Volume 41, Issue S1 p. E2051-E2072
RESEARCH ARTICLE
Open Access

The precipitation patterns and atmospheric dynamics of the Serengeti National Park

Josephine Mahony

Corresponding Author

Josephine Mahony

School of Geography and the Environment, University of Oxford, Oxford, UK

Department of Zoology, University of Oxford, Oxford, UK

Correspondence

Josephine Mahony, School of Geography and the Environment, Oxford University Centre for the Environment, University of Oxford, South Parks Road, Oxford, OX1 3QY UK.

Email: [email protected]

Search for more papers by this author
Ellen Dyer

Ellen Dyer

School of Geography and the Environment, University of Oxford, Oxford, UK

Search for more papers by this author
Richard Washington

Richard Washington

School of Geography and the Environment, University of Oxford, Oxford, UK

Search for more papers by this author
First published: 08 September 2020
Citations: 6

Funding information: Natural Environment Research Council, Grant/Award Numbers: NE/L002612/1, NE/M020207/1

Abstract

The unique precipitation patterns over the Serengeti National Park in East Africa form the foundation of an internationally important ecosystem. Quantifying these precipitation patterns, and identifying the causal atmospheric processes, can improve understanding of past and future changes to regional rainfall. Precipitation and reanalysis datasets (CHIRPS v2.0, TRMM 3B42, ERA5) were used to quantify the regional climatic conditions on annual, monthly and hourly timescales. Hierarchical cluster analysis identified regions with distinct annual cycles of precipitation. Annual and monthly precipitation over the wider Serengeti domain (1°–4°S, 33°–37°E) was spatially heterogeneous. Cluster analysis identified five sub-regions with distinct annual cycles, with differing rainfall totals during January–February and June–September, wet season peak rainfall months, and rainfall peak symmetry. Seasonality was broadly controlled by the biannual passage of the tropical rainfall belt. Low-level wind, humidity and convergence patterns were impacted by the topography and Lake Victoria. An afternoon convergence zone between tropical easterlies and lake breeze winds always ran through the park and was associated with ascending motion and convection. The spatial progression of diurnal rainfall over the Serengeti followed the direction of 750 hPa tropical easterlies. The majority of the park received a late-afternoon rainfall peak, but from October to March an early afternoon peak was present between the wind convergence line over the central Serengeti and the rift topography. We propose that interactions between tropical easterlies, lake breeze westerlies and the topography control the spatial distribution of Serengeti precipitation, and that the seasonally changing rainfall gradient over the Serengeti may be generated by storms forming at the lake front and propagating in the direction of tropospheric easterlies. We suggest the early precipitation peak in the eastern Serengeti may be due to variability in the position of the lake front, or small storms generated by localized solar heating.

1 INTRODUCTION

Precipitation is a primary driver of one of the world's most iconic ecological phenomena: the annual migration of 1.2 million wildebeest around the 30,000 km2 of the Serengeti-Mara ecosystem (Holdo et al., 2009). A rainfall gradient stretches across the ecosystem with the north-west receiving three times more annual rainfall than the south-east (Norton-Griffiths et al., 1975), and the continued presence of precipitation in the north-west from June to October provides a refuge for the migratory herds during the dry season (Thirgood et al., 2004).

These complex spatial rainfall patterns are typical of East Africa, an area renowned for having markedly heterogeneous precipitation patterns (Nicholson, 1996). Across the wider region, annual totals and seasonality often change dramatically over tens of kilometres (Nicholson, 2017), as the large scale atmospheric circulation interacts with complex topography, local lake circulations and maritime influences (Camberlin, 2018). This climatic complexity supports an impressive array of ecosystems and human occupations (Niang et al., 2014), and could account for the conflicting rainfall trends recorded in the region (Ritchie, 2008; Ogutu et al., 2008b; Kizza et al., 2009; Veldhuis et al., 2019). Unfortunately, it is also a key reason for a lack of certainty about how climate change will affect future precipitation in areas such as the Lake Victoria basin (Dosio et al., 2019). Simulations with high-resolution models are required to simulate mesoscale atmospheric processes, and have been rare due to high computational and financial costs. A new generation of high-resolution regional climate model projections are now becoming available (Thiery et al., 2016; Kendon et al., 2019), and it is likely that these will improve precipitation projections over East Africa, including the Serengeti. However, to be able to evaluate the capacity of these climate models to simulate a region's future, we first require a detailed understanding of that region's past.

Despite the complexity of rainfall over the Serengeti-Mara ecosystem, and the recognition of its importance to the ecological community (Mduma et al., 1999; Fryxell et al., 2015; Veldhuis et al., 2019), limited work has focused on the region's mean climate. To date, the most thorough investigation into the regional rainfall was carried out by Norton-Griffiths et al. (1975). The authors quantified the rainfall gradient across the ecosystem, and noted that the orientation of this gradient altered throughout the year. They attributed these precipitation patterns to a rain shadow in the south-east, dry season wind convergence over Uganda and western Kenya (illustrated by Findlater, 1971), and a low-level lake convergence zone which migrated between Lake Victoria and the western Kenyan highlands over the diurnal cycle (described by Fraedrich, 1972).

For decades, the work carried out by Norton-Griffiths et al. (1975) has provided a valuable foundation for understanding the ecosystem's unique climate, but this was undertaken using limited data by current standards. More recent satellite datasets, with high spatial and temporal coverage and resolution, now allow us to examine the region's climatological precipitation patterns in greater detail. Reanalysis circulation data provides the opportunity to test the mechanisms suggested by Norton-Griffiths et al. (1975), and better understand the dynamics of this region.

Here, we use these datasets to quantify the spatial distribution of rainfall over the Serengeti on annual, monthly and hourly timescales, and identify the atmospheric processes generating the unique rainfall patterns observed over this important ecosystem. We focus the majority of our analysis on the area shown in Figure 1 (1°–4°S, 33°–37°E), which we define as the wider Serengeti domain. However, when focusing on larger swathes of East Africa, we investigate the region between 3°N and 9. 5°S, 30° and 42°E, which includes Lake Victoria, the southern Lake Victoria basin, the western Indian Ocean and the eastern branch of the East African Rift Valley.

