116
Chapter 6: Analysing the Boreal Summer Relationship
JAS HF SAHEL = 0.00 - 0.12(a2)
significance of
each variable:
1 %
Multiple r of the model is: 0.60
1%
The null hypothesis that each regression coefficient is zero (separate test
for each coeflicient) is tested, and rejected at the 1% level for both variables,
showing that each explains a significant fr action of Sahel rainfall variability. This in fact was an expected result since the EOF time coeflicients are
almost uncorrelated (the all seasons coeflicients from the original analysis
1901-80 are perfectly uncorrelated). Thus, since both EOFs have significant
individual correlations with Sahel rainfall (Table 6.3), we can expect them to
combine weIl in a multiple regression. This is a particularly useful property
of EOF time-coeflicients for use in multiple regression. However, this whole
methodology relies on the key SST patterns being orthogonal. Here, progress
has been possible since ENSO and warm/cold events in the tropical Atlantic
are largely uncorrelated (at zero lag), and thus the EOF analysis was not significantly hampered by its orthogonality constraint. However, identification
of the types of ENSO that most strongly aifect Sahel rainfall (the ones with
strongest gradients of SST anomaly into the western Pacific) cannot be made
using this global EOF analysis of all years, since the two types of ENSO are
highly correlated, and only one EOF representing ENSO is possible.
The regression model above is one step from forming a seasonal forecast
technique for the Sahel. For seasonal forecasting, pre-rainfall season values
of the EOF coeflicients are used to form a regression model (see Folland et
al. , 1991, Ward and Folland, 1991).
6.5.4 Low Frequency Sahel Rainfall Variations
Inspection of the time-coeflicients of SST p3 (Figure 6.1 b) and Sahel rainfall
(Figure 6 .5b) is suflicient to identify a clear association between the two on
the LF timescale (Table 6.3). Consistent with this observation, the correlations between LF Sahel rainfall and the marine atmosphere (not shown)
are very similar to those with SST p3 (Figure 6.3). Thus multidecadal Sahel rainfall fluctuations may be part of the climate variation involving interhemispheric SST variations, and tropic-wide circulation anomalies, with teleconnections to the extratropics (see Section 6.4.3).
6.6 Conclusions
Progress in understanding climate variability has arisen through the combined eifort of theoretical studies, numerical modelling experiments and statistical analysis of climate observations. This chapter has concentrated on
one area of climate variability: the role of air-sea inter action in seasonal to
multidecadal climate variability. This area is particularly apt to illustrate the
Chapter 6: Analysing the Boreal Summer Relationship
JAS HF SAHEL = 0.00 - 0.12(a2)
significance of
each variable:
1 %
Multiple r of the model is: 0.60
1%
The null hypothesis that each regression coefficient is zero (separate test
for each coeflicient) is tested, and rejected at the 1% level for both variables,
showing that each explains a significant fr action of Sahel rainfall variability. This in fact was an expected result since the EOF time coeflicients are
almost uncorrelated (the all seasons coeflicients from the original analysis
1901-80 are perfectly uncorrelated). Thus, since both EOFs have significant
individual correlations with Sahel rainfall (Table 6.3), we can expect them to
combine weIl in a multiple regression. This is a particularly useful property
of EOF time-coeflicients for use in multiple regression. However, this whole
methodology relies on the key SST patterns being orthogonal. Here, progress
has been possible since ENSO and warm/cold events in the tropical Atlantic
are largely uncorrelated (at zero lag), and thus the EOF analysis was not significantly hampered by its orthogonality constraint. However, identification
of the types of ENSO that most strongly aifect Sahel rainfall (the ones with
strongest gradients of SST anomaly into the western Pacific) cannot be made
using this global EOF analysis of all years, since the two types of ENSO are
highly correlated, and only one EOF representing ENSO is possible.
The regression model above is one step from forming a seasonal forecast
technique for the Sahel. For seasonal forecasting, pre-rainfall season values
of the EOF coeflicients are used to form a regression model (see Folland et
al. , 1991, Ward and Folland, 1991).
6.5.4 Low Frequency Sahel Rainfall Variations
Inspection of the time-coeflicients of SST p3 (Figure 6.1 b) and Sahel rainfall
(Figure 6 .5b) is suflicient to identify a clear association between the two on
the LF timescale (Table 6.3). Consistent with this observation, the correlations between LF Sahel rainfall and the marine atmosphere (not shown)
are very similar to those with SST p3 (Figure 6.3). Thus multidecadal Sahel rainfall fluctuations may be part of the climate variation involving interhemispheric SST variations, and tropic-wide circulation anomalies, with teleconnections to the extratropics (see Section 6.4.3).
6.6 Conclusions
Progress in understanding climate variability has arisen through the combined eifort of theoretical studies, numerical modelling experiments and statistical analysis of climate observations. This chapter has concentrated on
one area of climate variability: the role of air-sea inter action in seasonal to
multidecadal climate variability. This area is particularly apt to illustrate the