Future southern African summer rainfall variability related to a southwest Indian Ocean dipole in HadCM3

Authors


Abstract

[1] Southern Africa is subject to strong interannual summer rainfall extremes associated with modes of climate variability. Interaction between these modes and the climate change signal make improving understanding of their future operation and links with southern African rainfall a priority. This paper examines future southern African summer rainfall variability related to a meridional dipole mode of southwest Indian Ocean SST variability (the SWIOD) in coupled climate model HadCM3. The SWIOD is central to southern African rainfall in a 1000-year control simulation, and remains so during the twenty-first century under two different CO2 forcing scenarios: SRES A2 and B1. Future rainfall variability connected with this mode is found to be sensitive to the emissions pathway, particularly in the second half of the century. Flooding episodes with anomalies near 4 mm day−1 in already wet regions and which are unprecedented in a 1000 year control run are simulated under A2 SRES forcing.

1. Introduction

[2] It is widely recognized that advances in seasonal climate prediction have the potential to benefit sub-Saharan Africa [Washington and Downing, 1999; Dilley, 2000; Goddard and Dilley, 2005], a region highly vulnerable to climate variability. Prediction on the seasonal timescale has been proposed as a vital component of managing climate change. If the necessary scientific and institutional capacity can be put in place to manage large-scale interannual climate variations, then this timescale could provide a key to longer-term development in the face of climate extremes [Washington et al., 2006]. A linked idea is that climate change will be manifest to some degree through modification of the existing modes of climate variability [Corti et al., 1999; Palmer, 1999; Stone et al., 2001]. Additionally, it is the modes of variability which impose large amplitude climate anomalies on the more linear climate change trend line. Therefore, understanding climate variability now and in the future must be a primary goal. Coupled climate models are the principal tools for projections into the future, and understanding how these models simulate climate variability must be regarded as intrinsic to this goal.

[3] The southern African summer rainfall region has a well-established relationship with the El Niño-Southern Oscillation (ENSO) [e.g., Lindesay, 1988; Reason et al., 2000], and the tropical Pacific has provided a source of predictability of rainfall on the seasonal timescale [Landman and Mason, 1999]. However, there is growing recognition of the crucial role played by sea surface temperature (SST) variability in the Indian Ocean, and in particular the southwest Indian Ocean (SWIO), in forcing rainfall variability over southern Africa [Rocha and Simmonds, 1997; Goddard and Graham, 1999; Reason and Mulenga, 1999; Reason, 2002; Washington and Preston, 2006]. Washington and Preston [2006] report that extremes in observed rainfall during the twentieth century were associated with a meridional pattern of opposing SST anomalies in the SWIO. Forcing the Hadley Centre atmosphere-only GCM HadAM3 with a similar SST pattern to the observed generated a large-scale circulation response favorable for extreme rainfall over the southern African subcontinent.

[4] Here, we show that the principal mode of southern African rainfall-related SST variability in the coupled version of the Hadley Centre model, HadCM3, is almost identical to that reported by Washington and Preston [2006]. Although the pattern currently has little to no skill at seasonal timescales, it remains central to southern African rainfall variability during the twenty-first century under two different climate change scenarios, SRES A2 and B1, and is associated with unprecedented rainfall extremes over the subcontinent in these simulations of forthcoming decades.

2. Modes of Rainfall and Indian Ocean SST Variability in HadCM3

[5] We study HadCM3 coupled climate model data acquired from the World Climate Research Programme's (WCRP's) Coupled Model Intercomparison Project phase 3 (CMIP3) multi-model data set [Meehl et al., 2007], which is archived at the Program for Climate Model Diagnosis and Intercomparsion (PCMDI). We present results from HadCM3 forced with SRES scenarios A2 and B1, where CO2 emissions in A2 are projected to rise steadily through the twenty-first century to 29.1 GtC yr−1 by 2100 and in B1, emissions are projected to rise slowly until the middle of the century before declining to 4.2 GtC yr−1 by 2100, a level lower than today [Nakicenovic et al., 2000]. Our control is a 1000-year integration of HadCM3 with CO2 set at pre-industrial levels, obtained from the Coupled Ocean-Atmosphere Processes and European Climate (COAPEC) project.

