Projected changes in rainfall over Uganda based on CMIP6 models

Information about likely future patterns of climate variables is important in climate change mitigation and adaptation efforts. This study investigates future (2021–2100) changes in rainfall based on CMIP6 datasets over Uganda. The projection period is divided into two sub-periods: 2021–2060 (near future) and 2061–2100 (far future), relative to the baseline period (1985–2014). Two emission scenarios: SSP2-4.5 and SSP5-8.5, are considered. The results reveal a larger decrease (increase) in rainfall during March–April (November–December) under both SSPs. Moreover, an enhanced decline (increase) is projected under SSP2-4.5 (SSP5-8.5). The spatial distribution of future changes in seasonal rainfall reveals a decrease in MAM rainfall in the near future over most parts of the country under both emission scenarios. However, a recovery is exhibited towards the end of the century with more increase in the south-western parts of the country, and a higher magnitude under SSP5-8.5. In contrast, SON rainfall reveals wetter conditions during both timelines and emission scenarios. Maximum (minimum) wet conditions are expected in the north-western parts of the country (around the Lake Victoria basin). The linear trend analysis shows a non-significant (z =  − 0.714) decreasing trend for MAM rainfall during the historical period. This pattern is reflected in the near future with z-scores of − 0.757 and − 1.281 under SSP2-4.5 and SSP5-8.5, respectively. However, a significant increase for MAM and annual rainfall (z-scores of 2.785 and 3.46, respectively) is projected towards the end of the century under SSP5-8.5. These findings provide guidance to policy makers in devising appropriate adaptation measures to cope with expected changes in the local climate. Given the increase in intensity and frequency of extreme rainfall over the study region, future work should focus on examining projected changes in rainfall extremes under different global warming scenarios with consideration of model performance and independence.


Introduction
Climate variability and change in Uganda and the entire Great Horn of Africa are areas of concern due to their impacts on the economy that largely depends on rainfed agriculture (Limantol et al. 2016;Onyutha 2018Onyutha , 2019. Uganda has witnessed extreme weather events manifesting as droughts and floods in the past years (Mulinde et al. 2016;Ojara et al. 2020). The events had devastating impacts on communities' livelihoods leading to the displacement of people, loss of lives, and destruction of property. The most recent incident is the flooding in late 2021 that significantly affected over 40,000 people (Reliefweb 2021). This, among other weather and climate hazards, slowdown socioeconomic development, threatening the realization of sustainable development goals.
Uganda and the East African region at large experience two distinct rainfall seasons, the March to May (MAM "long rains") and the September to November (SON "short rains") seasons (Camberlin and Okoola 2003;Ayugi et al. 2016Ayugi et al. , 2018Ongoma and Chen 2017;Ngoma et al. 2021a). The former season is characterized by more rainfall than the latter. The two coincide with the passage of the Intertropical Convergence Zone (ITCZ; Nicholson 2018). The former season is characterized by more rainfall than the latter. The SON rainfall has been reported to exhibit higher inter-annual and inter-decadal variability compared to MAM (Hastenrath et al. 1993;Saji et al. 1999), owing to its linkage to global teleconnections such as the El-Niño Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD) (Nicholson and Kim 1997;Ogwang et al. 2015;Ngarukiyimana et al. 2017;Endris et al. 2019;Ngoma et al. 2021b).
Past studies have reported a decreasing (increasing) trend in MAM (SON) rainfall (Onyutha 2016;Nsubuga et al. 2014, Nsubuga andRautenbach 2018;Ongoma and Chen 2017;Ngoma et al. 2021b). The drying trend in the MAM season has been attributed to anthropogenic processes and internal climate variability such as ENSO and Interdecadal Pacific Oscillation (Lyon 2014;Tierney et al. 2015). On the other hand, the wetting pattern for the short rains is linked to the large-scale weakening of the Walker Circulation (Williams and Funk 2011;Liebmann et al. 2014;Tierney et al. 2015). Contrary to the current climatic condition which is characterized by several drought events across East Africa, a number of studies using model simulations show that it is currently wet and will get wetter in the future, a situation that has been termed as the "East African climate paradox" (Rowell et al. 2015;Ongoma et al. 2018). This paradox has challenged policy makers on what ought to be considered in preparing for future changes in rainfall patterns.
