆P is largely related to that in ∆PShift (r = 0.95 for OND and 0.97 for DJF for land regions), such that uncertainty in the dynamical drivers of projected change is dominant. We consider the contribution of intermodel differences in SST structures to uncertainty in ∆PShift in section 3c.
The MMM projected drying is more robust than wetting over both land and ocean and in OND, but differences are small in DJF. The future signal with the greatest robustness is the continental drying over (especially western) SA during the OND season (Figs. 5 and 7), reinforcing the importance of this early wet season component of the larger-scale rainfall change dipole. During DJF (Figs. 6 and 7) the MMM drying signal is less robust, and there are “hotspots” of nonrobust change, which include parts of Malawi, Tanzania, Madagascar, and northern Mozambique, with potentially important implications for approaches to climate change adaptation. Areas of robust wetting include East Africa and the western Indian Ocean at the equator.
The local hotspots in intermodel ∆P standard deviation show that uncertainty in ∆P does not simply scale with the absolute magnitude of ∆P. These are located, for example, proximate to the African great lakes, regions of complex topography (e.g., Madagascar), and at the transition boundaries of wetting and drying over the Indian Ocean (Figs. 5a and 6a), which raises the possibility that intermodel differences in background climatology may project onto the uncertainty in ∆P. However, only some of these hotspots of uncertainty correspond to locations of high intermodel spread in mean P (Figs. 5h and 6h), such that most of the spatial pattern in uncertainty in ∆P is unrelated to background climatology. The high uncertainty in ∆P often apparent in wetting–drying transition zones suggest that caution needs to be attached to interpretation of near-zero MMM ∆P change.
Most of the total uncertainty in ∆P is contained in the ∆PShift component, whose uncertainty is typically greater by a factor of 2–4 than that of the thermodynamic residual. Intermodel standard deviation within the ∆PShift term is markedly higher over ocean than land, notably over the ∆P dipole region. While the MMM wetting–drying dipole over the Indian Ocean may result from a warmest-get-wetter response to changes in the SST structure (Figs. 5h and 6h), high uncertainty in ∆PShift suggests strong intermodel divergence in the form of the ∆SST patterns responsible (Chadwick 2016), which we explore in section 3c. Indeed, in many regions, notably the region of projected drying over the Indian Ocean at 10°–15°S, the model uncertainty in ∆PShift is higher than that in ∆P. Note that ∆Ptwrh is especially robust over the ocean ITCZ where the thermodynamic response is closest to that of ∆PT.
Since the total uncertainty can be less than the sum of the individual components (e.g., Fig. 7) we can infer that the intermodel components are anticorrelated across models and offset each other, which acts to constrain total uncertainty in ∆P. For the drying over continental SA (during OND) and to a lesser extent over the Indian Ocean, uncertainty in the dynamical ∆PShift and thermodynamic residual term ∆Ptwrh terms appear to offset each other, resulting in a total intermodel standard deviation of ∆P, which is much less than the sum of the component terms.
The thermodynamic change due to Clausius–Clapeyron-driven increases in specific humidity (∆PT) exhibits the largest magnitude of all the decomposed components (Fig. 7); however, this is largely offset by the weakening of the tropical circulation component (∆PWeak) and relative humidity component (∆PRH) over land regions.
Intermodel standard deviation (shaded) in (a) ∆P (mm day−1 K−1) and (b)–(g) mechanism of change components therein (mm day−1 K−1), (h) ∆T (K), and (i) P (mm day−1), for the OND season. Contours show MMM ∆P (values from −0.6 to 0.6 mm day−1 K−1 in intervals of 0.2, where dashed contours show negative values). Change quantities are for the period 2071–2100 minus 1971–2000 under the RCP8.5 scenario.
Citation: Journal of Climate 31, 12; 10.1175/JCLI-D-17-0311.1
Box-and-whisker plots of the future change in ∆P and various components therein from the 20 CMIP5 models, for the period 2071–2100 minus 1971–2000 for RCP8.5 of CMIP5 models for (top)–(bottom) drying and wetting (land only) and drying and wetting (ocean only) for (left) DJF and (right) OND. (Units are in mm day−1 K−1 global warming.)
