Serengeti–Masai Mara ecosystem dynamics inferred from rainfall extremes

The Serengeti–Masai Mara Ecosystem (SMME) is an iconic ecological and biodiversity resource in East Africa with a spectacular great animal migration feature. Environmental shocks like droughts, floods, and land degradation threaten the SMME’s ecological functioning. However, the spatiotemporal ecosystem dynamics during climate extremes are inadequately examined. Here we quantified water availability and vegetation changes during extreme wet (EWE) and dry (EDE) events over the SMME for wet and dry seasons in 1982–2020. We derived extreme events from seasonal mean precipitation anomaly exceeding standard deviation and applied composite and correlation techniques to assess their dynamics with animal populations and migration prospects. Wet season EWE increases vegetative and moist conditions over southern SMME, suggesting elongating migrants’ occupancy compared to normal periods and delayed northward migration. Wet season EDE significantly suppresses these features, heightening ecosystem competition and survival threats, proposing an earlier northward migration. Dry season EWE increases vegetation and water availability over northern SMME, prompting the late southward migration. Dry season EDE significantly reduces vegetation and water availability over northern SMME, suggesting an early southward migration towards regions with more vegetation and increased water availability. The SMME also experiences multiple EDE occurring in consecutive seasons, prolonging dry conditions that aggregate wildlife survival threats. Notably, two EDE prevailed consecutively from the 1993 dry season to the 1994 wet season, coupled with a population decline of wildebeest (1.2–0.9 million), buffalo (40–20 thousand), and lion (1.3–0.9 thousand). We also note a reversal from more EDE to EWE during the study period. Prevalent EWE can lessen wet and vegetative conditions distribution gradient, which is imperative for the functioning of the SMME migratory ecosystem. Our study unveiled hotspot areas of extremes-driven ecosystem changes for the sustainable SMME migratory functioning essential for framing meaningful conservation management policies under climate change.


Introduction
The Serengeti-Masai Mara Ecosystem (SMME) is an iconic ecological and biodiversity resource in East Africa, with over a million wildebeests migrating between the Serengeti National Park in northeastern Tanzania and the Masai Mara National Reserve in southwestern Kenya (Sinclair et al 2015, Larsen et al 2020).Along with them are zebras and gazelles forming large herds of migrating ungulates and attracting many predators (Boone et al 2006, Sinclair et al 2008).Mineral resources, vegetation and water availability dynamics principally drive the ecological migratory functioning of the SMME (Boone et al 2006, Strauch 2013).These ecosystem resources are predominantly regulated by the annual rainfall cycle across the SMME (Dobson 2009, Holdo et al 2009).The SMME rainfall seasonality creates a gradient of wet and dry seasonal ranges, which promote northward and southward animal migration during dry and wet seasons, making a round trip of one of the longest remaining terrestrial migrations in the world (Subalusky et al 2017, Joly et al 2019).This spectacular migration feature has developed high ecotourism demand with enormous regional economic benefits (Larsen et al 2020).However, environmental shocks like droughts, floods, land conversion and degradation pose serious threats to the ecological functioning of this ecosystem and conservation management.
Rainfall variability has demonstrated quasiperiodic influence on ungulate population dynamics in African savannas (Ogutu and Owen-Smith 2005), including the SMME (Ogutu et al 2008b).Meanwhile, the SMME lion population revealed sudden growth from 10 to 20 years of stable equilibria during periods of enhanced precipitation over the 1960s-2000s, which resulted in enough prey supply and increased cub survival during the 1960s-2000s period (Packer et al 2005).Rainfall influences food and water availability and affects the prey's susceptibility to predation based on the prevailing habitat conditions (Ogutu andOwen-Smith 2005, Bartzke et al 2018).Further, studies show that floods and droughts are the major environmental disturbances regulating migratory ungulates and bird survival, whereas predation regulation affects small carnivores and residential ungulates (Fryxell andSinclair 1988, Sinclair et al 2007).This reflects how the impacts of climate extremes can extensively affect wildlife and associated ecosystem dynamics.
Regionally, East Africa experiences detrimental impacts of climate variability and change, with a notable increase in the intensity and frequency of climate extremes (Nicholson 2017, Mtewele et al 2021, Omondi and Lin 2023).Moreover, East Africa reveals a continued increase in hydrological extremes throughout the 21st century (Haile et al 2020, Gebrechorkos et al 2023), and nonlinear vegetation trends and high vegetation variability in response to precipitation and temperature change (Hawinkel et al 2016, Kalisa et al 2019, Huang et al 2021).Climate change-driven impacts on the SMME vegetation dynamics exceed human-induced disturbance (Hunninck et al 2020).This demonstrates the longterm vulnerability of the East Africa region to climate variability coupled with amplified climate change impacts.
Although the SMME is a quite studied region, much focus has been on mean climatic distributions and associated drivers (Mahony et al 2021).Only limited station observations have been employed for rainfall variability and trends to study localized patterns (Ogutu et al 2008a, 2008b, Bartzke et al 2018).The ecosystem dynamics with precipitation during climate extremes across the SMME are inadequately examined.Following the SMME's ecological importance worldwide and regional socioeconomic benefits, we aimed to provide a novel framework of spatiotemporal changes of rainfall extremes and their potential impacts on ecosystem changes based on vegetation, soil moisture and temperature, and subsequent prospects to migratory functioning across the ecosystem.