General circulation models (GCMs) are used to project climate patterns as they give an insight into the Earth's likely response to warming induced by anthropogenic greenhouse gas (GHG) emissions . One set of GCMs is the Coupled Model Intercomparison Project (CMIP) outputs (Gates et al. 1999), used widely in climatological and hydrological studies. CMIP outputs have evolved over the years, with the latest being phase six (CMIP6) . Future changes in the CMIP6 data give a range of future GHG and land use change scenarios estimated from integrated assessment models and based on various assumptions regarding economic growth, climate mitigation efforts, and global governance .
Over Africa, Almazroui et al. (2020) employed 27 CMIP6 GCMs under three emission scenarios and demarcated the region into 8 zones as recommended by the Intergovernmental Panel on Climate Change (IPCC). The study projected an increase in rainfall over East Africa, where Uganda is located. However, the study considered a broader domain that may not clearly portray the local climate changes. Past studies conducted over East Africa, including Tierney et al. (2015) and Ongoma et al. (2018), based on CMIP5, projected an increase in rainfall which is higher during the short than long rains. Tierney et al. (2015) further reported a likely decrease in June to August (JJA) rainfall. A recent study based on CMIP6 over the East African region focused mainly on the evaluation of CMIP6 models in simulating mean or extreme rainfall (Akinsanola et al. 2021;Makula and Zhou 2021;Ngoma et al. 2021c) or projections of precipitation extremes over the broader East Africa ). However, IPCC (2013 and numerous studies have reported better performance of GCMs in simulating mean rainfall than extreme events (Sillmann et al. 2013;Zhang et al. 2016).
Studies conducted over Uganda (e.g., Nsubuga et al. 2014;GOU 2015;Onyutha et al. 2016;Nsubuga and Rautenbach 2018;Egeru et al. 2019;Nimusiima et al. 2019;Ngoma et al. 2021a) projected an increase in rainfall during SON and December to February (DJF). Ngoma et al. (2021a) employed five regional climate models (RCMs) derived from the Coordinated Regional Climate Downscaling Experiment (CORDEX)-Africa to project rainfall patterns for the period 2020-2050 under the worst-case scenario (representative concentration pathway, RCP8.5). The results revealed the likelihood of wet conditions in April-May and October, while drier conditions were projected in March. Furthermore, the study projected a decrease (increase) in MAM (SON) rainfall. Other studies focused on smaller sub-regions of Uganda to examine the projected changes in rainfall or temperature based on CMIP5 outputs (Nimusiima et al. 2014;Egeru et al. 2019). Many of the aforementioned studies focused on extreme rainfall and considered small sub-regions of the country. Occasionally, few studies have considered Uganda as a whole. Nevertheless, the region has quite a few distinct features from the broader East Africa region, such as mesoscale effects and the short rains season. The short rains in Uganda start earlier than most regions of East Africa where the rainfall season is dominant in October-December (OND) (Nsubuga and Rautenbach 2018;Ngoma et al. 2021a). Importantly, only one study at present has utilized the CMIP6 outputs over Uganda but focused only on few water management zones of the country . A number of previous studies on climate change impacts on precipitation in Uganda were based on old generation(s) of the GCMs, especially those from phases 3 (CMIP3) and CMIP5.
The CMIP6 models, which includes more comprehensive GCMs with generally more sophisticated physics and higher resolution, are anticipated to better represente real climate systems. Moreover, the shared socioeconomic pathways (SSPs; O'Neill et al. 2017) provide additional descriptions of socioeconomic development, unlike the previous versions of RCPs (Van Vuuren et al. 2011) that only captured the projections of radiative forcing components for use in assessing of changes in climate system. Subsequently, these SSPs were considered in the preparation of latest Sixth Assessment Report (AR6; IPCC 2021) where they are being utilized to examine how the societal actions will impact the emissions of GHGs and also how the climate goals under the Paris Agreement will be attained.