[6] First we constructed an index of southern Africa summer rainfall. In common with numerous coupled models, HadCM3 has a wet bias in the western Indian Ocean January to March (JFM) rainfall climatology, but the magnitude of the bias is reduced over land, and the structure of JFM rainfall across sub-Saharan Africa is similar to the observed. A widespread tendency in climate models is to underestimate rainfall extremes, and southern African rainfall variance in HadCM3 is 66% of the observed. The rainfall index was defined using Empirical Orthogonal Function (EOF) analysis of JFM rainfall anomalies land areas south of 5°N for the whole 1000-year period (and tested for sensitivity in time in 100-year samples) in the control. The resultant EOFs were rotated to maximize the signal. The spatial loadings pattern of rotated EOF2 (describing 12.6% variance) - centered on Zimbabwe and Mozambique - is comparable to the leading observed mode, derived using the Hulme 1900–1998 gridded precipitation data [Hulme, 1992; Hulme et al., 1998]. JFM precipitation anomalies averaged over the domain defined by the strongest spatial loadings - approx. 20°E–35°E; 15°S–25°S - form the rainfall index. This analysis was repeated for the climate change runs for the whole, early (2000–2049) and late (2050–2099) twenty-first century to establish the position of the southern African rainfall region, and whether it changes in time. Results show that the spatial loadings are almost identical to the control and remain stable over the twenty-first century in both scenarios, allowing the climate change rainfall indices to be constructed using the same domain.

[7] In order to identify modes of variability in the Indian Ocean related to southern African rainfall variability, an EOF analysis of Indian Ocean seasonal (JFM) linearly detrended SST anomalies was carried out over the domain 20°E–120°E, 50°S–20°N for the whole 1000-year period, rotated, and tested for sensitivity in time in 100-year samples. Rotated EOF5 (explaining 5% of total variance) is a dipole pattern in the SWIO (the SWIOD), characterized by opposing SST anomalies in the tropical and subtropical SWIO (Figure 1a) (higher order rotated Indian Ocean SST EOFs comprise a basin-wide ENSO mode (EOF1), as well as patterns in the central and eastern SIO (EOFs 2 and 4) and the extratropical SWIO (EOF3)). EOF5 is a) very reminiscent of the pattern of highest values in the spatial correlation map between the southern African rainfall index and global SSTs, and its time series is the most strongly correlated (r = −0.46) of all the Indian Ocean EOFs with the rainfall index, and b) remarkably similar to the observed mode argued to be responsible for extremes in the twentieth century rainfall record [Washington and Preston, 2006]. The positive (warm tropical anomalies, cool subtropical anomalies) phase of the SWIOD is associated with anomalously dry and negative with anomalously wet summers. The SWIOD is the most important mode of rainfall-related SST variability in the Indian Ocean in both of the climate change runs. Spatial loadings are very similar to the control in A2 (rotated EOF6, describing 4.4% of the variance) and B1 (rotated EOF8, describing 2.3% of the variance), and so the existence of the mode appears to be independent of the CO2 forcing applied here (Figure 1). A SWIOD index is constructed to represent the SST pattern in the time domain from the difference between SST anomalies in the regions of maximum loadings of opposing signs, centered at ∼15°S, 65°E and ∼34°S, 56°E.

Figure 1.

Rotated EOF spatial loadings: (a) REOF5 in HadCM3 1000-year control, (b) REOF8 in HadCM3 B1 twenty-first century, and (c) REOF6 in HadCM3 A2 twenty-first century.

