Witness Clay Massawe , Ziniu Xiao

a State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics, Institute of Atmospheric Physics, Chinese Academy of Sciences,Beijing, China

b University of Chinese Academy of Sciences, Beijing, China

Keywords:Rainfall variability Tanzania SST anomaly Dry and wet year Atmospheric general circulation

A B S T R A C T Based on site-observation data, NCEP—NCAR reanalysis data, and Climatic Research Unit gridded data, the rainfall variability over Tanzania during late austral summer (January—March, JFM) was analyzed for the period 1961—2011. Further, the associated atmospheric circulation and SST anomalies (SSTAs) were explored to understand the mechanisms of dry- and wet-year cases based on an interannual time scale. The correlation, Morlet wavelet power spectrum, and composite analysis methods were employed. The results showed that the JFM standardized rainfall anomaly time series exhibited significant time scales of variability at interannual (2—8 years) and quasidecadal (8—12 years). During dry years, anomalous anticyclonic northeasterly flow originating from western tropical Indian and southeast trades from the Indian Ocean to the southeast were associated with subsiding dry air, which resulted in suppression of rainfall as observed. In the typical wet-year cases, meanwhile, anomalous westerlies from the tropical and southeast Atlantic were strengthened over the Congo basin, delivering more precipitation to the region. Significant correlation was exhibited over the western tropical and southeast Indian Ocean, as well as the southeast and tropical Atlantic Ocean. These SSTA patterns favored atmospheric general circulation anomalies that were closely related to JFM rainfall over Tanzania.

1. Introduction

With about 75% of Tanzanian households dependent on agriculture as their primary economic activity, rainfall variability is of crucial importance to Tanzania’s socioeconomic development ( Mugabe, 2016 ).The year-to-year fluctuations in frequency and magnitude of rainfall influence the occurrence of weather-related extreme events such as drought and floods.

The annual rainfall climatological cycle is modulated by the movement of the ITCZ ( McHugh, 2004 ; Luhunga et al., 2016 ). Seasonally, the latitudinal ITCZ migrates towards the southern regions during October—December, reaching the southern parts of the country in January—February, and then reverses northwards from March to May. This movement of the ITCZ means there is a well-defined bimodal pattern in this region with two distinct rainfall seasons: a short rainy season concentrated from October to December, and a long rainfall season in March,April, and May. In other parts, over the southern, southwestern, central and western parts of Tanzania, a single ITCZ passage results in a unimodal distribution in which much of the rainfall is experienced between October and April.

Many previous studies, both observational and model-based, have been conducted on the rainfall variability over Tanzania. Some of them( Vizy, 2003 ; Yoo et al., 2006 ; Ihara et al., 2008 ) highlighted global and regional sea surface temperature anomalies (SSTAs) as playing an important role in the mechanisms related to the variability. El Ni?o—Southern Oscillation (ENSO), one of the large-scale SST-forced signals in the atmosphere, as well as the Indian Ocean Dipole (IOD), have been reported to have an influence on simultaneous precipitation during the short rainy season from October to December ( Rocha et al.,1997 ; Sun et al., 1999 ; Indeje and Semazzi, 2000 ; Ihara et al., 2008 ;Manhique et al., 2011 ; Taschetto et al., 2013 ; Kipkogei et al., 2017 ).These studies show that negative (positive) phases of the IOD and ENSO are accompanied by below-normal (drought) and above-normal (flood)conditions, respectively.

The present paper is focused on the rainfall during the late austral summer (January—March, JFM) period. This is not only because few studies have examined this period, but also owing to its importance.As Msongaleli et al. (2014) pointed out, JFM is an important growing season for cereal (sorghum) in some regions. Thus, it is necessary to improve our knowledge of the characteristics of JFM rainfall in Tanzania and to reveal the mechanisms behind the occurrence of dry and wet conditions.