Thus, there is a need for accurate and reliable information on future rainfall patterns for informed decision-making and planning in climate sensitive sectors such as agriculture, energy and health. Other than helping to formulate effective adaptation measures, climate information is key in devising relevant climate change mitigation strategies. The improved performance in CMIP6 as compared to CMIP5 presents promising reliability of the models in projecting the East African regions' climate. The key scientific questions to address are as follows: (1) What are the future characteristics of rainfall over Uganda as projected by the best-performing CMIP6 GCMs? (2) How will future rainfall patterns over Uganda vary under the worst-case scenarios? Section 2 gives a brief description of the study domain, the data, and methods employed in the study. Section 3 presents the findings, while Section 4 enumerates the discussions of the study. Lastly, the conclusion and possible recommendations are outlined in Section 5.
2 Study area, data, and methods

Study area
Uganda is situated in the east of Africa (Fig. 1). The country has a complex topography ranging from low-lying areas in the west along the Rift Valley, flat land in the north, and highland regions in the southwest and eastern parts of the country. The highest altitude is on the southwest (Mts. Rwenzori and Mufumbira) and east (Mts. Elgon and Moroto). The southern part of the country comprises tropical landscapes. However, moving northward towards South Sudan, the landscape becomes increasingly flat and dry, with widespread savannah and thorn bush landscapes. Other geographical features of Uganda include Africa's largest freshwater lake, Lake Victoria, as well as Lakes Kyoga, Albert, George and Edward, and Africa's longest river, River Nile. All these features influence the local weather and climate. The rainfall seasonality is mainly influenced by the ITCZ, which oscillates northward and southward, crossing the Equator twice a year.
Other systems influencing rainfall over the region include subtropical anticyclones, ENSO, quasi-biennial oscillation (QBO), tropical monsoons, IOD, and moist westerlies from the Congo Basin (Basalirwa 1995). The country's annual rainfall ranges between 750 and 2500 mm per year, with the highest amounts received by the Lake Victoria basin (NEMA 2008). The start of the rainy seasons can shift by 15 to 30 days from an average year, making it difficult for farmers to determine the best time for planting. Uganda's economy largely depends on rainfed agriculture and 96% of all farmers in Uganda are considered smallholders, contributing to 75% of the country's total agricultural production (WFP 2016). The leading crops produced in Uganda are beans, cassava, sweet potatoes, coffee, groundnuts, maize, millet, sorghum, sesame, and matoke -plantain banana (NBER 2010), grown in distinct climatic zones.

Data
The multi-model ensemble (MME) of six CMIP6 models are utilized in the present study. The models are chosen based on their ability to simulate rainfall over Uganda ). The use of MME reduces the uncertainties and makes climate projection more reliable (Knutti et al. 2010). The underlying assumptions of the MME mean is that all models are reasonably independent and equally plausible (Sanderson et al. 2015;Knutti et al. 2017). MME also assumes that the range of model projections is representative of what we believe is the uncertainty. Many studies have thus employed MME as a way of minimizing the uncertainty in the projections of rainfall over various regions (Tebaldi and Knutti 2007;Ayugi et al 2021b;Lim Kan Sian et al. 2021;2022). Table 1 summarizes the models employed, their developing institutions, horizontal resolution, and main reference(s). The first realization members (r1i1p1f1) of the models for both SSP scenarios are chosen for projections. The models are then re-gridded to a common grid of 1° × 1° resolution using a remapping procedure of distance weighted average. The aforesaid interpolation technique has better classification of diverse geography through data triangulation of nearest points and sub-regionalization of grid points by the nearest cell center input grids thereby suitable for comparative analysis from uniform grids (Vermeulen et al. 2017).
Rainfall projections are analyzed based on two emission scenarios: SSP2-4.5 and 5-8.5, representing medium forcing and policy, and worst-case and no policy scenario, respectively. Table 2 gives more details about the SSPs. Monthly rainfall data from the Climate Hazards Group Infra-Red Rainfall with Station (CHIRPS) (Funk et al. 2015) is used to validate the models' simulation of rainfall climatology over the region. The data covers a period of 1981-present and has been evaluated and appraised over the region (Dinku et al. 2018;Ngoma et al. 2021a).

