1. Introduction

Rainfall in East Africa mostly occurs during the boreal spring (long rains, March–May) and autumn (short rains, September/October–December) seasons as the intertropical convergence zone (ITCZ) migrates through the equator from south to north, and vice versa. Like other tropical regions, interannual variability of rainfall in East Africa results from complex interactions of forced and free atmospheric variations. These include interactions between sea surface temperature (SST) forcing, large-scale atmospheric patterns, and synoptic-scale weather disturbances including monsoon and trade winds, persistent mesoscale circulations, tropical cyclones, subtropical anticyclones, easterly/westerly wave perturbations, and extratropical weather systems.

Relative to the long rains, the short rains tend to have stronger interannual variability, stronger spatial coherence of rainfall anomalies across a large part of the region, and a substantial association with El Niño–Southern Oscillation (ENSO) (e.g., Ropelewski and Halpert 1987, 1996; Ogallo 1988; Ogallo et al. 1988; Beltrando 1990; Hastenrath et al. 1993; Nicholson and Kim 1997). These studies described that warm events of ENSO tend to be associated with above-average rainfall and cold events of ENSO are associated with below-average rainfall during the short rains. However, much remains to be understood about the global, regional, and local circulation mechanisms that generate rainfall anomalies in East Africa and how they are connected to ENSO. Hastenrath et al. (1993) showed a particularly close association between near-surface zonal wind anomalies in the equatorial Indian Ocean and East African rainfall, an association stronger than that with ENSO. There is a need to assess if such an enhanced correlation results from a local response to increased East African rainfall, or whether it is part of a large-scale ocean–atmosphere teleconnection structure partially independent of ENSO. Indeed, some severe droughts and floods observed over parts of East Africa appear to have had very little to do with ENSO. Some of the local circulation features associated with droughts and floods have been studied, emphasizing the role of local near-surface westerly wind anomalies in above-normal rainfall seasons (Nakamura 1969; Anyamba 1984; Davies et al. 1985) and intraseasonal wet spells (Camberlin and Wairoto 1997; Camberlin 1997; Okoola 1999). However, how these features relate to structures in the global Tropics on intraseasonal to interannual timescales is not known and is addressed in this paper.

Anomalous SST in the Atlantic and Indian Oceans has also been associated with rainfall fluctuations in eastern Africa (Reverdin et al. 1986; Nicholson and Entekhabi 1987). Predictive schemes based solely on ENSO do show skill in the region (Farmer 1988; Hutchinson 1992), but two of the three predictors used in Mutai et al. (1998) were largely independent of traditional ENSO measures, including aspects of SST variability in the northwestern tropical Pacific–Indian Ocean and in the equatorial–tropical South Atlantic. There is a need to diagnose the teleconnection circulation structures associated with these SST patterns as well as ENSO, to gain a better understanding of the SST–rainfall links, and to underpin confidence in prediction schemes based on SST forcing.

This paper is sensitive to the issues of temporal and spatial scales of climate anomalies. Traditionally, climate research in Africa has tended to focus on seasonal mean area-average rainfall totals, guided partly by available data and partly by a desire to diagnose the large-scale relations as a first step. However, it is now clear that for a full understanding and a maximum predictive utility, there is a need to study spatial and temporal variability within a season. Inspection of daily rainfall sequences for Kenya reveals a tendency for a season to be made up of a few isolated rainfall events combined with a few more major rainfall spells (5–10-day duration), punctuated by dry spells. As a first step toward better understanding the nature and role of the major rainfall spells, this paper diagnoses the global Tropics circulation before, during, and after the rainfall spells. The primary aim is to assess how the intraseasonal teleconnection structure relates to the seasonal atmospheric teleconnection structure. A secondary aim is to assess if there is any potential to foresee the likelihood of major rainfall spells a week or two in advance by monitoring circulation in the global Tropics (e.g., Waliser et al. 1999). The study is undertaken against the backdrop of the known strongest signal at the intraseasonal timescale in the Tropics, the Madden–Julian oscillation (MJO; Madden and Julian 1971, 1994), which propagates eastward around the global Tropics with a spectral signature of around 30–60 days, rather shorter and more coherent in atmospheric circulation and rather broader and longer in tropical convection (Salby and Hendon 1994). Thus, this paper generates some unique lead–lag results by focusing on the coherent short rains season over East Africa (centered on October–November) to construct intraseasonal global Tropics signals, at a time of the year when most convection is close to the equator from Africa to the western Pacific.

