atmosphere

Article
Did ERA5 Improve Temperature and Precipitation
Reanalysis over East Africa?
Stephanie Gleixner 1, * , Teferi Demissie 2,3 and Gulilat Tefera Diro 4
1 Potsdam Institute for Climate Impact Research (PIK), 14473 Potsdam, Germany
2 CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS), East Africa,
P.O. Box 5689, Addis Ababa 1000, Ethiopia; t.demissie@cgiar.org
3 NORCE Norwegian Research Centre, Bjerknes Centre for Climate Research, P.O. Box 22,
5838 Bergen, Norway
4 Canadian Network for Regional Climate and Weather Processes, ESCER Center, University of Quebec
at Montreal, 201 President-Kennedy Avenue, Montreal, QC H2X 3Y7, Canada; gulilattef@gmail.com
* Correspondence: gleixner@pik-potsdam.de




Received: 5 August 2020; Accepted: 14 September 2020; Published: 17 September 2020


Abstract: Reanalysis products are often taken as an alternative solution to observational weather and
climate data due to availability and accessibility problems, particularly in data-sparse regions such
as Africa. Proper evaluation of their strengths and weaknesses, however, should not be overlooked.
The aim of this study was to evaluate the performance of ERA5 reanalysis and to document the
progress made compared to ERA-interim for the fields of near-surface temperature and precipitation
over Africa. Results show that in ERA5 the climatological biases in temperature and precipitation are
clearly reduced and the representation of inter-annual variability is improved over most of Africa.
However, both reanalysis products performed less well in terms of capturing the observed long-term
trends, despite a slightly better performance of ERA5 over ERA-interim. Further regional analysis
over East Africa shows that the representation of the annual cycle of precipitation is substantially
improved in ERA5 by reducing the wet bias during the rainy season. The spatial distribution of
precipitation during extreme years is also better represented in ERA5. While ERA5 has improved
much in comparison to its predecessor, there is still demand for improved products with even higher
resolution and accuracy to satisfy impact-based studies, such as in agriculture and water resources.

Keywords: ERA5; ERA-interim; reanalysis; Africa; precipitation; near-surface temperature

1. Introduction
According to the fifth Intergovernmental Panel on Climate Change (IPCC) report, Africa is one of
the most vulnerable continents due to its high exposure to climate risks as well at its low adaptive
capacity [1]. In particular East Africa, where the majority of the countries are highly dependent on
rain-fed agriculture and have a high level of poverty and a low level of education, is vulnerable to
climate change and climate extremes [2–4]. The region is already experiencing droughts and floods
that have substantial socio-economic impacts [5] leading to serious food insecurity and resource-based
conflicts. The IPCC’s fifth report indicated that future climate change will lead to an increase in climate
variability and in the frequency and intensity of extreme events in the region [1]. Decisions about
climate and climate change are complex, costly, and have long-term implications, and it is essential that
such decisions are based on the best available evidence [5]. In this regard, reliable climate information
is the basis for developing a climate-resilient system and intervention mechanisms to minimize the
vulnerability of the region to various climatic risks.
Reliable and decision-relevant climate data at a national and local level are needed on all weather
and climate time scales. For instance, the assessment of climate-related baseline risks, early warning

Atmosphere 2020, 11, 996; doi:10.3390/atmos11090996 www.mdpi.com/journal/atmosphere