Methods
Rainfall projection for both SSPs is considered during two time slices: 2021-2060 (near future) and 2061-2100 (far future) relative to a baseline period in order to compare the changes in the near future and towards the end of the century. The changes are evaluated relative to the latest baseline period of 30 years . The models' historical simulation is first compared with CHIRPS data to validate their performance in simulating the monthly annual cycle over the region during the baseline period. Future changes in monthly rainfall are then assessed for both time slices under the two emission scenarios. The projected changes are    High end of the future pathways range in the IAM literature; updates the RCP8.5 pathway with 8.5 W/ m 2 forcing quantified in absolute units of the baseline mean rainfall. Furthermore, a projected spatial rainfall anomalies analysis is performed for the two rainfall seasons. Inter-annual and intra-seasonal variation of future changes in rainfall over the whole study period is computed and presented on a portrait diagram.
The modified Mann-Kendall (m-MK) test (Hamed and Rao 1998) is used to detect the significance of projected linear rainfall trends. The significance for m-MK is calculated at 95% confidence level. The m-MK test is preferred over the original MK test (Mann 1945;Kendall 1975) because it incorporates missing values in a time series and it also employs relative magnitudes rather than the actual numerical values that allows "trace" or "below" detection data (Hirsch et al. 1993). The approach has been employed by various hydrological and climatological studies (e.g., Tadeyo et al. 2020;Tan et al. 2020;Ongoma et al. 2021;Ngoma et al. 2021c;Ayugi et al. 2021a, b, c). The Theil-Sen's slope estimator (Theil 1950;Sen 1968) is used to measure and compare the magnitude of projected rainfall against the baseline period over the study domain. The method is considered effective since it is not influenced by any extreme distribution and does not entail any normal distribution of the residuals. Furthermore, the sequential Mann-Kendall (SQMK) test (Sneyers 1990) is carried out to determine abrupt changes in seasonal and annual projected rainfall.
Several climatological studies have successfully employed the technique (e.g., Mumo et al. 2019;Ngoma et al. 2021a, b;. Probability density functions (PDFs) for seasonal and annual mean monthly rainfall are computed for the baseline and projections periods under both emission scenarios. PDFs are used to show how variability, skewness, and distribution of a variable may change under a shifting climate (Stott et al. 2016). Figure 2 shows the monthly annual cycle of future rainfall under SSP2-4.5 and SSP5-8.5 scenarios, alongside the models' ensemble and observed historical period. It is apparent that the highest amount of rainfall in the CMIP6 models is received in November while the lowest amount is in July. However, CHIRPS records the maximum amount of rainfall in April (153.14 mm) while the minimum amount is in January (40 mm). Projections under SSP5-8.5 scenario for the far future point to higher rainfall than under the SSP2-4.5 scenario, relative to the baseline period. Remarkably, the models reproduce the expected bimodal rainfall pattern following the north-south oscillation of the ITCZ across the region (Nicholson 2018). Meanwhile, the models slightly overestimate January to April rainfall and highly underestimate (overestimate) June to August (October to December) rainfall during the historical period. January to May, and September reveal minimal model biases.

Monthly annual cycle
Projections under the SSP2-4.5 scenario reveal a decrease (increase) in rainfall between January and April (May to December) in the near future. Notably, the increase rate is higher than the projected decrease. In addition, SSP5-8.5 for the same period projects a slight increase in wet conditions for January (3.35 mm) and February (3.59 mm) while a decrease in March to April rains is observed at − 0.76 mm relative to the baseline period. Nevertheless, a large increase is noted in the remaining months under SSP2-4.5 and SSP5.85 scenarios. SSP2-4.5 projects relatively more increase in rainfall towards the end of the century than in the near future. Wetter conditions are depicted in all months except for March and April, revealing a reduction in rainfall during this period. On the other hand, SSP5-8.5 projects an increase in monthly rainfall in the far future. Overall, the models project an increase in rainfall over Uganda between 2021 and 2100, except for March and April. A more pronounced increase will be experienced under the high emission scenario towards the end of the century compared to the near future.

