The intra-seasonal variability of rainfall over the northeastern Highlands (NEH) of Tanzania during the March, April, May (MAM) rainfall season and associated circulation anomalies have been studied for selected wet and dry years. Station rainfall data were used to extract the onset, peak, and cessation dates of the rainfall season and National Center for Atmospheric Research/National Center for Environmental Prediction (NCEP/NCAR) data were used to study the atmospheric circulation features associated with the intra-seasonal variability. It was found that the anomalously wet MAM seasons over the NEH of Tanzania tend to result from an earlier onset of the rains and a late cessation and thus, a longer-than-average rainfall season, whereas, dry years were characterized by a shorter-than-average duration of the rains. The moisture flux plots indicate that the composite wet MAM seasons were associated with advection of moisture from both the Indian Ocean and the Congo Basin. For the dry MAM seasons, a cyclonic anomaly off the Somali coast leads to southerly moisture flux anomalies over Kenya and westerly anomalies over the Indian Ocean which, together with easterly moisture flux anomalies over the Congo Basin, act to divert and transport low-level moisture away from Tanzania.
Rainfall in Tanzania is highly variable in space and time which makes agricultural activities difficult. Small differences in the amounts and timing of rain at a particular region may determine the success or failure of agricultural crops. Like the rest of equatorial East Africa, the northeastern Highlands (NEH) of Tanzania studied here experiences a bimodal annual rainfall pattern, whereas, western and southern Tanzania exhibits a unimodal rainfall pattern extending from October to April (Mapande and Reason, 2005a). Considerable intra-seasonal variability and breaks in convective activity occur during the rainfall season so that the wet season is typically characterized by a number of wet and dry spells. Since the economy of Tanzania largely relies on rain-fed agriculture, the country is vulnerable to the impacts of rainfall variability, posing challenges for food security and planning.
Research into rainfall variability over East Africa has tended to consider anomalies in monthly and seasonal totals. Similarly, seasonal forecasts often predict rainfall over large regions focusing on whether the rainfall for the forthcoming season will be above, near or below average. However, feedback from various user groups (e.g. agriculture, health, forestry) at regional climate outlook fora (such as the Southern Africa Regional Outlook Forum (SARCOF) and the Greater Horn of Africa Regional Outlook Forum (GHARCOF) is that more specific information on the intra-seasonal characteristics of rainfall seasons is needed for use in the agricultural, health, and other socio-economic sectors.
Despite the importance of intra-seasonal rainfall variability to agriculture in Tanzania, relatively little work has been done. Preliminary work on the onset and cessation of the rainfall season over Tanzania has been done by Alusa and Gwange (1978) and Mhita and Nassib (1987). More recently, Kijazi and Reason (2005) examined the intra-seasonal oscillations responsible for short-term rainfall variability along the Tanzanian coast during ENSO years. These authors suggested that rainfall seasons tend to end considerably later during wet years than dry years, i.e. wet years tend to have longer-than-average rainfall seasons and vice versa for dry years.
Mapande and Reason (2005b) analysed the link between rainfall variability on intra-seasonal scales over western Tanzania and the associated regional circulation. In their study, they found strong links between regional circulation patterns and rainfall over the unimodal area of western Tanzania. Currently, there is an urgent need to investigate possible links over the bimodal areas of the NEH of Tanzania. Since the NEH region of Tanzania (Figure 1) is populous and its land and air transport networks, industry and tourism are important for the national economy, it provides an appropriate regional focus for this research.
In terms of inter-annual and seasonal scales, it has long been established that ENSO influences Tanzanian climate mainly during the OND season (e.g. Ogallo, 1988; Hastenrath et al., 1993; Kabanda and Jury, 1999; Janowiak, 1988; Nicholson and Kim, 1997; Indeje et al., 2000; Mutai et al., 1998; Kijazi and Reason, 2005). A detailed analysis of the evolution of the ENSO signal over the Indian Ocean together with associated rainfall signals over eastern and southern Africa may be found in Reason et al., (2000). Hastenrath et al. (2007) have shown that East African rainfall during the OND season is related to the strength of the equatorial surface westerlies over the Indian Ocean and the zonal pressure gradient over the basin. Both these variables are modulated during ENSO as well as during other modes such as the Indian Ocean zonal dipole mode (Saji et al., 1999; Webster et al., 1999). Both these modes mainly influence the OND ‘short rains’ over East African region; on the other hand, the MAM ‘long rains’ season has not been much studied for the region including NEH of Tanzania and forms the focus in this paper.
