The EA region comprises six countries namely Tanzania, Kenya, Uganda, Burundi, Rwanda and South Sudan. However, the focus of this study is on Tanzania, Kenya and Uganda, located within the 5˚N and 12˚S and 29˚E and 42˚E (Figure 1). The region is characterized by diverse climate patterns due to complex topographical features. These include the Great Rift Valley, highest mountain in Africa, Mount Kilimanjaro about 5895 m above sea level, Mount Kenya (5199 m), Mount Rwenzori (5109 m), Mount Elgon (4222 m) and Mount Meru

(4565 m). The region is also endowed with the presenceof abundant and varying size of water bodies including the Lake Victoria basin (69,485 km²) Lake Tanganyika (32,900 km²), Lake Nyasa (29,600 km²), Lake Albert (5594 km²), Lake Turkana (6405 km²) and Lake Edward (2325 km²). The climate pattern of the region and particularly its rainfall distribution is therefore typified by stronger spatial and temporal variations, which is also amplified by significant differences in relief and vegetation cover. The large areas within the region (including the northern parts of Tanzania and southern parts of Kenya and Uganda) receive bimodal rainfall pattern with “long rains” season in MAM and the “short rains” season in OND associated with the North-South seasonal movement of the main rain belt, the ITCZ [29]. The onset and cessation of rainfall in East Africa is also characterized by stronger variability and is largely influenced by the migratory pattern of the ITCZ. The ITCZ usually migrates southwards through Tanzania during OND, reaching the Southern parts of Tanzania in January and February and returning northwards during MAM. The south-western highlands (SWH), southern region (i.e., Songea), southern coastal areas, western and central parts of the country experience Uni-modal rainfall patterns that normally starts from November and ends in April. In Uganda, the “short rains” season starts a month earlier; running from September to October [29] while some areas around the Lake Victoria basin and eastern regions of the Rift Valley experience trimodal rainfall regime that peaks in June to July (JJA).

This study utilized the daily and monthly rain gauge data sets from twenty-one (21) selected synoptic stations scattered over Tanzania (Figure 1). The stations selection was to include all the bimodal stations since during OND season the region experiences a noticeable impact of the short rain in comparison to unimodal areas where the rain normally starts in November and ends in April. Following the complexity in obtaining the daily and monthly rain gauge data sets from the rest of the East African countries (e.g., Kenya and Uganda), the study made use of the monthly global precipitation climatology center (GPCC) with a 0.5˚ × 0.5˚ grid resolution [30] and climate hazards group infrared precipitation with station data (CHIRPS) with a 2.5˚ × 2.5˚ grid resolution [31]. The gridded rainfall data sets were used to study the monthly (i.e., October, November and December) and seasonal (i.e., OND) climatological rainfall distribution over the Eastern Africa. Meanwhile, the study made use of the monthly SST data sets from the National Oceanic and Atmospheric Administration (NOAA) with 2˚ × 2˚ grid resolution for a period extended from 1960-2019 [32] to study the influence of IOD and ENSO on enhanced rainfall during October 2019. The IOD Index is computed as the difference in SSTA between the western equatorial IO (50˚E - 70˚E, 10˚S - 10˚N) and the south-eastern equatorial IO (90˚E - 110˚E, 10˚S-Equator) [19]. The mean monthly ENSO indices were computed from the tropical Pacific SST fields at an area bounded with 5˚S - 5˚N, 170˚ - 120˚W (i.e., Nino 3.4). These indices stand for the normalized SSTA time series averaged from Niño 3.4. Zonal and meridional winds from the National Centers for Environmental Prediction and the National Center for Atmospheric Research (NCEP-NCAR) re-analysis data [33], were used to study the circulation anomalies associated with enhanced rainfall during positive IOD event. These data sets span with a spatial resolution of 2.5˚ × 2.5˚ grid from 1949 to present.