Details are in the caption following the image
Map of the orography and key landmarks of the Serengeti National Park and surrounding area. The area shown in this figure is termed as the “wider Serengeti domain” in this study. Shading and contour lines denote elevation (using Gtopo30 topography data), with the exception of lakes which are masked. The dashed line shows the path of atmospheric cross-sections analysed in Section 24

In this paper, a description of the observational datasets and methodology are presented in Section 2. In the results section, we quantify the spatiotemporal distribution of annual and monthly rainfall totals over the wider Serengeti domain in Section 10; examine regional wind, precipitation, and moisture patterns over East Africa in Section 20; then analyse the diurnal cycle of wind and precipitation patterns over the wider Serengeti domain, and how they change throughout the year, in Section 23. Section 27 contains a discussion of the results, with individual sections dedicated to the evaluation of the precipitation data and three key drivers of spatial heterogeneity in rainfall across the wider Serengeti domain (Sections 27, 30). We go on to discuss how interactions between these features may generate the seasonally changing rainfall gradient across the Serengeti (Section 31); the possible drivers of afternoon storm development (Section 32); and the potential causal mechanisms of the region's diverse annual cycles (Section 33). Finally, Section 35 contains the study conclusions.

2 DATA AND METHODS

2.1 Datasets

Precipitation, reanalysis and topography datasets were used to explore the atmospheric conditions over the Serengeti and East Africa. When data is shown for different points in the diurnal cycle, this study uses East Africa Time (EAT), which is 3 hours ahead of Coordinated Universal Time (UTC).

2.1.1 Precipitation

The satellite products CHIRPS v2.0 and TRMM 3B42 were used in this study. The Climate Hazards Group InfraRed Precipitation with Stations dataset (CHIRPS v2.0, hereafter CHIRPS) was chosen due its high spatial resolution, large temporal range and proven performance over East Africa (Dinku et al., 2018). It provides daily precipitation estimates at 0.05° resolution between 1981 and 2017, based on blended station and satellite data (Funk et al., 2014). Rain gauge measurements from the wider Serengeti domain (1°–4°S, 33°–37°E) were included in the CHIRPS dataset over the whole time period of the study, although the number dropped from 120 at the start of 1987 to 3 by the end of 2016 (information from the Climate Hazards Centre, 2017). Tropical Rainfall Measuring Mission (TRMM) 3B42 (hereafter TRMM) data is available at 0.25° resolution, between 1998 and 2014 (Huffman et al., 2007). The TRMM dataset was chosen for the availability of 3-hr measurements, and its strong agreement with in situ readings over the Lake Victoria basin (Haile et al., 2013), and Tanzania and Kenya (Camberlin et al., 2018). Climatological averages were calculated for CHIRPS from 1987–2016, and TRMM from 1998–2014.

2.1.2 Wind and specific humidity

ERA5 reanalysis wind and specific humidity data were used to examine the regional atmospheric dynamics. This dataset was chosen for its high temporal and spatial resolution, as it provides hourly global data on a 0.25° x 0.25° grid (Hersbach et al., 2020). Our study included monthly and hourly climatological averages of ERA5 horizontal wind, vertical velocity, and specific humidity, using data from 1987 to 2016. A 825 hPa specific humidity data was used in Section 21 as surface values were not available for this variable (Copernicus Climate Change Service [C3S], 2019).

2.1.3 Topography

A 2-minute resolution version of the Global 30 Arc-Second Elevation (GTOPO30) dataset was used to display topography (Gesch et al., 1999).

2.2 Methods

2.2.1 Cluster analysis

We used cluster analysis to identify areas with similar annual cycles of precipitation. The annual cycle of each grid cell was normalised prior to the cluster analysis, to ensure all monthly values fell between 0 and 1, using the formula:
urn:x-wiley:08998418:media:joc6831:joc6831-math-0001
xi, individual monthly climatological totals; xmax, wettest monthly total; xmin, driest monthly total.

Normalising the data meant the clustering algorithm detected differences between annual cycle structure, rather than annual rainfall totals, allowing us to better identify the impact of different atmospheric processes.

We calculated the Euclidean distance between each normalised annual cycle, then used this data to perform hierarchical cluster analysis using Ward's minimum variance method (Ward, 1963). This was achieved using the dist (method = “Euclidean”) and hclust (method = “ward.D2”) functions in R (Core Team, 2019). This analysis was performed separately on each precipitation dataset.

The number of clusters used in each dataset were chosen based on the size of the step changes in Euclidean distance between different numbers of clusters, and the spatial coherence of those clusters. Further details can be found in the Appendix S1.

2.2.2 Atmospheric and topographic cross-sections

Diagonal cross-sections of the atmosphere and topography were generated between 1°S, 33°E and 4°S, 36°E (Figure 1) using the MetPy library (May et al., 2020). Transects were generated using the cross_section() function, and horizontal wind vectors tangential to the transect line were calculated using the tangential_component() function. The transect line was chosen as it included both Lake Victoria and a south-eastern part of the Serengeti National Park: an area of crucial ecological importance where the wildebeest calve in February/March (Hopcraft et al., 2015).

3 RESULTS

3.1 Annual and monthly precipitation over the wider Serengeti domain

In the first section of our study, we mapped the spatial and temporal distribution of rainfall over the wider Serengeti domain on annual and monthly timescales.

3.1.1 Annual precipitation

We aimed to identify the key spatial patterns of regional precipitation, beginning with the climatological annual mean, and assess the stability of these patterns across two datasets (CHIRPS and TRMM). Figure 2 shows a south-east to north-west rainfall gradient across the Serengeti National Park in both datasets. Areas north-west of the park received totals approximately three times greater than those to the south-east (approximately 1,200 and 400 mm year−1, respectively in CHIRPS). In both datasets, the west of the wider Serengeti domain was wetter than the east, although high rainfall totals (1,200–1,500 mm) were recorded over south-eastern highlands. Rainfall generally increased with proximity to Lake Victoria.