The SSTAs of the neighboring ocean have been noted as possible external forcing factors. Several studies conducted in the southern Africa domain have demonstrated that JFM rainfall is affected by the western tropical Indian and Atlantic Ocean SST ( McHugh, 2004 ; Cook et al.,2004 ; Mapande and Reason, 2005 ; Manhique et al., 2011 ; Munday and Washington, 2017 ). Conversely, the subtropical South Indian Ocean Dipole (SIOD), which can often occur at the same time as a similar dipole in the South Atlantic and South Pacific oceans, is related to JFM rainfall by modulation of the atmospheric circulations ( Todd and Washington, 1999 ; Beal et al., 2011 ). Crucially, the occurrence of tropical cyclones in Africa are observed mostly between November and April in the subtropical southwestern Indian Ocean, and those tropical cyclones may be associated with significant enhancement of the westerly/northwesterly flow from the Congo basin in the southwestern Indian Ocean. Despite the studies mentioned above, more research is still needed, particularly over the Tanzania domain.

This study investigates the spatiotemporal variation of JFM rainfall,particularly over Tanzania, and its linkage with sea surface temperatures and associated atmospheric circulation. The emphasis is on interannual variability.

2. Data and methods

The rainfall dataset was from the Climate Research Unit (CRU), available from http://data.ceda.ac.uk/badc/cru/data/cru_ts/cru_ts_3.23/data/pre ,and described in detail by Harris et al. (2014) . The observed data of 20 synoptic stations from 1961 to 2017 across Tanzania obtained from the Tanzania Meteorological Agency were used to validate the CRU dataset. Atmospheric variables were obtained from the NCEP—NCAR dataset ( Collins et al., 1996 ) and the SST data were from HadISST1(Rayner, 2003). The climate indices used —the NAO (North Atlantic Oscillation) and ONI (Oceanic Ni?o Index) —were obtained directly from NCEP—NCAR, while the IOD and SIOD were calculated from HadISST1 SSTs. The computation of the intensity of the IOD and SIOD has been described previously by Saji et al. (1999) and Behera and Yamagata (2015) , respectively.

This paper focuses on the late austral summer rainfall season(JFM) for the period 1961—2017. The correlation, composite, and Morlet wavelet power spectrum analysis methods were employed, which have been used extensively in previous studies ( Shabbar et al., 1997 ;Guo et al., 2017 ).

3. Results

3.1. JFM rainfall over Tanzania

Tanzania is located in the east of Africa and south of the equator. The topography is low in the southeast and high in the northwest. Fig. 1 (a)is an elevation map of Tanzania (areas shaded in blue are water bodies),and includes the locations of the synoptic stations used to validate the gridded data. Fig. 1 (b) shows the annual variation of rainfall from the station observations and CRU dataset. As we can see, the annual cycle of rainfall could be divided into the wet season with much of the rainfall from October to May, and the dry season with less rainfall from June to September. In the wet season, two distinct rainfall peaks can be seen in December and March or April. Both the station data and CRU data are able to capture the annual rainfall cycle over Tanzania, which is similar to reported in several previous studies ( Camberlin and Philippon, 2002 ;Kijazi and Reason, 2005 ; Ogwang et al., 2015 ). The correlation coeffi-cient between the CRU and observed data estimated from the annual cycle time series is 0.96, which is statistically significant beyond the 99% confidence level. Therefore, the CRU rainfall dataset is suitable for further analysis of JFM rainfall variability over Tanzania.

Fig. 2 (a) shows the distribution of the JFM mean rainfall climatology over Tanzania. Overall, the amount of rainfall varies from about 150—200 mm/month over the south and west, and to less than 80 mm/month over northeastern Tanzania. Large centers of rainfall in JFM are found along the southern and western regions at around 9°—12°S, which decrease slowly toward the north and west regions. Note that during the JFM rainfall period, the continental ITCZ lies further south between southern Tanzania and Mozambique, coinciding with the intense concentration of precipitation over the southern regions. The northeastern region is characterized by low rainfall, as much of its rainfall is accumulated during the March—May rainfall period ( Mbululo and Nyihirani, 2012 ).

The standardized year-to-year time series of JFM rainfall can be seen in Fig. 2 (b), which shows substantial interannual variability during 1961 to 2017. Throughout the entire period, the first extremely wet year is 1979, followed by 1998, while the first extremely dry year is 1982,followed by 1992. Notably, the JFM rainfall demonstrates prominent decadal variation before the 1990s and thereafter features interannual variation. There was an anomalous wetter period before 1970 but an anomalous dryer period from 1970 to 1990, except in 1978 and 1979.

The Morlet wavelet power spectrum analysis result is given in Fig. 2 (c). The notable characteristics are a quasi-decadal 8—12-year periodicity from the 1970s to the late 1980s, and a 2—4-year periodicity from the late 1990s to 2000s. The latter variability pattern is possibly associated with the influence of ENSO, while the longer 8—12-year pattern is linked more to the NAO ( McHugh and Rogers, 2001 ).