The current study employed the singular value decomposition (SVD) to analyze the large scale inter-annual variability between the mean monthly rainfall over the EA and the mean SSTA over the Indian and Pacific Oceans during OND rainfall season with particular interest in 2019 October rainfall variability based on 1981 to 2010 climatology. SVD involves the matrix operation applied to covariance between grid point observations of two fields. It therefore decomposes the matrix of two fields into singular values with two sets of paired-orthogonal vectors namely the loading maps [34]. The SVD analysis in this study explored the leading climatic modes from the mean monthly rainfall over EA and SST over the Indian and Pacific Oceans and their associated principal components (PCs) (i.e. time series of SVD pattern over time), which further ascertained the likely influence of the IOD and ENSO modulation towards enhancing extreme rainfall during October, 2019. Since the PCs provide the strength of SVD pattern over time, the first principal component, which corresponds with the leading mode (i.e., PC1, in Figure 8(c)) was assigned to represent the rainfall indices.

Even though the IOD operates independently of the ENSO but we owe to study the interaction between the two indices towards enhancing rainfall during October 2019 since the proximity between the Indian and Pacific Oceans, and the existence of oceanic and atmospheric pathways, facilitate mutual interactions between these tropical climate modes [30]. Quantifying the likely association between IOD, ENSO and the mean October rainfall over the eastern Africa based on 1981-2010 climatology, we first defined their respective indices and later computed their respective correlation.

The ENSO index was computed by averaging the normalized SST from Niño 3.4 region (5˚S - 5˚N, 170˚ - 120˚W), while the IOD is expressed in terms of dipole mode index (DMI), which defines the difference in SSTA between the western equatorial IO (50˚E - 70˚E, 10˚S - 10˚N) and the south-eastern equatorial IO (90˚E - 110˚E, 10˚S-Equator). Revealing the circulation anomalies responsible for the enhanced rainfall during October 2019, the present study computed anomaly (i.e., the departure from the mean) fields for winds and SST. The nonparametric test, Mann-Kendall trend [35] [36] was also used to detect the trend in time series for the mean October rainfall over the bimodal areas of Tanzania and EA in general and tested for the corresponding significance at 95% confidence interval. Noted that, all the anomalies were calculated with respect to 1981-2010 climatology.

GPCC and CHIRPS data sets were validated against observed rain gauge data. This was achieved by comparing the GPCC and CHIRPS data and observed station data over Tanzania. The process was meant to identify a better dataset among the two in reproducing the annual rainfall over EA, to be used instead of observed data that is generally short and incomplete. In this case the GPCC and CHIRPS rainfall data sets were extracted as an areal average from Tanzania domain (i.e., 12˚S - 1˚S and 30˚E - 40˚E) and compared to the country average of observed station data sets. The results indicate that both GPCC and CHIRPS data sets are in agreement with the observed rainfall data sets, and the pattern from all data sets recapture the two rainfall peaks during MAM and OND seasons (Figure 2) and are consistent with the earlier findings [37] [38]. However, the pattern from August to December (i.e., including the period under study, October-December) shows that both GPCC and CHIRPS gridded data sets are more closely related with the observed data. The correlation coefficient between CHIRPS and observed rainfall data is nearly equivalent to that of GPCC and observed rainfall (r = 0.8194 and 0.8184) respectively. From these statistics, it is clear that both datasets are good in representing observed rainfall over EA. Thus they can be used to characterize rainfall over EA. However, CHIRPS rainfall data was chosen for further analysis to understand the long-term rainfall variability over EA due to its advantage of using the blend of rain gauge and Satellite data.

Generally, the monthly rainfall climatology indicates that, during October the region (i.e., mainly Tanzania and Kenya) receives less amount of rainfall (less than 50 mm) while the highest rainfall record between 100 mm and 250 mm is

