## 1. Introduction

June–September is the primary rainy season (locally known as *Kiremt*) for the northern two-thirds of Ethiopia (Fig. 1), accounting for more than 60% of the annual total rainfall (Fig. 1a of Segele and Lamb 2005). This monsoon over Ethiopia and more generally across the Horn of Africa (Ethiopia, Eritrea, Djibouti, northern Somalia, eastern Sudan; Fig. 1) exhibits variability from intraseasonal to interannual and longer time scales (e.g., Beltrando and Camberlin 1993; Camberlin 1997; Shanko and Camberlin 1998; Conway 2000; Gissila et al. 2004; Seleshi and Zanke 2004; Segele and Lamb 2005; Korecha and Barnston 2007; Block and Rajagopalan 2007; Diro et al. 2008; Riddle and Cook 2008; Segele et al. 2009a,b). Recently, Segele et al. (2009b) performed an observational analysis that identified the regional atmospheric circulation patterns and global sea surface temperature (SST) anomalies directly linked to Horn of Africa (especially Ethiopian) rainfall and documented the interannual variability of those linkages.

However, few studies have examined the coherent temporal modes of Horn of Africa rainfall variability and their associations with the large-scale atmospheric and oceanic patterns on individual time scales. Bowden and Semazzi (2007), for example, noted that previous studies of intraseasonal variability focus on case studies of wet or dry years linked with the El Niño–Southern Oscillation (ENSO) phenomenon. Furthermore, this research also was geographically limited to the equatorial East Africa–southern Horn of Africa region (e.g., Mutai and Ward 2000; Mwale and Gan 2005) and primarily investigated variability of the “short” October–December rainy season (e.g., Bowden and Semazzi 2007). Although October–November brings the second seasonal rains for southern and southeastern Ethiopia and adjoining Somalia, it is a dry period for much of the northern two-thirds of Ethiopia and adjacent Eritrea and Djibouti. Therefore, isolating the different temporal modes of summer monsoon rainfall variability for the Horn of Africa is essential for understanding the sources and mechanisms of this rainfall variability, identifying interactions among the different time scales and assessing the predictability of this monsoon (e.g., Webster and Hoyos 2004; Yang et al. 2007).

Over the past several decades, the Horn of Africa (especially Ethiopia) experienced few flood years but many drought years. According to Mwale and Gan (2005), such irregularity in flood and drought events and the associated large time variability constitutes nonstationarity. Nonstationarity hampers use of the Fourier transform, the most common tool for power frequency spectrum analysis, which assumes time series homogeneity and stationarity (e.g., Weng and Lau 1994; Baliunas et al. 1997; Torrence and Compo 1998).

The most appropriate available alternative tool is the wavelet transform (Weng and Lau 1994; Huang et al. 1998; Torrence and Compo 1998). Unlike the Fourier transform, the wavelet transform localizes a signal in both the frequency and time domains and is well suited to analysis of multiscale, nonstationary time series that result from nonlinear interactions between several physical processes occurring on a range of temporal and spatial scales (Lau and Weng 1995; Webster and Hoyos 2004). The wavelet transform uses generalized base functions (wavelets) that can be stretched and translated with a flexible resolution (e.g., Weng and Lau 1994; Torrence and Compo 1998) and is increasingly employed to decompose geophysical time series into time–frequency space to detect periodicities and trends (e.g., Wang and Wang 1996; Baliunas et al. 1997; Chapa et al. 1998, Barrett and Leslie 2009) and to develop statistical prediction models (e.g., Webster and Hoyos 2004; Mwale and Gan 2005).

This study employs wavelet analysis to isolate and examine the spectral and temporal characteristics of the coherent modes of rainfall variability over the Horn of Africa (especially Ethiopia). Here in Part I, we present a time–frequency quantification of the teleconnections between Ethiopian June–September rainfall and the large-scale atmospheric circulation and global SST patterns on seasonal-to-interannual time scales. We set the results in the context of key findings of Segele et al. (2009b), which established that the Ethiopian summer monsoon is enhanced by increased lower-tropospheric westerlies from west and central Africa and the strengthening of both the Somali low-level jet (SLLJ) and the upper-tropospheric tropical easterly jet (TEJ). This new temporal quantification of the most strongly related dynamical and thermodynamic atmospheric and SST features provides the basis for developing statistical prediction models for Ethiopian summer rainfall on monthly to seasonal time scales. The development and validation of those models is being reported in part II of the study (Z. Segele et al. 2009, unpublished manuscript).

## 2. Data and methodology

### a. Data

Several datasets were used in this study. The first dataset contains rain gauge measurements for Ethiopia for 1970–99. Daily rainfall data for 121 Ethiopian stations were obtained in hard copy from the National Meteorological Agency of Ethiopia. As reported in Segele and Lamb (2005), these data were digitized and quality controlled to construct the first research-quality rainfall dataset for Ethiopia. Of the 121 stations obtained, 100 were determined to have their main rainy season during boreal summer and were used in this study (Fig. 1). Individual 5-day (pentad) rainfall averages (mm d^{−1}) were computed for these stations for 31 May 31–2 October (also subsequently referred to as June–September) 1970–99. This averaging produced a discontinuous time series of 750 pentad rainfall values—25 pentads per season for 30 yr. Use of the period 31 May–2 October to compute pentad gauge averages ensured consistency with the pentad Climate Prediction Center Merged Analysis of Precipitation (CMAP) data discussed below. Monthly totals also were computed for the 100 Ethiopian stations for June–September 1970–99.

To supplement these Ethiopian rain gauge data and to broaden the spatial coverage over the Greater Horn of Africa, we used also the National Oceanic and Atmospheric Administration/National Weather Service (NOAA/NWS) CMAP data (Xie and Arkin 1997) for the larger Horn region (Fig. 1). The CMAP dataset contains pentad and monthly analyses of global precipitation in which observations from rain gauges are merged with precipitation estimates from several satellite-based algorithms. As for the Ethiopia rain gauge data, we used averages for the 25 pentads per year, between 31 May 31 and 2 October. CMAP analyses are on a 2.5° × 2.5° latitude–longitude grid and extend back to 1979. Only 13 of the 28 grid points used here were in Ethiopia (Fig. 1). (Data for the grid points available online at http://www.cdc.noaa.gov/data/gridded/data.cmap.html.)

For large-scale atmospheric circulation analyses for the region between 50°N and 40°S and between 30°W and 90°E, daily average products of the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis project were obtained for 1970–99 from the NOAA–Cooperative Institute for Research in Environmental Sciences (CIRES) Climate Diagnostics Center (Boulder, Colorado; available online at http://www.cdc.noaa.gov/). The daily averages included geopotential height, temperature, horizontal wind, vertical velocity, and specific humidity at standard pressure levels, and mean sea level pressure (MSLP). The reanalysis data have a spatial resolution of 2.5° latitude × 2.5° longitude and are described by Kalnay et al. (1996).

To examine the linkages between Ethiopian rainfall and SST variability, we used the Met Office Hadley Centre global SST dataset [Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST1); Rayner et al. 2003]. This SST dataset contains globally complete fields on an individual monthly-mean basis for a 1° latitude–longitude grid from 1871 to the present. The present study used data from June–September 1970–99. These SST data largely are based on ship observations from the Met Office’s Marine Data Bank (MDB). To enhance data coverage where there were no MDB data, Rayner et al. (2003) added monthly median SST for 1871–1995 from the Comprehensive Ocean–Atmosphere Data Set (COADS; Woodruff et al. 1987).

### b. Wavelet analysis

*x*is defined as the convolution of

_{n}*x*with a scaled and translated mother wavelet

_{n}*ψ*to give

*s*is the scale,

*n*= 0, …,

*N*− 1 is a localized time index, and

*δt*(1/25 for pentad and 1/4 for monthly) and

*N*(750 for pentad, 120 for monthly) are the time spacing and length of the time series

*x*, respectively.

_{n}*W*(

_{n}*s*)|

^{2}describes the spectral characteristics of the time series

*x*in a time–frequency domain (Yang et al. 2007). The wavelet transform also can be used to reconstruct the original time series or to obtain wavelet-filtered time series between any two scales (

_{n}*j*

_{1}and

*j*

_{2}). For the Morlet wavelet, scales

*j*

_{1}and

*j*

_{2}are nearly the same as Fourier periods (Torrence and Compo 1998). Filtered time series for such arbitrary scales are given by

*ψ*

_{0}(0) removes the energy scaling,

*C*is a constant, ℜ[ ] denotes the real part of the wavelet spectrum, and

_{δ}*δ*is a factor for scale averaging. Using the Torrence and Compo (1998) software (available online at http://paos.colorado.edu/research/wavelets), we obtained wavelet-filtered time series

_{j}*x*′ for the major temporal modes of Ethiopian rain gauge and Horn of Africa CMAP rainfall variability. The filtered data were used in further time series analysis in sections 3 and 4.

_{n}### c. Statistical significance

Regression, correlation, and composite analyses were used to relate regional atmospheric circulation and global SST features to Horn of Africa (especially Ethiopian) rainfall variability for key individual time scales. Wavelet-filtered anomalies of horizontal wind components and vertical velocity for 1000–100 hPa across the Atlantic–Africa–Indo–Pacific domain were regressed on wavelet-filtered anomalies of Ethiopian pentad rainfall and Horn of Africa pentad CMAP time series. These regressions were designed to have zero intercepts and hence slopes that give the change in horizontal wind or vertical velocity anomaly that corresponds to a unit change in rainfall anomaly.

Because wavelet-filtered time series for seasonal and longer time scales are highly autocorrelated, basic assumptions of independence and normality can be violated when evaluating the statistical significance of their regression and correlation coefficients (e.g., Hoover 2003; Schmith et al. 2007). To address this issue, we assessed the statistical significance of the regression and correlation results using a “matched block” bootstrap test (Carlstein et al. 1998; Srinivas and Srinivasan 2005), which is a modified version of the bootstrap method of Efron and Tibshirani (1993, 221–233). The matched-block bootstrap test is a data-driven nonparametric technique that can even be applied to weakly nonstationary time series (Carlstein et al. 1998). This method samples time series blocks or sets of fixed length *l* of consecutive data values. It preserves much of the correlation structure within the blocks (Wilks 1997), as well as at block boundaries, by applying a rank matching procedure to join blocks that were a priori more likely to be close to one another (Carlstein et al. 1998; Srinivas and Srinivasan 2005). The block lengths used here were estimated following Politis and White (2004) and from a publicly available MATLAB computer code (available online at http://www.economics.ox.ac.uk/members/andrew.patton/opt_block_length_REV_dec07.txt).

This test yields an achieved significance level (ASL) that is the probability of exceeding the observed correlation by chance (Efron and Tibshirani 1993, p. 203). Here, the ASL was obtained from Monte Carlo simulations of 1000 iterations at each grid point, in which wavelet-filtered rainfall and gridpoint atmospheric and SST time series were appropriately sampled with replacement (in blocks and also rank matched) and their regression and correlation coefficients determined. Then, each observed regression (correlation) coefficient was ranked among the regression (correlation) coefficients obtained from the matched-block bootstrap samples. The resulting two-sided ASL was the fraction of the random regression (correlation) coefficients at least as large in absolute value as the observed regression (correlation) coefficient at that grid point. This ASL gives probability values *p* for two-sided statistical significance of nonzero regression (correlation) coefficients (e.g., Efron and Tibshirani 1993, p. 212; Davison and Hinkley 1997, 266–270). (Results appear in Figs. 5a,b, 6, 8, 9, 12, 13.)