The East African region contains some of the most varied topography in the world, including large lakes, rift valleys, and snow-capped mountains on the equator. This heterogeneity gives rise to dramatic variations in climatological mean rainfall totals. In addition, seasonal rainfall anomalies can tend to have a coherence that is confined across small subregions. It is postulated that some of this is generated by interaction of the large-scale regional atmospheric forcing with the topography (Sun et al. 1999a,b). It is therefore possible for large-scale atmospheric changes associated with, for example, ENSO to have a stronger signal in some parts of the region. This is very difficult to prove statistically and likely requires a combination of statistics with regional numerical modeling. Nonetheless, we provide in this paper an indication of the extent to which the large-scale forcing is manifest in the observed rainfall anomalies of smaller subregions. For this, we use the eight homogeneous rainfall anomaly subregions defined recently by Indeje et al. (1999) on the basis of empirical orthogonal function (EOF) and point correlation analysis. These results are intended to serve as benchmarks for comparison with numerical model results and provide a practical indication of the extent to which teleconnections with SST are reflected in local regions.

The fundamental goal of this study is to improve existing climate monitoring and forecasting in East Africa. First, the study addresses the representativeness of the predictability of the large-scale area average at smaller spatial scales. Second, the study diagnoses atmospheric circulation structures associated with ENSO and non-ENSO variability, with a view to establishing the physical basis for remote teleconnections with SST and therefore improving reliability and confidence in SST-based prediction schemes. Third, the paper aims to make a contrast with the teleconnection structures associated with the long rains, to provide a basis for why the ENSO links in that season are weaker. Finally, the study takes a first look at the role of extended wet spells in East Africa in the seasonal rainfall anomalies and associated teleconnection structures. This is intended to shed light on the potential for anticipating intraseasonal rainfall events in East Africa.

In section 2, the data sources are given in more detail, along with an outline of analysis procedures. Section 3 focuses on the monthly/seasonal East African rainfall and tropical circulation/convection anomalies including structures that are independent of ENSO. In section 4, the intraseasonal lead–lag structures of wind fields in the global Tropics are defined using 5-day East African rainfall as the host index. Section 5 reevaluates the SST predictors used in Mutai et al. (1998), by assessing the atmospheric structures associated with these SST predictors and comparing them to the results in sections 3 and 4.

2. Data and analysis procedures

3. Fluctuations of the monthly/seasonal time means

4. Intraseasonal teleconnections with October–November East African rainfall

Correlation and composite atmospheric circulation structures have been computed for time lags ranging from −15 to +15 days relative to the East African area-average 5-day proxy for rainfall (OLR) in the 1979–96 period. To check for the reliability of the pentad OLR data, it was correlated with the pentad area-average rainfall anomaly over Kenya, as estimated from daily rainfall data. The correlation of −0.55 between the 216 pairs of October–November pentad values over 1979–96 suggests a good common signal in the two datasets, especially given the relatively small number of rainfall stations confined to a subregion of East Africa. Furthermore, when pentad values of OLR were replaced with pentad values of the Kenya rainfall index, lead–lag circulation patterns were found to be a little weaker but remarkably similar in pattern (not shown). No filtering was applied to any of the series because, as a first step, we wanted to study the full atmospheric circulation anomalies. Thus, the pictures contain aspects of circulation that are standing quasi-permanent anomalies through the season, combined with some aspects that are propagating intraseasonal disturbances, moving through the quasi-permanent seasonal time-mean circulation. Lead–lag teleconnection structures in the Tropics are shown for winds at 850 (Figs. 6a–g) and 200 hPa (Figs. 7a–g). Lead–lag structures for OLR (not shown) were largely consistent with the zones of divergence and convergence in the 200-hPa fields.