[8] Teleconnections between ENSO and southern Africa are central to empirical seasonal forecasting efforts for the region so it is worth demonstrating how this relationship is simulated by HadCM3 compared with the southern Africa/SWIOD connection in the control simulation in the model before going on to look at future variability. Over the 1000 years of the control, the SWIOD not only has a much stronger relationship with the summer rains than Niño-3 (correlation of r = −0.42 as opposed to r = −0.16), but it is also more stable. This result is consistent with work that reports on twentieth century low-frequency variability in the ENSO teleconnection with southern Africa [Richard et al., 2001]. A correlation analysis of the 1000-year control simulation time series over a 31-year sliding window shows that the expected negative relationship between Niño-3 and southern African rains is statistically significant only 20% of the time, compared with 82% of the SWIOD-rainfall correlation coefficients. Moreover, a similar proportion (16%) of the Niño-3-rainfall correlation coefficients is opposite in sign to the expected, which is never the case for the SWIOD-rainfall relationship. Regressing the two different SST indices onto the rainfall further demonstrates the relative importance of the SWIOD, which accounts for 18% of the variance in JFM rainfall over 1000 years, compared with a mere 3% accounted for by ENSO, a figure that gets smaller still in the future projections. These results underline the importance of the SWIOD to southern African rainfall in HadCM3 and justify the need to examine how this relationship is simulated in the future.

3. Twenty-First Century Rainfall Variance and the SWIOD

[9] Over the course of the twenty-first century, variance in southern African rainfall increases in both A2 and B1, but more markedly in the A2 run. Variance in the second half of the century more than doubles in A2 compared with the first, and increases by over 50% in B1 (Figure 2), demonstrating that extremes in the late twenty-first century will be more intense than in the earlier period.

Figure 2.

The 21-year running standard deviation (mm day−1) of southern African rainfall over the twenty-first century in (a) B1 and (b) A2.

[10] During the twenty-first century as a whole, the percentage of rainfall variance accounted for by the SWIOD is similar in the A2 and B1 runs at 22% and 26% respectively, both of which are higher than the control. However, the character of the rainfall variability is different in the early compared with the late twenty-first century. For the A2 run during the early twenty-first century, the SWIOD accounts for 14.4% of the variance in rainfall, rising to 26.6% in the second half of the century. The B2 SWIOD, on the other hand, accounts for 43.4% of variance in the early period, exceeding the maximum achieved in any 50-year period in 1000 years of the control and indicating a very strong teleconnection during this time, but subsequently declining to 15.8% in the late twenty-first century. These results point to a non-stationary teleconnection in the future, and also to differences in the rainfall response to SWIOD events between the two experiments.

4. Precipitation Anomalies During SWIOD Events

[11] Precipitation anomalies during SWIOD events (where the SWIOD index ≥∣1σ∣), divided into early and late twenty-first century, are displayed in Figure 3. These are compared with SWIOD-related precipitation anomalies for each distinct 50-year epoch of the control, and with the mean distribution for all epochs. There are between five and eleven events of both signs in each epoch of the combined control and experiment distributions.

Figure 3.

Distribution of precipitation anomalies during all SWIOD events in the experiments compared with a mean distribution of control epochs: (a) B1 2000–2049, (b) A2 2000–2049, (c) B1 2050–2099, and (d) A2 2050–2099. Red (pale red) and blue (pale blue) lines indicate experiment (control) positive and negative events, respectively.

[12] During the early twenty-first century in the B1 run (Figure 3a), there is a slight wet shift in the peak response to negative SWIOD events relative to the control, and a small dry shift in peak response to positive events. The extremes–both wet and dry–do not exceed the largest magnitude events of the control. In the latter part of the century (Figure 3c), the response to negative SWIOD events is very similar to the control, but there is a large spread in the response to positive events. Not only is there a high-magnitude dry anomaly (top 5% of control positive events), a strong wet anomaly of 2.1 mm day−1 occurs, which is exceeded only twice in over 150 control positive events. Nevertheless, anomalies of both signs fall within the envelope of variability defined by the control (not shown).