The inter-annual variability of seasonal rainfall over Tanzania is typified by intra-seasonal wet and dry events with varying characteristics. However, much of the available knowledge about this phenomenon is scanty and needs to be explored. To date, rainfall forecasting in Tanzania has been based on a seasonal understanding of the mechanisms involved and is mainly provided in a probabilistic manner, i.e. a percentage occurrence of above, below, and average conditions for a season. Such forecasts are not of much use to the agricultural sector because the application of rainfall variability studies needs more detailed information than a simple departure from the long-term average. For example, a season with above-average rainfall over an agricultural region may not be any better than a below-average season if the rains are not well distributed in either time or space. For agricultural practices, the consistency with which the basic minimum of needed rainfall is received is more important than the total received over time. Crops are more likely to do well with uniformly spread ‘light’ rains than with a few ‘heavy’ rains interrupted by dry periods (Usman and Reason, 2004; Hachigonta and Reason, 2006). The timing of rainfall onset, duration, and cessation relative to the cropping calendar is more fundamental to crop viability than focusing on total seasonal rainfall alone. With this in mind, the analysis in this paper aims to identify the characteristics of in-season variability, such as rainfall onset, cessation, and dry spells. To this end, the NCEP/NCAR reanalyses were used to investigate the regional circulation features associated with the in-season variability.
2. Data and methodology
To understand the temporal and spatial behaviour of rainfall over the NEH of Tanzania, daily and monthly rainfall for four stations scattered over the region were obtained from the Tanzania Meteorological Agency for the 1970–2005 period. Various authors have divided the East African region into homogeneous zones (e.g. Indeje et al., 2000). However, most of these studies combine the northern coast with parts of the NEH of Tanzania. Although rainfall over both the NEH and the north coast is mainly controlled by moisture advection from the Indian Ocean, the climate of the NEH is also highly modified by topography. Mutoni (2001) divided Tanzania into 10 homogeneous zones, thereby separating the northern coast from the NEH. Tanzania's Meteorological Agency has adopted these zones for various scientific applications over the country. This study considers one of these zones; namely, the NEH that includes the following weather stations Same, Moshi, Kilimanjaro International Airport, and Arusha as shown in Figure 1. Attention is focused on the March, April, May (MAM) rainfall season over the study region.
Daily rainfall data were first grouped into 5-day means (pentads) before being subjected to time series analysis. The grouping was performed on non-overlapping 5-day means starting at pentad 1 for each year (1–5 January) and ending at pentad 73 (27–31 December). Pentad 12 contains an additional day to include February 29 in the case of a leap year. The NCEP/NCAR reanalysis 1 data were used to assess the regional atmospheric circulation patterns associated with intra-seasonal rainfall variability over Tanzania. These data resulted from a joint project between the National Center for Environmental Prediction (NCEP) and the National Center for Atmospheric Research (NCAR) Kalnay et al. (1996) leading to atmospheric reanalyses at a horizontal resolution of 2.5° for 1948 to the present. This effort involves the recovery of land surface, ship, rawinsonde, pibal, aircraft, satellite, and other data sources. The output variables are classified into four categories, depending on the relative influence of the observational data and the model for gridded variable. This study uses category A data which indicates that the analysis variable is strongly influenced by observed data and, hence, it is the most reliable class (Kalnay et al., 1996). Various authors have used NCEP/NCAR reanalysis to study climate variability over East Africa; for example, Mapande and Reason, 2005a, b; Kijazi and Reason, 2005; Kijazi and Reason, 2009a, b. CMAP precipitation data at a horizontal resolution of 2.5° were used to correlate with moisture fluxes. The CMAP precipitation data used are the merged satellite, rain gauge, and model precipitation estimates discussed in Xie and Arkin (1997).