Citation: Journal of Climate 31, 12; 10.1175/JCLI-D-17-0311.1
c. Understanding potential causes of uncertainty in the dynamic component of projected change
The dynamical ∆PShift term provides the primary contribution to intermodel uncertainty in future projections of precipitation. Among the potential causes of this uncertainty in the location of tropical convection, a primary candidate is the varying patterns of model projected SST changes. We explore this using EOF analysis of the intermodel ∆PShift patterns. Over the Indian Ocean domain during both OND and DJF the leading mode of intermodel ∆PShift variability shows a loading pattern (Figs. 8a,b) that projects strongly onto the pattern of the future wetting–drying dipole in ∆PShift (Figs. 3b and 4b), and indeed in ∆P explaining 37% and 31.5% of variability, respectively. These EOFs therefore represent well the strength of the MMM ∆P wetting–drying dipole, oriented broadly north–south, in individual models.
(a),(b) Leading EOF loading patterns of intermodel ∆PShift for 20 CMIP5 models over the Indian Ocean domain for OND and DJF, respectively; and (c),(d) the respective correlation coefficients of the EOF component scores vs intermodel ∆SST. Significant correlations at the 90% percentile are stippled.
Citation: Journal of Climate 31, 12; 10.1175/JCLI-D-17-0311.1
The component scores of this leading EOF correlate moderately with intermodel ∆SST over the Indian Ocean in both OND and DJF (Figs. 8c,d). In both cases the correlation is such that a stronger north–south wetting–drying dipole in individual models is associated with lower rates of SST warming in the Indian Ocean south of the equator, and hence an increased north–south SST gradient across the Indian Ocean. This response across models is consistent with the warmest-get-wetter mechanism inferred from the MMM changes in precipitation and SST (section 3a). Moderate correlations also exist with ∆SST in the tropical Pacific broadly around the Niño-3.4 region. During OND there is an additional indication that the ∆PShift EOF dipole is related to changes in the mean east–west SST gradient reminiscent of the Indian Ocean dipole (IOD) mode of interannual variability active in this season (Saji et al. 1999). Previous analysis has related changes in long-term mean precipitation to a shift toward a preference for the positive mode of the IOD in some coupled models (Shongwe et al. 2011), which remains consistent with the warmest-get-wetter mechanism.
Over the domain centered on the SA landmass the intermodel ∆PShift EOF analysis for both the OND and DJF seasons reveals a dominant EOF loading pattern oriented diagonally northwest–southeast across SA toward the southwest Indian Ocean (Figs. 9a,b). The peak loadings lie to the southwest of the position of the mean SIOCZ feature such that the EOF represents the magnitude of the MMM drying signal to the south of the mean rainfall maximum. The MMM projected pattern of ∆PShift change across the region is more zonally oriented than the EOF SIOCZ-like orientation (Figs. 3b and 4b; see section 3a) such that the MMM change clearly masks considerable spread within the ensemble.
As in Fig. 8, but for the southern Africa land domain whereby ∆SST correlations are restricted only to the SWIO domain.
Citation: Journal of Climate 31, 12; 10.1175/JCLI-D-17-0311.1
The EOF pattern of intermodel ∆PShift appears associated with intermodel variability in SST, land heating and circulation and resulting moisture fluxes based on both correlations (Figs. 9c,d) and composites of diagnostic fields based on samples of the upper and lower 25% of models from the EOF component scores (Fig. 10). The EOF is rather weakly correlated to the intermodel structures of projected changes to SST across the southern Indian Ocean with an east–west dipole structure of correlation with the EOF component scores along ~(20°–25)°S (Figs. 9c,d) and corroborated in the composites (Fig. 10). For both the OND and DJF EOF, stronger future drying over SA in models is associated with weaker future warming in the Mozambique Channel–southwest Indian Ocean and stronger warming in the eastern subtropical Indian Ocean, as well as the eastern South Atlantic.
Composite mean fields of diagnostic variables based on samples of the uppermost 25% minus the lowermost 25% of models selected from the component scores of the leading EOF of the intermodel ∆PShift (Figs. 9a,b) for (a) OND and (b) DJF. Fields shown are changes in surface temperature (K; shaded), surface pressure (hPa; contours, interval 0.2), and 850-hPa OND winds (m s−1; vectors).