Details are in the caption following the image
Climatological mean annual precipitation in the Serengeti and surrounding region. Calculated from (a) CHIRPS data (1987–2016); and (b) TRMM data (1998–2014). Gtopo30 topography data overlaid

However, there were also differences between CHIRPS and TRMM. TRMM had a lower spatial resolution, less distinction between rainfall totals on opposing mountain sides, and a more linear, homogenous appearance to the driest area in the south-east of the Serengeti National Park. CHIRPS had lower regional maximum and minimum values than TRMM, and TRMM recorded higher rainfall totals over Lake Eyasi, Lake Manyara and the highland area north of the Serengeti. The two climatological averages cover different time periods (1987–2016 for CHIRPS; 1998–2014 for TRMM), which partially explains these discrepancies. However, similar patterns remained when TRMM and CHIRPS were compared over the same time period (1998–2014; results not shown). TRMM still recorded higher values over the majority of the wider Serengeti domain, including higher totals over Lake Manyara, Lake Eyasi and east of Lake Victoria, and lower totals over certain south-facing mountain slopes as well as a small region south-east of the Serengeti National Park.

3.1.2 Monthly precipitation

We next examined the evolution of the spatial distribution of monthly precipitation patterns through the annual cycle (Figure 3). We focused solely on CHIRPS data due to its higher spatial resolution. Precipitation was recorded across the majority of the wider Serengeti domain from November to April. In these months, elevated rainfall totals were often recorded over high ground, Lake Victoria, and its shoreline, although in January and February the highest precipitation values were found south of the lake. May and October saw high precipitation totals on mountain slopes and along the shores of Lake Victoria, and low rainfall totals in the rift valley and south-west of the Serengeti National Park. From June to September, rainfall was limited to Lake Victoria, Ngorongoro Crater, and the area north of the Serengeti National Park. The wettest and driest months were April and July, respectively.

Details are in the caption following the image
Monthly climatological mean precipitation across the wider Serengeti domain. Monthly rainfall climatologies calculated from the CHIRPS dataset (1987–2016) for each month, starting with January (a) and finishing with December (l). Gtopo30 topography data overlaid

The rainfall gradient identified in the annual climatology was present throughout the year, however, its orientation altered. From March to May, and October to November, the north-west received greater rainfall than the south-east. From June to September, the north received more rainfall than the south, and from December to February, the south-west received greater precipitation than the north-east. Another notable seasonal feature was exceptionally high rainfall totals on the eastern and southern slopes of the south-eastern mountains in April and May.

3.1.3 Variation in annual cycles

The monthly changes to the spatial distribution of rainfall suggested annual cycles would be distinctly different across the region. We therefore examined the diversity of annual cycles present in the wider Serengeti domain (Figure 4). The black lines show mean precipitation across the whole region. The means of both datasets reveal a dry season from June to October, and a wet season from November to May. This wet season had two rainfall peaks (one in November/December and a larger one in March/April) and a reduced amount of rainfall in the interceding January–February period. When comparing the mean annual cycles of TRMM and CHIRPS, the monthly totals were similar, which was unexpected given the higher regional maximum and minimum values recorded in TRMM (Figure 2). This suggests that in the TRMM dataset, there were a greater number of comparatively dry grid cells to compensate for the higher maximum and minimum precipitation values.

Details are in the caption following the image
Annual cycle of precipitation over the wider Serengeti domain (1–4°S, 33–37°E). The dashed and dotted line shows the annual cycle averaged across all of the CHIRPS grid cells, while the dashed line shows the average of the TRMM data. Solid lines show the climatological annual cycles of individual grid cells in the CHIRPS and TRMM datasets. Data (CHIRPS 1987–2016, and TRMM 1998–2014)

Examining the annual cycles of individual locations, there was a large range in the magnitude of precipitation totals. CHIRPS had the largest range in precipitation values, but also a greater number of data points. There was also variability in peak rainfall months, with different localities having peaks in November or December, and March or April.

Figure 5 shows that there are clear patterns to the spatial distribution of peak rainfall months, which appear partially linked to local topography. Southern areas, west of the East African Rift mountains, typically had a March instead of an April peak (Figure 5a). Later in the year, the majority of the Serengeti National Park had a December peak, while a November peak occurred in areas close to Lake Victoria and in northern areas east of the East African Rift mountains (Figure 5b).

Details are in the caption following the image
Spatial distribution of change in precipitation signal between consecutive core wet season months. Shading illustrates peak rainfall month during (a) the long rains, and (b) the short rains. Data: CHIRPS (1987–2016). Gtopo30 topography data overlaid

3.1.4 Cluster analysis

We performed cluster analysis on the normalised annual cycles represented in each of the precipitation datasets, to identify areas with similar annual precipitation cycles. While there were discrepancies between the datasets, distinct annual cycles were identified in the north-west, south-west, east and centre of the region (Figure 6). In the TRMM dataset, an additional cluster was identified in the north-east (Figure 6a). The mean normalised annual cycles of each cluster are shown in Figures 6b and d. The key differences that emerged between the annual cycles of each cluster were dry season precipitation, wet season peak months, the ratio between wet season and January–February rainfall totals, and the symmetry of the rainfall peaks (Table 1).

Details are in the caption following the image
Cluster analysis of normalised climatological annual cycles. For each gridcell the annual cycle of precipitation was clustered using hierarchical cluster analysis (Ward's criterion). The spatial distribution of the resulting clusters is shown for the (a) TRMM (1998–2014); and (c) CHIRPS (1987–2016) precipitation datasets. The normalised annual cycles for the identified clusters are shown for (b) TRMM; and (d) CHIRPS
TABLE 1. Summary of the key characteristics of annual precipitation cycles in different parts of the wider Serengeti domain
Cluster Proportional January-February precipitation Proportional June-October precipitation Wet season peaks Rainfall peak symmetry
Nov/Dec Mar/Apr
South-west High Low Dec Mar Symmetric
East Low Low Dec Apr Asymmetric
North-west ~ High Nov Apr ~
North-east Low Low Nov Apr Asymmetric
Central High ~ Dec ~ ~
  • Note: The ~ symbol represents disagreement between clusters in different datasets.