Based on the criteria of Kijazi and Reason (2005), the onset, peak, cessation, and major dry spells for each season were determined from the pentad rainfall time series for the anomalously wet/dry years over the NEH. These criteria are based on the specific rainfall thresholds using pentads (five-day totals) as defined below.
2.1. Long rain season (MAM)
Onset: Five-day rainfall exceeds 10 mm followed by three consecutive pentads having a rainfall amount of not less than 10 mm per pentad.
Peak: The pentad with the highest amount of rainfall in the season.
Cessation: If three consecutive pentads have a mean rainfall of ≤2 mm per day, the preceding pentad is considered to be the cessation of the rainy season.
Dry spell: Five-day rainfall of less than 10 mm.
To determine the wet and dry years, monthly rainfall data were used to develop a standardized rainfall anomaly index over the NEH for the MAM season. Various indices have been used to assess the intensity of rainfall anomalies in East Africa. These include the standardized rainfall anomaly index (Ogallo and Nasib, 1984; Kabanda, 1995; Kijazi and Reason, 2005, 2009a), the decile index, the weighted rainfall anomaly index (Ininda, 1987), and several others. The standardized rainfall anomaly index similar to that employed by Wagner and Da Silva (1994) was used herein.
A rainfall index was computed by averaging the monthly rainfall for the region (NEH) followed by calculating the standardized departure for each month over the region for 1970–2005. Based on the computed standardized rainfall index, a dry (wet) year was defined as when the seasonal rainfall was below (above) one standard deviation from the long-term mean (Kijazi and Reason, 2005). Otherwise, the year would be classified as ‘normal.’ Selected wet and dry years were then used for a detailed analysis of intra-seasonal variability over the study region.
Composite analyses were used to study the atmospheric moisture flux circulations associated with wet and dry years. The compositing procedure consists of adding each parameter value at the same level at each grid point in the domain for all cases and dividing by the sample size to get the mean value.
3. Time series analysis
Normalized rainfall departures (averaged over the region) for the NEH's MAM season are shown in Figure 2a. It is evident from the time series that 1978, 1979, 1981, and 1990 were years with anomalously wet MAM seasons (rainfall departures ≥ 1). Years 1973, 1993, 2000, 2003, 2004, and 2005 had anomalously dry MAM seasons (departures ≤− 1). It is evident that more dry than wet MAM seasons occurred and it is important to note that the dry anomalies are generally weaker than the wet anomalies. The recent drought in the region (1998–2005) that occurred during the OND season was also observed during the MAM season, although it began in 2000 rather than in 1998 as for the OND season (Kijazi and Reason, 2009b). Below, the intra-seasonal variability during the 1978, 1979, 1981, and 1990 wet years and 1973, 1993, 2000, 2003, and 2004 dry years are analysed.
Figures 2b and c, and 3a and c plot pentad time series for the 1973, 1993, 2000, 2003, and 2004 dry MAM seasons over the NEH. The rainfall onset, peak, and cessation dates were extracted and are tabulated in Table I. The last column of Table I indicates the climatological dates as obtained from Kijazi and Reason (2009b). In 1973, the onset of the MAM rainfall was late by about 6 pentads and the season itself ended about 4 pentads earlier. Thus, the MAM 1973 season was much shorter than average (by about 10 pentads) and the amount received was less than half the average. For 1993, the onset of the MAM rains was about 5 pentads late and the amount received was also less than half the average even though the cessation of the season was the same date as the average cessation date. For 2000, the onset of the MAM rains was about 1 pentad later, and the rainfall season ended about 3 pentads earlier, and the amount received was less than half the average. In 2003, the onset of the MAM rainfall was not defined but the peak pentad was 3 pentads late. The season itself ended about 1 pentad later and the amount received was just over half the average. For 2004, the onset of the MAM rainfall season was 1 pentad earlier and the cessation was about 5 pentads earlier. Thus, the MAM 2004 season was much shorter than average (by about 4 pentads) and the amount received was less than half the average. This greatly reduced rainfall was confirmed by the cumulative plots for Moshi and Arusha (Figure 5a and b). Similar differences in onset and cessation date for the dry MAM seasons noted here were also found by Kijazi and Reason (2009b) in their analysis of the prolonged 2000–2005 drought.