Following Mason and Mimmack (1992) and Nicholls (2001), the reliability of the sample correlation coefficient as a population parameter estimate also was assessed by calculating the confidence intervals of correlations between wavelet-filtered rainfall and atmospheric variables for the annual mode at selected grid points. This computation was based on the “bias corrected and accelerated” (BC* _{a}*) bootstrap technique of Efron and Tibshirani (1993, p. 178). Because of strong temporal dependence in time series filtered to retain the annual mode, the BC

*method was applied to 1000 rainfall–grid point variable correlation replicates obtained by matched-block bootstrapping the wavelet-filtered Ethiopian rainfall averages. (Results are given in Tables 2 –6.)*

_{a}To account for spatial redundancy among neighboring grid points, the field significance of correlation maps was tested using the Monte Carlo simulation method of Barnston and Livezey (1989) that employs a modified Student’s *t* statistic. The procedure is described fully in Segele et al. (2009b). The Monte Carlo simulation assesses the collective correlation statistical significance by evaluating the probability of exceeding the observed correlation by chance. Following Barnston and Livezey (1989), Monte Carlo simulations were carried out for (i) domain-averaged *z* values (after converting local modified Student’s *t* values to *z* values and then averaging over all grid points) and (ii) domain-averaged coefficients of determination *R*^{2} (after squaring local correlation coefficients and then averaging over all grid points). The statistics computed from the actual data again were compared with the results of 1000 Monte Carlo simulations, in which the filtered rainfall time series was reshuffled and sampled with replacement for each simulation. All correlation maps shown in section 4 (Figs. 5a,b, 6, 8, 9, 12, 13) are statistically field significant at the 5% level.

Finally, the ordinary Student’s *t* test (Tamhane and Dunlop 2000, 124–125) was used to assess the statistical significance of WET minus DRY (WET − DRY) composite difference fields of atmospheric variables for wavelet-filtered Ethiopian rainfall extremes, for which the time series exhibited sufficient independence. (Results appear in Figs. 5c,d, 7, 10.)

## 3. Coherent modes of June–September rainfall variability

Wavelet analysis was performed on time series of pentad and monthly rain gauge averages for Ethiopia and pentad CMAP rainfall estimates for the Greater Horn of Africa. Figure 2 shows the raw time series and its spectral characteristics of Ethiopian pentad rainfall for 1970–99. The time–frequency plot of the power associated with Ethiopian rainfall variability (Fig. 2b) shows the distribution of the rainfall variance in time (abscissa) for all Fourier periods (ordinate). The thick dashed line in Fig. 2b delineates the cone of influence (COI), below which the amplitude of the power spectrum is impacted by edge effects (Torrence and Compo 1998; Mwale and Gan 2005). The global wavelet spectrum (Fig. 2c) is the time average of the local power (Fig. 2b) for each Fourier period. Both the local and global power spectra are normalized by total variance. On the basis of the average global power, nonoverlapping frequency bands were estimated for the major power peaks. Table 1 gives the estimated time scales of the major modes and their temporal variance fractions. These variance fractions were obtained by applying the standard statistical variance formula to each wavelet-filtered time series and are close to spectral variance fractions obtained by integrating Eq. (14) of Torrence and Compo (1998) over each frequency band.

The annual cycle (0.57–1.42 yr or 210–517 days; Table 1) contains the dominant power in the Ethiopian rainfall wavelet spectra and accounts for 66% of the raw (unfiltered) Ethiopian pentad variance (Table 1) and a peak of more than 12% of the total global power for all frequencies (Figs. 2b,c). The amplitude of the annual cycle fluctuates from year to year (Figs. 2b). For example, the maximum local power varied from 3.0*σ*^{2} in 1987 to 7.4*σ*^{2} in 1981, where *σ*^{2} = 4.47 (mm d^{−1})^{2} is the variance of the pentad rainfall time series for 1970–99. The power associated with rainfall variability at shorter intraseasonal time scales (less than 0.20 yr or 75 days) is small but includes statistically significant (5% level) irregular spectral peaks with a maximum power of 0.6*σ*^{2} (Fig. 2b). Power peaks also occur at seasonal (0.20–0.57 yr or 75–210 days), quasi-biennial (QB; 1.42–3.04 yr or 517–1109 days), and ENSO (3.04–4.60 yr or 1109–1680 days) time scales (Fig. 2c; Table 1). These periodicity bands account for 12%, 5%, and 2% of the raw Ethiopian pentad rainfall variance, respectively (Table 1). On the QB time scale, relatively large power is evident in both locally (mid-1970s, 1980s, 1990s; Fig. 2b) and globally (Fig. 2c). Also, the spectral bands for monthly Ethiopian rainfall time series (used in section 4 to investigate Ethiopian rainfall–SST relationships) are very close to the above bands obtained for pentad rain gauge time series (Table 1), with the small spectral bandwidth differences being primarily due to their sampling interval mismatch. Because all the above modes are mutually exclusive (Table 1), removing one mode from the raw rainfall data (e.g., the annual cycle) does not affect the power characteristics of the remaining modes, and thus the spectrally isolated time series for all the other modes remain unchanged.

To compare the power spectrum for the Horn of Africa CMAP data with that considered above for Ethiopian rain gauge data, wavelet analysis was also performed on CMAP pentad rainfall estimates (Fig. 3; Table 1). The starting times for the two datasets differ (section 2a), and the CMAP values are generally of smaller magnitude (Figs. 2a, 3a). Despite these differences, the rain gauge and CMAP power spectra bear overall strong resemblance in the frequency distribution of their global power (Figs. 2b,c; 3b,c; Table 1). However, there also are noticeable contrasts in the detailed temporal structure of the local (Figs. 2b,c; 3b,c) and global (Figs. 2c, 3c) power spectra for the two time series. These differences include particularly the absence of broadly distributed CMAP variance on the annual time scale (annual cycle peak power is only 4% of the total global power), larger CMAP variance for the intraseasonal (less than 0.25 yr or 91 days) and ENSO (3.74–6.51 yr or 1365–2377 days) modes, and a period of quiescence in the early 1990s for the QB (1.32–3.74 yr or 225–483 days) mode.

Figure 4 presents wavelet-filtered Ethiopian pentad rain gauge time series for the different temporal modes identified in Fig. 2 and Table 1. For the annual mode, separate times series were produced for the 0.57–1.42- and 0.95–1.05-yr bands to show the effects of a broadened annual spectral bandwidth (Fig. 4d). Although the annual time series produced by the broader filter has much more variable and larger amplitudes, the two annual time series correlate strongly (+0.99), which indicates near-perfect temporal covariability between both spectral bandwidths. Although the wide bandwidth annual mode contains the largest amplitudes, the combined effects of multitime-scale signals in Fig. 4 determine the overall strength of the Ethiopian summer monsoon. For example, the wide bandwidth annual cycle amplitude for 1996 was very small (Fig. 4d), but the country (especially central Ethiopia) experienced excessive rainfall and floods. This wetness resulted from a combination of large positive early-season rainfall anomalies due to intraseasonal-to-seasonal time-scale oscillations (Figs. 4a–c) and season-long positive rainfall anomalies that were part of QB (especially Fig. 4e) and ENSO (Fig. 4f) time-scale oscillations. Such multitime-scale constructive superposition also leads to other extremely wet monsoons (e.g., 1988—wettest since 1970, Fig. 2a; Segele et al. 2009a, their Fig. 9) and to extremely dry monsoons (e.g., 1987—driest since 1970, Fig. 2a; Segele et al. 2009a, their Fig. 9). For 1988 (1987), the amplitude of the wide bandwidth annual cycle was large (small), the intraseasonal-to-seasonal time-scale signals possessed larger (smaller) positive rainfall anomalies than negative anomalies and, most importantly, the amplitudes of the QB and ENSO time-scale oscillations remained large positive (negative) throughout the season (Figs. 4a–f).

The identification in Figs. 2, 4 and Table 1 of the 1.42–3.04- and 3.04–4.60-yr (1.46–3.13 and 3.13–4.43 yr) bands for Ethiopian pentad (monthly) time series as QB and ENSO modes, respectively, is consistent with the same representation of 1.3–3.0- and 3–7-yr spectral peaks in global sea level pressure and SST (e.g., Barnett 1991; Terray 1995; Webster et al. 1998; Torrence and Webster 1999) and regional precipitation (e.g., Webster et al. 1998; Fasullo 2004; Yang et al. 2007). However, dominance of these variability modes differs regionally. For example, Shen and Lau (1995) found that the QB mode is dominant for East Asian monsoon rainfall and SST in the eastern Indian Ocean and the western, northern, and southern Pacific Ocean, while much of central and eastern Pacific and northern Indian Oceans’ SSTs have major spectral peaks between 3 and 6 yr. As will be shown in section 4d, the 3.04–4.60-yr band for Ethiopian pentad rain gauge (3.13–4.43-yr band for monthly time series) here clearly isolates ENSO effects on the Ethiopian *Kiremt*.

## 4. Associations of rainfall variability with atmospheric circulation and SST patterns

On the basis of the above frequency bands identified for pentad Ethiopian rain gauge averages (CMAP estimates), counterpart wavelet analysis was performed for June–September 1970–99 pentad time series of MSLP and tropospheric horizontal wind, vertical velocity, geopotential height, and temperature (all 1000–100 hPa), and specific humidity (1000–300 hPa). Wavelet-filtered time series were extracted for each 2.5° latitude–longitude grid point for 50°N–40°S and 30°W–90°E. For SST, the analysis was performed on a monthly 1° latitude–longitude basis between 60°N and 60°S across all longitudes for 1970–99. Regression, correlation, and composite analyses were then performed on the time series of filtered pentad or monthly rainfall and identically filtered pentad-gridded atmospheric variables and monthly SST fields to quantify the climate system linkages on individual time scales. Only variations from seasonal to interannual time scales (75 days–4.6 yr) are presented because of their strong roles for Ethiopian drought. Intraseasonal variability was less important in that regard (Fig. 2c). Unless otherwise stated, all correlation values discussed below are statistically significant with an ASL of 5%, according to the matched-block bootstrap test described in section 2c.

### a. Seasonal time-scale variability (75–210 days)

The seasonal atmospheric circulation cycle exhibits moderate correlations with Ethiopian summer rainfall. Figure 5 presents the correlation patterns for selected reanalysis fields or levels, along with the large-scale circulation anomalies for DRY and WET Ethiopian rainfall composites defined below. The correlation analysis (Fig. 5a) shows a maximum positive correlation (+0.36) between Ethiopian pentad rainfall and Gulf of Guinea MSLP near 10°S and 12°, and a maximum negative correlation (−0.28) over the Arabian Peninsula near 30°N and 40°E. Thus, wet Ethiopian monsoons are characterized by enhanced southwest-northeast-directed MSLP gradients resulting from an Arabian Peninsula MSLP fall and a Gulf of Guinea MSLP rise at the seasonal time scale. Consistent with this pressure gradient anomaly, westerly flow from the tropical Atlantic and from the Congo basin rain forest increases atmospheric water vapor over the Horn of Africa, where Ethiopian rainfall correlates maximally with vertically integrated (surface to 300 hPa) water vapor (+0.60; not shown). This pattern becomes especially strong at 700 hPa (Fig. 5a), where the correlation between Ethiopian pentad rainfall and zonal wind maximizes (+0.51) over the Horn of Africa.

Lower-tropospheric temperatures in the surface-to-850-hPa layer over Ethiopia and the Gulf of Guinea (near 5°N, 10°E) are negatively correlated (−0.40 to −0.45) with Ethiopian pentad rainfall at the seasonal time scale (not shown). These correlations extend to 500 hPa over Ethiopia. In the upper troposphere, the correlations between Ethiopian pentad rainfall and horizontal wind components or geopotential height are moderate, with the strongest negative correlation for the zonal wind being −0.35 over the Arabian Peninsula at 150 hPa (Fig. 5b). Thus, enhanced upper-level easterlies and a cooler local midtroposphere in the tropics are associated with a wetter Ethiopian monsoon on the seasonal time scale.

In addition to the deepening of the surface heat low across the Arabian Peninsula, strong anticyclones over southern Europe, the western Mediterranean, and North Africa and the southern Indian Ocean (south of Madagascar) enhance the meridional flow from both hemispheres and resulting convergence over the Horn of Africa on the seasonal time scale. These features are evident in the WET − DRY composite difference maps in Fig. 5 (bottom panels). The WET and DRY composite patterns were constructed by averaging atmospheric fields for the wettest and driest 2% of the 75–210 days filtered June–September 1970–99 Ethiopian pentad rain gauge averages, respectively. Consistent with the correlation analyses, the WET − DRY difference maps also show (i) increased lower-tropospheric westerlies from West Africa to the Horn of Africa and a stronger SLLJ (Findlater 1969); (ii) enhanced mid-to-upper-tropospheric easterlies across tropical regions; and (iii) warmer upper-tropospheric temperatures in the northern (especially) subtropics and cooler temperatures in the equatorial regions. Most of these features also appeared strongly in the raw WET − DRY difference maps of Segele et al. (2009b, their Fig. 12), which underlines the importance of the seasonal time scale.