observed over the larger domain of Uganda and a small area over the central Kenya (Figure 3). The coastal strip of Kenya and Tanzania (mainly the northern coast), areas to the north-west of Tanzania (i.e., around Lake Tanganyika and Lake Victoria basin (LVB)) only receive a considerable amount of rainfall (between 50 mm and 150 mm) compared to other areas in Kenya and Tanzania. This is in agreement with the expected rainfall pattern since these areas benefit from the moisture influx from the Lakes and the IO [39]. It is almost wet all over the region with an exception of some areas over the north-west Kenya and central Tanzania during November (Figure 3). The higher amount of rainfall between 100 mm and 250 mm is mainly observed over northwestern Tanzania, Most of Uganda and southern parts of Kenya extending to the coastline of northern Tanzania. The observed scenario is consistent with the earlier finding by [11] that during OND season, rainfall over the EA region peaks in November. In December the majority of the areas in Tanzania (Figure 3) receive enhanced amount of rainfall (between 150 mm and 250 mm) over the western, South Western Highlands (SWH) extending to southern region of Tanzania while less

amount of rainfall in the range between 50 mm and 150 mm are observed over the northern part of Tanzania and southern parts of Kenya and Uganda. (Figure 3), this is due to the fact that the ITCZ is far south [40]. The rainfall distribution in OND (Figure 3) shows that the western sector of Tanzania, areas around the LVB, the large area of Uganda, central Kenya and the coastal belt of Kenya and Tanzania record high amounts of rainfall between 300 mm and 650 mm.

The present study investigated the variability of the 2019 OND rainfall pattern over EA region and revealed a noticeable significant change in its distribution especially during October (Figure 4). The results reveal that most parts of the EA region recorded above normal rainfall, while areas around Lake Victoria Basin, central Kenya and the coastal strip of Kenya and Tanzania received the highest rainfall amounts exceeding their respective long-term mean by between 100 mm and 450 mm. Most of the meteorological stations in Tanzania recorded well above normal rainfall, with stations such as Arusha, Moshi, Babati, Same and Kilimanjaro over the North-eastern highland Tanzania, and Pemba, Ilonga and Morogoro over the northern coast recording between 500% to 1100% of their long-term mean (Figure 5).

To investigate further the temporal distribution of the October rainfall pattern over the EA, the inter-annual variability of IOD index and the October rainfall over the bimodal areas of Tanzania is shown in Figure 6. The highest frequency of the October rainfall over the entire climatology (Figure 6) is observed in 2019. The observation of the October 2019 rainfall anomaly in Figure 6 records the highest in history compared to the rest of the years (e.g., 1982 and 1997) where both the warm phase of ENSO and positive IOD coincided [41] [42].

Interestingly, even though a detailed discussion on the physical features associated with the October 2019 anomalous rainfall will be given later in this study, but a close linkage between the SST variability over the IO and the October rainfall anomalies is clearly noted (Figure 6). Thus, the enhanced (suppressed) rainfall anomalies during October are found to be associated with positive (negative) phase of the IOD index and the 2019 positive October rainfall anomaly recorded the highest positive phase IOD event in history.

The MK-trend test in Figure 7 is carried out aiming at verifying the temporal distribution of the October rainfall pattern. The sequential MK-trend test in the present study graphically illustrates the forward, u(F) and backward, u(B) trends of the October rainfall over the EA domain and the bimodal areas in Tanzania (Figure 7). It is revealed that, when a set of two series, a forward and backward one cross each other and diverge beyond the specific threshold value (i.e. at 95% confidence interval for the current study), then the point is said to be significant change point.

A typical sequential MK-trend test for the October rainfall over the bimodal areas of Tanzania (Figure 7(a)) shows that, generally, the region is in favor of increasing trend with abrupt changes marked from 1982 to 2019 window. However, both the trend and the changing points are insignificant at 95% confidence interval. Figure 7(a) indicates that the October rainfall trend started with a decreasing trend from 1982 until 2000-2001 where a significant decrease in trend noted and thereafter learned towards increasing trend. Over the eastern Africa (Figure 7(b)), the October rainfall pattern shows an increasing trend with abruptly changes observed from 2008 to 2016. Likewise, the progressive series, u (F) for both the rainfall trend and changing points over the EA domain are not above the critical limit suggesting the fact that are not significant at 95% confidence interval.