At a lead of 15–10 days, there is an intensification of anomalous 850-hPa divergence over the Maritime Continent, with anomalous easterlies strengthening across the equatorial Indian Ocean (Figs. 6a,b). At 5 days’ lead (Fig. 6c), the easterly wind anomalies turn into Arabia and also southward into the off-equator convection anomalies that form the southern rim of the horseshoe structure. At lag 0, low-level circulation anomalies across the Indian Ocean are generally weaker (Fig. 6d).

The precursor signal in the 200-hPa wind field is particularly striking. At 15 days before the East African rainfall event (Fig. 7a), anomalous equatorial westerlies extend from the Greenwich meridian to the western Indian Ocean. These slowly migrate eastward and reach the eastern Indian Ocean at zero lag, when anomalous easterlies extend over equatorial Africa and into the western Atlantic (Figs. 7b–d). Anticyclonic curvature in westerly anomalies is found at 20°N and 20°S at the longitudes of East Africa, suggesting partly formed Rossby anticyclonic gyres in response to convection anomalies over East Africa. These westerly anomalies turn equatorward over the Maritime Continent and form the equatorial westerly component of compensating anomalous cyclonic Rossby gyres north and south of the Maritime Continent (the interpretation is that the reduced Maritime Continent convection actually leads to reduced anticyclonic gyres, leading to the cyclonic anomalies). At lag T + 5, the circulation anomalies migrate farther eastward, with upper-level divergence centered over the western Indian Ocean. This signature is very indicative of a juxtaposed positive (East Africa–western Indian Ocean), negative (Maritime Continent) diabatic heating anomaly forcing in a Gill (1980) dynamical system, consistent with interpretations of MJO structure in this region (Hendon and Salby 1994; Hendon and Glick 1997). Anomalous easterlies at 200 hPa are well established over East Africa at lag +5 days (Fig. 7e), similar to the seasonal mean teleconnection pattern over East Africa (Fig. 4b). We propose that wet seasons in East Africa are marked by favorable conditions for convection over East Africa and the western Indian Ocean, and the circulation picture during the intraseasonal phase of enhanced convection over the western Indian Ocean (lag T + 5 in East Africa, Fig. 7e) tends to dominate the seasonal mean picture (Fig. 4b), perhaps simply due to the stronger magnitude of anomalies in this intraseasonal state, or due to the climate system residing in the state of enhanced convection over the western Indian Ocean on more days than that with the enhanced convection over East Africa.

It has been noted before that, during boreal autumn, the leading mode of interannual variability in the region of the Indian Ocean and Maritime Continent has a dynamic structure that is similar to that of the MJO when its ascending branch is located over the Indian Ocean or (reversed sign) the Maritime Continent (Gutzler and Harrison 1987). It is possible that the MJO is unaltered by the background atmospheric and SST anomalies (the dipole across the Indian Ocean), and standing quasi-permanent anomalies of equatorial convection over the western Indian Ocean (one sign) and Maritime Continent (opposite sign) make the complete contribution to the seasonal mean anomalies in the region. However, the pulsing nature of the convection in this region makes this unlikely. Furthermore, there is mounting evidence that the MJO may be a coupled ocean–atmosphere phenomenon (e.g., Krishnamurti et al. 1988; Wang and Li 1994; Hendon and Glick 1997; Lau et al. 1997). These observations, combined with the analyses here, raise some important questions regarding interannual variability of the MJO and East African rainfall. Instead of a dominance of quasi-permanent convection anomalies, a more plausible hypothesis may be that in years favorable to convection over East Africa and the western Indian Ocean, convection bursts associated with the development of MJO tend to be stronger and/or occur more often. However, the unfavorable background anomalies in the Maritime Continent can be expected to reduce the strength of MJO convection anomalies in this region. This is expected to be one reason why the anomalies that propagate across the Indian Ocean in Figs. 7a–g become very weak as they approach the Maritime Continent. Such scenarios could give rise to the observed seasonal mean teleconnection structures purely through a regionally modified MJO structure. Thus, these results suggest that as a next step, the intraseasonal structures should be stratified according to the background October–November circulation and SST anomalies across the Indian Ocean and Maritime Continent.