For monthly Ethiopian rain gauge totals, the seasonal (<232 days; Table 1) wavelet power spectrum is weak. As a result, the correlations between monthly June–September Ethiopian rain gauge totals and identically filtered grid square global SST are generally not strong (not shown) but exhibit a distinct hemispheric contrast with positive (negative) correlations covering the Northern (Southern) Hemisphere. Although correlations in the northern Indian Ocean are weakly negative, they are positive in the central equatorial Indian Ocean. The largest negative correlations (−0.60) occur in localized areas across the southern Indian and southeastern Pacific Oceans.

### b. Annual time-scale variability (210 days–1.42 yr)

Dominance of the annual mode for Ethiopian rainfall variability is evidenced by exceptionally strong correlations between 210 days–1.42 yr wavelet-filtered Ethiopian pentad rainfall averages and several identically filtered atmospheric fields (Fig. 6). As explained in section 2c, because of the importance of this association, the matched-block bootstrap significance assessments (Figs. 6 –8) were complemented by the use of the BC* _{a}* bootstrap technique to examine the reliability of the sample correlation coefficient as a population parameter estimate (Tables 2 –6).

At the annual time scale, Ethiopian pentad rain gauge averages are strongly negatively correlated with MSLP (Fig. 6a; Table 2) over western Mauritania (−0.92), the eastern Mediterranean Sea (−0.90), the Arabian Peninsula (−0.84), and northern India and Pakistan (−0.92). Positive correlations occur over the equatorial and southern Atlantic (+0.82) and the southwestern Indian Ocean (+0.84). For wet Ethiopian monsoons, the regression of wavelet-filtered 700-hPa horizontal wind components against identically filtered Ethiopian pentad rain gauge averages (Fig. 6a, Table 4) shows the importance of well-developed cross-equatorial southwesterlies from West Africa and the equatorial Atlantic and also over the western Indian Ocean. The correlation between wavelet-filtered Ethiopian pentad rainfall averages and vertically integrated specific humidity was strong (>+0.90) from West Africa to the Arabian Sea (not shown), further emphasizing the key role of water vapor transport from the Atlantic Ocean for Ethiopian monsoon rainfall.

At and above 500 hPa, tropical wind anomalies become easterly and strengthen with height to 200–150 hPa, where the prevalence of strong easterly anomalies across much of Africa and southwestern Asia indicates that a strengthened TEJ enhances the Ethiopian monsoon (Fig. 6b; Tables 3 –4). This causation is consistent with our earlier findings concerning the role of the TEJ for monsoon onset (Segele and Lamb 2005, their Figs. 5, 7) and for convection promoting upper-tropospheric divergence (Segele et al. 2009b, their Fig. 3). Also, consistent with this situation, the correlations between wavelet-filtered Ethiopian pentad rain gauge averages and gridpoint temperature (Fig. 6b) are high and positive (negative) in the Northern Hemisphere north of 20°N (to the south). The largest positive (negative) correlation of +0.95 (−0.94) occurs over the eastern Mediterranean (south-central Ethiopia) at low levels (Table 5). Also, the annual time-scale rainfall–geopotential height correlations are very strong and spatially coherent (Table 6).

Figures 6c,d uses vertical cross sections to further depict the annual mode correlations between Ethiopian pentad rainfall averages and tropospheric geopotential height (Fig. 6c) and temperature (Fig. 6d). The transects feature very strong correlation magnitudes with a clear latitudinal correlation contrast. In the lower troposphere, the rainfall–height correlation patterns (Fig. 6c) reflect the rainfall–MSLP relationships (Fig. 6a), but the correlations reverse sign in the upper troposphere, where positive (negative) geopotential height anomalies in the northern subtropics (near equatorial zone) are associated with enhanced Ethiopian rainfall. Unlike for geopotential height, the correlations between Ethiopian pentad rainfall averages and tropospheric temperatures do not change sign with height but remain strong positive (negative) north (south) of 20°N in the surface-200-hPa layer (Fig. 6d). This correlation pattern indicates that anomalously warm (cool) tropospheric temperatures north (south) of 20°N are associated with increased Ethiopian rainfall. Above 200 hPa, the hemispheric temperature contrast weakens and the rainfall–temperature correlations reverse sign. The 50°N–15°S latitudinal correlation contrast in Fig. 6d is similar to the spatial correlation patterns for the raw (unfiltered) time series discussed in Segele et al. (2009b), except that the present correlations are very strong [maximum positive (negative) correlation reaching +0.96 (−0.95); Tables 5 –6].

The regional zonal and meridional circulations associated with the annual cycle are almost identical in pattern to the climatological circulation features identified in Segele et al. (2009b; their Fig. 4) and so are not discussed here. The present focus is on the flow patterns characteristic of the high and low amplitudes of the annual cycle that correspond to strong and weak Ethiopian monsoons. The stratification used identified Ethiopian monsoons for which filtered peak season pentad rainfall values were greater than 2.6 mm d^{−1} (WET) and less than 1.5 mm d^{−1} but greater than 0.8 mm d^{−1} (DRY). The minimum 0.8 mm d^{−1} threshold eliminated influence of pentads far from monsoon peaks. There are 23 (25) pentads in the WET (DRY) category. The WET pentads occurred in 1970, 1981, 1985, 1992, 1998, 1999 (each with at least three pentads); and 1971, 1980, 1994 (one pentad each). The DRY pentads were concentrated in 1984 (9 that year), 1986 (8), and 1996 (8). WET − DRY difference maps for MSLP, geopotential height, and horizontal wind components are shown in Fig. 7.

Large amplitudes of Ethiopian annual rainfall cycle occur in association with MSLP weakening across northern Africa and the eastern Mediterranean, and MSLP intensification over the southern Atlantic and southern Indian Oceans (Fig. 7a). The annual cycle of the Ethiopian monsoon is also enhanced when midlatitude cyclone intrusions occur to the south of South Africa. Consistent with St. Helena high intensification and MSLP weakening between 20° and 30°N, lower-tropospheric anomalous westerlies prevail across near-equatorial Africa during wet Ethiopian monsoons. Furthermore, in association with geopotential height elevation in the Northern Hemisphere subtropics and across southern Africa, mid-to-upper-tropospheric easterlies strengthen and become more dominant across much of tropical Africa (Fig. 7b).

These findings are consistent with the climatological circulation results of Segele et al. (2009b), which noted enhancement of Ethiopian peak summer rainfall in relation to intrusion of tropospheric midlatitude cyclones in the southern Atlantic, increased lower-tropospheric near-equatorial westerlies, and strengthened upper-tropospheric easterlies. Interestingly, both the WET and DRY composites featured large negative MSLP anomalies over the Arabian Peninsula, so their difference was not statistically significant at the 5% level. This indicates that there is little annual mode variability in the intensity of the surface monsoon low over the Arabian Peninsula and emphasizes further the importance for Ethiopian rainfall time-scale variability of subtropical MSLP or geopotential height variations and the resulting tropical wind pattern changes.

Rainfall–SST relationships on the annual time scale (232 days–1.46 yr) show a pronounced interhemispheric contrast (Fig. 8). Unlike for the seasonal mode, the monthly rainfall–SST correlations for the annual mode are strong and spatially homogeneous. The strongest and most coherent correlations occur over the northwestern Pacific (+0.91), northwestern (+0.91) and equatorial (−0.87) Atlantic, and western Indian Oceans (−0.87). Thus, Ethiopian monsoon rainfall is enhanced (reduced) in association with annual time-scale cold (warm) SST over the equatorial Atlantic and the western Indian Oceans. The above conclusion is consistent with results of previous studies that linked basin- and global-scale SSTs to monsoon rainfall variability over many parts of Africa and India (e.g., Lamb, 1978a,b; Rao and Goswami 1988; Folland et al. 1991; Lamb and Peppler 1992; Hastenrath et al. 1995; Ju and Slingo 1995; Kawamura 1998; Ward 1998; Giannini et al. 2003).

### c. QB time-scale variability (1.42–3.04 yr)

Ethiopian summer rainfall also exhibits strong associations with regional monsoon systems and remote large-scale atmospheric circulation and SST patterns on the QB time scale. Regionally (Fig. 9a), large correlations occur between Ethiopian pentad rain gauge averages and the Azores and Saharan high (+0.65), the monsoon trough over the Arabian Peninsula (−0.87), and the St. Helena high (+0.59). The surface monsoon low deepening over the Arabian Peninsula strengthens the anomalous cyclonic circulation there and increases moisture convergence across the Gulf of Aden and the Arabian Peninsula, where Ethiopian rainfall is maximally correlated with vertically integrated (surface to 300 hPa) water vapor (+0.80; not shown). At 150 hPa (Fig. 9b), the largest positive Ethiopian rainfall–geopotential height correlation also occurs over northern Africa (+0.66). Tropical easterlies dominate the tropics between the subtropical highs. The highest correlation between Ethiopian pentad rain gauge and the zonal wind component (−0.84) occurs over the Arabian Sea (7.5°–10°N, 57.5°–62.5°E) at 150 hPa, indicating that TEJ strengthening enhances Ethiopian rainfall at this time scale also.

The above MSLP deepening over the Arabian Peninsula is linked to negative lower-tropospheric pressure anomalies (Fig. 9c) extending across the eastern Indian Ocean (−0.75) to Indonesia (−0.72). Farther east across the Pacific Ocean, the rainfall–MSLP correlation reverses sign and intensifies eastward, with positive correlations maximizing (+0.79) just east of Tahiti (17.6°S, 149.6°W). This dipolar correlation pattern east and west of 160°E closely resembles the classical Southern Oscillation signal (e.g., Barnett 1991). Consistent with this situation, the 1000-hPa regression wind vectors (Fig. 9c) indicate that QB Ethiopian rainfall is favored more by strong easterly anomalies over the equatorial central and western Pacific than by weak (statistically not significant) westerly anomalies over the eastern Indian Ocean.

In the upper troposphere (Fig. 9d), the QB Ethiopian rainfall–geopotential height correlation is less strong, with the largest correlations of −0.47 (−0.57) occurring over the northeastern (southeastern) tropical Pacific. To the west of these rainfall–geopotential correlation maxima lie anomalous cyclonic circulation cells (10°N, 137°W, 15°S, 137°W) that are evident when the entire regression wind vector field is plotted (not shown). The strongest 200-hPa wind results in Fig. 9d are concentrated along the equator over the western and central Pacific (fast westerlies) and Indian Oceans west of 80°E (fast easterlies) as in Fig. 9b, suggesting Walker circulation strengthening occurs during wet Ethiopian monsoons.

The atmospheric flow structures associated with extreme dry and wet Ethiopian monsoons on the QB time scale further illustrate the regional and teleconnection patterns that affect Ethiopian rainfall. Stratification into DRY and WET QB composites was based on the 15 driest and 15 wettest pentads of the 1.42–3.04-yr filtered June–September Ethiopian pentad rainfall, respectively. The DRY composite pentads occurred in 1984 (9) and 1997 (6), while the WET composite contained pentads in 1983 (5) and 1988 (10). Except for 1984, these QB composite years are different from their annual mode counterparts, which is consistent with time-scale preferences noted in section 3. The WET − DRY composite results (Figs. 10a,b) largely reflect the overall tropospheric correlation patterns. In this regard, the major features of the difference fields are (i) MSLP intensification (deepening) over the subtropical Atlantic and southeastern Pacific (Arabian Peninsula, eastern Indian Ocean, and western Pacific); (ii) enhanced lower-tropospheric westerlies (northerlies) across central Africa and western Indian Ocean (Red Sea); (iii) enhanced lower-tropospheric easterlies across the equatorial Pacific; (iv) anomalous upper-tropospheric anticyclonic circulation over southwestern Europe and Asia; (v) upper-tropospheric easterly (westerly) anomalies across the western Indian Ocean (equatorial Pacific); and (vi) a strengthened TEJ over the Arabian Sea.