Study area
The SMME traverses an area of about 25 000 km 2 across the Kenya-Tanzania border and is surrounded by several protected areas acting as buffer zones of the ecosystem (Thirgood et al 2004, Sinclair et al 2015).The SMME lies on a plateau sloping between 2000 m (east) and 1200 m (west) above the mean sea level (figure 1(a)).The SMME extends between 33.8 • -35.4 • E and 1.2 • -3.3 • S.Here we used a slightly extended area of 33 • -36.2 • E, 0 • -4 • S for spatially based analyses, while temporal series analyses are only limited to the SMME boundary (figure 1).The ecohydrology of the SMME is supported by about four rivers (Mara, Grumeti, Mbalageti and Simiyu) which all drain westward to Lake Victoria, although only the Mara River flows throughout the year, making it the major wildlife water source during the dry season (Gereta et al 2009, Kihwele et al 2021).
The SMME features varied vegetation classes, partly accounted for by rainfall gradients and heterogeneous soil properties across the ecosystem (Holdo et al 2009, Huang et al 2021).Southeastern plains feature short grasslands with shallow and highly fertile soils, producing high-quality forage during wet periods that attract migrant herbivores (Sinclair et al 2008).On traversing westward and northward, the ecosystem features Acacia dominated woodlands vegetation coverage and tall grasses.The acquired land cover classification depicts dominant grassland coverage with trees and shrublands throughout the ecosystem (figure 1(b)).The SMME annual rainfall cycle and vegetation characteristics prompt ungulates migratory ecosystem functioning (figure 1(c)) (Sinclair et al 2008).In particular, the annual rainfall cycle characterizes the SMME with a wet season in November-May and a dry season in June-October (Serneels andLambin 2001, Bartzke et al 2018), which we have also adapted for seasonal-based analyses in this study.

Precipitation
For precipitation, we employed the Climate Hazards Group InfraRed Precipitation with Stations (CHIRPS) dataset, which is reliable for environmental monitoring, especially on seasonal droughts and rainfall trend analysis (Funk et al 2015).We used a 0.05 • monthly CHIRPS product over 1981-2020.CHIRPS datasets have demonstrated good performance and are recommended for early warning applications and climate extremes analysis in Africa (Funk et al 2015, Cattani et al 2018).We also used the Climatic Research Unit (CRU) precipitation dataset with monthly and 0.5 • resolutions for comparative analysis with population dynamics from the 1950s, taking advantage of its 1901-present long period (Harris et al 2020).The comparative analysis of CRU and CHIRPS precipitation datasets reveals good agreement over the SMME at both annual and wet and dry seasons (r > 0.78, SD ≈ 1, figure S1), corroborating their good performance in East Africa (Ongoma andChen 2017, Mbigi andXiao 2021).

Vegetation
We employed the Normalized Difference Vegetation Index (NDVI) as a proxy for vegetation which is inferred from the surface reflectance of Advanced Very High-Resolution Radiometer (AVHRR) aboard the National Oceanic and Atmospheric Administration (NOAA) polar-orbiting satellites (Vermote and Noaa 2019).NDVI is derived from the surface reflectance of red and near-infrared (NIR) radiations (equation (1)), which is a good indicator of vegetation condition as healthier vegetation reflects and absorbs more NIR and red radiations, respectively, and vice-versa (Pettorelli et al 2005).The AVHRR NDVI is available on a daily temporal resolution from 1981 to near-present with 0.05 • grid resolution.We employed a maximum value composite (MVC) procedure to generate monthly NDVI from the daily dataset.The MVC retains only the maximum value, which is likely free of cloud cover and with less atmospheric interference, hence capturing the actual surface vegetation state well.Moreover, it is considered that vegetation does not vary that much daily.This MVC technique is often used in satellite data observations and vegetation indices processing (Zhou et al 2014, Wu et al 2015) NDVI = NIR−RED NIR+RED . (1)