Atmosphere 2020, 11, 996 2 of 19

systems, and science-informed adaptation and mitigation planning to climate change in Africa all
depend on available data [6]. Such climate data have to fulfill various requirements. Statistical
forecast models and early warning systems need near-real time data as model input. For assessing
long-term climatic changes and trends, identifying the seasonality of climate or crop cycles, assessing
historical impacts, and providing a reference against which to compare current and anticipated climate
conditions, data with a long record are necessary. In order to use climate models for future projections,
the model simulations of the historical climate have to be assessed and validated against reliable
observational data with a good spatial coverage to understand the skill and uncertainty of the models.
For any investigation of extreme events, climate data with a high temporal resolution are needed.
The overall demand for any application of climate data is that the data have high-spatial resolution
and accuracy. In summary, there is a demand for very accurate climate data with high spatial and
temporal resolution published in near-real time and covering several decades. These requirements are
particularly difficult to meet in a region like the African continent, where the network of meteorological
stations is sparse due to the number of weather stations as well as the uneven distribution of weather
stations (not many stations in rural areas), which leads to serious gaps in observations [1,7].
In the last decade a variety of high-resolution precipitation products, which combine satellite
data and in situ measurements, has become available. Data products like the Tropical Applications
of Meteorology using SATellite and ground-based observations (TAMSAT) rainfall estimate [8] and
the African Rainfall Climatology version 2 (ARC2) [9] have been developed specially for the African
continent. However, a recent study by Dinku et al. [10] has shown that the semi-global Climate
Hazards Group (CHG) InfraRed Precipitation with Stations data (CHIRPS) [11] captured the East
African precipitation better.
By contrast, temperature datasets have been less in the spotlight, despite the fact that temperature
plays a key role in modulating surface hydrology and the severity of droughts (see, e.g., SPEI—the
Standardized Precipitation-Evapotranspiration Index). Furthermore, even recent temperature products
like the Climatic Research Unit (CRU) Time-Series (TS) Version 4 (CRU TS4.02)) [12], Climatic
Research Unit TEMperature, version 4 (CRUTEM4) [13], the NASA GISS Surface Temperature Analysis
(GISTEMP) [14], NOAA MLOST [15] or UDEL [16] are spatially and temporally coarser than the
precipitation counter parts.
A convenient solution for these limitations is reanalysis data. Reanalysis products provide
comprehensive and coherent climate datasets to overcome these data limitations. Reanalysis data is
produced by combining forecast model estimates with observations via data assimilation, therefore
providing optimized global estimates of climate data without spatial or temporal gaps. However,
the accuracy of reanalysis data varies strongly between regions and variables [17,18]. In regions with
few observations or complex terrain, reanalysis products may suffer from large biasses [19–22] and
some variables like precipitation are purely forecasted with no observational input [23,24]. Therefore,
temperature data are generally more reliable than precipitation data [17,19]. While precipitation is the
most important climate variable to directly affect humans according to the WMO [25], the sub-grid
processes determining precipitation have to be parametrized and make it a difficult variable to
model [26,27].
Several studies have been conducted to assess the performance of reanalysis datasets over Africa,
in particular for precipitation. Generally it was found that the seasonality of precipitation is captured
well in the reanalysis products [28–30], but temporal and spatial correlation with observations is low
in many regions of East Africa [28,29,31,32]. Additionally, local biasses are present in all reanalysis
datasets. For instance, NCEP/NCAR, ERA-40 and ERA-interim reanalyses all tend to overestimate
precipitation in the Ethiopian highlands [21,28–30].
These studies evaluated the previous generation of reanalyses and payed less attention to the
temperature reanalysis product. In 2019, ECMWF published the first completed version ERA5 [33],
the successor of ERA-interim. It provides hourly data with a 0.25◦ spatial resolution from 1979
to near-real time and is easy to access via Copernicus’ Climate Data Store. Therefore it offers the
Atmosphere 2020, 11, 996 3 of 19

opportunity for a wide range of applications, not only for research but also for operational use.
In particular in lesser developed countries, the ease of access plays an important role when choosing
suitable climate data. ERA5 underwent some substantial changes in the assimilation system in terms
of the model as well as the included observation data from its predecessor ERA-interim, and first
evaluation studies have shown its improvements in terms of surface energy fluxes [34], surface
irradiance [35], and surface climate in North America [36,37]. These results are encouraging, suggesting
that also the African climate representation may have improved in ERA5 reanalysis. A comprehensive
evaluation of temperature and precipitation, which are the most widely used climate variables, in terms
of reproducing the temporal and spatial variability as well as the observed long-term trend is necessary
to judge the reliability of the data for African climate research and operational use.
The goals of this study were to assess the climatology, the long-term trend, and the inter-annual
variability across the different generations of ECMWF reanalysis datasets over Africa. Further
investigation on the performance of the reanalysis during extreme wet and dry years will also be
carried out on a regional scale.
The rest of the paper is arranged as follows: Section 2 describes the datasets and methods used in
this study. The performance of reanalysis at the continental, regional, and country scales is discussed
in Section 3. Finally, the summary and conclusions are provided in Section 4.