Next, the evolution of circulation over the equatorial Atlantic and equatorial Africa is focused upon. An association between local near-surface westerly anomalies and rainfall bursts in East Africa is known (e.g., Camberlin and Wairoto 1997; Camberlin 1997; Okoola 1999). However, the extension of the westerly anomaly back to the equatorial Atlantic, and some evidence of it beginning the pentad before the East African rainfall event (Figs. 6c,d) is previously unknown. It is, however, known from satellite images that wet spells in East Africa are often associated with synoptic disturbances that migrate eastward into the region in association with westerly near-surface wind anomalies (E. Mutoni, Tanzania Directorate of Meteorology 1998, personal communication) and so this result frames further investigation of the interaction of the disturbances with the westerly anomalies. To see in more detail the changes in African circulation that this evolution corresponds to, Figs. 8a–f show the composite evolution of actual and anomalous 850-hPa winds over East Africa. It can be seen that the westerly anomalies at lag 0 correspond to an actual westerly wind of about 2–5 m s⁻¹ extending from the Atlantic coast across equatorial Africa at latitudes 0°–5°S. The meridional arm of convergence is clearly enhanced and shifted east (Fig. 8f). Over North and West Africa, the mass flow has a different kinematic structure at −10 days compared to lag 0. At −10 days, anticyclonic trades extend from the eastern Sahel into equatorial West Africa and the Atlantic even slightly stronger than over Africa. At lag 0, these trades are weaker over the tropical Atlantic and west Africa while they have a stronger northerly component over central Africa (around 0°–10°N, 15°–25°E) that feeds into the westerlies south of the equator. This component of circulation adds to the converging northeast and southeast trades from the Indian Ocean previously emphasized in East African rainfall. Indeed, Indian Ocean anomalies are weak during the rainfall event itself (Fig. 8e). These results also raise the prospect for an Atlantic influence on East African rainfall. First, the Atlantic SST anomalies could modify the nature of the Congo air mass that converges from west to east. Second, the SST anomalies over the equatorial Atlantic may change the likelihood of spells of westerly anomalies through the October–November season.

Teleconnection structures are also evident in the central Pacific Ocean 850-hPa winds at considerable lead time. From 10 to 15 days (Figs. 6a,b) before the East African rainfall event, westerly anomalies are found in the central Pacific, connected with an anomalous burst of near-surface divergence over the Maritime Continent and its counterpart strengthening of easterly anomalies across the Indian Ocean. At lag 0, these westerly anomalies in the central Pacific have almost completely disappeared (Fig. 6d). A weaker, but perfectly out-of-phase pulse occurs at 200 hPa (Figs. 7a–d) and similar temporal evolution of convection anomalies is also seen in OLR fields (not shown), with enhanced convection accompanying the westerly anomalies at 10–15 days before the East Africa event, decaying to near-zero anomalies during the rainfall event.

5. Associations of SST eigenvector predictors with tropical convection and circulation anomalies

Mutai et al. (1998) developed an empirical prediction scheme for East Africa OND rainfall based on three precursor global SST rotated EOF patterns (Figs. 9a–c). This section diagnoses the atmospheric circulation and convection anomalies associated with these predictors, and interprets them in the light of the results in sections 3 and 4.

One predictor (SST REOF4) was noted by Mutai et al. (1998) to closely reflect the ENSO signal (Fig. 9b), and teleconnections with winds (850 hPa, Fig. 10a; 200 hPa, Fig. 10b) and OLR (Fig. 10c) confirm this. Horseshoe structures in the OLR are accompanied by the 850- and 200-hPa wind anomalies noted in Figs. 3 and 4, especially in the Maritime Continent region, stronger in Figs. 10a–c in the central and eastern tropical Pacific. This SST predictor is clearly measuring the state of ENSO in September and showing how that correlates with circulation in OND. It can be noted that the circulation and convection teleconnections in the Indian Ocean do not extend so strongly into the East African region, compared to the simultaneous OND structures (Figs. 3a and 4a). This suggests that some development of ENSO-related indices occurs from September to October–December across the Indian Ocean, such that ENSO indicators measured in OND teleconnect even stronger into East Africa.