Soil moisture and soil temperature
The soil moisture and soil temperature datasets used in this study are sourced from the fifth generation of the European Centre for Medium-Range Weather Forecasting Reanalysis of Land datasets (ERA5-Land (Muñoz-Sabater et al 2021)).In particular, we employed the soil layer-1 product of the ERA5-Land monthly datasets, which span from 1950 to near-present with global coverage at 0.1 • grid resolution.This product gives the volume of water and temperature in a soil layer from the surface to 7 cm depth.

Animal population
The SMME animal population data were synthesized from several studies (table S1).Such datasets were acquired through various methods like aerial helicopter photographic mapping, installed camera shots, or driving across the park, which requires enough resources for long-term monitoring.As such, both past studies and updated records are coupled to continuously monitor population dynamics within the ecosystem.Here, we retrieved the abundance data of six mammals (wildebeest, gazelle, zebra, buffalo, elephant, and lion) across the SMME based on their availability.In particular, we used the Engauge Digitizer method (Mitchell et al 2020) to extract population data points from the identified graphs and their corresponding period, which is also used in other ecosystem studies for data extraction (Bao et al 2022).The population data extracted from each study for each mammal were then composited to generate the overall evolution during the 1950s-2010s, although some years had no records.

Extreme events analysis
We defined extreme events based on the observed temporal rainfall variability averaged over the SMME boundary during the 1982-2020 period.In particular, we generated the seasonal mean and temporal anomaly series, computed its standard deviation (SD) and identified the events deviating from the mean.Extreme wet events (EWE) are years with seasonal mean rainfall anomaly exceeding at least one-fold +SD, while those below one-fold −SD are defined as extreme dry events (EDE).According to CHIRPS, the wet season reveals six EDE ravaged the SMME in 1984SMME in , 1994SMME in , 1997SMME in , 1999SMME in , 2000SMME in and 2017, of , of  We further analyzed the probability of spatial distribution of precipitation across the SMME during EWE, EDE and normal rainfall periods as identified on CHIRPS.Moreover, we employed composite analysis (Lovison et al 1994, Nicolai-Shaw et al 2017) to examine extreme-driven changes from normal conditions for precipitation, vegetation, soil moisture and temperature distribution patterns.The Student's t-test was employed (Tarasińska 2005, Li andNadarajah 2020) to examine the statistical significance of the observed changes based on the 95% confidence level.We further evaluated the field significance for the obtained local p-values by applying the false discovery rate (α FDR = 0.05) as detailed in Wilks (2016) and stippling only fields passing this field test.

Correlation analysis
We performed the Pearson correlation analysis (Kornbrot 2005, Benesty et al 2009) to investigate the dynamic interaction of precipitation, vegetation, soil moisture, and temperature during the study period across the SMME.We also used this method to examine the dynamics of population changes with precipitation within the ecosystem, particularly for the corresponding years of available datasets.We employed the Student's t-test to evaluate their statistical significance at 95% and 99% confidence levels.