2. Data and Study Approach

2.1. Study Region

Although the focus of our study is on East Africa, validation of the reanalysis datasets in
representing the climatology, the long term trend, and their inter-annual variability was carried
out for the entire African continent. This allowed us to assess the performance of the reanalysis
products over East Africa in comparison to the rest of the continent. The detailed regional and local
analysis was limited to the East Africa region from 25◦ E to 50◦ E and 13◦ S to 17◦ N (Figure 1).
East Africa has a complex topography from low coastal areas up to peaks of 4500 m above sea
level (Figure 1). It includes a multitude of climatic regions of tropical, arid, and moderate characters,
with very different temperature and precipitation conditions [38]. The northern parts of east Africa
experience the “long rains” during the northern hemisphere summer season (June to September) and
the “short rains” during spring (February–May). In most of Equatorial East Africa, the prevailing
two rainy seasons are the “long rains” in March–May and the “short rains” in October to December.
While this bimodal precipitation pattern was classically explained by the north–south migration of the
Inter Tropical Convergence Zone (ITCZ), more recent studies attributed it to a complex interaction of
sea surface temperature pattern and low-level winds [39,40].
The complexity of the terrain, the frequent presence of heavy cloud cover, which complicates
satellite measurements of surface conditions [41], and the sparsity of in situ observations make East
Africa a particularly challenging region for climate data. Observational precipitation datasets show
large discrepancies [8,42–44] and previous reanalysis products have been found to perform poorly
in East Africa with regard to rainfall [21]. As the climate of East Africa is particularly challenging
to capture with reanalysis data, the region provides a great opportunity to evaluate the quality
improvements from ERA-interim to ERA5.
Our particular focus in this paper is the evaluation of ERA5 in the four focus countries of
the CGIAR (Consultative Group on International Agricultural Research) research program CCAFS
(Climate Change, Agriculture and Food Security), namely Ethiopia, Tanzania, Uganda, and Kenya.
These countries are also included in the East African study region.
Atmosphere 2020, 11, 996 4 of 19

Figure 1. Topography in the study region of East Africa (25◦ E to 50◦ E and 13◦ S to 17◦ N).

2.2. ERA5 and ERA-interim Reanalysis Data

The two reanalysis datasets we compare in this paper, ERA-interim and ERA5, are the two most
recent reanalysis products of the European Center for Medium-Range Weather Forecasts (ECMWF),
which are produced by combining a numerical weather prediction model with observational data from
satellites and ground observations. ERA-interim was introduced in 2007 [45,46] and provided daily
climate information until August 2019. Since 2019 it was replaced by ERA5 [33], which provides hourly
meteorological conditions back to 1979 and is expected to be extended back to 1950. Both versions
of the ECMWF reanalysis are based on the Integrated Forecasting System (IFS) and include a
four-dimensional variational analysis (4D-Var). There are several substantial differences between the
two datasets, concerning the forecast model, the observational input, and the estimation of uncertainty.
ERA5 is run with the version Cy41r2 of IFS, in comparison to the Cy31r2 version of ERA-interim.
Therefore ERA5 data is available in higher spatial as well as temporal resolution. ERA5 data is available
on a 0.25◦ grid with hourly intervals, while ERA-interim data is available on a 0.75◦ grid with 6 h
intervals. Additionally, the vertical resolution increased from 60 levels in ERA-interim to 137 levels in
ERA5 [33,47]. The number of observational datasets that serve as input for the assimilation system
was increased and a major difference is the consideration of satellite estimates of precipitation in ERA5.
ERA-interim did not include an estimation of the uncertainty of the data, while ERA5 includes such an
estimate based on a 10-member ensemble run with 63 km resolution. While the ERA-interim product
started in 1979, ERA5 is planned to cover 1950 to near-real time.

2.3. Observational Data for Evaluation

Ideally, an evaluation of a reanalysis product would be based on a comparison to in situ
observations. As station data are scarce and often unevenly distributed in Africa, we relyed on
gridded observational datasets as a best estimate of the truth.
For near-surface temperature observations we used the high-resolution data Climatic Research
Unit (CRU) Time-Series (TS) version 4.02 (CRU TS4.02) [12]. This station-based gridded dataset is
produced by angular-distance weighting interpolation of station observations onto a 0.5◦ grid [7].
The dataset is available from 1901 to 2017 on a monthly time scale. CRU temperature is widely used and
have also been applied for correcting reanalysis data in order to apply them to impact modeling [48].
We regridded the data onto the 0.75◦ ERA-interim grid. As CRU is a purely station-based dataset,
the reliability of CRU temperature varies spatially, due to station sparsity. In Africa, the station
Atmosphere 2020, 11, 996 5 of 19

coverage is lowest in the region of Northern Namibia/Southern Angola, but also in East Africa some
regions have low station density (Figure 2).
For precipitation observations we used the Climate Hazards Group InfraRed Precipitation with
Stations version 2 (CHIRPS V2.0) [11]. CHIRPS combines a variety of satellite products and in situ
observations. The dataset is available in a 0.05◦ and a 0.25◦ spatial and a daily temporal resolution
from 1981 until today. CHIRPS has been shown to perform well over Eastern Africa [10,31,49].

Figure 2. Average number of stations used per timestep in the interpolation of CRU (Climatic
Research Unit) TS4.02 near-surface temperature from 1981 to 2017. Eight is the maximum number of
stations used.