A further predictor (SST REOF5) contained SST variance in the Pacific–Indian Ocean sectors. The July–August–September (JAS) time coefficients of SST REOF5 correlate very weakly with JAS values of SOI (−0.26) and Niño-3 SST (0.00) over the 1951–96 period. Correlation of JAS SST REOF5 with OND values of SOI and Niño-3 SST is a little higher (r = −0.34 and r = +0.17), suggesting a weak tendency for JAS SST REOF5 to anticipate ENSO development in OND. The SST REOF5 predictor has a pattern of weights with a relative minimum in the equatorial western Pacific (Fig. 9a). It was speculated that this could excite near-surface divergence out of the western Pacific [visible in correlation maps with near-surface divergence in Mutai et al. (1998), their Fig. 6a], generating conditions that favored above-normal rainfall in East Africa. The association of JAS SST REOF5 with OND 850-hPa circulation is also divergent along the equator out of the western Pacific in the reanalysis data (Fig. 11a), with strong easterly anomalies across the Indian Ocean extending all the way to the East African coast. The upper-tropospheric wind teleconnections (Fig. 11b) mirror the low-level patterns over the Maritime Continent with anomalous convergence. In addition, there are anomalously strengthened northeasterly trades in the northwestern Pacific (Fig. 11a) and suppressed convection (Fig. 11c) that gives rise to a horseshoe rim structure in this region. Indeed, this suppressed convection is surrounded to the west and east by enhanced convection anomalies, giving rise to a sequence of three trailing rims from equatorial latitudes into the subtropics. The equatorial pole over East Africa is particularly strong, consistent with the strong correlations between this EOF and East African rainfall (Table 1, column 7). The OLR and circulation anomalies in the northwestern Pacific and north Indian Ocean appear plausible with the pattern of SST gradients in the SST EOF. Southeastward trailing rims from the East Africa and Maritime Continent convection poles are also apparent, though weaker than their northern counterparts. The extent to which this arrangement of ocean–atmosphere associations will be repeated in the future is still of some concern, since it is not a well-known structure and it does not have a sound theoretical basis in ocean–atmosphere coupled dynamics at this stage. Therefore, its further monitoring and study is recommended.

The final SST EOF related to East African rainfall measures warming and cooling events in the equatorial and tropical South Atlantic (Fig. 9c). SST REOF2 in JAS is independent of SOI (r = 0.00). The association of September SST REOF2 with tropical OND OLR shows that anomalous warm water tends to draw the ITCZ to be anomalously active over the open ocean just south of the West African coastline, with enhanced convection extending westward into Amazonia and eastward into equatorial and eastern Africa (Fig. 12c). This pattern of correlations has been confirmed using the CRU precipitation dataset, including into the Amazon basin (not shown). The picture is almost identical to that of the East Africa rainfall–OLR signature once ENSO was removed from the data (Fig. 3d). It is strong evidence for an influence of the equatorial and tropical South Atlantic on East African rainfall. The location of the OLR anomalies connecting from the Atlantic through East Africa is centered at about 0°–5°S, the latitude where westerly winds develop near the surface before and during major rainfall events in East Africa. The relationship between the SST REOF2 and near-surface wind anomalies exhibits a cyclonic anomaly in the equatorial and South Atlantic (Fig. 12a), consistent with the convection anomalies. Associated 200-hPa wind anomalies (Fig. 12b) are very clear, with anomalous easterlies extending from equatorial Africa across the equatorial Atlantic.