Rainfall climatology and extremes of the SMME
The annual cycle of the SMME reveals an apparent north-south spatiotemporal variation of rainfall, soil moisture and vegetation, with a characteristic dry season, in which the wet and vegetative conditions are only evident over the northern region (figures 3(a)-(c)).Meanwhile, the wet season exhibits general wet, moist and greenish coverage across the entire SMME.Furthermore, the southern SMME receives about 600 mm of annual rainfall, mainly in the wet season (>80%) (figures 3(d)-(f)).The annual rainfall gradually increases to 1000 mm on traversing northwestward, characterizing the northern SMME with up to 70% and 30% of wet and dry seasons accumulations, respectively, providing a dry range refuge for the northward migratory ecosystem.Similar features are observed for soil moisture and vegetation (NDVI) with more than 0.2 m 3 m −3 and 0.5, respectively, throughout the SMME during the wet season and heightening in magnitude westward and northward (figures 3(g) and (i)).During the dry season, moist and vegetative characteristics are mostly limited over northern SMME (figures 3(h) and (j)).
Distribution analysis of the rainfall extremes unveils as much as 400 mm month −1 of wet season EWE, widely distributed over 10% of the SMME (figure 4(a)).Meanwhile, EDE and normal rainfall periods depict most areas receive 60 and 80 mm month −1 , distributed over 9% and 10.5%, respectively.Nonetheless, a few places experience as low as 20 and 40 mm month −1 of wet season rainfall during EDE and EWE, respectively.Besides, dry season EDE characterizes much of the SMME with 20-60 mm month −1 , 15%-2%, respectively (figure 4(b)).For EWE, most areas receive 20-120 mm month −1 , 11%-2%, but up to 200 mm can be observed over some places (<0.5%).On the other hand, comparative analysis in extremesinduced rainfall changes relative to near-normal rainfall periods illustrates wet season EWE expounds a significant increase of 30%-50% throughout the SMME, whereas a substantial reduction (up to 30%) prevails during EDE (figures 4(c) and (d)).Similarly, dry season EWE reveals a 30%-60% significant rainfall increase across the SMME, while EDE suppresses by up to 45%, with a notable increase outside of the domain, southeast of SMME (figures 4(e) and (f)).These observed extreme rainfall changes have significant implications for the ecological functioning of the SMME.

SMME changes inferred from rainfall extremes
The SMME rainfall extremes induced changes expound hotspots of ecosystem changes and potential impacts on ecological migratory functioning.Wet season EWE is coupled with a significant increase in vegetation and soil moisture (>10%) and reduced soil temperature (>4%) across much of the southern plains of SMME (figures 5(a)-(c)).These changes indicate enhanced availability of ecosystem resources for the migratory herbivores that tend to occupy this region during the wet season and move away northward as the resources diminish.As such, EWE may lead to a late northward relocation of these migratory herbivores compared to average rainfall periods (figure 5(d)).Conversely, during EDE, the southern SMME suffers a significant reduction in vegetation and soil moisture (>12%, figures 5(e) and (f)), coupled with soil temperature rise (>3%, figure 5(g)), which promotes scarcity of migratory ecosystem resources and can lead to an early northward migration (figure 5(h)).The SMME wet season rainfall exhibits significant correlations with vegetation (0.33, p < 0.05), soil moisture (0.78, p < 0.01) and soil temperature (−0.51, p < 0.01) (figure S2(a)), suggesting impactful rainfall extremes induced changes in other ecosystem dynamics.
During the dry season, EWE corresponds to higher vegetation greenness and soil moisture, and lower soil temperature, mainly over the northern SMME (figures 6(a)-(c)), which is the dry season range for the SMME migratory herbivores.These herbivores can thus elongate their stay duration across these ranges before relocating southward following EWE-induced changes in resource availability (figure 6(d)).In contrast, dry season EDE suppresses vegetation and soil moisture and amplifies soil temperature across much of the SMME except over the southern SMME (figures 6(e)-(g)).These changes demonstrate a heightened scarcity of migratory ecosystem resources in the northern SMME dry range.This can prompt an early southward relocation to the southern edges of SMME and the surrounding buffer zone (Ngorongoro Conservation Area) with greener vegetation, less dry soil and low soil temperature environments (figure 6(h)).Nonetheless, here, we also note a significant correlation of the SMME dry season rainfall with soil moisture (0.7, p < 0.01) and insignificant correlations with soil temperature (−0.25) and vegetation (0.01), which may partly be accounted for by the lead-lag responses (figure S2(b)).