To identify the anomalous QB regional meridional circulation linked with Ethiopian and Horn of Africa rainfall, we employed the regression approach of Krishnamurthy and Goswami (2000). The 1.42–3.04-yr filtered horizontal wind components and vertical velocity for 1000–100 hPa between 30°N–20°S and 30°–50°E were regressed on identically filtered pentad Ethiopian (Fig. 11a) and pentad Horn of Africa CMAP (Fig. 11b) rainfall averages. The regression coefficients, signifying responses in each dependent variable (*u*, *υ*, –*ω* anomalies) to changes in the independent variable (rainfall anomaly), were then averaged over 30°–50°E to obtain the mean meridional circulation (*υ*, –*ω*) anomalies. As explained further in Segele et al. (2009b, their section 3), this approach yields normalized meridional–vertical vectors, where the meridional wind components are represented in meters per second (standard deviation)^{−1}, and the vertical (pressure) velocity components are in pascals per second (standard deviation)^{−1}; therefore, both are dimensionless. Thus, the vector magnitudes in Fig. 11 indicate the local strength of the circulation relative to that across the rest of the 30°N–20°S and 1000–100-hPa domain for the wet Ethiopian monsoon phase. The dry monsoon phase counterpart is obtained by multiplying the regression vectors by −1.0.

The QB time-scale meridional circulation for wet Ethiopian monsoons generally resembles that associated with the wider Horn of Africa CMAP estimates (Figs. 11a,b). In both cases, reflecting the westward shift of the equatorial Pacific Walker circulation, the maximum ascending vectors are located in the lower-to-middle troposphere, with the largest ascending motion occurring north of 12°N. However, for Ethiopian rainfall (Fig. 11a), the anomalous ascending motion extends high into the upper troposphere, tilting southward with height and bringing much of Ethiopia under enhanced middle-to-upper-tropospheric ascending motion. The anomalous descending motion is also shallow. In contrast, the anomalous ascending (descending) motion for Horn of Africa rainfall is shallow (deep; Fig. 11b). These differences largely reflect the generally lower rainfall totals in the CMAP data (e.g., Segele et al. 2009a,b), and also the monsoon being weaker in the CMAP countries outside of Ethiopia (Fig. 1). Thus, for a wet Ethiopia/Horn of Africa phase of the QB cycle, the large-scale meridional flow intensifies the strong climatological ascending motion over the northern Horn of Africa and the weak descending motion to the south (Segele et al. 2009b, their Fig. 4).

The regional circulation for a dry Ethiopia/Horn of Africa rainfall phase is opposite to the circulation in Fig. 11. The eastward shift in the Walker-type circulation in the Pacific forces the ascending motion in the regional meridional circulation to shift to southern Ethiopia or equatorial regions. This shift brings descending motion over much of the Horn of Africa and thus weakens the climatological ascending motion and consequently the monsoon rainfall.

Equatorial Pacific monthly SST anomalies exhibit strong associations with Ethiopian monthly rain gauge anomalies on the QB time scale (Fig. 8b), some of which tend to counteract the annual mode relations over the western equatorial and southern Pacific (Fig. 8a). The QB rainfall–SST correlations possess clear ENSO signatures, being negative across the central and eastern equatorial Pacific and positive over parts of the equatorial western Pacific. Accordingly, a region of strong negative correlation (−0.83) coincides with a tongue of anomalous SST in the eastern Pacific that typically occurs during El Niño events. Over the western Pacific Ocean, a horseshoe-shaped region of opposite correlation (+0.84) straddles the equator. In general, a warmer equatorial western (eastern) Pacific tends to be associated with enhanced (reduced) Ethiopian summer rainfall on the QB time scale. The QB correlations with Ethiopian rainfall lack spatial homogeneity over the Indian Ocean. They are negative off the coast of Somalia (−0.47) and positive farther east over the central Indian Ocean (+0.43) and off the southeastern coast of Africa (+0.46). Ethiopian rainfall also correlates positively with SST over the equatorial Atlantic (+0.49) on this time scale. Compared to the rainfall–SST correlations on the annual time scale, QB SST variations in the equatorial Atlantic, the central and southwestern Indian Ocean, and the western Pacific counteract those occurring for the annual mode.

### d. ENSO time-scale variability (3.04–4.60 yr)

There are similarities and distinct differences in the Ethiopian rainfall–lower tropospheric circulation relationships at the QB and ENSO time scales. For both modes, Ethiopian rainfall is associated with MSLP intensification over the northeastern Atlantic and northwest Africa, anomalous westerlies (dry northerlies) across central Africa (along the Red Sea), and surface monsoon low deepening over the Arabian Peninsula. However, although the QB MSLP correlations are strongest over the Arabian Peninsula and Arabian Sea, the largest ENSO mode correlations occur more locally over the Red Sea (Fig. 12a) for MSLP (−0.65) and from West Africa (–0.78) to Ethiopia (−0.83) for 700-hPa geopotential height (not shown). Also, unlike for the QB mode, ENSO-related Ethiopian rainfall enhancement is linked with (i) Mascarene high intensification and midlatitude cyclone intrusion over the southeastern Atlantic Ocean (Fig. 12a) and (ii) a large upper-tropospheric geopotential height decrease and strengthened easterlies from Sudan to West Africa (Fig. 12b).

The ENSO mode pentad regression and correlation patterns for Ethiopian rainfall exhibit the major ENSO signatures in the wind and pressure fields across the Pacific and Indian Oceans (Figs. 12c,d). Positive (negative) rainfall anomalies are associated with the cold (warm) ENSO phase (e.g., Rasmusson and Carpenter 1982; Webster et al. 1998; Garreaud and Battisti 1999; Kirtman and Shukla 2000; Lau and Nath 2000; Lau and Wu 2001). Especially prominent are the east–west Southern Oscillation dipole pattern in the MSLP correlation field and the related tropospheric flow structure. For the positive rainfall phase, well-developed anomalous lower-tropospheric westerlies in the eastern Indian Ocean and easterlies in the central and eastern Pacific Ocean (Fig. 12c) lead to a region of strongly convergent low-level flow over the Maritime Continent. In this case, the ENSO flow pattern at 200 hPa (Fig. 12d) is also more vigorous than for the QB mode (Fig. 9d), featuring strong anomalous easterlies (westerlies) over the eastern Indian (Pacific) sector and hence enhanced Walker circulation. Opposite anomalies prevail for the dry Ethiopian phase.

The tropospheric structure associated with the extreme positive and negative Ethiopian ENSO rainfall phases was assessed further by compositing several atmospheric variables for the 15 driest (DRY) and 15 wettest (WET) pentads of the 3.04–4.60-yr wavelet-filtered Ethiopian rainfall time series. The driest Ethiopian rainfall pentads occurred in 1976 (4) and 1987 (11), while the wettest pentads were in 1988 (6) and 1989 (9). The WET-minus-DRY difference fields support the above regression and correlation analyses by showing the ENSO wet Ethiopian phase is associated with (i) Azores and Mascarene high intensification (Fig. 13a); (ii) anomalous anticyclones over the Mediterranean and North Africa (Fig. 13a); (iii) Arabian Peninsula monsoon low deepening and St. Helena high weakening (Fig. 13a); (iv) strengthened southerlies over the western Indian Ocean and near-equatorial westerlies across Africa (Fig. 13a); and (v) a large upper-tropospheric geopotential height decrease in the tropics (Fig. 13b). Further east, across the Indian and Pacific Oceans, the composite results (Figs. 13c,d) are again in good agreement with their regression or correlation analysis counterparts (Figs. 12c,d). Over the tropical Pacific Ocean, there is also good qualitative correspondence between the QB and ENSO composite difference fields across the tropical Pacific Ocean, although ENSO anomalies are larger than their QB counterparts. The most outstanding QB–ENSO difference occurs over the equatorial Indian Ocean, where the QB lower-tropospheric westerly anomalies are much weaker, more spatially limited, and produce weaker low-level convergence over the Maritime Continent. In contrast, upper-tropospheric easterly QB (ENSO) anomalies (Figs. 12d, 13d) are stronger than their ENSO (QB) counterparts over the western Indian Ocean (eastern Indian Ocean and northwest of Ethiopia).

As expected, SST variability has the greatest influence on Ethiopian rainfall at the ENSO time scale (3.13–4.43 yr; Table 1). Accordingly, negative monthly rainfall–SST correlations (Fig. 8c) are large across the central and eastern equatorial Pacific (−0.91). This belt is enclosed to the west and poleward by a horseshoe-shaped area of positive correlations that maximize near the western Pacific equator (+0.92). The complete tropical Pacific pattern closely resembles (the negative of) the SST anomalous composite structure for mature and transition phases of El Niño (e.g., Rasmusson and Carpenter 1982; Wang 1995). An equally strong relationship was found for most of the Indian Ocean (Fig. 8c), with the correlation maximizing (+0.92) over the central equatorial region, in strong contrast to the fragmented and weak QB pattern (Fig. 8b). The coherence of the Indo-Pacific ENSO correlations is consistent with the fact that a significant fraction of SST variability over the Indian Ocean is related to ENSO (e.g., Kawamura 1998; Klein et al. 1999, Baquero-Bernal et al. 2002; Lau et al. 2005; Terray and Dominiak 2005). Thus, the considerable Ethiopian rainfall increase (decrease) during La Niña (El Niño) events is captured more strongly by the lower-frequency ENSO time scale than by the QB mode.

## 5. Summary and discussion

This paper has investigated the time-scale dependence of the climate system causation of monsoon variability over the Horn of Africa, especially Ethiopia. Wavelet analysis was used to spectrally decompose 1970–99 time series of 5-day averages of Horn of Africa rainfall and regional atmospheric variables, and monthly global SST, and thus identify the coherent modes of variability over the region. The wavelet transform localizes signals in both the frequency and time domains and is well suited to analyze multiscale, nonstationary time series that reflect nonlinear interactions between several physical processes occurring on a range of temporal and spatial scales.

Four major temporal modes are important for Ethiopian monsoon variability and especially for drought: seasonal (75–210 days), annual (210 days–1.42 yr), quasi-biennial (QB; 1.42–3.04 yr), and ENSO (3.04–4.60 yr) time scales. The annual time scale explained 66% of the total Ethiopian pentad rainfall variance, indicating that Ethiopian monsoon rainfall variation largely is associated with annual time-scale circulation patterns involving variability in the major components of the monsoon system. Although the seasonal (12%), QB (5%), and ENSO (2%) time scales have much lower explained variances, they can modify the annual cycle in individual years and significantly affect the total Ethiopian monsoon rainfall variability. QB- and ENSO-related rainfall modulations can be substantial because of their season-long persistence. Thus, the major climate system processes contributing to the annual mode can be augmented or suppressed by seasonal, QB, and ENSO time-scale variability, as shown in Fig. 14 for the wet Ethiopian June–September rainfall phase.

Enhanced annual mode Ethiopian monsoon rainfall (Fig. 14a) is associated with (i) mean sea level pressure (MSLP) intensification (deepening) across the southern Atlantic and Indian Oceans (from West Africa to southwest Asia and over the eastern Mediterranean Sea); (ii) well-developed cross-equatorial southwesterlies from West Africa and the equatorial Atlantic; (iii) a strengthened Somali low-level jet (SLLJ); (iv) an enhanced tropical easterly jet (TEJ); (v) large upper-tropospheric geopotential height increases (decreases) in the northern subtropics (near-equatorial zone); (vi) anomalous surface-to-200-hPa warm (cool) temperatures north of 30°N (through the near-equatorial zone); (vii) anomalously cool SST off the eastern coast of Africa and across the Arabian Sea, equatorial Atlantic, and southwestern Pacific; and (viii) positive SST departures in the East China Sea and southwestern North Atlantic. The opposite climate system anomalies characterize the dry phase of annual mode Ethiopian rainfall.