Citation: Journal of Climate 31, 12; 10.1175/JCLI-D-17-0311.1
While the spatial patterns are similar, the correlations of EOF and SST are notably stronger in DJF than OND (Figs. 8c,d and 9c,d). The implication is that SSTs are most significant in austral summer than the transition season over the southern African region, particularly over the South Atlantic Ocean and central Indian Ocean. OND rainfall does not exhibit sufficient evidence of a significant link with changing SST gradients. Reason et al. (2006) state that the onset of the wet season over southern Africa appears to be associated with anomalous ridging of the South Atlantic high pressure, which requires further investigation and may shed more light on this research.
Circulation features associated with the intermodel spread are identified by compositing projected changes in diagnostic fields in the top and bottom 25% of models from the EOF coefficients (Fig. 10). We find that in both seasons more intense versus weaker drying over SA in models is associated with cyclonic low-level circulation over the southwest Indian Ocean (Fig. 10) around a relatively weaker subtropical high, resulting in weaker Indian Ocean easterlies, and therefore an effective northeastward shift in the axis of the SIOCZ (Lazenby et al. 2016). In DJF (Fig. 10b) intense drying is also favored in models in which strong land heating occurs at ~20°S. This may be a result of reduced soil moisture but the heating drives an apparent intensification of the continental low-pressure center (the “Angolan low” feature) and cyclonic low-level circulation. While this might be expected to favor rainfall over the continent, the composite of the drier models (not shown) indicates a marked anomalous southward shift of the low with implications for moisture transport. There is some indication of an intensified South Atlantic anticyclonic circulation in the models with more intense drying whose link with the SST changes requires further investigation.
These structures of intermodel differences in ∆SST, land temperature, and low-level circulation that may drive the leading pattern of intermodel spread in ∆PShift over SA have some resonances with modes and structures of interannual variability in the region. Notable among these are the following. (i) The first is the regional response to ENSO characterized by the displacement in the SIOCZ associated with cyclonic low-level anomalies over the southwest Indian Ocean during El Niño events (Cook 2001), similar to that in Fig. 10. However, it is cautionary to note that the physical mechanisms and influence of local Indian Ocean SST (e.g., Goddard and Graham 1999) versus remote Pacific SSTs (e.g., Cook 2001; Ratnam et al. 2014) in driving this in the observed climate remain to be fully resolved. (ii) The second feature is the south Indian Ocean dipole (SIOD) mode of east–west SST gradient in the subtropical Indian Ocean (Behera and Yamagata 2001). In this case, although the form of the association of ∆PShift over SA and ∆SST matches that of interannual variability in SA rainfall and the phase of the SIOD (Reason 2001), the circulation responses diverge (Fig. 10). (iii) Finally, there is the intensity of the Angolan low and associated circulation changes, known to be an important control on present day precipitation in observations (e.g., Manhique et al. 2011) and in models (Munday and Washington 2017). During wet (dry) years the strength of the Angola low tends to be stronger (weaker) with enhanced (suppressed) circulation bringing in moisture from the Atlantic and the Indian Ocean (Cook et al. 2004; Munday and Washington 2017). Differential rates of future warming over land versus ocean may be expected to intensify and displace the Angolan low and we see evidence of differential response across models strongly related to intermodel uncertainty in the structure of future dynamical precipitation responses.
Of course, we would not expect the dynamical processes of intermodel spread in projected rainfall change to be consistent with all aspects of current interannual variability, given (i) the complex suite of interannual modes affecting the SA region and (ii) the highly variable and mixed ability of coupled models to represent the mean state and variability, both through local processes (Lazenby et al. 2016) and remote teleconnections (e.g., Rowell 2013). Nevertheless the link between processes of climate variability and projected change has proved to be a fruitful focus in recent years in explaining why models make the future changes they do, providing some basis for assessing potential credibility of projected change.