Responses of mammals population to rainfall extremes in the Serengeti-Masai Mara
The SMME mammals' population analysis demonstrates significant changes with remarkable compound rainfall extremes, i.e. consecutive EDE in two seasons (figure 7).The two EDE prevailed consecutively from the 1993 dry season to the 1994 wet season, coupled with a population decline of wildebeest (1.2-0.9 million), buffalo (40-20 thousand), and lion (1.3-0.9 thousand).This decline may have also been aggravated by the notable vegetation decline in the   1992 wet season, limiting lactation, newborns' survival and juvenile recruitment.Similarly, two EDE occurred consecutively in the 1997 wet and dry seasons, coupled with significant vegetation decline.However, this compound EDE was preceded by a 1996 dry season EWE and later followed by a 1998 wet season EWE.These EWE played a key role in restoring the wildebeest population as they also featured a remarkable increase in vegetation.Moreover, the SMME also suffered a prolonged compound EDE in 1999-2000, spanning four consecutive wet and dry seasons, resulting in a decrease in vegetation and wildebeest abundance (1.3-1.2 million).Meanwhile, the 2007 wet season EWE followed the 2006 dry season EDE, restoring suppressed ecosystem resources and promoting wildebeests regaining population.For EWE, successive events only prevailed from the 2019 dry season to the 2020 wet season, indicating that the SMME suffers more from compound EDE, surpassing multiple seasons in a row.However, such compound EDE has become less prevalent over the last decade, in which EWE features an increasing tendency.
Apart from compound EDE, the 1984 wet season EDE coupled and low vegetation also led to a remarkable wildebeest decline (1.3-1.1 million).Moreover, a large increase in the lion population (1300-1700) during 1985-1988 also exacerbated a prolonged wildebeest decline.This top-down lion-wildebeest population regulation also promoted bottom-up regulation, eventually restoring both almost to their mean levels by 1990.Population-rainfall dynamics also reveal positive correlations with wildebeest (0.39, p < 0.05), buffalo (0.42, p < 0.1), zebra (0.24) and gazelle (0.27) (figure S3).Specifically, the wet season shows high correlations with wildebeest and zebra, while the dry season exhibits high correlations with buffalo and gazelle, demonstrating respective seasonal extreme influence for their abundance.