The seasonal time scale largely acts in phase with the wet annual mode by strengthening the SLLJ and enhancing the lower-tropospheric southwesterlies from the equatorial Atlantic and the Congo rainforest to Ethiopia (Fig. 14b). However, unlike for the wet annual mode, the largest seasonal MSLP deepening occurs over the Arabian Peninsula. This generates seasonal southerlies along the Red Sea that partly counteract the annual mode southwesterlies that extend farther north to northeastern Sudan. Development of a large anomalous seasonal anticyclone from southern Europe to northern Africa also contrasts with the annual mode (Fig. 14b). These wet phase seasonal anomalies were particularly characteristic of 1981 and 1991. Again, the opposite regional anomalies occurred during the dry phase of the seasonal mode, especially in 1984 and 1987.

QB time-scale variability modulates annual circulation anomalies through wet phase Azores or Saharan high intensification and Arabian Peninsula or Indian Ocean MSLP deepening (Fig. 14b). The QB MSLP intensification over the northeast Atlantic and across northern Africa produces northerlies along the Red Sea that promote enhanced Ethiopian moisture convergence, in association with QB monsoon trough deepening over the Arabian Peninsula and strengthened upper-tropospheric easterlies over the Arabian Sea (e.g., 1983, 1988). Consistent with the above enhanced QB moisture convergence, Ethiopian QB rainfall correlated strongly with vertically integrated surface-to-300-hPa water vapor over the Gulf of Aden (+0.80). The QB Arabian Peninsula MSLP deepening is the most northwestern extent of a wet Ethiopian monsoon-phase MSLP deficit that extends from the western Pacific to the southern Indian Ocean. This MSLP deficit further opposes the annual mode Mascarene high MSLP intensification over the southwestern Indian Ocean (Fig. 14b). Dry phase QB climate system anomalies were opposite to the above wet phase QB anomalies (e.g., 1984, 1997).

The wet ENSO time-scale anomaly patterns share many of the wet annual mode regional atmospheric and some SST features, including midlatitude cyclone intrusion over the southeastern Atlantic Ocean, Mascarene high intensification, and cooler Indian Ocean SST (Fig. 14b). The ENSO vertical–zonal circulation also augments its annual mode counterpart, although the ENSO descending motion over the Arabian Sea (60°–70°E) is much stronger. These anomalies were well developed in 1988 and 1989. On the other hand, annual mode SST cooling over the equatorial and southwestern Pacific is opposed by ENSO-related western Pacific warming that enhances Ethiopian rainfall (Fig. 14b). The opposite regional anomalies prevailed during the dry ENSO phase (e.g., 1976, 1987).

This study has shown that Ethiopian interannual rainfall variability is determined by the combined effects of multiscale oscillations of local and regional atmospheric circulation mechanisms and global SST anomaly patterns. These relationships have been used to develop and validate statistical prediction models for Ethiopian seasonal rainfall variability, as reported in the companion paper (Z. Segele et al. 2009, unpublished manuscript).

## Acknowledgments

This research was supported by the NOAA Cooperative Institute for Mesoscale Meteorological Studies (CIMMS) and the International Activities Office of the U.S. National Weather Service through NOAA Grant NA17RJ1227. The computations for this project were performed at the OU Supercomputing Center for Education and Research (OSCER) at The University of Oklahoma. We appreciate the clarifications Dr. Phil Arkin provided concerning the CMAP dataset. The first author received partial financial support from the National Meteorological Agency of Ethiopia during the early part of this research. We thank two anonymous reviewers for their comments, which helped to improve the manuscript.

## REFERENCES

Baliunas, S., P. Frick, D. Sokoloff, and W. Soon, 1997: Time scales and trends in the central England temperature data (1659-1990): A wavelet analysis.

,*Geophys. Res. Lett.***24****,**1351–1354.Baquero-Bernal, A., M. Latif, and S. Legutke, 2002: On dipolelike variability of sea surface temperature in the tropical Indian Ocean.

,*J. Climate***15****,**1358–1368.Barnett, T. P., 1991: The interaction of multiple time scales in the tropical climate system.

,*J. Climate***4****,**269–285.Barnston, A. G., and R. E. Livezey, 1989: A closer look at the effect of the 11-year solar cycle and the quasi-biennial oscillation on Northern Hemisphere 700 mb height and extratropical North American surface temperature.

,*J. Climate***2****,**1295–1313.Barrett, B. S., and L. M. Leslie, 2009: Links between tropical cyclone activity and Madden–Julian oscillation phase in the North Atlantic and Northeast Pacific basins.

,*Mon. Wea. Rev.***127****,**727–744.Beltrando, G., and P. Camberlin, 1993: Interannual variability of rainfall in the eastern Horn of Africa and indicators of atmospheric circulation.

,*Int. J. Climatol.***13****,**533–546.Block, P., and B. Rajagopalan, 2007: Interannual variability and ensemble forecast of Upper Blue Nile basin

*Kiremt*season precipitation.,*J. Hydrometeor.***8****,**327–343.Bowden, J. H., and F. H. M. Semazzi, 2007: Empirical analysis of intraseasonal climate variability over the Greater Horn of Africa.

,*J. Climate***20****,**5715–5731.Camberlin, P., 1997: Rainfall anomalies in the source region of the Nile and their connection with the Indian summer.

,*J. Climate***10****,**1380–1392.Carlstein, E., K-A. Do, P. Hall, T. Hesterberg, and H. R. Künsch, 1998: Matched-block bootstrap for dependent data.

,*Bernoulli***4****,**305–328.Chapa, S. R., V. B. Rao, and G. S. S. D. Prasad, 1998: Application of wavelet transform to Meteosat-derived cold cloud index data over South America.

,*Mon. Wea. Rev.***126****,**2466–2481.Conway, D., 2000: The climate and hydrology of the Upper Blue Nile River.

,*Geogr. J.***166****,**49–62.Davison, A. C., and D. V. Hinkley, 1997:

*Bootstrap Methods and Their Application*. Cambridge University Press, 582 pp.Diro, G. T., E. Black, and D. I. F. Grimes, 2008: Seasonal forecasting of Ethiopian spring rains.

,*Meteor. Appl.***15****,**73–83.Efron, B., and R. J. Tibshirani, 1993:

*An Introduction to the Bootstrap*. Chapman and Hall, 436 pp.Fasullo, J., 2004: Biennial characteristics of Indian monsoon rainfall.

,*J. Climate***17****,**2972–2982.Findlater, J., 1969: A major low-level air current near the Indian Ocean during the northern summer.

,*Quart. J. Roy. Meteor. Soc.***95****,**362–380.Folland, C., J. Owen, M. N. Ward, and A. Colman, 1991: Prediction of seasonal rainfall in the sahel region using empirical and dynamical methods.

,*J. Forecast.***10****,**21–56.Garreaud, R. D., and D. S. Battisti, 1999: Interannual (ENSO) and interdecadal (ENSO-like) variability in the Southern Hemisphere tropospheric circulation.

,*J. Climate***12****,**2113–2123.Giannini, A., R. Saravanan, and P. Chang, 2003: Oceanic forcing of Sahel rainfall on interannual to interdecadal time scales.

,*Science***302****,**1027–1030.Gissila, T., E. Black, D. I. F. Grimes, and J. M. Slingo, 2004: Seasonal forecasting of the Ethiopian summer rains.

,*Int. J. Climatol.***24****,**1345–1358.Hastenrath, S., L. Greischar, and J. van Heerden, 1995: Prediction of summer rainfall over South Africa.

,*J. Climate***8****,**1511–1518.Hoover, K. D., 2003: Nonstationary time series, cointegration, and the principle of the common cause.

,*Br. J. Philos. Sci.***54****,**527–551.Huang, N. E., and Coauthors, 1998: The empirical mode decomposition and the Hilbert Spectrum for nonlinear and non-stationary time series analysis.

,*Proc. Roy. Soc. London***A454****,**903–995.Ju, J., and J. M. Slingo, 1995: The Asian summer monsoon and ENSO.

,*Quart. J. Roy. Meteor. Soc.***121****,**1133–1168.Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project.

,*Bull. Amer. Meteor. Soc.***77****,**437–471.Kawamura, R., 1998: A possible mechanism of the Asian summer monsoon-ENSO coupling.

,*J. Meteor. Soc. Japan***76****,**1009–1027.Kirtman, B. P., and J. Shukla, 2000: Influence of the Indian Ocean summer monsoon on ENSO.

,*Quart. J. Roy. Meteor. Soc.***126****,**213–239.Klein, S. A., B. J. Soden, and N-C. Lau, 1999: Remote sea surface temperature variations during ENSO: Evidence for a tropical atmospheric bridge.

,*J. Climate***12****,**917–932.Korecha, D., and A. G. Barnston, 2007: Predictability of June–September rainfall in Ethiopia.

,*Mon. Wea. Rev.***135****,**628–650.Krishnamurthy, V., and B. N. Goswami, 2000: Indian monsoon–ENSO relationship on interdecadal timescale.

,*J. Climate***13****,**579–595.Lamb, P. J., 1978a: Large-scale tropical Atlantic surface circulation patterns associated with Subsaharan weather anomalies.

,*Tellus***30****,**240–251.Lamb, P. J., 1978b: Case studies of tropical Atlantic surface circulation patterns during recent sub-Saharan weather anomalies: 1967 and 1968.

,*Mon. Wea. Rev.***106****,**482–491.Lamb, P. J., and R. A. Peppler, 1992: Further case studies of tropical Atlantic surface atmospheric and oceanic patterns associated with sub-Saharan drought.

,*J. Climate***5****,**476–488.Lau, K-M., and H. Weng, 1995: Climate signal detection using wavelet transform: How to make a time series sing.

,*Bull. Amer. Meteor. Soc.***76****,**2391–2402.Lau, K-M., and H. T. Wu, 2001: Principal modes of rainfall–SST variability of the Asian summer monsoon: A reassessment of the monsoon–ENSO relationship.

,*J. Climate***14****,**2880–2895.Lau, N-C., and M. J. Nath, 2000: Impact of ENSO on the variability of the Asian–Australian monsoons as simulated in GCM experiments.

,*J. Climate***13****,**4287–4309.Lau, N-C., A. Leetmaa, M. J. Nath, and H-L. Wang, 2005: Influences of ENSO-induced Indo–western Pacific SST anomalies on extratropical atmospheric variability during the boreal summer.

,*J. Climate***18****,**2922–2942.Mason, S. J., and G. M. Mimmack, 1992: The use of bootstrap confidence intervals for the correlation coefficient in climatology.

,*Theor. Appl. Climatol.***45****,**229–233.Mutai, C. C., and M. N. Ward, 2000: East African rainfall and the tropical circulation/convection on intraseasonal to interannual timescales.

,*J. Climate***13****,**3915–3939.Mwale, D., and T. Y. Gan, 2005: Wavelet analysis of variability, teleconnectivity, and predictability of the September–November East African rainfall.

,*J. Appl. Meteor.***44****,**256–269.Nicholls, N., 2001: Commentary and analysis: The insignificance of significance testing.

,*Bull. Amer. Meteor. Soc.***82****,**981–986.Politis, D. N., and H. White, 2004: Automatic block-length selection for the dependent bootstrap.

,*Econ. Rev.***23****,**53–70.Rao, K. G., and B. N. Goswami, 1988: Interannual variations of sea surface temperature over the Arabian Sea and the Indian monsoon: A new perspective.

,*Mon. Wea. Rev.***116****,**558–568.Rasmusson, E. M., and T. H. Carpenter, 1982: Variations in tropical sea surface temperature and surface wind fields associated with the Southern Oscillation/El Niño.

,*Mon. Wea. Rev.***110****,**354–384.Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003: Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century.

,*J. Geophys. Res.***108****,**4407. doi:10.1029/2002JD002670.Riddle, E. E., and K. H. Cook, 2008: Abrupt rainfall transitions over the Greater Horn of Africa: Observations and regional model simulations.

,*J. Geophys. Res.***113****,**D15109. doi:10.1029/2007JD009202.Schmith, T., S. Johansen, and P. Thejll, 2007: Comment on “A semi-empirical approach to projecting future sea-level rise.”.

,*Science***317****,**1866.Segele, Z. T., and P. J. Lamb, 2005: Characterization and variability of Kiremt rainy season over Ethiopia.