2.4. Methodology
For this study, we used average near-surface (2 meter) temperature data and total precipitation
from 1981 to 2017. In order to compare the datasets of different spatial resolutions, we conducted this
study on the coarsest spatial resolution, which is ERA-interim with 0.75◦ , by regridding all data onto
the ERA-interim grid using first-order conservative remapping.
Similarly, for comparison purposes, both reanalysis products and validation datasets were
aggregated to the monthly time scale, as CRU temperature is only available as monthly means.
Correlations were calculated based on monthly means for both temperature and precipitation.
The trends were calculated based on annual mean (temperature) and annual sums (precipitation) and
the East African precipitation cycle was also based on monthly means. The country level analysis was
also based on the absolute monthly means, not removing any variability modes such as trends or the
annual cycle, as we are interested in the performance of the data including all variability. This option
was chosen with regard to the application of reanalysis data for impact studies, where absolute values
are needed.
Significance was tested using the scipy python suite. In particular, we applied a t-test for the means
of two independent samples to the annual means/sums to test the significance of the bias, while the
significance of the correlations as well as the trends was based on a Wald test with t-distribution.

3. Results and Discussion

3.1. All-Africa Evaluation

3.1.1. Spatial and Temporal Variability of Temperature and Precipitation

While the focus of our study is on East Africa, we assessed the spatial and temporal variability
of temperature and precipitation in the two reanalysis datasets as well as the trends over all of
Atmosphere 2020, 11, 996 6 of 19

Africa to allow for comparison with the rest of the continent. Figure 3 shows the spatial distribution
of the near-surface temperature of CRU and the difference between the reanalyses (Era-interim
and ERA5) and observations (CRU). Observations show the highest mean temperatures over the
Sahara reaching up to 32 ◦ C. The lowest mean temperatures occur over Southern Africa and over
the Ethiopian highlands. Both reanalysis datasets capture the overall temperature distribution well
(not shown), which is reflected in the strong pattern correlation between the reanalysis and CRU.
The pattern correlation improved from 0.91 for ERA-interim to 0.97 for ERA5 (significant at the 99%
level). ERA-interim exhibits a substantial bias in near-surface temperature over large parts of the
continent. The strongest bias is the distinct warm bias in Northwest Africa and the Sahara of up to
5 ◦ C. In addition, around the western border of Namibia and Angola there is a strong warm bias,
but as this region has the minimum of CRU stations, there is some uncertainty in the observations.
By contrast, there is a pronounced cold bias of up to 2 ◦ C in Western Equatorial Africa and along most
of the East African coast. The reduction of the bias from ERA-interim to ERA5 is striking, as in most of
the continent the near-surface temperature bias is less than 1.5 ◦ C. In particular, the strong warm bias
in the Northwest is reduced to less than 1 ◦ C in most of the region. The cold bias in Western Equatorial
Africa is even reduced to less than half a degree. Our focus region of East Africa show only weak and
patchy differences between ERA5 and observations. However, some small regions still display biases
of more than 1.5 ◦ C, such as the cold bias in the borderlands of Egypt, Sudan, Libya, and Chad and the
warm bias at the coastal borderlands of Namibia and Angola.
Temporal correlation between monthly mean temperature of reanalysis with observation show a
latitudinal dependence (Figure 4). While the correlation over Northern and Southern Africa is higher
than 0.95 for both reanalyses, it is lower towards central Africa from both hemispheres. In the region of
northwest Angola and the west of the Democratic Republic of Congo the correlation is even negative.
It must be noted that this region also overlaps with the most data-sparse region over Africa [50],
in which there exists a large uncertainty among observational datasets [51]. With the exception of
a small region in Ethiopia, the correlation of ERA5 with observations increased from ERA-interim.
In particular at the northern coast of Angola the correlation improved from negative to more than 0.8.
The mean rainfall distribution on the African continent show a very dry Sahara and Northern
Africa and maximum rainfall rates around Equatorial Central Africa and over the regions of the West
African monsoon (Figure 5a). We masked deserts (<250 mm of annual rainfall) in these figures, as the
relative bias in these regions is disproportionally large. Both reanalysis datasets reproduce the spatial
distribution of observed rainfall well (not shown) and the pattern correlation improved from 0.94
for ERA-interim to 0.96 for ERA5 (significant at the 99% level). In ERA-interim, there is a strong wet
bias over central Africa and the coastal areas of East Africa of up to 60–80% (Figure 5b). This wet
bias over Africa in ERA-interim was also noted in the study by Dee et al. [45], though they stated
that the bias was already better than in ERA-40 due to an improved moist boundary layer scheme
and better assimilation of humidity. There is a clear reduction of the wet bias in ERA5 over much of
Africa (Figure 5c). For instance, the bias over central Western Africa is reduced from up to 80% in
ERA-interim to less than 20% in ERA5. Similarly, the wet bias over the Somali coast is substantially
reduced. These changes of biases over Africa across the three generations of ECMWF reanalyses
suggest that improvements in model physics and the data assimilation procedure keep reducing the
precipitation bias. Despite these reductions in the precipitation bias, ERA5 still shows substantial
disagreements with observations in complex terrain, such as the Ethiopian highlands or above Lake
Victoria. However, it should be noted that there is general disagreement between observational
datasets in East Africa [21,52].
The temporal correlation between monthly precipitation in reanalysis and observations is
generally considerably lower than that of temperature (Figure 6). However, in East Africa precipitation
show higher values than temperature. ERA-interim shows the highest correlation in the southern
hemisphere with values beyond 0.95. Also in sub-Saharan Africa, correlation reaches values beyond
Atmosphere 2020, 11, 996 7 of 19