The likely cause for the enhanced rainfall anomaly during October and its possible linkage to SST were analyzed using the SVD, correlation and composite analysis methods. The results in Table 1 give a summary of the statistics for the first two leading modes in SVD expansion between the mean October rainfall

and SST anomalies over the Indian and Pacific Oceans. The squared covariance fraction (SCF) results in Table 1 strongly concentrate in the first two modes and account for about 85.42% of the total squared covariance. Notably, from the first mode the fractions of the variances corresponding to rainfall and SSTA field time series for the expansion coefficient component are 24.7% and 36.6% respectively. Generally, the expansion coefficients for both modes as indicators in strength of coupling, exhibit correlation coefficients of greater than 0.5 (see Table 1). The statistics indicated in Table 1 clearly signify how important the first two modes are in explaining the association between EA October rainfall and SST anomalies variability over the Indian and Pacific Oceans.

The first SVD (i.e., SVD1) coupled modes (in Figure 8(a) and Figure 8(b)) which dominate the co-variability between the October rainfall over the EA region and SSTA over the Indian and Pacific Oceans, shows a monopole positive co-variability with rainfall over the entire EA domain (Figure 8(a)). However, the corresponding spatial pattern for the SSTA in Figure 8(b) clearly indicate the establishment of the dipole like pattern over the IO with positive loadings to the western tropical Indian Ocean (WTIO) and negative to the eastern tropical India Ocean (ETIO) near Sumatra and Java. The scenario recaptures the positive phase of IOD as noted earlier by a number of researchers [41] [42] [43]. Likewise, the central equatorial Pacific Ocean (i.e., over Niño 3.4 region in Figure 8(b)) reveals a monopole positive loading clearly signifying the warm phase of ENSO signals [45]. It is therefore noted that, positive rainfall anomalies in EA during October are associated with increasing SST over the central and eastern equatorial Pacific, a typical warm phase of ENSO phenomenon. Furthermore, positive October rains anomalies are associated with a clear SST gradient in the IO (i.e., increasing SST to the west and decreasing to the east), which is a typical positive IOD event.

Furthermore, assessing the coupling strength of the association between the October rainfall over the eastern Africa and the different phases of IOD and ENSO, the current study computes the correlation with their respective indices. The calculation of the IOD index follows the definition by [44], that is the difference between the SSTA in the western (10˚S - 10˚N; 50˚ - 70˚E) and the eastern (10˚S-Equator; 90˚ - 110˚E) tropical IO. On the other hand, the ENSO index is the SSTA averaged over the “Niño 3.4” region (5˚S - 5˚N; 120˚ - 170˚W) of the equatorial Pacific. The rainfall index was computed based on PC1 time series for the October rainfall over the EA as indicated in Figure 8(c). It should be noted

that, all the indices above are divided by their respective standard deviation for comparison’s sake based on the 1981-2010 climatology.

Figure 9 displays the inter-annual variability between October rainfall over the EA, IOD and ENSO indices. There is a moderate inter-annual fluctuation between the October rainfall and ENSO index (i.e., r = 0.4212 in Figure 9(a)) while a strong fluctuation is revealed with the IOD index (i.e., r = 0.7084 in Figure 9(b)). With some specific years, all the three indices seem to be in phase and strong positive (i.e., 1982, 1997 and 2006), which suggest the fact that positive IOD event usually, coincides with El Niño and becomes much stronger as the intensity of El Niño increases [46]. Exceptionally, some years (i.e., 2011 and 2019 in Figure 9) the IOD and rainfall indices noted to be out of phase with the ENSO index and even substantially higher in magnitude. In particular, the year 2019 (Figure 9(b)) shows that the October rainfall over the EA and IOD indices are strongly higher and independent with the ENSO event. At this stage, it reveals the fact that the IOD coupled mode (particularly the positive phase of IOD) remained the dominant mode controlling the October rainfall variability over the EA region in expenses of the ENSO condition. Understanding the unique feature of the IOD event associated with the enhanced rainfall during October 2019, the present study analyses and compare the SSTA and wind patterns to that of October 1997 which peaked the second in rainfall record (Figure 6) based on 1961 to 2019 climatology.