Spatio-temporal variability of projected change in rainfall
To assess the spatio-temporal variability of future rainfall, seasonal change in rainfall for MAM and SON is analyzed by establishing the difference in future rainfall patterns relative to the baseline period of 1985-2014. Figures 3 and 4 show the projected changes in the spatial distribution of rainfall for MAM and SON, respectively, for the near and far future under SSP2-2.5 and SSP5-8.5. Under SSP2-4.5, MAM rainfall is projected to decrease over the entire country in the range of 0-4 mm (Fig. 3). The highest reduction in rainfall will be experienced in the eastern parts of the country around Mts. Elgon and Moroto, while the least will be in the south-western parts of the Kabale Highlands. Most parts of the country are expected to experience less MAM rains (< 4-8 mm) relative to the reference period. Although the SSP5-8.5 scenario equally projects a decrease in rainfall in the near future over most parts of the country, the magnitude of the reduction is less than that under SSP2-4.5. An increase in the western part of the study domain is also expected. Similarly, the south-western region exhibits more increase compared to other parts of Uganda. Towards the end of the century, SSP2-4.5 projects an increase in rainfall of up to 8 mm in the southern and western parts of the country, and a decrease in the eastern part, except around the Lake Victoria region. In addition, it is evident that under Projected changes for SON rainfall depict wetter conditions over the entire country in both timelines under both scenarios (Fig. 4). For the near future, the increase in rainfall increases westwards. The intensity under SSP5-8.5 is higher than that under SSP2-4.5. SSP2-4.5 (SSP5-8.5) exhibits an   increase of 0-30 (5-40) mm. Drier conditions are noted in the Lake Victoria region, which has been known to receive more rainfall than other parts of the country. Towards the end of the century, both scenarios project wetter conditions over the region as compared to the near future. More wetness will be experienced in the northern parts. These regions are arid and semi-arid lands (ASALs), receiving relatively low rainfall than other parts of the country. Furthermore, similar patterns to the near future are exhibited in the southern parts, where the smallest increase is depicted. These results agree with the findings of Nsubuga and Rautenbach (2018) that reported wet conditions over the west and north-western parts of the country and a drop in rainfall over the Lake Victoria region.
Further analysis for future rainfall changes in the next 80 years is conducted on a temporal scale to investigate rainfall inter-annual and intra-seasonal variability. Figure 5 shows the portrait diagram for future changes in rainfall relative to the reference historical period under the two scenarios. The results show high inter-annual and intra-seasonal variability under both scenarios. Under the SSP2-4.5 scenario, January-April show drier conditions, especially in the near future. More decrease in rainfall is projected during March and April. April shows the highest decrease with wet spells oscillation periods of 5-7 years. This could be linked to ENSO, which exhibits a similar cycle (Nicholson 2015). September to December show the wettest conditions, and the intensity and frequency increase towards the end of the century. Furthermore, November depicts a higher increase of > 100 mm/month for many years, except 2030, 2031, and 2061, which reveal drier conditions relative to the reference period. October 2029 also projects a high increase in rainfall (134 mm/month). June-August project minimal changes in rainfall throughout the study period. Under SSP5-8.5, wetter conditions are expected than under SSP2-4.5. November and December show wetter conditions with few dry conditions throughout the century. Drier conditions are evident during March and April, but they will get wet in the last decade. January is also projected to become wetter in the far future, which might extend the SON season. Overall, more rainfall is projected towards the end of the century under both emission scenarios.

Projected trend of rainfall
Future spatial rainfall trends are analyzed and compared against the historical baseline for MAM and SON seasons (Figs. 6 and 7). The historical period reveals a decreasing trend of 0 to − 4 mm in rainfall during MAM over most parts of the country, except a small region in the southwest which shows an increasing trend of up to 0.2 mm. Under SSP2-4.5, rainfall in the near future is expected to decrease, but the magnitude will be higher in the southern parts of the country. On the contrary, SSP5-8.5 projects larger rainfall reduction in the range of 0-1.2 mm and a slight increase in the north-western part. Towards the end of the century, both scenarios project an increasing trend in MAM rainfall over the whole country. However, the increase is higher under SSP5-8.5 and the magnitude increases westward. On the other hand, SON rainfall exhibits a slight increasing trend during the baseline period over the country. SSP2-4.5 projects almost similar patterns as the historical period in the near future. In addition, SSP5-8.5 reveals an increasing trend which is higher in the north-eastern parts of the country. For the far future, SSP2-4.5 projects a slight increasing (decreasing) trend over most (southern and north-western) parts of the country. Furthermore, SSP5-8.5 reveals an increasing trend with the highest magnitude in the north-western parts of the country.