South-west

The annual cycles in the south-west (cluster 1 in TRMM and CHIRPS) consistently had a symmetrical profile, with peaks in December and March. The ratios between peak months and January/February totals were low; the annual cycle had an appearance closer to a single rather than a double-peaked wet season. The proportion of annual rainfall occurring in the dry season was low.

East

In the east (cluster 2 in TRMM and CHIRPS), December and April were peak rainfall months. There was an asymmetry between these rainfall peaks, with much higher totals in April. Only slightly more rainfall occurred in November compared to January and February. In this region, a low proportion of annual precipitation fell during the dry season.

North-west

Cluster 4 in the TRMM data covered an area over the north-west. Cluster 3 in the CHIRPS data included the north-east of the region, and a large area on the shores of Lake Victoria. Both had rainfall peaks in November and April, and proportionally high dry season rainfall totals. However, the symmetry between the rainfall peaks, as well as ratios between peak rainfall months and January–February totals, differed between datasets.

North-east

The annual cycle of TRMM's cluster 5 in the north-east, was similar to TRMM's cluster 4 in the north-west. It had November and April peaks, and relatively high rainfall totals in these months compared to January and February. The main differences were that in the north-east the April peak was higher than the November peak, and the proportion of rain falling in the dry season was lower.

Central

While a central cluster was present in both datasets, the spatial extent and structure of the mean annual cycles differed. This may be an artefact of the analysis, or alternatively suggests that local annual cycles changed rapidly over short distances in this area. The L-shaped central TRMM cluster (cluster 3) had wet season peak months in December and March, and a high proportion of rainfall in the dry season. There was a moderate difference between January–February totals and peak month totals. The central CHIRPS cluster (cluster 4) extended from 2° to 3°S and 33° to 35°E. It had asymmetric December and April peaks, high levels of dry season rainfall, and a moderate ratio between January–February totals and wet season rainfall peaks.

3.2 Seasonal and diurnal wind and precipitation over East Africa

The next stage of our study explored the large-scale atmospheric mechanisms that could be causing the spatial distribution of regional rainfall, and the distinct annual cycles identified by the cluster analysis.

3.2.1 Wind and precipitation patterns on monthly and seasonal timescales

Figures 7 and 8 show climatological 750 hPa winds over East Africa. Easterlies were present throughout the year. The meridional wind direction switched from northerly in March (Figure 7a) to southerly in May (Figure 7e), and from southerly in October (Figure 7b) to northerly in December (Figure 7f). Higher rainfall totals were recorded over Lake Victoria and much of the topography in the Lake Victoria basin. The wettest periods in the Serengeti were similar to the rest of equatorial East Africa, coinciding with the migration of the tropical rainfall belt over the equator.

Details are in the caption following the image
Climatological precipitation and 750 hPa wind patterns during the equatorial East African wet seasons. Precipitation (CHIRPS, 1987–2016) and wind (ERA5, 1986–2017) climatologies shown for (a) March; (b) October; (c) April; (d) November; (e) May; and (f) December. Wind vectors were regridded to a resolution of 0.75° for ease of viewing. Gtopo30 topography data overlaid
Details are in the caption following the image
Climatological precipitation and 750 hPa wind patterns during the equatorial East African dry seasons. Precipitation (CHIRPS, 1987–2016) and wind (ERA5, 1986–2017) climatologies shown for (a) January–February; and (b) June–September. Winds regridded to 0.75 ° resolution. Gtopo30 topography data overlaid

The equatorial East African “dry seasons” occurred in January–February and June–September when the tropical rainfall belt was furthest from the equator. However, large areas of East Africa received substantial rainfall during these “dry seasons,” including much of Tanzania in January–February, and Uganda and south-western Kenya in June–September (Figure 8). The Serengeti National Park is at the edge of both wet areas, and portions of the park continued to receive rainfall during these months. Moisture provided by Lake Victoria appeared a plausible reason for increased rainfall over the Serengeti during “dry season” months. To investigate how moisture could travel eastwards, against tropical easterlies, we next examined surface winds.

3.2.2 Wind, divergence and moisture patterns on annual, monthly and diurnal timescales

Figure 9 shows surface wind patterns were strongly influenced by topographic features, the East African coastline, and land-lake breezes. The winds diverged and converged over the largest lakes at 06:00 and 18:00, respectively, while the topography of the East African rift both channelled and blocked the tropical easterlies from the Indian Ocean. Figure 9b shows that at 18:00 convergence occurred over the Serengeti National Park, and regions where diverging lake breeze winds from Lake Victoria met tropical easterlies or topographic boundaries. Complex patterns occurred over the East African rift highlands, with strong convergence over the highest topography and divergence in the valleys.

Details are in the caption following the image
Mean annual surface wind (vectors) and divergence (shading) over east African for (a) early morning (06:00 EAT) and (b) early evening (18:00 EAT). Wind vectors were regridded to a 0.5° resolution for ease of viewing, while divergence values are displayed at the native resolution of 0.25°. Data: ERA5 (1987–2016). Gtopo30 topography data overlaid

Analysis of monthly 18:00 wind and divergence patterns showed that they changed throughout the year. Divergence associated with Lake Victoria extended up to 2° of latitude north (Figure 10b, c) or south (Figure 10a, d) of the lake, depending on the meridional component of the tropical easterlies. However, convergence was always present over the Serengeti (Figure 10).

Details are in the caption following the image
Mean surface wind (vectors) and divergence (shaded) at 18:00 EAT in (a) February, (b) April, (c) August, and (d) November. Wind vectors were regridded to a resolution of 0.5° for ease of viewing, while divergence values are displayed at the native resolution of 0.25°. Data: ERA5 (1987–2016). Gtopo30 topography data overlaid

These wind and convergence patterns played a large part in the distribution of moisture from Lake Victoria (Figure 11). Relatively high levels of specific humidity were present across the whole of East Africa from November to May. In other months, the rift topography acted as a barrier to moisture being transported westward from the Indian Ocean, illustrating the rain shadow generated by these highlands. Within the Lake Victoria basin, increased moisture levels were associated with diverging winds from Lake Victoria.