6. Summary and conclusions

This paper began by establishing the representativeness of an area-average rainfall index for East Africa during the short rains. It was shown that rainfall anomalies in smaller subregions within East Africa strongly covary with the area-average rainfall index. Furthermore, teleconnections with the broad-scale ENSO indices were also reflected in smaller regions, generally with slightly reduced magnitude of teleconnection. In some cases though, especially subregions near lakes and mountains, the teleconnection strength was more substantially reduced. These results provide a framework for the interpretation of the rest of the paper, which used the area-average rainfall index. The results also provide a framework for further studies to investigate if the low teleconnection signals in the given subregions can be reproduced in high-resolution numerical models so that the mechanisms altering the teleconnection signals in the subregion can be diagnosed. Near-surface flows modified by topography may interact in special ways with the large-scale teleconnection signals entering the broader East African region and the interaction may give rise to local teleconnection signals (Mukabana and Pielke 1996; Sun et al. 1999a,b).

For warm ENSO phase, September rainfall in East Africa tends to be part of the July–September reduced rainfall in India, the Ethiopian highlands, and through to the western Sahel (Camberlin 1995, 1997; Ward 1998). The teleconnection structure is quite different in October–December, when rainfall in East Africa is enhanced in warm ENSO phase. A sequence of three horseshoe structures are evident in convection anomalies, with western tips (i.e., the center point of curvature) on the equator and outer rims trailing to the northeast and southeast into the subtropics centered over East Africa (positive sign), Maritime Continent (negative sign), and central Pacific (positive sign). The equatorial nodes over East Africa and the Maritime Continent are connected by an anomalous Walker circulation across the Indian Ocean, evident in near-surface and upper-tropospheric winds. One interpretation of the anomalies is in terms of equatorial dynamics driven by equatorial poles of upper-level diabatic heating anomalies associated with the anomalous convection (Gill 1980; Reverdin et al. 1986). However, forcing of the marine boundary layer by anomalous SST gradients is likely to play an initiating role too (Lindzen and Nigam 1987;Neelin and Held 1987), with, in warm ENSO phase, anomalously cold SST over the Maritime Continent and warm SST over the western Indian Ocean generating anomalous east–west sea level pressure gradients that drive the observed near-surface easterly wind anomalies. This scenario of SST anomalies, which usually sets up in ENS0 years, may account for much of the ENSO–East Africa teleconnection. Indeed, a strong role for the Indian Ocean SSTs associated with ENSO in generating East Africa short rains variability was recently found in GCM experiments by Goddard and Graham (1999).

ENSO-free teleconnections contained good structure from the eastern Indian Ocean westward to Amazonia, indicating that some of the variance not explained by ENSO was still part of coherent large-scale climate processes. From East Africa westward through to the Atlantic and Amazonia, sympathetic convection anomalies were found, suggesting a potential role for the tropical Atlantic SST to excite such fluctuations. Sympathetic convection anomalies were also found southeastward from East Africa, like the horseshoe rim in the full teleconnection pattern, but not extending so far. This suggests that the first half of the rim could be attributed to internal atmospheric variability and/or SST forcing not connected to ENSO.

In March and April, ENSO signal was found to be very different from OND in the Indian Ocean–central Pacific region. Suppressed convection in warm phase ENSO over the Maritime Continent was located farther east and the horseshoe equatorial node over the western Indian Ocean was completely absent, seemingly detaching East Africa from ENSO. However, the horseshoe structure in the western Indian Ocean convection anomalies returns in May and East Africa regains a teleconnection structure with some similarities to the OND pictures.

It was previously known that seasonal and intraseasonal wet conditions were associated with local westerly anomalies near the surface (Anyamba 1984; Davies et al. 1985; Camberlin and Wairoto 1997; Okoola 1999) and it was reassuring to recover this result using the OLR and reanalysis data. In fact, the reanalysis data suggest the westerlies may start to develop during the 5-day period before the rainfall event in the equatorial Atlantic and during the event, extending through equatorial Africa into East Africa. At 0°–5°S, these correspond to actual westerlies of about 2–5 m s⁻¹ in the pentad reanalysis data. The possible interaction of these winds with synoptic-scale disturbances that are known to propagate eastward into East Africa represents a line for further investigation. Before the rainfall event, anomalous near-surface easterlies grow over the Indian Ocean, giving the northeast and southeast trades a stronger onshore component as they approach East Africa, though during the event itself, these wind anomalies weaken. Other aspects of circulation prior to the event included anomalous near-surface westerlies in the central Pacific (which decay during the event and may be connected to the development of westerlies in the Atlantic and over Africa) and an anomalous upper-level cyclonic wind component over Arabia with associated enhanced convection before the rainfall event.