The temporal trends are further assessed at seasonal (MAM and SON) and annual scales. Figure 8 shows the temporal variation and trends for both scenarios at seasonal and annual scales relative to the reference period. MAM season depicts a decreasing trend in the historical period, which is expected to slightly decrease under both scenarios in the near future. However, in the far future, a small and sharp increase is projected under SSP2-4.5 and SSP5-8.5 scenarios, respectively. On the other hand, SON rainfall has been increasing historically and is projected to follow the same trend in the near future under both emission scenarios. Moreover, contradicting trends are projected by the two scenarios towards the end of the century. SSP2-4.5 shows a slight decreasing trend, whereas SSP5-8.5 reveals an increasing trend. On an annual scale, the historical baseline period simulates an increasing trend, which is projected to persist in the near future under the two scenarios. However, towards the end of the century, a sharp increasing trend is expected in the annual rainfall over the region.
Results in Fig. 8 show that trend slope can change in magnitude and sign based on the period selected for analysis. In other words, if the trend is fitted to the entire 2021-2100 future climatic data, one would miss out on the relevant information regarding the sub-trends (Onyutha 2021). The variation in rising and falling sub-trends over the various sub-periods indicates random large-scale fluctuations in the climatic data (Onyutha 2021). This means there is a need to consider both climate variability and change when analyzing climate variables. Furthermore, the characterization of natural variation in terms of the statistical dependence is relevant in determining how well the climate models reproduce the variability in the observed rainfall and other climatic variables .
The m-MK test and Theil-Sen's slope estimator are employed to test the significance and magnitude of projected rainfall trends, respectively, at α = 0.05. Table 3 summarizes the mean, slope, and z-score of seasonal and annual rainfall for the baseline period, and near and far future under both scenarios. The MAM season depicts decreasing trend for the baseline period and in the near future for both emission scenarios. However, an increasing trend is projected in the far future under the two scenarios. SSP2-4.5 (SSP5-8.5) projects an insignificant (significant) trend (z-score 0.92 (2.785) or p > 0.05 (p < 0.05)). In addition, SON rainfall reveals an insignificantly increasing trend for all periods under both scenarios, including the baseline period. Annual rainfall also shows increasing trends for all cases, which is only significant towards the end of the century under SSP5-8.5 with a z-score of 3.46 (p < 0.01). Overall, rainfall is projected to increase during MAM, SON, and at annual scale, except for MAM near future. However, MAM and annual rainfall project a significant increase in the rainfall trend. These findings agree with previous studies (Nimusiima et al. 2014(Nimusiima et al. , 2019Ongoma et al. 2018;Egeru et al. 2019) conducted over East Africa or smaller regions of Uganda. Figure 9 shows the results of the SQMK test statistics for MAM and SON rainfall under SSP2-4.5 and SSP5-8.5 scenarios. Overall, the results depict an upward trend during both seasons for the two scenarios, though with varying amplitudes. Under SSP2-4.5, the MAM season projects a decreasing trend from the start until mid-decade, where there is a reversal up to 2029. The trend then decreases again in Fig. 6 Spatial pattern of projected linear trend for MAM rainfall over Uganda based on CMIP6 ensemble mean during 2021-2060 and 2061-2100 relative to the baseline period  the second decade, with an abrupt change projected to occur around 2043, significant at α = 0.05, implying a significant (p < 0.05) decline in rainfall. The trend after that increases throughout the projection period with abrupt but insignificant changes at five different periods. For SSP5-8.5, an increasing trend is initially projected, after which a prolonged reversal decrease occur until 2060. Later in the far future, a sharp increasing trend is expected towards the end of the century. No abrupt change is projected throughout the period. The SON season and annual rainfall project increasing trends under both emission scenarios without any abrupt change.