Details are in the caption following the image
Mean monthly early evening (18:00 EAT) surface wind patterns (vectors) and 825 hPa specific humidity (shading) from ERA5 data (1987–2016). Specific humidity values at levels below surface pressure are masked. Gtopo30 topography data overlaid

3.3 Diurnal wind and precipitation over the wider Serengeti domain

The results in Section 20 showed an association between rainfall and afternoon surface wind patterns, and illustrated the importance of the diurnally changing lower level winds to moisture availability within the Lake Victoria basin. Therefore, the final stage of our study investigated the evolution of rainfall and wind patterns over the wider Serengeti domain on an hourly timescale.

3.3.1 Diurnal precipitation and surface wind patterns

Figure 12 illustrates the time of peak annual rainfall across the wider Serengeti domain. Across the majority of this area, the 3-hr window of peak rainfall represented a significant contribution to total diurnal rainfall (μ = 33.6%, σ = 10.4%), although this value varied based on location and time of year (range = 15–75%). The shape of the rainfall peak was also relatively sharp in most locations, although this too varied with location and month.

Details are in the caption following the image
Time of maximum annual precipitation. Analysis based on precipitation climatologies that were calculated for each 3-hr timestep of the TRMM dataset (1998–2014). Gtopo30 topography data overlaid

Maximum rainfall occurred between 03:00–09:00 over Lake Victoria, around Lake Manyara and to the east of Ngorongoro Crater and the rift highlands. A 21:00 maximum was found east of the Loita Hills, near Lake Natron, and south of Ngorongoro Crater. The majority of the rest of the region had a late-afternoon/early evening rainfall maximum of 18:00, including half of the Serengeti National Park. However, along the eastern edge of the Serengeti National Park, maximum rainfall occurred in the afternoon at 15:00.

This mid-afternoon rainfall peak over the Serengeti National Park was situated where a rain shadow from the highlands would be expected. This mid-afternoon rainfall peak was not present all year (Figure 13), and instead appeared correlated to the southerly extent of the diverging lake breeze winds. From October to March, the lake westerlies extended below 4°S, and there was a relatively close (although not exact) match between the wind convergence line and the boundary between the areas receiving mid- and late-afternoon rainfall peaks. This relationship partially broke down in April, May and September; months when the lake breeze winds did not reach below 3°S. From June to August, when the surface lake breeze winds did not reach any part of the Serengeti National Park below 2°S, mid-afternoon rainfall peaks were absent from most of the wider Serengeti domain.

Details are in the caption following the image
Monthly time of maximum precipitation and 15:00 (EAT) surface wind patterns. Plots (a)-(l) represent January–December respectively. Precipitation climatologies were calculated for each 3-hourly timestep using TRMM data (1998–2014), and wind climatologies were calculated using surface wind vectors from the ERA5 dataset (1987–2016). Gtopo30 topography data overlaid

To better understand the role of the lake breeze westerlies in generating rainfall at different points in the annual and diurnal cycle, we inspected the afternoon climatological rainfall and wind patterns in a wet and dry season month (March and August) (Figure 14).

Details are in the caption following the image
Mean afternoon precipitation and surface wind patterns during a wet and dry season month. Mean hourly rainfall and wind vectors were calculated during March for (a) 15:00 (13:30–16:30 precipitation), (b) 18:00 (16:30–19:30 precipitation), and (c) 21:00 (19:30–22:30 precipitation), and in August for (d) 15:00 (13:30–16:30 precipitation), (e) 18:00 (16:30–19:30 precipitation), and (f) 21:00 (19:30–22:30 precipitation). Times in East Africa time. Data: TRMM (1998–2014) and ERA5 (1987–2016). Gtopo30 topography data overlaid

In both March and August, rainfall over the Serengeti National Park was more widespread at 15:00 (Figure 14a, d) than at 18:00 (Figure 14b, e), and 15:00 rainfall did not concentrate in areas where the winds were converging. By 18:00 the lake breeze winds had penetrated slightly further east. The maximum rainfall totals recorded in the park increased, but the spatial distribution of that rainfall shifted to the west of the converging winds (Figure 14b, e). This pattern remained at 21:00, although the wind convergence line and areas of highest rainfall moved westward towards the lake (Figure 14c, f).

When comparing 15:00 rainfall in the 2 months, rainfall was widely spread across the Serengeti National Park in March (including in the “rain shadow” of the rift topography), whereas limited rainfall occurred in August. At 18:00 and 21:00, high rainfall totals were recorded over the north-west of the wider Serengeti domain in both March and August. The main differences were in the spatial extent of rainfall, with minimal totals recorded in August away from the northern highlands and southern lakes.

3.3.2 Monthly afternoon vertical wind profiles

To further understand the temporal evolution of rainfall-generating atmospheric conditions, we examined the vertical structure of the atmosphere over a diagonal transect between 1°S, 33°E and 4°S, 36°E (path shown in Figure 1). Figure 15 shows afternoon/evening vertical atmospheric cross-sections during March (Figures 15a, e) and August (Figures 15b, d, f), and the topography of the transect line (Figures 15g, h). Distinct differences were present between March and April, both above and below 600 hPa.

