CMIP5 Simulations of West African Monsoon Multidecadal Variability
The aim of this section is to assess the ability of the CMIP5 models to simulate multidecadal variability of Sahel rainfall in order to more completely understand results from the decadal hindcast simulations. It is hypothesized that models must represent the processes involved in multidecadal variability of Sahel rainfall in order to be successful with decadal hindcasts. This is done using both historical (20th century) and control (pre-industrial) simulations that have a low pass filter applied to isolate multi decadal signals. Simulations are being acquired as they are made available on the ESG.
Figure 1 shows the fraction of the total variance explained by decadal variance in July-August-September Sahel rainfall. In observations, this fraction is close to 45 %. In the CMIP5 multi-model ensemble of historical simulations, only one simulation exceeds 30 % with the multi-model mean less than 15 %. Values are marginally higher in the small sample of control simulations that have been examined so far.
Figure 1 shows the fraction of the total variance explained by decadal variance in July-August-September Sahel rainfall. In observations, this fraction is close to 45 %. In the CMIP5 multi-model ensemble of historical simulations, only one simulation exceeds 30 % with the multi-model mean less than 15 %. Values are marginally higher in the small sample of control simulations that have been examined so far.
Figure 1: Contribution of decadal variation to total variance of JAS Sahel rainfall in observations (black) and individual historical CMIP5 simulations (blue bars).
Examination of this reduction in decadally variability in the CMIP5 simulations has focused on SSTs in both the Atlantic and Indian Ocean. The decadal variability of the AMO is well simulated in most historical simulations. The significant positive correlation between Sahel rainfall and North Atlantic SSTs (or the Atlantic Multidecadal Oscillation, AMO) is simulated in the majority of historical and control simulations. The simulations of teleconnections between multidecadal Sahel rainfall and Indian Ocean SSTs are much less successful that those with the Atlantic Ocean so focus is placed on the Atlantic.
To investigate why some models successfully simulated this teleconnection and others did not despite having similarly large AMO variability, two groups of models were selected. Models with relatively large AMO multidecadal variability were highlighted as good (or poor) by their ability to simulate relatively high (low) Sahel multidecadal variability and have significant (not significant) correlation between multidecadal Sahel rainfall and the AMO index.
To investigate why some models successfully simulated this teleconnection and others did not despite having similarly large AMO variability, two groups of models were selected. Models with relatively large AMO multidecadal variability were highlighted as good (or poor) by their ability to simulate relatively high (low) Sahel multidecadal variability and have significant (not significant) correlation between multidecadal Sahel rainfall and the AMO index.
Figure 2: JAS SST regressed onto the leading principal component of North Atlantic low- pass filtered SST (◦C per standard deviation) from HadISST observations (a) with stippling showing regions where the regression coefficient is significant at 95 %. The multi-model mean from the good models is shown in b) and from the poor model group in c). In b) and c) stippling indicates regions where the mean significance level across the selected ensembles is at least 95 %.
Poor models fail to capture the teleconnection between the AMO and Sahel rainfall, as the spatial distribution of SST multidecadal variability across the North Atlantic is incorrect. This is illustrated in Fig. 2. In the poor models (Fig. 2c) a weak SST signal in the tropical North Atlantic reduces the interhemispheric SST gradient and through circulation changes, the rainfall variability in the Sahel (Fig. 3). Increases in rainfall across the Sahel with a warm AMO were evident in observations and good models but little response to AMO variability was evident in the poor models (Fig, 3c). Changes in circulation such as vertical wind shear, an important environmental variable for tropical cyclones also responded to the AMO incorrectly in the poor models.
Poor models fail to capture the teleconnection between the AMO and Sahel rainfall, as the spatial distribution of SST multidecadal variability across the North Atlantic is incorrect. This is illustrated in Fig. 2. In the poor models (Fig. 2c) a weak SST signal in the tropical North Atlantic reduces the interhemispheric SST gradient and through circulation changes, the rainfall variability in the Sahel (Fig. 3). Increases in rainfall across the Sahel with a warm AMO were evident in observations and good models but little response to AMO variability was evident in the poor models (Fig, 3c). Changes in circulation such as vertical wind shear, an important environmental variable for tropical cyclones also responded to the AMO incorrectly in the poor models.
Figure 3: Same as for Fig. 2 but for JAS precipitation regressed onto the leading principal component of North Atlantic low-pass filtered SST (mm/day per standard deviation). Observations (a) are from the CRU dataset.
The SST and rainfall patterns were also evident in the control simulations, where SST and Sahel rainfall variability were significantly weaker than historical simulations, suggesting external forcings such as sulfate aerosols my be amplifying the signal. Errors in SST variability were suggested to result from a combination of poorly simulated cloud amounts and feedbacks in the stratocumulus regions of the Eastern Atlantic, dust-SST-rainfall feedbacks and sulfate aerosol interactions with clouds. By understanding the deficits and successes of the CMIP5 historical simulations, future projections and decadal hindcasts can be examined with additional confidence.
- Paper accepted in J. Climate entitled "The Multidecadal Atlantic SST - Sahel Rainfall Teleconnection in CMIP5 Simulation"