Discussion
Our findings show that the SMME wet and vegetative conditions exhibit a north-south transition consistent with the movement of the tropical rainfall belt, which is the first-order driver of the annual rainfall cycle across the ecosystem (Bartzke et al 2018, Mahony et al 2021).This gives the SMME distinct wet and dry seasons (Musiega et al 2006, Ogutu et al 2008a, 2008b, Bartzke et al 2018).Further, Lake Victoria westerlies, topography, and tropical westerlies coupled interactions characterize the SMME rainfall gradient (Sinclair et al 2008, Mahony et al 2021).The low rainfalls over eastern SMME are due to overshadowing from the leeward side of the Ngorongoro highlands, while western and northern SMME benefits more wet characteristics from Lake Victoria (Mahony et al 2021).
The south-north and east-west spatial variations in vegetative and wet characteristics dictate migrant herbivores to move consistently with the observed variations throughout the year (Wolanski and Gereta  2013).In the late wet period, forage dries up, elevating surface water demand among herbivores, which raises the ecosystem competition and forces them to migrate northward to access quality water and green grasses (Strauch 2013).Moreover, the SMME wildebeests tend to avoid areas of reduced intake of green grasses along the season as the dry grasses increase enormously (Holdo et al 2009).Noteworthy, southern SMME river streams dry out during the dry season, whereas the Mara River flows across northern SMME throughout the year (Gereta et al 2009, Kihwele et al 2021).
The SMME suffers frequent rainfall extremes during both wet and dry seasons.However, these extremes expound a reversal from more EDE to EWE during the study period.This pattern corroborates with the reported turning of downward to upward rainfall trends over East Africa (Lyon and Vigaud 2017, Walker et al 2020, Mtewele et al 2021).This would infer a relative stabilization of animal population dynamics as EDE exacerbates survival threats more than EWE.However, multiple EDE aggregated into compound extremes, unlike the EWE, which would lessen wet and vegetative conditions distribution gradient imperative for the SMME migratory ecosystem.The 1997/1998El Niño (McPhaden 1999) led to the 1998 wet season EWE preceding the compound EDE in 1999-2000, which was also coupled with the multiyear La Niña events (Shabbar and Yu 2009, Okumura et al 2017, Raj Deepak et al 2019, Anderson et al 2023), usually developing after a strong El Niño (Iwakiri and Watanabe 2021).Principally, the El Niño Southern Oscillations (ENSO) and Indian Ocean Dipole (IOD) actively modulate the East African rainfall variability by developing wetter conditions during their positive phases and drier conditions during their negative phases for both short (October-December) and long (March-April) rains (Anderson et al 2023, Palmer et al 2023).This demonstrates that both wet and dry season EWE in SMME can be associated with positive ENSO and IOD while EDE with the negative phases.
During the wet season EWE, southern SMME features more vegetation coupled with wetter and cooler soil environments, which can support the wildlife ecosystem functioning relatively longer than during average rainfall years.This can delay the northward movement of the migrating herbivores, considering that increased rainfall and cooler temperatures lower water salinity, which also drives the migration timing toward the wet season-ending (Wolanski andGereta 2001, Strauch 2013).This finding is also consistent with the framework for migration-route patterns prediction, which demonstrated a delay in the northwestward movement of migrating wildebeest herds from the wetter-than-normal southern plains of SMME due to increased food abundance (Musiega et al 2006).Wet season EWE may also promote the growth of tall grasses, shrublands and woodlands, which provides better habitat conditions for carnivores' camouflage and higher herbivores' susceptibility to predation.Nonetheless, excessive wetness may lead to vegetation nutrient dilution and more favorable habitats for vectors of diseases across the ecosystem (Anyamba et al 2009, Bartzke et al 2018).
During the wet season EDE, we observed dry and warm conditions across much of the SMME, implying high mammal concentration near the prevailing water streams and rivers.This can raise competition and reduce population, especially for herbivores, falling into limited food and water availability and predation regulation (Sirot et al 2016).Past studies have also documented that before increased settlements and agricultural land conversion, wetlands downstream of the Mara River and Lake Victoria shores (western SMME) used to serve as refugia for the SMME migrants during droughts like that of 1993 (Mati et al 2008, Sinclair et al 2008).These anthropogenic changes led the migratory ecosystem to rely mainly on the Mara River during both the wet season EDE and dry season, threatening the ecosystem's sustainability (Dybas 2011, Mnaya et al 2017).Meanwhile, the observed dry season EWE with enhanced vegetation, soil moisture and precipitation over the northern SMME suggests a relatively delayed southward migration due to an extended availability of green forage and water.On the other hand, the southeastern SMME heightened wet and vegetation patterns during the dry season EDE may prompt an earlier southward migration than normal occupancies over the northern ranges with notable dry conditions.However, a further southward extension of these vegetative features up to Ngorongoro Conservation Area and Maswa Game Reserve (MGR) shows their potential to serve as migrants' refuge occupancy during dry season EDE, although the MGR is also reported as refugia for the wet season EDE (Sinclair et al 2008).
Generally, the long-term (1950s-2000s) SMME population analysis unveils that wildebeests and buffalos had a remarkable population growth until the early 1970s following the rinderpest virus eradication, which mainly decimated the ruminants' abundance (Sinclair et al 2008, Mariner et al 2012).Afterwards, the wildebeest population remained at a relative equilibrium of about 1.3 million, with notable changes following ecosystem regulation by predation, competition, and climate extremes.Besides, the buffalo population later declined due to increased competition, predation, and even poaching, also marked by the elephant population (Dublin and Ogutu 2015).Improved anti-poaching management and an international convention to stop the ivory trade enlisting elephants as endangered species resulted in gradually regaining the elephant population (Sinclair et al 2008, Wasser et al 2008, Wittemyer et al 2014).Meanwhile, the lion population exhibited a relatively consistent pattern with that of wildebeest, but its highest increase prevailed during the prolonged average rainfall period of [1985][1986][1987][1988][1989][1990].Compared to other mammals, the zebra population featured a relative stable pattern, with notable increase during average rainfall periods.However, it is worth noting that the prevailing discontinuities in population data also limited our detailed analysis.Nevertheless, correlation magnitudes are generally relatively higher during the wet than dry season.Birth synchronization, lactation, and juvenile recruitment for most SMME mammals happen during the wet season, and thus its variability can primarily affect the newborns' survival and account for the population dynamics (Sinclair et al 2008, Ogutu et al 2008b).

Conclusion
This study reveals that wet season EWE demonstrates increased moist and vegetative features over southern SMME ranges, suggesting elongating migrants' occupancy and subsequent late northward migration compared to average rainfall years.Conversely, wet season EDE expounds a significant suppression of these features, heightening ecosystem competition and survival threats, proposing an earlier northward migration.During the dry season, EWE heightens vegetation and water availability over northern SMME, favoring a delayed southward animal migration.For EDE, a significant reduction prevails, suggesting an early southward animal migration relocating to notable moist and vegetative characteristics.However, the availability of long-term animal migration data would offer further insights into the changes in exact timings for which the migrant herds delay or move early during these extreme events.Meanwhile, compound EDE aggravates detrimental impacts into consecutive seasons, significantly declining wildlife populations.Our findings are meaningful for enhancing and framing SMME conservation management policies.For further SMME studies, we recommend integral research observing and modeling shifting migration timing with climate change and the sustainability of the potential migratory refugia.