Details are in the caption following the image
Topographic and afternoon atmospheric cross-sections over the Serengeti during a wet and dry month of the year. The cross-sections run in a diagonal line from 1°S, 33°E in the north-west (NW) to 4°S, 36°E in the south-east (SE) (path shown in Figure 1). Individual panels show atmospheric conditions in March at (a) 15:00 EAT, (b) 18:00 EAT, (c) 21:00 EAT, and August at (d) 15:00 EAT, (e) 18:00 EAT, and (f) 21:00 EAT. Panels (g) and (h) are identical and depict the topography along the transect line (Gtopo30 data). In panels (a)-(f) shading illustrates ascending (negative vertical velocity values) and descending (positive vertical velocity values) motion. Values below the surface pressure are masked with black. The x-components of the vectors show the magnitude of winds tangential to the transect line (m·s−1). The diagonal nature of the transect means that vectors with a positive (negative) x-component represent the north-westerly (south-easterly) component of the winds. The y-components show vertical velocity (pa·s−1). The vertical component of the vectors is scaled by −10 for ease of viewing. The 10-fold multiplication of the vectors allows the vertical motion to be more easily visible, while the negative multiplication allows areas with ascending motion to have upward-pointing arrows. Data: ERA5 (1987–2016)

Above 600 hPa, there was near-ubiquitous descending motion in August, with weak south-easterly winds and the presence of winds with a north-westerly component between 500 and 650 hPa (Figures 15b, d and e). March had stronger south-easterly winds but more variable patterns of vertical motion. At 15:00 descending motion was present north-west to the lake shoreline (33.7°E) and south-east of 35.7°E, with ascending motion present in between (Figure 15a). This region of ascending motion (which extended from the surface to 400 hPa) coincided with the areas receiving rainfall in Figure 14a. By 18:00, descending motion was present over the lake and stretched diagonally from the lake shoreline at the surface (33.7°E) to approximately 35.2°E at 400 hPa (Figure 15c). By 21:00 ascending motion had returned to the majority of the troposphere above 600 hPa (Figure 15e).

Below 600 hPa, there was a clear lake circulation cell in the north-west, strong ascending motion over the centre, and topographic effects in the south-east. South-easterly winds were stronger in August than March. The lake circulation cell occurred over Lake Victoria and the adjoining land, with air descending and ascending in the north-western and south-eastern portions, respectively. The position and size of the cell changed depending on month and time of day. Between 15:00 and 18:00, the centre of the lake cell moved inland, then retreated back towards the lake by 21:00. The height of the circulation cell also decreased to below 700 hPa by 21:00. The centre of the circulation cell reached further inland in March (33.8°E, 34.0°E and 34.0°E at 15:00, 18:00 and 21:00, respectively) than in August (33.6°, 33.7° and 33.5°E at 15:00, 18:00 and 21:00, respectively), and the orientation of the cell was steeper in August.

A peak of ascending motion was always present at approximately 700–800 hPa over 34.5°E, a position just north-west of an increase in the topography of approximately 260 m over 7 km. At 15:00, this peak was predominantly present between 34.4° and 35.1°E, which is slightly north-west of a topographic plateau (34.6°–35.2°E). This region of ascending motion moved towards the north-west/lake at 18:00, and again at 21:00. This was despite the surface lake breeze winds reaching further inland at 18:00 than 15:00. In addition, the magnitude of peak ascending motion increased between 15:00 and 21:00 in August, and decreased in March.

Finally, there were topographic effects present below 700 hPa in the south-east. Ascending motion was present north-west of 35.2°E and south-east of 35.7°E, approximately corresponding to the escarpments next to Lake Eyasi and Lake Manyara, respectively. Descending motion was present between these regions of ascent. This matched a gap in the rift topography between Lake Eyasi and Lake Manyara (Figure 1, Figure 15g), making it likely that the descending motion was due to zonal winds being funnelled through this channel.

4 DISCUSSION

The complex precipitation patterns over the Serengeti National Park are an integral part of the internationally important Serengeti-Mara ecosystem. Our results suggest that these diverse precipitation conditions are caused by interactions between tropical easterlies, land-lake breezes generated by Lake Victoria, and regional topography. Here, we consider the performance of the precipitation data used in our study and the roles of the above three key features in controlling regional precipitation. We then discuss their combined effect on the spatial distribution of rainfall, local storm development, and seasonality over the Serengeti.

4.1 Evaluation of precipitation data

If we compare our results to monthly rain gauge measurements reported by Norton-Griffiths et al. (1975), we see that both TRMM and CHIRPS captured the rainfall gradient and spectrum of annual cycles previously reported over the Serengeti National Park. The magnitude of the CHIRPS measurements closely matched those reported by Norton-Griffiths et al. (1975), whereas TRMM values were higher across the majority of the national park, particularly east of Lake Victoria. These high totals may be partially due to the shorter time period used to calculate the TRMM climatology (1998–2014), as it contained two exceptionally wet years over the northern Serengeti (1998 and 2001) (Ogutu et al., 2008a, 2008b). However, TRMM also differed from CHIRPS readings taken from 1998 to 2014, with patterns similar to those found by Haile et al. (2013); TRMM 3B42 overestimated totals over lake surfaces and underestimated totals over mountaintops. Overall, this suggests both TRMM and CHIRPS provide accurate representations of the spatial and temporal patterns of monthly precipitation over the wider Serengeti domain, but the absolute precipitation totals in TRMM should be interpreted with caution.

4.2 Large-scale circulation patterns

Examination of annual cycles across the wider Serengeti domain showed that all locations had: (a) a drier period from June to September/October; (b) a double-peaked wet period from October/November to May; and (c) rainfall peaks in November/December and March/April. These results were expected, since it is well documented that the wettest periods in East Africa occur between March to May and October to December during the biannual passage of the tropical rainfall belt (Nicholson, 2018). In these months, the tropical easterly winds are weak and have a low meridional component, while Indian Ocean sea surface temperatures (SSTs) are high, resulting in an unstable atmosphere (Yang et al., 2015). Conversely, low SSTs, strong descending motion in the mid-troposphere, and enhanced cross-equatorial easterly winds during the dry seasons of January–February and June–September, result in a stable atmosphere which suppresses moist convection (Yang et al., 2015). Our results confirm the major determinants of seasonality in the Serengeti are the tropical rainfall belt, and the large-scale circulation patterns that dominate East African climate. Future changes to large-scale regional circulation would therefore likely impact Serengeti rainfall.