Inspection of previously published intraseasonal oscillation of equatorial convection (e.g., Rui and Wang 1990; Anyamba and Weare 1995) and velocity potential (Slingo et al. 1996; Sperber et al. 1997) show that bursts often pull off the East African coast and intensify over the Indian Ocean, where they can develop into full MJO events. Thus, the precursory features discussed above are of interest not just to anticipating wet events over East Africa, but also perhaps to MJO development. Indeed, MJO-like structures are seen to migrate across the Indian Ocean in the fields immediately after the East African rainfall event, but in the analyses here, the signal weakens as it approaches the Maritime Continent. This raises issues of interactions between the background ocean–atmosphere seasonal anomaly and the propagating intraseasonal anomalies, which themselves can partially modify the SST field (Emanuel 1987; Neelin et al. 1987; Krisnamurti et al. 1988).

The intraseasonal analyses represent a stringent examination of the utility of the OLR and NCEP–NCAR reanalysis data at pentad timescales. It is therefore a positive endorsement of both datasets when intraseasonal signals are identified by correlating OLR with reanalysis winds [see also Sperber at al. (1997) for case studies in selected years]. In particular, it is encouraging to recover previously anticipated associations, like westerly near-surface anomalies during a rainfall event over East Africa. Further work is, however, still needed to check details, like the exact location of convergence zones over East Africa and the magnitudes of wind anomalies. Also, diagnosis of sub-5-day features, like individual synoptic systems, has not been attempted here. It represents a considerable further demand on the accuracy of the reanalysis data over Africa.

In the light of the above diagnostic studies, the atmospheric teleconnections with the three SST predictors used by Mutai et al. (1998) were also explored. Interpretation of the ENSO predictor was clear, though with a suggestion that ENSO indicators in September may not capture so strongly the extension of ENSO activity right across the Indian Ocean during OND into East Africa. The SST predictor measuring gradients of SST from the northwestern Pacific into the Maritime Continent and northern Indian Ocean teleconnected with a train of convection anomalies from the northwestern tropical Pacific across India into East Africa, representing the northern rims of the horseshoe structures. Such a finding could be consistent with the SST gradients in the predictor, but careful monitoring of the performance of this predictor is recommended. The teleconnection results with the Atlantic predictor, measuring warming events in the equatorial and tropical South Atlantic, were particularly insightful and now suggest with some confidence an additional role for Atlantic SST forcing of East African rainfall variance that is independent of ENSO (at least at zero lag). The warm phase of this predictor was clearly associated with enhanced convection south of West Africa in the ITCZ, extending westward to Amazonia and eastward through equatorial Africa and into East Africa. The pattern closely resembled the East Africa rainfall teleconnection with enhanced convection once ENSO was removed. Possible interaction between this SST pattern and the westerly bursts across equatorial Africa bringing rainfall to East Africa at the intraseasonal timescale is an important avenue for further study.

This paper has therefore enhanced the basis for monitoring East African rainfall at interannual and (for October–November) at intraseasonal timescales. This monitoring can be rooted in the knowledge of the teleconnection structures and some of the underlying mechanisms for their generation. In addition, it has diagnosed a set of SST predictors in the light of the circulation anomalies associated with the East African rainfall anomalies and given greater confidence in the potential of these prerainfall season SST patterns to be connected to East African rainfall. At the same time, the intraseasonal analyses, when compared to the interannual analyses, have started to raise issues of scale interaction that are critical to a better understanding of interannual variability in the Tropics. In particular, it is now desirable to stratify intraseasonal variability in the Indian Ocean during October–November according to the phase of the seasonal mean Maritime Continent–East Africa convection anomaly dipole.