Table I. Rainfall onset, peak, cessation, and total amount received for the NEH (MAM) dry seasons (in brackets the percentage of seasonal total). The climatological dates as obtained from Kijazi and Reason, 2009b are shown in the last column
For the selected wet MAM seasons (Table II), each year seems to have its own characteristics in terms of rainfall onset, peak date, cessation, and total amount received except for 1978 and 1990 when the rainfall started at the same time (16–20 February). Figure 4 shows the temporal distribution of the pentad rainfall for each anomalously wet season. Of these, MAM 1979 was particularly wet with a standardized rainfall anomaly value of about 2.7 (Figure 2a). Comparison with climatology (Table I) indicates that the rainfall season was longer than average by about 3 pentads and the amount of rainfall received was more than double the average. The cumulative rainfall plot for Moshi and Arusha further confirms how wet this season was (Figure 5a and b). For 1978, the rainfall season was much longer than climatology (mainly due to a very early start in mid-February) and the amount received was just less than double the average. In 1981, the MAM season was 2 pentads shorter than average, but the amount of rainfall received was over a third greater than average with peak values of about 15 mm/day. On the other hand, the 1990 MAM season was about 2 pentads longer than climatology. The amount of rainfall received in this season was more than three quarters above the mean. Generally, the wet MAM seasons were characterized by longer-than-average rainfall seasons with rainfall totals about 1.5–2 times the average.
Table II. Rainfall onset, peak, cessation, and total amount received for the NEH (MAM) wet seasons (in brackets, the percentage of seasonal total)
(P + 1)
The onset, peak, and cessation dates of selected wet and dry seasons are used in the next section to analyse the associated anomalous moisture flux circulation.
4. Moisture flux circulations associated with onset, peak and cessation
This section examines NCEP/NCAR reanalysis moisture flux circulations related to the rainfall over the NEH during each anomalous MAM season. Attention has been given to the lower-level moisture flux circulations to understand the evolution of moisture transport during the onset, peak, and cessation pentads of the selected wet and dry seasons. Goddard and Graham (1999) observed the 850 hPa level to be the most representative of the behaviour of the vertically integrated moisture flux over East Africa. However, considering the high-altitude topography typical of the NEH, the 700 hPa level was used in this section to study the intra-seasonal moisture flux circulations during selected wet and dry MAM seasons over the NEH.
In the tropics, there tends to be a near-surface wind discontinuity with horizontal velocity convergence and upward vertical motion on large scales. There is often cloudiness near this region of low-level convergence typically aligned in a nearly east–west direction that oscillates north–south in response to the relative position of the Sun (Asnani, 1993). This region is referred to as the Intertropical Convergence Zone (ITCZ) and is the main system controlling weather over much of the East African region. The long rain season (MAM) is associated with the northward movement of the ITCZ. Figure 6 shows the long-term mean moisture flux patterns at 700 hPa for the onset, peak, and cessation pentads of the MAM rainfall season over the NEH. At the onset of the rainfall season (Figure 6a), most of Tanzania is covered by easterly moisture flux. Between 5°S and 10°S over the Indian Ocean there is a zonal band of westerly moisture flux (at the surface, these equatorial westerlies lead to the ocean Wyrtki jets (Wyrtki 1973)) with northeasterly moisture flux closer to the Tanzanian coast. To the north and south of this westerly band, easterly moisture flux dominates. The easterly moisture flux transport moisture from the Indian Ocean towards the NEH is consistent with the onset of the rainfall season.
At the peak of the rainfall season (Figure 6b), strong southeasterly moisture flux exists over the tropical south west Indian Ocean and over much of Tanzania. This period is about one or two months prior to the onset of the Somali jet that feeds into the Indian monsoon (Halpern et al., 1998; Mafimbo and Reason, 2010). The strong southeasterly moisture flux (Figure 6b) occurs over a large western Indian Ocean region and is fed by an equally intense flux across most of the subtropical South Indian Ocean suggesting a strong moisture fetch towards Tanzania. Such moisture flux circulation patterns result in enhanced easterly moisture flux convergence over Tanzania and increased rainfall during this peak pentad of the rainfall season.