4.3 Topography

Despite overarching similarities amongst annual cycles over the wider Serengeti domain, there were also distinct differences, many of which were associated with the rift topography. Our results show that high rainfall totals were recorded on mountain slopes facing tropical easterlies, and low rainfall totals were recorded in their lees. Different slopes received this topographic rainfall throughout the year depending on the orientation of the tropical easterlies. Particularly high precipitation totals were recorded on slopes that faced the tropical easterlies in April, which is the wettest month of the year across much of East Africa (Yang et al., 2015). These findings were expected, as the generation of topographic rainfall through orographic uplift on windward slopes, and the presence of rain shadows in their lees, is a well-known phenomenon (Houze, 2012). The impact of topographic rainfall is also well documented in East Africa: the rain shadow generated by the rift valley is a recognised regional feature (Camberlin, 2018); exposure is known to be one of the key determinants of annual rainfall totals over the rift highlands (Oettli and Camberlin, 2005); and Norton-Griffiths et al. (1975) highlighted the rain shadow caused by Ngorongoro Crater highlands as a key cause of low rainfall totals in the south-eastern Serengeti.

However, our results suggest that the topography has additional effects over the Serengeti that have not previously been discussed. At 15:00, a region of ascending motion was present over a topographic plateau in the south-eastern Serengeti National Park. This region of ascending air moved towards the north-west between 15:00 and 18:00, despite the surface winds from the lake penetrating further inland. The mid-afternoon ascent over, and evening movement away from, the highest topography suggests this pattern may be generated by diurnal heating. Anabatic-katabatic flows are well documented in the highlands of East Africa (Anyah et al., 2006; Nicholson, 2016), and the diurnal heating of mountain peaks is known to play an important role in triggering orographic precipitation (Houze, 2012). Therefore, the impact of topographic flows on precipitation generation in the southern and central Serengeti should be considered.

4.4 Lake breeze winds

Our results showed high rainfall totals over Lake Victoria and the surrounding areas. They also showed diverging daytime lake breeze winds associated with high moisture levels in the lower troposphere. In the eastern Lake Victoria basin, a wind convergence line was present between the westerly lake breeze winds and the tropical easterlies. This wind convergence line always passed through the Kenya highlands and the north of the Serengeti National Park; these regions received high-annual rainfall totals. The impact of the convergence of lake westerlies over the Kenya Highlands is well documented (Fraedrich, 1972; Okeyo, 1986; Anyah et al., 2006), with moisture-rich lake breeze winds known to combine with anabatic winds to generate thermally-triggered afternoon storms over the high ground. These storms drift westward over the afternoon and evening, and become reinvigorated once over Lake Victoria by the reversal of the lake breeze, and addition of katabatic winds overnight (Anyah et al., 2006; Finney et al., 2019). Over the Serengeti National Park, Norton-Griffiths et al. (1975) correctly suggested that storms generated by the lake front were the reason that the north-west received heavier and less variable rainfall than the rest of the park throughout the year.

However, less attention has been paid to the seasonal impact of lake westerlies to the south of the Kenya highlands. Our results show that a convection cell between the lake breeze winds and tropical easterlies was always present over part of the Serengeti National Park. During months when the tropical easterlies have a northerly or neutral meridional component, the lake breeze winds extended further south into the Lake Victoria basin and Serengeti National Park. Logically, the ascending branch of these convection cells would provide favourable conditions for triggering convective storms, similar to those seen over the Kenya highlands. Thus, these results suggest that the lake plays a role in determining precipitation over a larger proportion of the Serengeti than was suggested by Norton-Griffiths et al. (1975), who described its influence as being restricted to the north-west of the ecosystem.

4.5 Rainfall gradient over the Serengeti National Park

We propose that the rainfall gradient across the Serengeti may be generated by interactions between three features: tropical easterlies, lake breeze westerlies, and the rift topography. Our results support the findings of Norton-Griffiths et al. (1975), that the lake breeze is responsible for providing year-round rainfall in the north-west of the Serengeti National Park, and the topography generates a rain shadow in the east by blocking Indian Ocean moisture from reaching this area. Norton-Griffiths et al. (1975) also identified that the seasonally changing rainfall gradient across the Serengeti is related to the position of the tropical rainfall belt, but were not able to suggest a mechanism to explain how this occurred. Our results suggest it may be due to afternoon storms forming at the convergence line of lake breeze westerlies and tropical easterlies. These storms could then be blown in the direction of the tropical easterlies, the orientation of which change on a seasonal basis. This theory would explain both the enhanced precipitation totals to the west of the maximum longitudinal extent of the lake breeze westerlies, and the seasonally changing orientation of the rainfall gradient.

4.6 Afternoon storm development

Our results show that peak rainfall always occurred during the day over the Serengeti National Park, and in the dry season, it occurred in the late afternoon/early evening (18:00). However, during many of the wettest months of the year, the eastern Serengeti National Park had an early afternoon rainfall peak (15:00). This early afternoon peak typically occurred between the surface wind convergence line and the rift topography, suggesting the influence of both features.

Camberlin et al. (2018) found that early afternoon (15:00) rainfall maximums are a common feature around lakes in East Africa, caused by their diverging lake breeze winds. Therefore, the late afternoon (18:00) rainfall maximums found on the east coast of Lake Victoria are regionally anomalous. This late afternoon peak may be caused by the greater strength of the Lake Victoria westerlies, or positive feedbacks between anabatic winds and the lake breeze, resulting in an increased intensity and duration of the storms. However, this does not explain the early afternoon maximum we found to the east of the mean position of the lake breeze winds. A potential explanation may be variability in the position of the lake front. Our study has focused on the mean atmospheric conditions over the wider Serengeti domain, but the position of the lake front in fact varies from day to day (Woodhams et al., 2019). Rainfall in the south-eastern part of the Serengeti National Park may only occur on the rare occasions when the lake front penetrates this far south-east. Given the distance from the lake, the storm may then be blown west by mid-tropospheric (>750 hPa) easterly winds by late-afternoon, resulting in an early afternoon rainfall peak in the east and a later afternoon rainfall peak in the west. Variability in the position of the lake front may also explain why the 18:00 wind convergence line does not always match the boundary between the areas with peak rainfall at 15:00 and 18:00. Alternatively, this mismatch could be due to the datasets having different temporal resolutions, as the ERA5 reanalysis dataset and TRMM precipitation dataset represent one and 3-hour periods, respectively.