,*Meteor. Atmos. Phys.***89****,**153–180.Segele, Z. T., L. M. Leslie, and P. J. Lamb, 2009a: Evaluation and adaptation of a regional climate model for the Horn of Africa: Rainfall climatology and interannual variability.

,*Int. J. Climatol.***29****,**47–65.Segele, Z. T., P. J. Lamb, and L. M. Leslie, 2009b: Large-scale atmospheric circulation and global sea surface temperature associations with Horn of Africa June-September rainfall.

, doi:10.1002/joc.1751, in press.*Int. J. Climatol.*Seleshi, Y., and U. Zanke, 2004: Recent changes in rainfall and rainy days in Ethiopia.

,*Int. J. Climatol.***24****,**973–983.Shanko, D., and P. Camberlin, 1998: The effects of the southwest Indian Ocean tropical cyclones on Ethiopian drought.

,*Int. J. Climatol.***18****,**1373–1388.Shen, S., and K. M. Lau, 1995: Biennial oscillation associated with the East Asian monsoon and tropical sea surface temperatures.

,*J. Meteor. Soc. Japan***73****,**105–124.Srinivas, V. V., and K. Srinivasan, 2005: Matched block bootstrap for resampling multiseason hydrologic time series.

,*Hydrol. Processes***19****,**3659–3682.Tamhane, A. C., and D. D. Dunlop, 2000:

*Statistics and Data Analysis: From Elementary to Intermediate*. Prentice Hall, 722 pp.Terray, P., 1995: Space–time structure of monsoon interannual variability.

,*J. Climate***8****,**2595–2619.Terray, P., and S. Dominiak, 2005: Indian Ocean sea surface temperature and El Niño–Southern Oscillation: A new perspective.

,*J. Climate***18****,**1351–1368.Torrence, C., and G. P. Compo, 1998: A practical guide to wavelet analysis.

,*Bull. Amer. Meteor. Soc.***79****,**61–78.Torrence, C., and P. J. Webster, 1999: Interdecadal changes in the ENSO–monsoon system.

,*J. Climate***12****,**2679–2690.Wang, B., 1995: Interdecadal changes in El Niño onset in the last four decades.

,*J. Climate***8****,**267–285.Wang, B., and Y. Wang, 1996: Temporal structure of the Southern Oscillation as revealed by waveform and wavelet analysis.

,*J. Climate***9****,**1586–1598.Ward, M. N., 1998: Diagnosis and short-lead time prediction of summer rainfall in tropical Africa at interannual and multidecadal timescales.

,*J. Climate***11****,**3167–3191.Webster, P. J., and C. Hoyos, 2004: Prediction of monsoon rainfall and river discharge on 15–30-day time scales.

,*Bull. Amer. Meteor. Soc.***85****,**1745–1765.Webster, P. J., V. O. Magaña, T. N. Palmer, J. Shukla, R. A. Tomas, M. Yanai, and T. Yasunari, 1998: Monsoons: Processes, predictability, and prospects for prediction.

,*J. Geophys. Res.***103****,**14451–14510.Weng, H., and K-M. Lau, 1994: Wavelets, period doubling, and time–frequency localization with application to organization of convection over the tropical western Pacific.

,*J. Atmos. Sci.***51****,**2523–2541.Wilks, D. S., 1997: Resampling hypothesis tests for autocorrelated fields.

,*J. Climate***10****,**65–82.Woodruff, S. D., R. J. Slutz, R. L. Jenne, and P. M. Steurer, 1987: A comprehensive ocean–atmosphere data set.

,*Bull. Amer. Meteor. Soc.***68****,**1239–1250.Xie, P., and P. A. Arkin, 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs.

,*Bull. Amer. Meteor. Soc.***78****,**2539–2558.Yang, S., X. Ding, D. Zheng, and Q. Li, 2007: Depiction of the variations of Great Plains precipitation and its relationship with tropical central-eastern Pacific SST.

,*J. Appl. Meteor. Climatol.***46****,**136–153.

Spectral characteristics of Ethiopian June–September 1970–99 rainfall averaged across stations in Fig. 1 (dots). (a) Time series of 5-day (pentad) mean station rainfall rates. (b) Local wavelet spectra of time series in (a), normalized by 1/*σ*^{2}, where *σ*^{2} = 4.47 (mm d^{−1})^{2} is pentad rainfall variance. Area below thick broken line is the COI. (c) Global wavelet spectra (local power averaged over 1970–99) expressed as percentage of total global power summed over all frequencies. Insets in (c) magnify the major peaks (color coded) at indicated time scales. Here, (a) and (b) use the same abscissa for easy identification of the power at any given time. Thin solid black contours in (b) enclose areas of greater than 95% confidence for a red-noise process estimated from *α* = 0.5(*α*_{1} + *α*_{2}*α*_{1} and *α*_{2} are lag-1 and lag-2 autocorrelations of the time series (Torrence and Compo 1998). Year check marks in (a) and (b) indicate start of first pentad of season.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spectral characteristics of Ethiopian June–September 1970–99 rainfall averaged across stations in Fig. 1 (dots). (a) Time series of 5-day (pentad) mean station rainfall rates. (b) Local wavelet spectra of time series in (a), normalized by 1/*σ*^{2}, where *σ*^{2} = 4.47 (mm d^{−1})^{2} is pentad rainfall variance. Area below thick broken line is the COI. (c) Global wavelet spectra (local power averaged over 1970–99) expressed as percentage of total global power summed over all frequencies. Insets in (c) magnify the major peaks (color coded) at indicated time scales. Here, (a) and (b) use the same abscissa for easy identification of the power at any given time. Thin solid black contours in (b) enclose areas of greater than 95% confidence for a red-noise process estimated from *α* = 0.5(*α*_{1} + *α*_{2}*α*_{1} and *α*_{2} are lag-1 and lag-2 autocorrelations of the time series (Torrence and Compo 1998). Year check marks in (a) and (b) indicate start of first pentad of season.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spectral characteristics of Ethiopian June–September 1970–99 rainfall averaged across stations in Fig. 1 (dots). (a) Time series of 5-day (pentad) mean station rainfall rates. (b) Local wavelet spectra of time series in (a), normalized by 1/*σ*^{2}, where *σ*^{2} = 4.47 (mm d^{−1})^{2} is pentad rainfall variance. Area below thick broken line is the COI. (c) Global wavelet spectra (local power averaged over 1970–99) expressed as percentage of total global power summed over all frequencies. Insets in (c) magnify the major peaks (color coded) at indicated time scales. Here, (a) and (b) use the same abscissa for easy identification of the power at any given time. Thin solid black contours in (b) enclose areas of greater than 95% confidence for a red-noise process estimated from *α* = 0.5(*α*_{1} + *α*_{2}*α*_{1} and *α*_{2} are lag-1 and lag-2 autocorrelations of the time series (Torrence and Compo 1998). Year check marks in (a) and (b) indicate start of first pentad of season.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

As in Fig. 2, but for pentad CMAP rainfall estimates for Greater Horn of Africa for 1979–99 averaged over crosses in Fig. 1. Variance of time series in (a) is 2.3 (mm d^{−1})^{2}. Here, (b) and (c) use same ordinate for easy identification of power at any period. Thin solid contours in (b) enclose area of greater than 95% confidence for a red-noise process estimated from *α* = 0.5(*α*_{1} + *α*_{2}*α*_{1} and *α*_{2} are lag-1 and lag-2 autocorrelations of pentad CMAP rainfall.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

As in Fig. 2, but for pentad CMAP rainfall estimates for Greater Horn of Africa for 1979–99 averaged over crosses in Fig. 1. Variance of time series in (a) is 2.3 (mm d^{−1})^{2}. Here, (b) and (c) use same ordinate for easy identification of power at any period. Thin solid contours in (b) enclose area of greater than 95% confidence for a red-noise process estimated from *α* = 0.5(*α*_{1} + *α*_{2}*α*_{1} and *α*_{2} are lag-1 and lag-2 autocorrelations of pentad CMAP rainfall.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