4. Summary and conclusions
Human influence is expected to drive considerable changes to the hydrological cycle, of particular concern in regions such as southern Africa that are currently vulnerable to climate. There is a need to quantify and understand the projections of future precipitation change and associated uncertainty to inform the possible use of climate information in adaptation planning (Hackenbruch et al. 2017). Our analysis of end-of-twenty-first-century projections from a sample of 20 models from the CMIP ensemble reveals a dominant dipole pattern of precipitation change over SA and the SWIO with a wetting (drying) response to the north (south) of the ITCZ, with therefore an effective northward shift in the ITCZ. Pronounced drying is exhibited over land in OND, implying a delay in the onset of the wet season, an important and relatively robust signal.
A decomposition framework is applied to separate the mechanisms of change in both the MMM and in individual models to shed light on the relative importance of the multiple, coincident, and often competing mechanisms that may drive future precipitation changes in a warmer world. We show that the dynamical component ∆PShift representing spatial shifts in convection explains most of the dipole structure of ∆P, as evidenced by high spatial correlations no less than 0.94 over the study domain. The thermodynamic component of increased moisture is offset by the weakening of the tropical circulation leading to a relatively weak wet-get-wetter contribution. Considering the robustness of projected change, represented by the intermodel dispersal in response, we find that ∆PShift holds the most uncertainty, whereas the thermodynamic component is typically more robust. Drying is more robust than wetting over the continent, notably in the early summer OND season.
For both the OND and DJF seasons MMM future precipitation projections show, at the broadest scale, a wetting–drying dipole across the continent and Indian Ocean. Over land the magnitude, extent, and robustness of drying is greater during in OND than DJF, with implication of a later onset in the wet season over southern Africa. Greater uncertainty during DJF is apparently associated with a stronger role of intermodel differences in the structure SST changes in the adjacent oceans.
Over the Indian Ocean sector, we relate the dominant dynamically driven ∆PShift component to changes in SST structures. The wetting–drying dipole is associated with patterns of SST change, both in the MMM and across the model ensemble, that is consistent with a warmest-get-wetter mechanism driving the northward shift in the location of convection and convergence. Much of the model uncertainty is therefore associated with intermodel differences in future SST patterns in the tropical Indian and Pacific Ocean and the atmospheric teleconnection response, such that better understanding of these projected changes and in the teleconnections linking these to rainfall may provide some basis for constraining uncertainty in projections. This inference, however, is particularly applicable to DJF as OND does not exhibit the same magnitude of significance over the region.
Changes in rainfall over land are likely more complex, first because over the continental interior of SA the contribution to mean drying of the thermodynamic reduction in relative humidity is roughly equal to that of the dynamical component. Reduced relatively humidity is likely itself to be related to the effects of enhanced land–sea temperature contrasts and physical mechanisms for suppression of convection through raised lifting condensation levels have been proposed (e.g., Fasullo 2010). Second, regarding the dynamical component of change, over the SA landmass we find intermodel uncertainty to be associated with low-level circulation patterns associated with zonal gradients in SST changes in the subtropical Indian Ocean and the intensity of continental heating and the thermal low pressure center. The land–sea heating contrast has been hypothesized by Bayr and Dommenget (2013) to be a driver of enhanced convergence over land with implications for convection at the continental scale, but our results indicate that the response can be complex in this case leading to a northward shift of the ITCZ/SIOCZ and wetting–drying at the regional scale. Over both the ocean and land examination of intermodel dispersal yields structures with less zonal uniformity of change than represented by the MMM response. The sensitivity of ∆PShift across models to the pattern of SST structures and their resemblance to contemporary modes of variability suggests that further analysis of future SST patterns and assessment of model representation of present rainfall–SST teleconnections and mean state biases may be useful to inform our interpretation of the credibility of projected changes.
This paper provides new insight of projected precipitation mechanisms of change over southern Africa by identifying the dominant pattern of change over southern Africa (the wetting–drying dipole). Drivers of this projected change are evaluated and attempts are made to quantify the associated uncertainty. Regional circulation analysis is additionally performed to shed light on the patterns driving future dynamic precipitation changes in OND and DJF, which are in agreement with previous studies over other regions (e.g., Kent et al. 2015).