Towards the withdrawal of the rainfall season (Figure 6c), a month later, the belt of strong southeasterly moisture flux emanating from the Indian Ocean shifts north to be more influential over Kenya. The veering of the flow near the Tanzanian coast, which starts to signal the transition towards the Somali jet strengthening and the onset of the Indian monsoon, implies a relative transport of moisture away from the NEH. The equatorial Indian Ocean between the equator and 10°N is covered by westerly moisture flux thereby further transporting moisture away from Tanzania consistent with the cessation of the rainfall season.
Moisture flux circulation features for the onset, peak, and cessation pentads of the composite dry and wet years were analysed. Although each anomalously wet/dry year seems to have its own characteristics in terms of the date of the MAM rainy season onset, peak, and cessation dates, the moisture flux circulation patterns associated with each of these stages of the season seems to be similar and, therefore, a composite approach is justified. As in Section 3, only anomalous dry/wet years having rainfall anomaly value of at least one standard deviation (Figure 2a) are considered for compositing; i.e. the 1973, 1993, 2000, 2003, and 2004 dry and the 1978, 1979, 1981, and 1990 wet MAM seasons.
Moisture flux circulation features for the onset, peak, and cessation pentads of the composite dry years are shown in Figure 7. During the onset of the MAM dry composite rainfall season (Figure 7a), an anticyclonic moisture flux anomaly occurs over the Indian Ocean east of Madagascar coupled with westerly anomalies over the equatorial Indian Ocean between 5°N and 5°S. The position of the descending limb of the Walker-type circulation was over the equatorial western Indian Ocean closer to the East African coast as implied by westerly anomalies over the ocean and easterly anomalies over the land. Over the Lake Victoria Basin, there is relative convergence of westerly anomalies emanating from the Congo Basin and easterly anomalies from Kenya. The occurrence of the descending limb of Walker-type circulation closer to the East African coast tends to suppress convection over the NEH consistent with the observed dry conditions during this onset pentad of anomalously dry rainfall season.
During the peak of the MAM dry composite rainfall season (Figure 7b), a cyclonic moisture flux anomaly occurs near the Somali coast. The easterly moisture flux anomaly over the southwestern Indian Ocean veers to a more southerly anomaly over the East African coast thereby denying inland penetration of moisture consistent with the below-average rainfall received over the NEH. Thus, compared to climatology (Figure 6b), there is a stronger-than-average Somali Jet developing over Kenya which tends to divert more of the marine sourced moisture away from the NEH. Over the Congo Basin, the easterly moisture flux anomalies represent a strengthening of the mean flow (Figure 6b) and, thus, further divert moisture away from the country, again consistent with the observed dry conditions. At the cessation of the MAM dry composite rainfall season (Figure 7c), the westerly moisture flux anomalies over much of the tropical Indian Ocean act to transport more moisture away from Tanzania than in the climatology (Figure 6c), thus marking the cessation of the rainfall season.
Figure 8 shows the corresponding wet composite determined from the 1978, 1979, 1981, and 1990 MAM wet seasons. During the onset (Figure 8a), strong easterly moisture flux anomalies occur over the Indian Ocean between 5°S and 15°S that extend over the entire country. Comparison with climatology (Figure 6a) indicates that the composite pattern implies a reduction in moisture flux away from Tanzania over the ocean between 5°S and 10°S, and a stronger transport towards southern Tanzania near 10°S–14°S. Over the Congo Basin along the equator, a southwest moisture flux anomaly occurs west of 25°E and acts to enhance the mean flow and, hence, the relative moisture flux convergence over northern Tanzania as suggested by Goddard and Graham (1999). These patterns are, therefore, favourable for the onset of an anomalously wet rainfall season.
During the peak of the MAM wet composite rainfall season (Figure 8b), the easterly moisture flux anomalies between 5°N and 5°S over the Indian Ocean imply less moisture transport away from East Africa relative to the climatology (Figure 6b). Similarly, relative to the mean, the band of westerly moisture flux anomaly that emanates from the eastern equatorial Atlantic Ocean and extends over the Congo Basin and Tanzania implies more moisture availability over Tanzania and less moisture transport away from the country. Hence, there is relative moisture flux convergence over the NEH. Such moisture flux circulation patterns are thus consistent with the anomalously wet conditions during this peak pentad.