As in Fig. 2, but for pentad CMAP rainfall estimates for Greater Horn of Africa for 1979–99 averaged over crosses in Fig. 1. Variance of time series in (a) is 2.3 (mm d^{−1})^{2}. Here, (b) and (c) use same ordinate for easy identification of power at any period. Thin solid contours in (b) enclose area of greater than 95% confidence for a red-noise process estimated from *α* = 0.5(*α*_{1} + *α*_{2}*α*_{1} and *α*_{2} are lag-1 and lag-2 autocorrelations of pentad CMAP rainfall.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Time series of pentad June–September Ethiopian wavelet-filtered rainfall anomalies (mm d^{−1}) for 1970–99. The nonoverlapping frequency bands shown contain the major power peaks and were delineated from the global power spectra in Fig. 2c. Year check marks indicate start of first pentad of season.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Time series of pentad June–September Ethiopian wavelet-filtered rainfall anomalies (mm d^{−1}) for 1970–99. The nonoverlapping frequency bands shown contain the major power peaks and were delineated from the global power spectra in Fig. 2c. Year check marks indicate start of first pentad of season.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Time series of pentad June–September Ethiopian wavelet-filtered rainfall anomalies (mm d^{−1}) for 1970–99. The nonoverlapping frequency bands shown contain the major power peaks and were delineated from the global power spectra in Fig. 2c. Year check marks indicate start of first pentad of season.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spatial correlation patterns and WET − DRY composite difference maps for 75–210 days wavelet-filtered June–September 1970–99 pentad time series. Correlation between filtered Ethiopian pentad rain gauge averages (Fig. 1; dots) and filtered pentad gridpoint (a) MSLP (contours) and 700-hPa horizontal wind (vectors) and (b) 200-hPa geopotential height (contours) and 150-hPa horizontal wind (vectors). (c) WET − DRY composite difference fields for MSLP (hPa; contours) and 700-hPa horizontal wind (m s^{−1}; vectors). (d) As in (c), but for 200-hPa temperature (K; contours) and 150-hPa horizontal wind (vectors). (top) Correlations between rainfall and horizontal wind components are shown as vectors. Unit vector in the bottom right corner of each panel represents a correlation magnitude of 1.0; arrows point eastward (westward) for positive (negative) zonal wind correlations and northward (southward) for positive (negative) meridional wind correlations. Vectors are shown when either the (top) zonal or meridional correlations and (bottom) wind differences are significant at the ≤5% level according to the (top) matched-block bootstrap and (bottom) Student’s *t* tests, while shading in top (bottom) depicts correlation (difference) values significant at the 5% level according to the matched-block bootstrap (Student’s *t*) test for the contoured fields (section 2c). (bottom) Compositing is based on the driest and wettest 2% of 75–210 days wavelet-filtered June–September 1970–99 Ethiopian pentad rain gauge averages. Reference vectors in bottom-right corners of (c),(d) are in meters per second. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spatial correlation patterns and WET − DRY composite difference maps for 75–210 days wavelet-filtered June–September 1970–99 pentad time series. Correlation between filtered Ethiopian pentad rain gauge averages (Fig. 1; dots) and filtered pentad gridpoint (a) MSLP (contours) and 700-hPa horizontal wind (vectors) and (b) 200-hPa geopotential height (contours) and 150-hPa horizontal wind (vectors). (c) WET − DRY composite difference fields for MSLP (hPa; contours) and 700-hPa horizontal wind (m s^{−1}; vectors). (d) As in (c), but for 200-hPa temperature (K; contours) and 150-hPa horizontal wind (vectors). (top) Correlations between rainfall and horizontal wind components are shown as vectors. Unit vector in the bottom right corner of each panel represents a correlation magnitude of 1.0; arrows point eastward (westward) for positive (negative) zonal wind correlations and northward (southward) for positive (negative) meridional wind correlations. Vectors are shown when either the (top) zonal or meridional correlations and (bottom) wind differences are significant at the ≤5% level according to the (top) matched-block bootstrap and (bottom) Student’s *t* tests, while shading in top (bottom) depicts correlation (difference) values significant at the 5% level according to the matched-block bootstrap (Student’s *t*) test for the contoured fields (section 2c). (bottom) Compositing is based on the driest and wettest 2% of 75–210 days wavelet-filtered June–September 1970–99 Ethiopian pentad rain gauge averages. Reference vectors in bottom-right corners of (c),(d) are in meters per second. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spatial correlation patterns and WET − DRY composite difference maps for 75–210 days wavelet-filtered June–September 1970–99 pentad time series. Correlation between filtered Ethiopian pentad rain gauge averages (Fig. 1; dots) and filtered pentad gridpoint (a) MSLP (contours) and 700-hPa horizontal wind (vectors) and (b) 200-hPa geopotential height (contours) and 150-hPa horizontal wind (vectors). (c) WET − DRY composite difference fields for MSLP (hPa; contours) and 700-hPa horizontal wind (m s^{−1}; vectors). (d) As in (c), but for 200-hPa temperature (K; contours) and 150-hPa horizontal wind (vectors). (top) Correlations between rainfall and horizontal wind components are shown as vectors. Unit vector in the bottom right corner of each panel represents a correlation magnitude of 1.0; arrows point eastward (westward) for positive (negative) zonal wind correlations and northward (southward) for positive (negative) meridional wind correlations. Vectors are shown when either the (top) zonal or meridional correlations and (bottom) wind differences are significant at the ≤5% level according to the (top) matched-block bootstrap and (bottom) Student’s *t* tests, while shading in top (bottom) depicts correlation (difference) values significant at the 5% level according to the matched-block bootstrap (Student’s *t*) test for the contoured fields (section 2c). (bottom) Compositing is based on the driest and wettest 2% of 75–210 days wavelet-filtered June–September 1970–99 Ethiopian pentad rain gauge averages. Reference vectors in bottom-right corners of (c),(d) are in meters per second. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Regression and correlation of 210 days–1.42 yr wavelet-filtered June–September 1970–99 pentad time series. (a) Correlation between Ethiopian pentad rain gauge averages (Fig. 1; dots) and pentad gridpoint MSLP (contours), and regression of 700-hPa horizontal wind components against the same rainfall (vectors; m s^{−1}). (b) As in (a), but for 200-hPa temperature (contours) and horizontal wind components (vectors). (c) Latitude–height cross section of correlation between Ethiopian pentad rain gauge averages and gridpoint geopotential height averaged over 35°–45°E. (d) As in (c), but for temperature. (a),(b) Regression vectors indicate wind direction and magnitude for each millimeter of positive pentad rainfall anomaly and are depicted when either the zonal or meridional regression coefficients are statistically significant with an ASL of ≤5% according to the matched-block bootstrap test (section 2c). Reference vectors in bottom-right corners of (a),(b) are in meters per second. Light gray shading depicts correlation values with ≤5% ASL according to same test for the contoured fields. Ethiopian boundary is bold black line. (c),(d) Correlation averages are square roots of spatial averages of coefficients of determination *R*^{2}, with their signs being specified according to the average correlation for 35°–45°E. Solid circles show observed surface pressure for Addis Ababa (central Ethiopia). Dark gray shading at bottom depicts average orography estimated from mean surface pressure for June–September. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Regression and correlation of 210 days–1.42 yr wavelet-filtered June–September 1970–99 pentad time series. (a) Correlation between Ethiopian pentad rain gauge averages (Fig. 1; dots) and pentad gridpoint MSLP (contours), and regression of 700-hPa horizontal wind components against the same rainfall (vectors; m s^{−1}). (b) As in (a), but for 200-hPa temperature (contours) and horizontal wind components (vectors). (c) Latitude–height cross section of correlation between Ethiopian pentad rain gauge averages and gridpoint geopotential height averaged over 35°–45°E. (d) As in (c), but for temperature. (a),(b) Regression vectors indicate wind direction and magnitude for each millimeter of positive pentad rainfall anomaly and are depicted when either the zonal or meridional regression coefficients are statistically significant with an ASL of ≤5% according to the matched-block bootstrap test (section 2c). Reference vectors in bottom-right corners of (a),(b) are in meters per second. Light gray shading depicts correlation values with ≤5% ASL according to same test for the contoured fields. Ethiopian boundary is bold black line. (c),(d) Correlation averages are square roots of spatial averages of coefficients of determination *R*^{2}, with their signs being specified according to the average correlation for 35°–45°E. Solid circles show observed surface pressure for Addis Ababa (central Ethiopia). Dark gray shading at bottom depicts average orography estimated from mean surface pressure for June–September. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Regression and correlation of 210 days–1.42 yr wavelet-filtered June–September 1970–99 pentad time series. (a) Correlation between Ethiopian pentad rain gauge averages (Fig. 1; dots) and pentad gridpoint MSLP (contours), and regression of 700-hPa horizontal wind components against the same rainfall (vectors; m s^{−1}). (b) As in (a), but for 200-hPa temperature (contours) and horizontal wind components (vectors). (c) Latitude–height cross section of correlation between Ethiopian pentad rain gauge averages and gridpoint geopotential height averaged over 35°–45°E. (d) As in (c), but for temperature. (a),(b) Regression vectors indicate wind direction and magnitude for each millimeter of positive pentad rainfall anomaly and are depicted when either the zonal or meridional regression coefficients are statistically significant with an ASL of ≤5% according to the matched-block bootstrap test (section 2c). Reference vectors in bottom-right corners of (a),(b) are in meters per second. Light gray shading depicts correlation values with ≤5% ASL according to same test for the contoured fields. Ethiopian boundary is bold black line. (c),(d) Correlation averages are square roots of spatial averages of coefficients of determination *R*^{2}, with their signs being specified according to the average correlation for 35°–45°E. Solid circles show observed surface pressure for Addis Ababa (central Ethiopia). Dark gray shading at bottom depicts average orography estimated from mean surface pressure for June–September. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

WET − DRY composite difference maps for 210 days–1.42 yr wavelet-filtered June-September 1970–99 pentad time series. (a) MSLP (hPa; contours) and 700-hPa horizontal wind (m s^{−1}; vectors); (b) 150-hPa geopotential height (gpm; contours) and horizontal wind (vectors). Wind arrows in bottom right corner of each panel indicate direction of vector difference (length proportional to magnitude) and are shown when either the zonal or meridional wind differences are statistically significant at the 5% level according to a two-tailed Students’ *t* test (section 2c). Shading indicates parts of contoured fields with above level of significance for same test. See section 4b for description of compositing procedure. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

WET − DRY composite difference maps for 210 days–1.42 yr wavelet-filtered June-September 1970–99 pentad time series. (a) MSLP (hPa; contours) and 700-hPa horizontal wind (m s^{−1}; vectors); (b) 150-hPa geopotential height (gpm; contours) and horizontal wind (vectors). Wind arrows in bottom right corner of each panel indicate direction of vector difference (length proportional to magnitude) and are shown when either the zonal or meridional wind differences are statistically significant at the 5% level according to a two-tailed Students’ *t* test (section 2c). Shading indicates parts of contoured fields with above level of significance for same test. See section 4b for description of compositing procedure. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

WET − DRY composite difference maps for 210 days–1.42 yr wavelet-filtered June-September 1970–99 pentad time series. (a) MSLP (hPa; contours) and 700-hPa horizontal wind (m s^{−1}; vectors); (b) 150-hPa geopotential height (gpm; contours) and horizontal wind (vectors). Wind arrows in bottom right corner of each panel indicate direction of vector difference (length proportional to magnitude) and are shown when either the zonal or meridional wind differences are statistically significant at the 5% level according to a two-tailed Students’ *t* test (section 2c). Shading indicates parts of contoured fields with above level of significance for same test. See section 4b for description of compositing procedure. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Patterns of concurrent correlation between June–September 1970–99 time series of wavelet-filtered monthly anomalies of Ethiopian station rain gauge averages (Fig. 1; dots) and gridpoint SSTs for (a) annual (232 days–1.46 yr), (b) QB (1.46–3.13 yr), and (c) ENSO (3.13–4.43 yr) time scales (Table 1). Thick red lines delineate correlation values with an ASL of ≤5% according to the matched-block bootstrap test (section 2c). Solid green line denotes border of Ethiopia.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Patterns of concurrent correlation between June–September 1970–99 time series of wavelet-filtered monthly anomalies of Ethiopian station rain gauge averages (Fig. 1; dots) and gridpoint SSTs for (a) annual (232 days–1.46 yr), (b) QB (1.46–3.13 yr), and (c) ENSO (3.13–4.43 yr) time scales (Table 1). Thick red lines delineate correlation values with an ASL of ≤5% according to the matched-block bootstrap test (section 2c). Solid green line denotes border of Ethiopia.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Patterns of concurrent correlation between June–September 1970–99 time series of wavelet-filtered monthly anomalies of Ethiopian station rain gauge averages (Fig. 1; dots) and gridpoint SSTs for (a) annual (232 days–1.46 yr), (b) QB (1.46–3.13 yr), and (c) ENSO (3.13–4.43 yr) time scales (Table 1). Thick red lines delineate correlation values with an ASL of ≤5% according to the matched-block bootstrap test (section 2c). Solid green line denotes border of Ethiopia.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spatial regression and correlation patterns for QB frequency (1.42–3.04 yr) wavelet-filtered June–September 1970–99 pentad time series. (a) Regional correlation between Ethiopian pentad rain gauge averages (Fig. 1; dots) and pentad gridpoint (contours) and regression of 700-hPa horizontal wind components against the same rainfall (vectors; m s^{−1}). (b) Same as (a) except for 150-hPa geopotential height (contours) and horizontal wind components (vectors). (c) Correlation between Ethiopian pentad rain gauge averages and pentad gridpoint MSLP (contours) and regression of 1000-hPa horizontal wind components against the same rainfall (vectors) across the tropical Indian and Pacific Oceans. (d) Same as (c) except for 200-hPa geopotential height (contours) and horizontal wind components (vectors). Regression vectors indicate wind direction and magnitude for each millimeter of positive pentad rainfall anomaly and are depicted when either the zonal or meridional regression coefficients are statistically significant with an ASL of ≤5% according to the matched-block bootstrap test (section 2c). (a)–(d) (bottom right) Reference vectors are in meters per second. Light gray shading depicts correlation values with ≤5% ASL according to same test for the contoured fields. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spatial regression and correlation patterns for QB frequency (1.42–3.04 yr) wavelet-filtered June–September 1970–99 pentad time series. (a) Regional correlation between Ethiopian pentad rain gauge averages (Fig. 1; dots) and pentad gridpoint (contours) and regression of 700-hPa horizontal wind components against the same rainfall (vectors; m s^{−1}). (b) Same as (a) except for 150-hPa geopotential height (contours) and horizontal wind components (vectors). (c) Correlation between Ethiopian pentad rain gauge averages and pentad gridpoint MSLP (contours) and regression of 1000-hPa horizontal wind components against the same rainfall (vectors) across the tropical Indian and Pacific Oceans. (d) Same as (c) except for 200-hPa geopotential height (contours) and horizontal wind components (vectors). Regression vectors indicate wind direction and magnitude for each millimeter of positive pentad rainfall anomaly and are depicted when either the zonal or meridional regression coefficients are statistically significant with an ASL of ≤5% according to the matched-block bootstrap test (section 2c). (a)–(d) (bottom right) Reference vectors are in meters per second. Light gray shading depicts correlation values with ≤5% ASL according to same test for the contoured fields. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spatial regression and correlation patterns for QB frequency (1.42–3.04 yr) wavelet-filtered June–September 1970–99 pentad time series. (a) Regional correlation between Ethiopian pentad rain gauge averages (Fig. 1; dots) and pentad gridpoint (contours) and regression of 700-hPa horizontal wind components against the same rainfall (vectors; m s^{−1}). (b) Same as (a) except for 150-hPa geopotential height (contours) and horizontal wind components (vectors). (c) Correlation between Ethiopian pentad rain gauge averages and pentad gridpoint MSLP (contours) and regression of 1000-hPa horizontal wind components against the same rainfall (vectors) across the tropical Indian and Pacific Oceans. (d) Same as (c) except for 200-hPa geopotential height (contours) and horizontal wind components (vectors). Regression vectors indicate wind direction and magnitude for each millimeter of positive pentad rainfall anomaly and are depicted when either the zonal or meridional regression coefficients are statistically significant with an ASL of ≤5% according to the matched-block bootstrap test (section 2c). (a)–(d) (bottom right) Reference vectors are in meters per second. Light gray shading depicts correlation values with ≤5% ASL according to same test for the contoured fields. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