An alternative explanation for the different peak precipitation times observed over the Serengeti National Park could be cloud formation over the higher topography in the east resulting in light rainfall. Our results show a region of lower tropospheric ascending motion was present at 15:00 over the high plateau (34.5°–35°E) in the eastern Serengeti in both March and August. By 18:00, the magnitude of peak ascent had strengthened in August, but weakened in March. If the ascending motion over the south-eastern Serengeti high ground is controlled by land surface temperatures, the formation of rain clouds in March may lower the surface temperature, which would then reduce ascending motion in the lower troposphere and could prevent smaller storms from growing. It is known that over the East African mountains, maximum surface temperatures can occur early in the day due to the formation of afternoon clouds, which block further solar heating (Duane et al., 2008; Camberlin, 2018). In addition, over the African continent, early afternoon rainfall is often associated with weaker shallow storms, while late afternoon rainfall is associated with deeper storms and heavier rainfall (Geerts and Dejene, 2005). Therefore, it is possible that during wetter months, when higher levels of atmospheric moisture are present, rainfall in the two halves of the national park are generated in separate ways. Cloud formation over the higher topography in the east may generate light topographic storms with an early rainfall peak, while larger storms in the west caused by the converging lake westerlies and tropical easterlies result in a late afternoon peak. However, this scenario would not explain why the early afternoon rainfall peak is largely absent in April, as this is the month with the highest specific humidity values across the wider Serengeti domain, and would therefore provide favourable conditions for the formation of topographic storm clouds.

To fully understand the mechanisms generating rainfall in the south-eastern Serengeti National Park, further research would need to examine local wind and precipitation patterns on daily timescales, focusing on afternoons when rainfall occurred in the south-east. Ideally, this would use a combination of ground-based observations and remotely sensed datasets.

4.7 Key differences between annual cycles of precipitation

Despite limits to our understanding of rainfall-generating processes in the south-east, we propose that interactions between tropical easterlies, topography and lake breeze westerlies can explain key differences in the structure of annual cycles of precipitation. Cluster analysis identified these key differences as precipitation during January–February and June–September, the symmetry between rainfall peaks, and wet season peak months.

We suggest rainfall totals during January to February and June to October depend on the presence of westerlies from Lake Victoria. This results in relatively high totals in the northwest during June to October, and the south-west during January and February.

Similarly, the relative symmetry between rainfall peaks depends on the availability of moisture from Lake Victoria. Regions that do not receive westerlies from Lake Victoria have less symmetric annual cycles, similar to the rest of East Africa (Yang et al., 2015). This asymmetry is amplified over topographic features with slopes facing the prevailing tropical easterlies during April.

The position of the tropical rainfall belt is the first-order control on the distribution of peak rainfall months. However, our results suggest that the rift topography and lake breeze winds slightly augment the spatial distribution of November and December rainfall peaks. During November, monthly rainfall totals are likely higher to the west of the wind convergence line due to storms generated at the lake front. The cause of the November precipitation peak in the north-east is less clear, but its spatial distribution suggests it could be due to orographic rainfall generated on windward slopes by the tropical easterlies meeting the rift topography. Areas in the lee of this high topography (such as the northern Serengeti in the lee of the Loita hills) would not receive extra orographic precipitation, and would therefore have a December peak, as precipitation is more prevalent across the whole southern Lake Victoria basin in this month.

Similarly, the first-order control on the spatial distribution of the March and April rainfall peaks is the northward progression of the tropical rainfall belt. However, in the south, there is a clear difference in peak rainfall month between areas east and west of the rift topography. Norton-Griffiths et al. (1975) suggested that the lack of an April rainfall peak in the south-eastern Serengeti National Park was due to a rain shadow blocking the moisture from the Indian Ocean. However, our results show that high moisture levels and afternoon convergence are still present in the south-eastern Serengeti National Park during April, suggesting that lack of moisture is not the main reason for this decreased rainfall. We propose instead that the March rainfall peak may be due to the Lake Victoria westerlies penetrating further south and enhancing rainfall in the southern Serengeti National Park during March. In April, the increased southerly component of the tropical easterly winds prevents moisture from Lake Victoria reaching this area. Simultaneously, exceptionally high rainfall totals are produced by orographic uplift on the south-facing rift topography during April. The combination of these features, results in a March rainfall peak in the southern Lake Victoria basin, and an April rainfall peak in the northern basin, and along the rift topography.

5 CONCLUSIONS

Wildlife in the Serengeti National Park depends on rainfall. To understand past rainfall conditions, and predict future changes, it is essential to quantify the region's mean precipitation patterns, and identify the causal atmospheric processes. We found that the local rainfall is determined by interactions between Indian Ocean tropical easterlies, Lake Victoria daytime westerlies and the complex topography of the East African rift valley. The relative importance of the three key features depended on location, time of year, and time of day. The passage of the tropical rainfall belt is the first-order determinant of seasonality, generating a dry season and double-peaked wet season across the whole region. The topography augments lower tropospheric vertical motion through orographic channelling and solar heating of high ground, while orographic uplift generates topographic precipitation and rain shadows. Lake Victoria affects local rainfall through the seasonal provision of moist westerlies. Storms are likely generated at the lake front where these daytime westerlies meet the tropical easterlies, then are blown in the direction of the tropical easterlies, generating a seasonally changing rainfall gradient across the Serengeti National Park. Further research focusing on daily changes to the position of this wind convergence zone would likely improve our understanding of rainfall-generation over the south-eastern plains of the Serengeti. Overall, this study provides an updated understanding of the Serengeti's climate. In addition, by highlighting the importance of the topography and lake breeze winds, we show that research aiming to understand past or future changes to rainfall in this ecologically significant area, must consider the dominant roles of these key mesoscale features.

ACKNOWLEDGEMENTS

Josephine Mahony was supported by the U.K. Natural Environment Research Council (NERC) through the DTP in Environmental Research (Grant NE/L002612/1). Richard Washington acknowledges the Future Climate for Africa UMFULA project, supported by the U.K. Natural Environment Research Council (NERC), NE/M020207/1, and the U.K. Government's Department for International Development (DFID).