Figure 9 displays the inter-annual variability between October rainfall over the EA, IOD and ENSO indices. There is a moderate inter-annual fluctuation between the October rainfall and ENSO index (i.e., r = 0.4212 in Figure 9(a)) while a strong fluctuation is revealed with the IOD index (i.e., r = 0.7084 in Figure 9(b)). With some specific years, all the three indices seem to be in phase and strong positive (i.e., 1982, 1997 and 2006), which suggest the fact that positive IOD event usually, coincides with El Niño and becomes much stronger as the

intensity of El Niño increases [46]. Exceptionally, some years (i.e., 2011 and 2019 in Figure 9) the IOD and rainfall indices noted to be out of phase with the ENSO index and even substantially higher in magnitude. In particular, the year 2019 (Figure 9(b)) shows that the October rainfall over the EA and IOD indices are strongly higher and independent with the ENSO event. At this stage, it reveals the fact that the IOD coupled mode (particularly the positive phase of IOD) remained the dominant mode controlling the October rainfall variability over the EA region in expenses of the ENSO condition. Understanding the unique feature of the IOD event associated with the enhanced rainfall during October 2019, the present study analyses and compare the SSTA and wind patterns to that of October 1997 which peaked the second in rainfall record (Figure 6) based on 1961 to 2019 climatology.

During October 1997, the SSTA fields over the Pacific Ocean shows that, the El Niño event is characterized by high SSTA over the central and eastern tropical sector while low SSTA to the western sector (Figure 10(a)). The condition recaptures the warm phase ENSO condition and thus likely to be the cause of enhanced October rainfall anomaly over the EA. On the other hand, a remarkable feature of the dipole mode condition with warm SSTA to the western and cool SSTA to the eastern tropical IO is observed (Figure 10(a)) over the IO recapturing the positive phase of the IOD mode. Meanwhile the warm phase of ENSO and positive IOD event depicted in Figure 10(a) coupled the strengthened easterly wind anomaly (Figure 11(a)) along the Equatorial IO, which decelerate as it approaches the coastal belt of the EA region. It is noted that in October 1997, the enhanced rainfall anomaly was firmly associated with the in-phase influence from the positive phase of IOD event and the warm phase of ENSO. During October 2019 (Figure 10(b)) the Pacific Ocean was mostly characterized by relatively weak or slightly warmer SSTA over the central and eastern tropical sector (i.e., over Niño 3.4 region). The scenario recaptures the neutral ENSO condition with less influence to the October rainfall pattern over the EA region. However,

the conditions over the IO (Figure 10(b)) reveal the positive SSTA to the western while negative SSTA to the eastern tropical IO. The scenario recaptures the positive IOD event with strong easterlies anomalies along the equatorial IO decelerating and converges with the southwesterly wind (i.e., moisten wind sourced from Congo basin) along the EA coastal belt. Notably, the anomalous rainfall recorded in October 2019 was influenced by the strong positive IOD event (Figure 6, Figure 9(b) and Figure 10(b)) opposite to the dual influence (i.e., in phase condition of the IOD and ENSO) revealed during October 1997. While the IOD operates independently of the ENSO, the proximity between the Indian and Pacific Oceans, and the existence of oceanic and atmospheric pathways, still facilitate mutual interactions between these tropical climate modes [46].

The socio-economic impacts of extreme rainfall events in October 2019 were widespread and consequential ranging from loss of life and properties, to severe destruction of infrastructure including destruction of houses, roads, bridges, crop farms, and displacement of people. More than 32 people were reported dead as a result of the flooding in Tanzania. A detailed description of the extreme rainfall events and the associated impacts for Tanzania is provided in Table 2, and it provide description of the meteorological stations, which set new records of the highest 24 hours rainfall since the establishment of the station. This observation is consistent with the findings from the IPCC special report on global warming of 1.5˚C, which indicates an increase in frequency and intensity of heavy precipitation in many regions [22] [23] [24].

Report from Kenya Red Cross (KRC) confirmed that, in early October 2019 flooding events associated with heavy rains had resulted in riverine, rock falls,

mudslides and landslides. More than 101,000 people were affected mainly in the north-eastern, central and coastal region of Kenya with at least 29 flood related death toll.