3.2. Associated atmospheric circulation anomalies with JFM rainfall

Next, we examine the features of atmospheric circulation using composite analysis to understand the relevant variations of JFM rainfall. The wet years ( ≥ 0.5) and dry years ( ≤ - 0.5) were determined based on the standardized rainfall anomaly time series. A total of 12 typical wet years(1962, 1963, 1964, 1968, 1970, 1978, 1979, 1998, 2002, 2007, 2010,and 2016) and 12 typical dry years (1967, 1971, 1974, 1982, 1983,1984, 1992, 1997, 2000, 2003, 2012, and 2015) were obtained for JFM.

The composite wind anomalies at 850 hPa in the typical dry and wet years are presented in Fig. 3 (a, b), respectively. For the typical JFM dry years, the conspicuous features in the low-level wind field are the easterly anomaly located in the southern tropical Indian Ocean and the northeast wind anomaly over northwestern tropical Africa ( Fig. 3 (a)).Meanwhile, an anomalous easterly occurs over the southern tropical Indian Ocean and a westerly appears over the regions from the tropical Atlantic to Africa in the typical JFM wet years ( Fig. 3 (b)). Accordingly, the anomalous easterly from the tropical Atlantic to Africa, and the anomalous westerly over the tropical Indian Ocean, bring about divergence over eastern tropical Africa in the low-level atmosphere in dry years,and vice versa in wet years. The anomalous westerly from the tropical and southeast Atlantic strengthens the occurrence of moist air over the Congo basin, while the anomalous easterly brings moisture from the eastern Indian Ocean to the East African coast. The circulation pattern characterized by the anomalous southeast trades and northeasterly flow results from an anomalous anticyclonic circulation over the southern Indian Ocean, which is consistent with the findings of McHugh (2004) .

Fig. 1. (a) Topographic map showing the elevation (units: m) above the sea level of Tanzania (blue areas are water bodies) along with the distribution of the synoptic stations used to validate the gridded data (red triangles). (b) Mean annual cycle of rainfall over Tanzania (units: mm).

Fig. 2. The (a) JFM climatology, (b) standardized rainfall anomalies, and (c) wavelet power spectrum for the Tanzanian JFM rainfall season. Standardized rainfall anomalies exceeding 0.5 and - 0.5 are delineated by the red dashed lines.

The anomalous vertically integrated water vapor flux from 1000 to 300 hPa is displayed in Fig. 3 (c, d). During the dry-year cases, the vertically integrated water vapor flux pattern ( Fig. 3 (c)) shows anomalous positive divergence over the western Indian Ocean near coastal areas of Tanzania where easterly flow dominates over the region, indicating there is a net outflow of water vapor from that region. In wet years, however, the vertically integrated water vapor flux ( Fig. 3 d) shows anomalous convergence across the Tanzania domain extending from the Arabian Sea over the Indian Ocean.

Fig. 3 (e, f) illustrates the composite 500-hPa omega anomalies in the typical dry and wet years, respectively. It can be seen that the Tanzania region is dominated by negative anomaly values corresponding to ascending vertical motions during the wet years, while it is characterized by positive anomalies that depict descending motion associated with below-normal rainfall in the dry years.

Fig. 3. The composite anomalous (a, b) winds vectors (units: m s - 1 ) at 850 hPa, (c, d) vertically integrated water vapor flux (units: kg (m s - 1 ) - 1 ) from 1000 hPa to 300 hPa and its divergence (shading; units: 10 - 4 kg m - 2 s - 1 ), and (e, f) omega field at 500 hPa (units: Pa s - 1 ) in JFM. Black dots represent values at or above the 90% confidence level.

3.3. SSTAs associated JFM rainfall

To understand the general relationship between SST and JFM rainfall over Tanzania, we performed simultaneous correlation analysis to identify regions of SST that are directly or indirectly associated with the rainfall, as shown in Fig. 4 . Areas of significant positive correlation are apparent over the western tropical Indian Ocean, tropical Atlantic and the southeastern Atlantic Ocean, tropical Pacific and South Pacific Ocean. Possible key SST regions were selected, as shown by the rectangular frame in Fig. 4 . A description of the regions and the regional mean SST correlation coefficients with JFM rainfall are given in Table 1 . Almost all of the selected regions demonstrate a remarkable correlation with JFM rainfall over Tanzania with a 99% level of confidence level.Although there is no significant correlation between JFM rainfall and SST in Region-A, which is located in the southwest Indian Ocean, it will be combined with Region-B and Region-C as the future research object because the two adjacent sea areas Region-B and Region-C present significantly highly correlation with JFM rainfall.