Acknowledgments
This work was supported financially by the Peter Carpenter Scholarship for African Climate Change at the University of Sussex and the Future Climate for Africa (FCFA) regional consortium project UMFULA funded by NERC (Grant NE/M020258/1) and the U.K. government’s Department for International Development (DfID). We also acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling responsible for CMIP5 model data, which was provided by the Program for Climate Model Diagnosis and Intercomparison (PCMDI). More information on this model data can be found at the PCMDI website (https://pcmdi.llnl.gov/).
REFERENCES
Ackerman, A. S., O. B. Toon, D. E. Stevens, A. J. Heymsfield, V. Ramanathan, and E. J. Welton, 2000: Reduction of tropical cloudiness by soot. Science, 288, 1042–1047, https://doi.org/10.1126/science.288.5468.1042.
Allen, C. D., and Coauthors, 2010: A global overview of drought and heat-induced tree mortality reveals emerging climate change risks for forests. For. Ecol. Manage., 259, 660–684, https://doi.org/10.1016/j.foreco.2009.09.001.
Archer, E. R. M., W. A. Landman, M. A. Tadross, J. Malherbe, H. Weepener, P. Maluleke, and F. M. Marumbwa, 2017: Understanding the evolution of the 2014–2016 summer rainfall seasons in southern Africa: Key lessons. Climate Risk Manage., 16, 22–28, https://doi.org/10.1016/j.crm.2017.03.006.
Basher, R., and S. Briceño, 2006: Climate and disaster risk reduction in Africa. Climate Change and Africa, P. S. Low, Ed., Cambridge University Press, 271–286.
Baudoin, M. A., C. Vogel, K. Nortje, and M. Naik, 2017: Living with drought in South Africa: Lessons learnt from the recent El Niño drought period. Int. J. Disaster Risk Reduct., 23, 128–137, https://doi.org/10.1016/j.ijdrr.2017.05.005.
Bayr, T., and D. Dommenget, 2013: The tropospheric land–sea warming contrast as the driver of tropical sea level pressure changes. J. Climate, 26, 1387–1402, https://doi.org/10.1175/JCLI-D-11-00731.1.
Behera, S. K., and T. Yamagata, 2001: Subtropical SST dipole events in the southern Indian Ocean. Geophys. Res. Lett., 28, 327–330, https://doi.org/10.1029/2000GL011451.
Bony, S., G. Bellon, D. Klocke, S. Sherwood, S. Fermepin, and S. Denvil, 2013: Robust direct effect of carbon dioxide on tropical circulation and regional precipitation. Nat. Geosci., 6, 447–451, https://doi.org/10.1038/ngeo1799.
Byrne, M. P., and P. A. O’Gorman, 2013: Link between land–ocean warming contrast and surface relative humidities in simulations with coupled climate models. Geophys. Res. Lett., 40, 5223–5227, https://doi.org/10.1002/grl.50971.
Chadwick, R., 2016: Which aspects of CO2 forcing and SST warming cause most uncertainty in projections of tropical rainfall change over land and ocean? J. Climate, 29, 2493–2509, https://doi.org/10.1175/JCLI-D-15-0777.1.
Chadwick, R., I. Boutle, and G. Martin, 2013: Spatial patterns of precipitation change in CMIP5: Why the rich do not get richer in the tropics. J. Climate, 26, 3803–3822, https://doi.org/10.1175/JCLI-D-12-00543.1.
Chou, C., and J. D. Neelin, 2004: Mechanisms of global warming impacts on regional tropical precipitation. J. Climate, 17, 2688–2701, https://doi.org/10.1175/1520-0442(2004)017<2688:MOGWIO>2.0.CO;2.
Chou, C., J. D. Neelin, C. A. Chen, and J. Y. Tu, 2009: Evaluating the “rich-get-richer” mechanism in tropical precipitation change under global warming. J. Climate, 22, 1982–2005, https://doi.org/10.1175/2008JCLI2471.1.
Christensen, J. H., and Coauthors, 2013: Climate phenomena and their relevance for future regional climate change. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 1217–1308.