Figure 1 .
Figure 1.Study domain.(a) Topographical mapping, water resources and the Serengeti-Masai Mara Ecosystem (SMME) region (red polygon).On the top left is an insert of the African map locating the study domain (red rectangle).(b) Land cover across the study domain is adapted from the ESA World Cover (Global-2020-10m, (Zanaga et al 2021)) via the FAO Hand-in-Hand Geospatial Platform (https://data.apps.fao.org/).© ESA WorldCover project / Contains modified Copernicus Sentinel data (2020) processed by ESA WorldCover consortium.(c) The Great animal migration route across the SMME.

Figure 2 .
Figure 2. Extreme rainfall events during the wet (left panel) and dry (right panel) seasons over the SMME, as inferred from the CHIRPS (top panel, 1982-2020) and CRU (bottom panel, 1955-2020) datasets based on the 1982-2020 climatology.Years with extreme wet/dry events (EWE/EDE) exceeding/below standard deviation (SD) of precipitation by at least 1.0/−1.0and 1.5/−1.5 are marked by blue/red circles and rectangles, respectively.For years of climatological near-normal rainfall events within the SD range of −0.5-0.5 are marked by the black diamonds.
which the most severe (>1.5 SD) prevailed in 1984 and 2000 (figure2(a)).The ecosystem was also hit by six EWE in the wetseason, notably in 1998, 2007, 2010,  2016, 2018 and 2020, with 2007, 2018  and 2020 events depicting the highest magnitudes.The wet season EDE were more prevalent during the first half of the study period than the latter, in which EWE became more frequent.On the other hand, about six EDE and five EWE hit the SMME throughout the 1982-2020 dry season (figure 2(b)).The EDE prevailed more during the mid-study period, with the most severe events observed in 1993 and 1997 and others in 1999, 2000, 2005 and 2006.For EWE, the first was observed in 1996, but they occurred more often in the 2010s, notably in 2011, 2012, 2014 and 2019.This indicates a reversal from more EDE to EWE during the study period.We also identified EWE and EDE based on the CRU dataset during 1955-2020, protracted to animal population data span.We employed the 1982-2020 climatology for comparative analysis with CHIRPS.During the overlapping period, all CRU wet season EWE were also identified in CHIRPS, while coidentified wet season EDE are marked in 1984, 1999 and 2000.For the dry season, EWE concurrent in both datasets were in 2014 and 2019, while for EDE were in 1999 and 2000.Remarkably, contemporary EDE in CRU and CHIRPS extended from wet to dry seasons, creating compound EDE prolonging drought in 1999-2000.Similarly, a consecutive 1993 dry season-1994 wet season CHIRPS EDE was compounded by the 1994 dry season CRU EDE.For the full spectrum of CRU EWE and EDE up to an extended period, refer to figures 2(c) and (d).

Figure 3 .
Figure 3. SMME zonal annual cycle of (a) rainfall, (b) soil moisture, and (c) NDVI.Spatial distribution of (d) mean annual rainfall and its percentage accumulation from (e) wet and (f) dry seasons.Climatological mean spatial distribution of soil moisture during (e) wet and (f) dry seasons and the corresponding NDVI patterns, (g) and (h), respectively.

Figure 4 .
Figure 4. Extreme rainfall events.(a) Rainfall distribution for EWE, EDE and near-normal rainfall events across the SMME during the wet season.(b) As (a) but for the dry season.(c), (d) Wet season rainfall changes during EWE and EDE, respectively, relative to near-normal rainfall events.(e), (f) As (c), (d) but for the dry season.Stippling points mark significant fields passing the Student's t-test 5% significance level and the false discovery rate (αFDR = 0.05).

Figure 5 .
Figure 5. Wet season extreme rainfall-driven changes for vegetation (first column), soil moisture (second column), soil temperature (third column), and hypothesized impact on migration (fourth column) during EWE (top row) and EDE (bottom row) relative to near-normal events.Stippling points mark significant fields passing the Student's t-test 5% significance level and the false discovery rate (αFDR = 0.05).

Figure 6 .
Figure 6.As figure 5 but for the dry season.