Probability density functions
The distribution of future seasonal and annual rainfall over the study domain is compared with that of the baseline period using PDFs. Figure 10 shows the distribution of MAM, SON, and annual rainfall displayed as normalized curves under the two scenarios. Under SSP2-4.5, a negative shift with higher variation in MAM rainfall is projected in the near future, whereas the distribution is expected to persist with a slight increase in variation towards the end of the century relative to the reference period. However, SSP5-8.5 projects similar variations with a smaller negative shift in the near future and a greater positive shift towards the end of the century. For SON, SSP2-4.5 reveals a positive shift with an increase in the maximum for both time periods, implying an increase in mean rainfall amount and variability. In addition, the shift and variability will be higher in the far future. Under SSP5-8.5, both time slices reveal a shift to the right in the far future. The peak in the near future is higher; thus, more variability is expected. Furthermore, mean annual rainfall depicts a smaller (larger) positive shift in the near (far) future. The near future reveals a sharp increase in the peak, pointing to less increase in mean annual rainfall but with high variability. SSP5-8.5, on the other hand, projects a positive shift in mean rainfall, but the peak is reduced more in the far future, which means an increase in magnitude. These results are congruent with previous studies conducted over East Africa (Anyah and Qiu 2012;Ongoma et al. 2018).

Discussion
Information on future changes in rainfall is valuable in developing preparedness measures through early warning systems. This study employs MME of the best-performing models ) to assess future changes in rainfall in the near future (2021-2061) and towards the end of the century under SSP2-4.5 and SSP5-8.5 scenarios. The results reveal that there will be a decline in the MAM rainfall in the near future, while an increase in SON and annual rainfall is projected for all the time periods under both emission scenarios (Fig. 2). Notably, rainfall is projected to increase with the increase in atmospheric radiative forcing. The MAM rainfall also reveals a recovery period with a projected increase towards the end of the century under both scenarios (Fig. 2). The findings disagree with previous studies, including Ongoma et al. (2018), which projected an increase in rainfall during March over East Africa based on CMIP5 but agree with Ngoma et al. (2021a). This points out the need for focusing on homogeneous rainfall zones when analyzing their variability. Patricola and Cook (2011) attributed the projected decrease to the weakening of the Somali Jet Stream and Indian Monsoon, and reduction in evaporation. The projected wet patterns in other months, especially OND, are, on the other hand, linked to the weakening of the Walker Circulation over the Indian Ocean (King et al. 2021).
Overall, SON rainfall is projected to increase more compared to the MAM rainfall. The results further show that there will be an increase in rainfall from October to December. This might shift or extend the short rainfall season of SON to December in the near future. Previous studies link the models' October to November rainfall overestimation to the poor representation of global teleconnections such as the Walker Circulation over the Indian Ocean in the parameterization schemes (Yang et al. 2015;King et al. 2021), and the prominent modes of inter-annual and inter-decadal rainfall variability (Giannini et al. 2018). According to Sylla et al. (2012), some of these features and systems are poorly understood due to limited access to observations and research attention. None of the current generations of GCMs was developed in Africa and the relevant processes over the continent have not always been a priority during model development. On the other hand, James et al. (2018) argue that the evaluation of climate models over Africa needs to move from scalar metrics, validation, and performance checks to investigating how models simulate processes on a regional scale. This would help in improving the accuracy of future projections. Nevertheless, despite the observed model uncertainties, future projections provide a picture of the expected climate conditions that are needed for long-term planning. The models can therefore be employed to project changes in rainfall over the study region.