Collins, M., R. E. Chandler, P. M. Cox, J. M. Huthnance, J. Rougier, and D. B. Stephenson, 2012: Quantifying future climate change. Nat. Climate Change, 2, 403–409, https://doi.org/10.1038/nclimate1414.
Cook, C., C. J. C. Reason, and B. C. Hewitson, 2004: Wet and dry spells within particularly wet and dry summers in the South African summer rainfall region. Climate Res., 26, 17–31, https://doi.org/10.3354/cr026017.
Cook, K. H., 2000: The south Indian convergence zone and interannual rainfall variability over southern Africa. J. Climate, 13, 3789–3804, https://doi.org/10.1175/1520-0442(2000)013<3789:TSICZA>2.0.CO;2.
Cook, K. H., 2001: A Southern Hemisphere wave response to ENSO with implications for southern Africa precipitation. J. Atmos. Sci., 58, 2146–2162, https://doi.org/10.1175/1520-0469(2001)058<2146:ASHWRT>2.0.CO;2.
Dessai, S., M. Hulme, R. Lempert, and R. Pielke, 2009: Do we need better predictions to adapt to a changing climate? Eos, Trans. Amer. Geophys. Union, 90, 111–112, https://doi.org/10.1029/2009EO130003.
DiNezio, P. N., G. A. Vecchi, and A. C. Clement, 2013: Detectability of changes in the walker circulation in response to global warming. J. Climate, 26, 4038–4048, https://doi.org/10.1175/JCLI-D-12-00531.1.
Dong, B., J. M. Gregory, and R. T. Sutton, 2009: Understanding land–sea warming contrast in response to increasing greenhouse gases. Part I: Transient adjustment. J. Climate, 22, 3079–3097, https://doi.org/10.1175/2009JCLI2652.1.
Emori, S., and S. J. Brown, 2005: Dynamic and thermodynamic changes in mean and extreme precipitation under changed climate. Geophys. Res. Lett., 32, L17706, https://doi.org/10.1029/2005GL023272.
Fasullo, J. T., 2010: Robust land–ocean contrasts in energy and water cycle feedbacks. J. Climate, 23, 4677–4693, https://doi.org/10.1175/2010JCLI3451.1.
Flato, G., and Coauthors, 2013: Evaluation of climate models. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 741–866, https://doi.org/10.1017/CBO9781107415324.020.
Goddard, L., and N. E. Graham, 1999: Importance of the Indian Ocean for simulating rainfall anomalies over eastern and southern Africa. J. Geophys. Res., 104, 19 099–19 116, https://doi.org/10.1029/1999JD900326.
Hackenbruch, J., T. Kunz-Plapp, S. Müller, and J. W. Schipper, 2017: Tailoring climate parameters to information needs for local adaptation to climate change. Climate, 5, 25, https://doi.org/10.3390/cli5020025.
Hallegatte, S., A. Shah, C. Brown, R. Lempert, and S. Gill, 2012: Investment decision making under deep uncertainty—Application to climate change. World Bank Policy Research Working Paper 6193, 39 pp.
Hansen, J., M. Sato, and R. Ruedy, 1997: Radiative forcing and climate response. J. Geophys. Res., 102, 6831–6864, https://doi.org/10.1029/96JD03436.
Held, I. M., and B. J. Soden, 2000: Water vapor feedback and global warming. Ann. Rev. Energy Environ., 25, 441–475, https://doi.org/10.1146/annurev.energy.25.1.441.
Held, I. M. and B. J. Soden, 2006: Robust responses of the hydrological cycle to global warming. J. Climate, 19, 5686–5699, https://doi.org/10.1175/JCLI3990.1.
Huang, P., S.-P. Xie, K. Hu, G. Huang, and R. Huang, 2013: Patterns of the seasonal response of tropical rainfall to global warming. Nat. Geosci., 6, 357–361, https://doi.org/10.1038/ngeo1792.
Huang, Y., W. L. Chameides, and R. E. Dickinson, 2007: Direct and indirect effects of anthropogenic aerosols on regional precipitation over East Asia. J. Geophys. Res., 112, D03212, https://doi.org/10.1029/2006JD007114.