Figure 7 .
Figure 7. (a) Population dynamics of wildebeest, zebra, gazelle, buffalo, elephant, and lion over the SMME.Datasets were retrieved from several studies enlisted in table S1.(b) Standardized NDVI anomalies over the SMME during the wet and dry seasons.(c) Extreme dry (EDE, red) and wet (EWE, blue) events over the SMME.Extremes based on the CHIRPS dataset during the wet (CHIRPS_WS) and dry (CHIRPS_DS) seasons are marked by downward and upward triangle symbols, respectively (figures 2(a) and(b)).Extremes based on the CRU dataset during the wet (CRU_WS) and dry (CRU_DS) seasons are marked by star and diamond symbols, respectively (figures 2(c) and (d)).The light blue polygon shade highlights the population regaining following rinderpest virus eradication.The light red polygons shade highlights compound EDE notable on both CHIRPS and CRU datasets.
Anderson W, Cook B I, Slinski K, Schwarzwald K, McNally A and Funk C 2023 Multiyear La Niña events and multiseason drought in the horn of Africa J. Hydrometeorol.24 119-31 Anyamba A, Chretien J-P, Small J, Tucker C J, Formenty P B, Richardson J H, Britch S C, Schnabel D C, Erickson R L and Linthicum K J 2009 Prediction of a Rift Valley fever outbreak Proc.Natl Acad.Sci.106 955-9 Bao T, Jia G and Xu X 2022 Warming enhances dominance of vascular plants over cryptogams across northern wetlands Glob.Change Biol.28 4097-109 Bartzke G S, Ogutu J O, Mukhopadhyay S, Mtui D, Dublin H T and Piepho H-P 2018 Rainfall trends and variation in the Maasai Mara ecosystem and their implications for animal population and biodiversity dynamics PLoS One 13 e0202814 Benesty J, Chen J, Huang Y and Cohen I 2009 Pearson correlation coefficient Springer Topics in Signal Processing (Springer) vol 2 pp 37-40 Boone R B, Thirgood S J and Hopcraft J G C 2006 Serengeti wildebeest migratory patterns modeled from rainfall and new vegetation growth Ecology 87 1987-94 Cattani E, Merino A, Guijarro J and Levizzani V 2018 East Africa rainfall trends and variability 1983-2015 using three long-term satellite products Remote Sens. 10 931 Dobson A 2009 Food-web structure and ecosystem services: insights from the Serengeti Phil.Trans.R. Soc.B 364 1665-82 Dublin H T and Ogutu J O 2015 Population regulation of African buffalo in the Mara-Serengeti ecosystem Wildl.Res.42 382-93 Dybas C L 2011 Saving the Serengeti-Masai Mara Bioscience 61 850-5 Fryxell J M and Sinclair A R E 1988 Causes and consequences of migration by large herbivores Trends Ecol.Evol. 3 237-41 Funk C et al 2015 The climate hazards infrared precipitation with stations-A new environmental record for monitoring extremes Sci.Data 2 1-21 Gebrechorkos S H, Taye M T, Birhanu B, Solomon D and Demissie T 2023 Future changes in climate and hydroclimate extremes in East Africa Earth's Future 11 1-21 Gereta E, Mwangomo E and Wolanski E 2009 Ecohydrology as a tool for the survival of the threatened Serengeti ecosystem Ecohydrol.Hydrobiol.9 115-24 Haile G G, Tang Q, Hosseini-Moghari S M, Liu X, Gebremicael T G, Leng G, Kebede A, Xu X and Yun X 2020 Projected impacts of climate change on drought patterns over East Africa Earth's Future 8 1-23 Harris I, Osborn T J, Jones P and Lister D 2020 Version 4 of the CRU TS monthly high-resolution gridded multivariate climate dataset Sci.Data 7 1-18 Hawinkel P, Thiery W, Lhermitte S, Swinnen E, Verbist B, Van Orshoven J and Muys B 2016 Vegetation response to precipitation variability in East Africa controlled by biogeographical factors J. Geophys.Res.Biogeosci.121 2422-44 Holdo R M, Holt R D and Fryxell J M 2009 Opposing rainfall and plant nutritional gradients best explain the wildebeest migration in the Serengeti Am.Nat.173 431-45 Huang N, Mondal P, Cook B I and McDermid S 2021 Moisture and temperature influences on nonlinear vegetation trends in Serengeti National Park Environ.Res.Lett.16 094049 Hunninck L, May R, Jackson C R, Palme R, Røskaft E and Sheriff M J 2020 Consequences of climate-induced vegetation changes exceed those of human disturbance for wild impala in the Serengeti ecosystem Conserv.Physiol.8 1-14