Fig. 4. Spatial pattern of correlation between the JFM rainfall anomalies and the JFM SST. Contours represent the 90% significance level.

Table 1 Description of the selected regional SST indices and their temporal correlation ( r ) values with the JFM rainfall index, based on HadISST1.1.

It is interesting to observe two tripole-like SST patterns related to the JFM rainfall variability in Tanzania. One is associated with the Atlantic and Indian Ocean, featuring a positive SST anomaly over the southeast Indian and Atlantic oceans but a negative SST anomaly over the southwestern Indian Ocean. This tripole pattern is similar to the region where tropical temperate troughs have been recognized to play a role in the summer synoptic rainfall-producing systems over southern Africa. We defined an index, SI, to represent this tripole SSTA pattern as follows:

The other tripole SST pattern anomaly pattern is observed over the Pacific Ocean. Similarly, we defined an index, CP, to represent it as follows:

In the above formulae, A, B, C, D, E, and F refer to coordinates described in Table 1 .

Both SI and CP yield a significant positive correlation coefficient value (0.46 and 0.5, respectively) with JFM rainfall over Tanzania.As we know, there are already some SSTA pattern indices to represent ocean and climate correlation, and the significant regions revealed above could be related to some of these notable climate indices. In particular, those in the tropical Atlantic Ocean are similar to that used to calculate the NAO index. The regions found in the Indian Ocean may be related to the IOD and SIOD index. As the tropical western Indian Ocean has a teleconnection with ENSO through perturbations with the Walker circulation, the ONI was used to examine ENSO’s relation with JFM rainfall. The correlation coefficient between JFM rainfall and some indices used in previous studies, such as NAO, IOD, ONI, and SIOD, yield correlation coefficients of 0.25, 0.30, 0.14, and - 0.22, respectively. Notably,SI and CP have a statistical relationship with JFM rainfall featuring the highest correlation coefficient value among all of the indices. Among the previous indices, only DMI and NAO appear to have a significant correlation with JFM rainfall. ONI, one of the ENSO indices, has a positive correlation but it is not statistically significant; plus, SIOD shows no significant correlation with the JFM rainfall index over Tanzania. Comparing the correlation coefficients between JFM rainfall and the indices discussed, it is worth noting that the SSTA over the regions related to SI and CP play an important role in influencing rainfall over Tanzania. This implies that the indices proposed in this paper are valuable for studying rainfall and for operational climate prediction.

4. Summary and conclusion

The rainfall during the JFM season over Tanzania is important for local agriculture but has not been well studied in the past. In this work,we assessed the late austral summer (JFM) rainfall variability over Tanzania. It was found that the JFM precipitation is characterized by significant interannual and interdecadal variation, with a periodicity of 8—12 years before the 1990s, followed by one of 2—4 years in the early 2000s. The anomalous zonal flows from the subtropical Indian Ocean and the southern Atlantic Ocean also have an impact on JFM rainfall.By analyzing the relationship between SST and JFM precipitation, we defined some SST anomaly indices for the regions in the tropical Indian Ocean, tropical Atlantic Ocean, southern Indian Ocean, and the central Pacific Ocean. Compared with the SST anomaly pattern indices commonly used in the past, the new indices defined here show a better correlation with JFM precipitation. Among them, the SI and CP indices have the best correlation, and can therefore be used as a reference and signal for future predictions of JFM rainfall. However, the mechanisms involved still need to be further studied on different time scales. Besides,JFM seasonal rainfall has a strong interdecadal variability, which is also worthy of further study.

Funding

This study was supported by the Strategic Priority Research Program of the Chinese Academy of Sciences [ XDA20060501] and the National Natural Science Foundatin of China [ 91637208 ].

Declaration of Competing Interest

The authors declare that there is no conflict of interest regarding the publication of this paper.

Acknowledgments

We thank Chang Sun for improving the quality of figures significantly.