IPCC, 2007: Climate Change 2007: The Physical Science Basis. S. Solomon et al., Eds., Cambridge University Press, 996 pp.
IPCC, 2013: Climate Change 2013: The Physical Science Basis. T. F. Stocker et al., Eds., Cambridge University Press, 1535 pp, https://doi.org/10.1017/CBO9781107415324.
Kent, C., R. Chadwick, and D. P. Rowell, 2015: Understanding uncertainties in future projections of seasonal tropical precipitation. J. Climate, 28, 4390–4413, https://doi.org/10.1175/JCLI-D-14-00613.1.
Knutti, R., and J. Sedláček, 2013: Robustness and uncertainties in the new CMIP5 climate model projections. Nat. Climate Change, 3, 369–373, https://doi.org/10.1038/nclimate1716.
Knutti, R., R. Furrer, C. Tebaldi, J. Cermak, and G. A. Meehl, 2010: Challenges in combining projections from multiple climate models. J. Climate, 23, 2739–2758, https://doi.org/10.1175/2009JCLI3361.1.
Kusangaya, S., M. L. Warburton, E. A. van Garderen, and G. P. Jewitt, 2014: Impacts of climate change on water resources in southern Africa: A review. Phys. Chem. Earth, 67, 47–54, https://doi.org/10.1016/j.pce.2013.09.014.
Lazenby, M. J., M. C. Todd, and Y. Wang, 2016: Climate model simulation of the South Indian Ocean Convergence Zone: Mean state and variability. Climate Res., 68, 59–71, https://doi.org/10.3354/cr01382.
Lempert, R. J., and M. T. Collins, 2007: Managing the risk of uncertain threshold responses: Comparison of robust, optimum, and precautionary approaches. Risk Anal., 27, 1009–1026, https://doi.org/10.1111/j.1539-6924.2007.00940.x.
Lohmann, U., and J. Feichter, 2005: Global indirect aerosol effects: A review. Atmos. Chem. Phys., 5, 715–737, https://doi.org/10.5194/acp-5-715-2005.
Ma, J., and S.-P. Xie, 2013: Regional patterns of sea surface temperature change: A source of uncertainty in future projections of precipitation and atmospheric circulation. J. Climate, 26, 2482–2501, https://doi.org/10.1175/JCLI-D-12-00283.1.
Ma, J., S.-P. Xie, and Y. Kosaka, 2012: Mechanisms for tropical tropospheric circulation change in response to global warming. J. Climate, 25, 2979–2994, https://doi.org/10.1175/JCLI-D-11-00048.1.
Manhique, A. J., C. J. C. Reason, L. Rydberg, and N. Fauchereau, 2011: ENSO and Indian Ocean sea surface temperatures and their relationships with tropical temperate troughs over Mozambique and the southwest Indian Ocean. Int. J. Climatol., 31, 1–13, https://doi.org/10.1002/joc.2050.
McSweeney, C. F., and R. G. Jones, 2013: No consensus on consensus: The challenge of finding a universal approach to measuring and mapping ensemble consistency in GCM projections. Climatic Change, 119, 617–629, https://doi.org/10.1007/s10584-013-0781-9.
Meadows, M. E., 2006: Global change and southern Africa. Geogr. Res., 44, 135–145, https://doi.org/10.1111/j.1745-5871.2006.00375.x.
Meehl, G. A., and Coauthors, 2007: Global climate projections. Climate Change 2007: The Physical Science Basis, S. Solomon et al., Eds., Cambridge University Press, 747–845.
Munday, C., and R. Washington, 2017: Circulation controls on southern African precipitation in coupled models: The role of the Angola low. J. Geophys. Res. Atmos., 122, 861–877, https://doi.org/10.1002/2016JD025736.
Neelin, J. D., C. Chou, and H. Su, 2003: Tropical drought regions in global warming and El Niño teleconnections. Geophys. Res. Lett., 30, 2275, https://doi.org/10.1029/2003GL018625.
Niang, I., O. C. Ruppel, M. A. Adrabo, A. Essel, C. Lennard, J. Padgham, and P. Urquhart, 2014: Africa. Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part B: Regional Aspects. V. R. Barros et al., Eds., Cambridge University Press, 1199–1265.