Spatial variability reveals a higher positive change during SON as compared to MAM, with notable changes over the north-western parts of the country (Figs. 3 and  4). A reduction in rainfall around the Lake Victoria basin is also revealed. These results align with previous studies conducted either over the East Africa region  or at national and sub-regional levels (Egeru et al. 2019;Nsubuga and Rautenbach 2018). The decrease in March and April rainfall has been attributed to a strong southerly Somali Jet over the Horn of Africa that turns westerly over the Arabian Sea, thereby transporting moisture over East Africa (Hastenrath et al. 2011). Other studies link the MAM rainfall drying pattern to the weakening of the Walker Circulation over the Indian Ocean (Williams and Funk 2011). MAM is the main crop growing season for seasonal crops such as maize, beans, potatoes, and ground nuts. A decrease in March and April rainfall could have far-reaching effects   on the communities' livelihoods. This shift in the MAM season might require a change in the timing of crop planting, and raise the need for adopting drought-resistant or fast-growing crops. Overall, an increase in rainfall is likely to boost agricultural productivity, especially in the ASALs that are generally characterized by fertile soils and limited rainfall. This would support farming across the whole country, as the main producers are located in the southern part of the country. Crops such as beans, bananas, potatoes, and cassava, which are only grown in wet areas, are likely to survive in the northern parts of the region under the projected increase in rainfall. On the other hand, the wetter conditions projected in the south-western parts of the country might increase the occurrence of floods as this region comprises highlands that has been prone to floods in recent years (Nsubuga et al. 2014). Meanwhile, it should be noted that the findings of the study might not give accurate climate conditions as the models may under/overestimate rainfall for both scenarios. This is mostly attributed to the low resolution of the models and the poor representation of systems influencing inter-annual and intra-seasonal variability of rainfall over the study domain. Notably, both the current models and their predecessors (CMIP3/5) tend to overestimate the short rains while underestimating the long rains over East Africa (Anyah and Qiu 2012;Ongoma et al. 2018;Ngoma et al. 2021c). Nevertheless, the results of this study would contribute to the ever-present debate on expected projected changes in the region's rainfall variation. The projections continue to draw much attention because they give opposite patterns to the observed rainfalls. The results can be utilized by relevant stakeholders, including policymakers and farmers. Policymakers should design appropriate policies that reflect all possible outcomes as various studies are inconclusive on the exact direction of future climate in the wake of global warming and climate change. Although it is imperative to examine the changes in mean rainfall over the study region, the ever-increasing occurrences of extreme rainfall events in recent years call for a shift in analysis to examine extreme events in Uganda. Thus, future studies are recommended to focus on projected changes in rainfall extremes under different global warming scenarios while considering model performance and independence for accurate projections.

Conclusion
Annual climatology from the CMIP6 models shows that the highest amount of rainfall is received in November while the lowest amount is in July. Projections under the SSP2-4.5 emission scenario reveal a decrease (increase) in rainfall between January and April (May to December) in the near future, whereas under the SSP5-8.5 scenario, higher rainfall occurrence is expected. The models project an increase in rainfall over Uganda between 2021 and 2100, except during March and April. However, a more pronounced increase will be experienced under the high emission scenario towards the end of the century.
Spatial changes during MAM show that the highest reduction in rainfall will be experienced in the eastern parts of the country around Mts. Elgon and Moroto while the lowest will be in the south-western parts of the Kabale Highlands. Towards the end of the century, SSP2-4.5 projects an increase in rainfall of up to 8 mm in the southern and western parts of the country and a decrease in the eastern part, except around Lake Victoria. In comparison, the highest decrease is expected in the south-western parts, while an increase of 8 to 24 mm is expected westwards under SSP5-8.5 scenario. During the SON season, projected changes depict wetter conditions over the entire country in both timelines under the two scenarios.
Spatial trends depict an increasing pattern in MAM rainfall over the country relative to the baseline period that showed a declining tendency of 0 to − 4 mm. SSP5-8.5 scenario reveals an increasing trend with the highest magnitude in the north-western parts of the country. Temporal changes based on the Theil-Sen's slope estimator show the SSP2-4.5 (SSP5-8.5) projects an insignificant (significant) rainfall trend during MAM (z-score 0.92 (2.785) or p > 0.05 (p < 0.05)). The sequential Mann-Kendall test shows no abrupt change in projected rainfall throughout the period. SON and annual rainfall display increasing trends under both emission scenarios without any abrupt change.
Lastly, a comparison of future seasonal and annual rainfall over the study domain relative to the baseline period is conducted using PDFs. Under SSP2-4.5, a negative shift with larger variation in the MAM rainfall is projected in the near future, whereas the distribution is expected to persist with a slight increase in variation towards the end of the century relative to the reference period. For SON, SSP2-4.5 reveals a positive shift with an increase in the maximum for both periods, implying an increase in mean rainfall amount and variability. Mean annual rainfall depicts a slight (larger) positive shift in the near (far) future.