Toward the cessation of the MAM wet composite rainfall season (Figure 8c), the anomalies are relatively weak and, therefore, close to the climatological patterns for the cessation of the rains as shown in Figure 6c.
5. Discussion and summary
Evidence of a relationship between MAM rainfall over coastal Tanzania and the zonal wind over the Congo Basin has previously been presented (Kijazi and Reason, 2005). Similarly, the previous section suggests that moisture flux anomalies over the Congo Basin can influence the MAM rainfall over the NEH. Westerly (easterly) moisture flux anomalies over the Congo Basin seem to be associated with increased (decreased) rainfall. In order to consider how the correlation of rainfall and moisture might vary spatially, a gridded rainfall product is needed. Figure 9 correlates rainfall with 700 hPa zonal wind for the dry and wet composite using CMAP data (Xie and Arkin, 1997) since previous work has found that this product compares reasonably well with Tanzanian station data (e.g. Kijazi and Reason, 2005). These figures indicate that the eastern Congo Basin and the western equatorial Indian Ocean are areas where the correlation is roughly of the same magnitude for the dry and wet composites but opposite in sign. Thus, the statistical relationship is more robust for these areas than elsewhere and, given the previous result of Kijazi and Reason (2005), a time series of the zonal wind averaged between 850 and 500 hPa over the Congo Basin (0–8°S; 20–29°E) was correlated with MAM rainfall over the NEH to see whether the zonal wind over the Congo Basin might offer some predictability for the MAM rainfall season over the NEH. The time series extends from 1971 to 2005.
Lag correlations between the rainfall time series over the NEH and the mid-level Congo Basin zonal wind were calculated for January and February or, respectively, two and one month before the climatological onset of the rainfall season. For January, the two series are correlated at 0.59 which is significant at the 95% level using Student's t-test (Figure 10a). In particular, a phase locking of the two variables occurred during the years with larger rainfall anomalies. For February, the correlation between mid-level Congo Basin zonal wind and MAM rainfall increased slightly to 0.63, again statistically significant at 95% (Figure 10b). These lagged relationships between the two variables suggests that it may be possible to get an indication of the MAM seasonal rainfall total as early as the previous January using the NCEP reanalysis winds over the Congo Basin for that month.
These statistical results further confirm the observations obtained in the previous section where wet conditions over the NEH of Tanzania were associated with westerly wind anomalies over the Congo basin and vice versa for dry conditions. The potential predictability offered by these results is important since the strong ENSO relationships that exist for OND rainfall over Tanzania do not apply to the MAM rainy season (Ogallo, 1988; Indeje et al., 2000).
In summary, this study has analysed the intra-seasonal variability over the NEH of Tanzania and its links with regional circulation patterns. It was found that the increased rainfall during wet years over the NEH of Tanzania tends to be associated with an earlier onset and, thus, a longer-than-average rainfall season while dry years were characterized by a shorter-than-average season.
It was evident from the moisture flux circulation anomalies that increased rainfall over the country during the MAM season may be associated with an increased fetch of low-level moisture from the tropical Indian Ocean. For the composite wet MAM season, the increased rainfall was associated with enhanced advection of moisture from the Indian Ocean and either reduced export away from Tanzania over the Congo Basin or relative moisture advection towards the country from this source. For the dry MAM seasons, a cyclonic anomaly off the Somali coast leads to southerly moisture flux anomalies over Kenya and westerly anomalies over the Indian Ocean which, together with easterly moisture flux anomalies over the Congo Basin, act to divert and transport low-level moisture away from Tanzania.
The correlation between time series of January or February zonal wind over the Congo Basin (0–8°S; 20–29°E) and MAM rainfall over the NEH of Tanzania indicated a phase locking of the two variables that occurred during the years with larger values of rainfall anomalies. This significant correlation at up to two months before the onset of the rainfall season in January suggests that some predictability of MAM seasonal rainfall totals may exist.
The authors gratefully acknowledge the Third World Organization for Women in Science (TWOWS) for funding the first author's PhD thesis from which this paper was derived. We thank colleagues at the Oceanography Department, University of Cape Town, and Tanzania Meteorological Agency for helpful discussions.