[13] In the early twenty-first century A2 run (Figure 3b), the peak precipitation response to negative SWIOD events is the same as the control. The wet extremes are close to the upper limit of the control, but remain within the envelope of variability defined by the 1000-year simulation (not shown). The peak precipitation response to positive SWIOD events has a slight wet shift relative to the control, and the extreme dry events are not as severe. This changes in the late twenty-first century (Figure 3d) when, although the peak response is the same as the control, the extremes associated with positive SWIOD events become more intense. The most extreme dry event falls within the top 5% of the greatest SWIOD-related droughts of the control. During negative events, a large wet shift in peak response occurs relative to the control, and in addition, extremes become much more intense. Thus, there is symmetry in the increase in variability apparent in the response to SWIOD events, making larger extremes of both signs more likely under A2 forcing. This is consistent with the notion that climate change will be manifest through modification of modes of variability. A positive rainfall anomaly of nearly 4 mm day−1 occurs during one event. The magnitude of this anomaly is unprecedented within 1000 years of the control.

5. Summary

[14] Modes of variability interact with the climate change signal, and therefore, understanding the current and future operation of these modes and their links with rainfall variability must be regarded as a priority for southern Africa. A meridional pattern of opposing SST anomalies in the SWIO was associated with extremes in observed southern African austral summer rainfall during the twentieth century and has been shown to exert strong control over the atmospheric circulation favorable for extreme rainfall in atmospheric GCM experiments [Washington and Preston, 2006]. Here, we show that an almost identical dipole pattern of SST anomalies in the SWIO, which we term the SWIOD, is central to southern African rainfall variability in a 1000-year control run of HadCM3. Positive SWIOD events are associated with dry conditions over southern Africa and negative events with wet.

[15] Southern African rainfall variability increases through the twenty-first century in HadCM3 under two different SRES CO2 emissions scenarios, indicating greater extremes in the decades to come. We find that the SWIOD remains strongly linked with southern African rainfall variability in the future. However, the proportion of rainfall variance accounted for by the SWIOD changes over the century and according to the emissions scenario. This implies a non-stationary teleconnection, and a difference in rainfall response between the two climate change experiments.

[16] The distributions of rainfall anomalies demonstrate the range of future rainfall responses to the SWIOD events under each forcing scenario and in the control run. During the early part of the twenty-first century, discrepancies between the two climate change runs are relatively small and all extremes lie within the envelope of variability defined by the control.

[17] During the second half of the twenty-first century, divergence between the experiments becomes more apparent. There is a wide range in response to positive SWIOD events in the B1 simulation, including a strong wet anomaly that is exceeded only twice in the positive events of the 1000-year control. However, while this is an extremely unusual event, it cannot be said that the climate in the B1 run has been perturbed beyond the bounds of natural variability. On the other hand, a positive area-average rainfall anomaly of just under 4 mm day−1 occurs in the A2 scenario in conjunction with a negative SWIOD event. An anomaly of this magnitude is unprecedented in a millennium of control data. The greatest JFM rainfall anomalies across southern Africa during this extreme year (up to 6 mm day−1) are of a similar magnitude as those during the devastating floods in Mozambique in 2000. And as noted in Section 2, HadCM3 underestimates rainfall extremes. In addition, if the intensity of rainfall events over the region is to increase in the future [Rocha et al., 2008] then the impact of these large rainfall anomalies will be compounded. The results presented here indicate that rainfall extremes in a coupled climate model, particularly flooding, over southern Africa hinge sensitively on the nature of the future global economic development trajectories and associated carbon budgets.

Acknowledgments

[18] The authors gratefully acknowledge the following for providing their data for analysis: UK Meteorological Office Hadley Centre, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) and the WCRP's Working Group on Coupled Modeling (WGCM) for the WCRP CMIP3 multi-model data set. Support of this data set is provided by the Office of Science, U.S. Department of Energy. The British Atmospheric Data Centre (http://badc.nerc.ac.uk/data/) provided access to the HadISST and COAPEC data. The observed precipitation data set was constructed and supplied by Mike Hulme at the Climatic Research Unit, University of East Anglia, Norwich, UK, and that work was supported by the UK Department of the Environment, Transport and the Regions.

Ancillary