Pitman, A. J., 2003: The evolution of, and revolution in, land surface schemes designed for climate models. Int. J. Climatol., 23, 479–510, https://doi.org/10.1002/joc.893.
Ratnam, J. V., S. K. Behera, Y. Masumoto, and T. Yamagata, 2014: Remote effects of El Niño and Modoki events on the austral summer precipitation of southern Africa. J. Climate, 27, 3802–3815, https://doi.org/10.1175/JCLI-D-13-00431.1.
Reason, C. J. C., 2001: Subtropical Indian Ocean SST dipole events and southern African rainfall. Geophys. Res. Lett., 28, 2225–2227, https://doi.org/10.1029/2000GL012735.
Reason, C. J. C., W. Landman, and W. Tennant, 2006: Seasonal to decadal prediction of southern African climate and its links with variability of the Atlantic Ocean. Bull. Amer. Meteor. Soc., 87, 941–955, https://doi.org/10.1175/BAMS-87-7-941.
Rowell, D. P., 2012: Sources of uncertainty in future changes in local precipitation. Climate Dyn., 39, 1929–1950, https://doi.org/10.1007/s00382-011-1210-2.
Rowell, D. P., 2013: Simulating SST teleconnections to Africa: What is the state of the art? J. Climate, 26, 5397–5418, https://doi.org/10.1175/JCLI-D-12-00761.1.
Rowell, D. P., B. B. Booth, S. E. Nicholson, and P. Good, 2015: Reconciling past and future rainfall trends over East Africa. J. Climate, 28, 9768–9788, https://doi.org/10.1175/JCLI-D-15-0140.1.
Saji, N. H., B. N. Goswami, P. N. Vinayachandran, and T. Yamagata, 1999: A dipole mode in the tropical Indian Ocean. Nature, 401, 360–363, https://doi.org/10.1038/43854.
Seager, R., N. Naik, and G. A. Vecchi, 2010: Thermodynamic and dynamic mechanisms for large-scale changes in the hydrological cycle in response to global warming. J. Climate, 23, 4651–4668, https://doi.org/10.1175/2010JCLI3655.1.
Shepherd, T. G., 2014: Atmospheric circulation as a source of uncertainty in climate change projections. Nat. Geosci., 7, 703–708, https://doi.org/10.1038/ngeo2253.
Shongwe, M. E., G. J. van Oldenborgh, B. van den Hurk, B. de Boer, C. Coelho, and M. van Aalst, 2009: Projected changes in mean and extreme precipitation in Africa under global warming. Part I: Southern Africa. J. Climate, 22, 3819–3837, https://doi.org/10.1175/2009JCLI2317.1.
Shongwe, M. E., G. J. van Oldenborgh, B. van den Hurk, and M. van Aalst, 2011: Projected changes in mean and extreme precipitation in Africa under global warming. Part II: East Africa. J. Climate, 24, 3718–3733, https://doi.org/10.1175/2010JCLI2883.1.
Soden, B. J., and I. M. Held, 2006: An assessment of climate feedbacks in coupled ocean–atmosphere models. J. Climate, 19, 3354–3360, https://doi.org/10.1175/JCLI3799.1.
Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485–498, https://doi.org/10.1175/BAMS-D-11-00094.1.
Vecchi, G. A., B. J. Soden, A. Wittenberg, I. Held, A. Leetmaa, and M. Harrison, 2006: Weakening of tropical Pacific atmospheric circulation due to anthropogenic forcing. Nature, 441, 73–76, https://doi.org/10.1038/nature04744.
Widlansky, M. J., A. Timmermann, K. Stein, S. McGregor, N. Schneider, M. England, M. Lengaigne, and W. Cai, 2013: Changes in South Pacific rainfall bands in a warming climate. Nat. Climate Change, 3, 417–423, https://doi.org/10.1038/nclimate1726.
Xie, S.-P., C. Deser, G. A. Vecchi, J. Ma, H. Teng, and A. T. Wittenberg, 2010: Global warming pattern formation: Sea surface temperature and rainfall. J. Climate, 23, 966–986, https://doi.org/10.1175/2009JCLI3329.1.