WET − DRY composite difference maps for QB frequency (1.42–3.04 yr) wavelet-filtered June–September 1970–99 pentad time series. (a) MSLP (hPa; contours) and 700-hPa horizontal wind (vectors; m s^{−1}). (b) 150-hPa geopotential height (contours; gpm) and horizontal wind (vectors). (c) MSLP (contours) and 1000-hPa horizontal wind (vectors) across the tropical Indian and Pacific Oceans. (d) Same as (c) except for 200-hPa geopotential height (contours) and horizontal wind (vectors). (a)–(d) Wind arrows indicate direction of vector difference (length proportional to magnitude) and are shown when either the zonal or meridional wind differences are statistically significant at the 5% level according to a two-tailed Students’ *t* test (section 2c). Shading indicates parts of contoured fields with the above level of significance for the same test. Compositing is based on the driest and wettest 2% wavelet-filtered June–September 1970–99 Ethiopian pentad rain gauge averages (section 4c). Reference vectors in bottom-right corners of (a)–(d) are in meters per second. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

WET − DRY composite difference maps for QB frequency (1.42–3.04 yr) wavelet-filtered June–September 1970–99 pentad time series. (a) MSLP (hPa; contours) and 700-hPa horizontal wind (vectors; m s^{−1}). (b) 150-hPa geopotential height (contours; gpm) and horizontal wind (vectors). (c) MSLP (contours) and 1000-hPa horizontal wind (vectors) across the tropical Indian and Pacific Oceans. (d) Same as (c) except for 200-hPa geopotential height (contours) and horizontal wind (vectors). (a)–(d) Wind arrows indicate direction of vector difference (length proportional to magnitude) and are shown when either the zonal or meridional wind differences are statistically significant at the 5% level according to a two-tailed Students’ *t* test (section 2c). Shading indicates parts of contoured fields with the above level of significance for the same test. Compositing is based on the driest and wettest 2% wavelet-filtered June–September 1970–99 Ethiopian pentad rain gauge averages (section 4c). Reference vectors in bottom-right corners of (a)–(d) are in meters per second. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

WET − DRY composite difference maps for QB frequency (1.42–3.04 yr) wavelet-filtered June–September 1970–99 pentad time series. (a) MSLP (hPa; contours) and 700-hPa horizontal wind (vectors; m s^{−1}). (b) 150-hPa geopotential height (contours; gpm) and horizontal wind (vectors). (c) MSLP (contours) and 1000-hPa horizontal wind (vectors) across the tropical Indian and Pacific Oceans. (d) Same as (c) except for 200-hPa geopotential height (contours) and horizontal wind (vectors). (a)–(d) Wind arrows indicate direction of vector difference (length proportional to magnitude) and are shown when either the zonal or meridional wind differences are statistically significant at the 5% level according to a two-tailed Students’ *t* test (section 2c). Shading indicates parts of contoured fields with the above level of significance for the same test. Compositing is based on the driest and wettest 2% wavelet-filtered June–September 1970–99 Ethiopian pentad rain gauge averages (section 4c). Reference vectors in bottom-right corners of (a)–(d) are in meters per second. Bold black line denotes Ethiopian boundary.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Latitude–height cross section of regression patterns of wavelet-filtered June–September 1970–99 pentad time series averaged over 30°–50°E for the QB time scale (1.42–3.04/1.32–3.74 yr; Table 1). Regression of meridional wind (*υ*) and negative vertical velocity (−*ω*) against (a) Ethiopian pentad rain gauge averages (Fig. 1; dots) and (b) Horn of Africa pentad rainfall estimates (Fig. 1; crosses). Horizontal wind anomalies are in meters per scond (standard deviation)^{−1}, and vertical (pressure) velocity anomalies are in pascal per second (standard deviation)^{−1}. Contours show actual regression coefficient for pressure vertical velocity anomalies (−*ω*, Pa s^{−1}). See text for construction and interpretation of wind vector anomalies (scale under bottom right of each panel). Vertical dashed lines mark the bounding latitudes for the Horn of Africa. Solid circles show observed surface pressure for Addis Ababa (central Ethiopia; Fig. 1). Shading at bottom depicts average elevation profile estimated from mean NCAR/NCEP surface pressure reanalysis for June–September 1970–99.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Latitude–height cross section of regression patterns of wavelet-filtered June–September 1970–99 pentad time series averaged over 30°–50°E for the QB time scale (1.42–3.04/1.32–3.74 yr; Table 1). Regression of meridional wind (*υ*) and negative vertical velocity (−*ω*) against (a) Ethiopian pentad rain gauge averages (Fig. 1; dots) and (b) Horn of Africa pentad rainfall estimates (Fig. 1; crosses). Horizontal wind anomalies are in meters per scond (standard deviation)^{−1}, and vertical (pressure) velocity anomalies are in pascal per second (standard deviation)^{−1}. Contours show actual regression coefficient for pressure vertical velocity anomalies (−*ω*, Pa s^{−1}). See text for construction and interpretation of wind vector anomalies (scale under bottom right of each panel). Vertical dashed lines mark the bounding latitudes for the Horn of Africa. Solid circles show observed surface pressure for Addis Ababa (central Ethiopia; Fig. 1). Shading at bottom depicts average elevation profile estimated from mean NCAR/NCEP surface pressure reanalysis for June–September 1970–99.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Latitude–height cross section of regression patterns of wavelet-filtered June–September 1970–99 pentad time series averaged over 30°–50°E for the QB time scale (1.42–3.04/1.32–3.74 yr; Table 1). Regression of meridional wind (*υ*) and negative vertical velocity (−*ω*) against (a) Ethiopian pentad rain gauge averages (Fig. 1; dots) and (b) Horn of Africa pentad rainfall estimates (Fig. 1; crosses). Horizontal wind anomalies are in meters per scond (standard deviation)^{−1}, and vertical (pressure) velocity anomalies are in pascal per second (standard deviation)^{−1}. Contours show actual regression coefficient for pressure vertical velocity anomalies (−*ω*, Pa s^{−1}). See text for construction and interpretation of wind vector anomalies (scale under bottom right of each panel). Vertical dashed lines mark the bounding latitudes for the Horn of Africa. Solid circles show observed surface pressure for Addis Ababa (central Ethiopia; Fig. 1). Shading at bottom depicts average elevation profile estimated from mean NCAR/NCEP surface pressure reanalysis for June–September 1970–99.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spatial regression and correlation patterns for lower-frequency ENSO (3.04–4.60 yr) wavelet-filtered (Table 1) June–September 1970–99 pentad time series. Details same as for Fig. 9.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spatial regression and correlation patterns for lower-frequency ENSO (3.04–4.60 yr) wavelet-filtered (Table 1) June–September 1970–99 pentad time series. Details same as for Fig. 9.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spatial regression and correlation patterns for lower-frequency ENSO (3.04–4.60 yr) wavelet-filtered (Table 1) June–September 1970–99 pentad time series. Details same as for Fig. 9.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

WET − DRY composite difference maps for lower-frequency ENSO (3.04–4.60 yr) wavelet-filtered (Table 1) June–September 1970–99 pentad time series. Details same as for Fig. 10, with section 4d providing more information on compositing procedure.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

WET − DRY composite difference maps for lower-frequency ENSO (3.04–4.60 yr) wavelet-filtered (Table 1) June–September 1970–99 pentad time series. Details same as for Fig. 10, with section 4d providing more information on compositing procedure.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

WET − DRY composite difference maps for lower-frequency ENSO (3.04–4.60 yr) wavelet-filtered (Table 1) June–September 1970–99 pentad time series. Details same as for Fig. 10, with section 4d providing more information on compositing procedure.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Schematic diagram showing the major large-scale atmospheric and SST anomaly patterns associated with the wet Ethiopian June–September rainfall phase. (a) Annual time-scale anomalies, where pink (blue) shading locates warm (cold) SSTs, MSLP intensification/deepening or 200-hPa geopotential height increase (decrease) indicated by H (L), 200-hPa warm (cool) temperatures indicated by W (C), southwest monsoon is denoted as SWM. Also shown are TEJ and SLLJ. (b) Seasonal (purple), QB (yellow), and ENSO (green) enhancements and modifications to above annual anomalies, where broken (solid) black lines enclose QB (ENSO) SST anomalies. Anomalies for the dry Ethiopian rainfall phase generally are opposite to those for the wet phase. See Table 1 for time-scale stratification.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Schematic diagram showing the major large-scale atmospheric and SST anomaly patterns associated with the wet Ethiopian June–September rainfall phase. (a) Annual time-scale anomalies, where pink (blue) shading locates warm (cold) SSTs, MSLP intensification/deepening or 200-hPa geopotential height increase (decrease) indicated by H (L), 200-hPa warm (cool) temperatures indicated by W (C), southwest monsoon is denoted as SWM. Also shown are TEJ and SLLJ. (b) Seasonal (purple), QB (yellow), and ENSO (green) enhancements and modifications to above annual anomalies, where broken (solid) black lines enclose QB (ENSO) SST anomalies. Anomalies for the dry Ethiopian rainfall phase generally are opposite to those for the wet phase. See Table 1 for time-scale stratification.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Schematic diagram showing the major large-scale atmospheric and SST anomaly patterns associated with the wet Ethiopian June–September rainfall phase. (a) Annual time-scale anomalies, where pink (blue) shading locates warm (cold) SSTs, MSLP intensification/deepening or 200-hPa geopotential height increase (decrease) indicated by H (L), 200-hPa warm (cool) temperatures indicated by W (C), southwest monsoon is denoted as SWM. Also shown are TEJ and SLLJ. (b) Seasonal (purple), QB (yellow), and ENSO (green) enhancements and modifications to above annual anomalies, where broken (solid) black lines enclose QB (ENSO) SST anomalies. Anomalies for the dry Ethiopian rainfall phase generally are opposite to those for the wet phase. See Table 1 for time-scale stratification.

Citation: Journal of Climate 22, 12; 10.1175/2008JCLI2859.1

Spectral bands for June–September pentad and monthly Ethiopian rain gauge (Horn of Africa CMAP rainfall estimates) for 1970–99 (1979–99). These nonoverlapping frequency bands were delineated from the global power spectra in Fig. 2c (Ethiopian pentad rain gauge) and Fig. 3c (Horn of Africa CMAP rainfall estimates) and for monthly Ethiopian rain gauge time series (not shown). Percentages in parentheses give temporal variance fractions accounted for by each mode.

Summary statistics and 95% confidence intervals for annual time-scale (210 days–1.42 yr wavelet filtered) correlations at selected grid points between Ethiopian pentad rainfall and gridpoint MSLP for 1970–99. Confidence intervals were computed by applying the bootstrap BC* _{a}* method to 1000 rainfall–gridpoint MSLP correlation replicates obtained by matched-block bootstrapping the wavelet-filtered Ethiopian pentad rainfall (section 2c). The mean and standard error columns under the bootstrap heading contain the average and standard deviation of 1000 matched-block bootstrap correlation replicates, while the bias column gives the average deviation of the observed correlation from the above 1000 sampled values.

Summary statistics and 95% confidence intervals for annual time-scale (210 days-1.42 yr wavelet-filtered) correlations at selected grid points and tropospheric levels between Ethiopian pentad rainfall averages and gridpoint tropospheric zonal wind component for 1970–99. Confidence intervals were computed by applying the bootstrap BC* _{a}* method to 1000 rainfall–zonal wind correlation replicates obtained by matched-block bootstrapping the wavelet-filtered Ethiopian pentad rainfall averages (section 2c). The mean and standard error columns under the bootstrap heading contain the average and standard deviation of the 1000 matched-block bootstrap correlation replicates, while the bias column gives the average deviation of the observed correlation from the above 1000 sampled values. Correlations that are not statistically significant at the 5% level are bounded by parentheses, and those with marginal statistical significance are in italics.

As in Table 3, but for the meridional wind component.

As in Table 3, but for temperature.

Same as Table 3 except for geopotential height.