0.85. A local minimum of correlation is found directly at the equator. In most regions of Africa
correlation did increase from ERA-interim to ERA5.

Figure 3. Observed near-surface temperature (CRU) averaged over 1981–2017 (a) and difference
between ERA-interim (b)/ERA5 (c) near-surface temperatures averaged over 1981–2017 and observed
near-surface temperature averaged over 1981–2017. Hatching represents regions where differences are
statistically significant at the 1% level.

Figure 4. Correlation between monthly observed (CRU) and monthly ERA-interim (a)/ERA5
(b) near-surface temperatures from 1981–2017. Hatching represents regions where correlation values
are statistically significant at the 1% level.

Figure 5. Observed precipitation (CHIRPS) averaged over 1981–2017 (a) and difference between
ERA-interim (b)/ERA5 (c) precipitation averaged over 1981–2017 and CHIRPS precipitation averaged
over 1981–2017. Hatching represents regions where differences are statistically significant at the
1% level.

Figure 6. Correlation between monthly observed (CHIRPS) and monthly ERA-interim (a)/ERA5
(b) precipitation from 1981–2017. Hatching represents regions where correlation values are statistically
significant at the 1% level.
Atmosphere 2020, 11, 996 8 of 19

3.1.2. Linear Trends

Figure 7 shows the spatial pattern of the annual mean near surface temperature trend according
to the CRU, ERA-interim, and ERA-5 reanalyses datasets. While all three datasets agree on the positive
temperature trend over Northern Africa, ERA-interim displays a negative trend over Equatorial East
Africa, along the coast from Sierra Leone to Senegal and in much of Botswana, Zambia, and Zimbabwe.
This negative trend of ERA-interim is consistent with the result of Simmons et al. [53], who found a
decrease in temperature when comparing the 1998–2008 period to the 1989–1998 period in ERA-interim.
As will be seen later in this section, these regions of negative temperature trends are closely linked to
positive precipitation trends, suggesting that an increase in cloud cover in ERA-interim results in the
cooling of the surface. Figure 7 also indicates that in most of Northern Africa the positive temperature
trend of ERA-interim is stronger than in CRU, whereas the temperature trend of ERA5 is much closer
to CRU. In ERA5, the positive temperature trend in Northern Africa is reduced from ERA-interim,
and there are no regions with a negative temperature trend. In particular, over East Africa, the slope of
the trend is much closer to observations in ERA5 compared to ERA-interim. Nevertheless, overall the
warming trend is stronger in ERA5 than in CRU.
The trends of annual precipitation show even larger differences (Figure 8). While CHIRPS shows
an increase in rainfall in much of sub-Saharan Africa, while both reanalysis products display a very
strong drying trend in that region. This negative precipitation trend in reanalysis datasets is consistent
with the study of Lin et al. [54], who highlighted that the negative trend is a common feature in other
reanalysis datasets as well. In Equatorial Africa, observations tend to a drying trend, which is captured
but strongly overestimated by both reanalysis products, though less so in ERA5. The observed trend
pattern of the southern hemisphere precipitation is captured quite well in reanalysis. In East Africa,
particularly over Southastern Ethiopia, Somalia, Kenya, and Tanzania, CHIRPS shows a weak and
mostly drying trend in annual sum precipitation. By contrast, ERA-interim show a strong wetting
trend in this region. ERA5 captures the negative sign of the trend, even though it overestimates the
magnitude of the drying trend in the region. Overall, the reduction of the differences to observations
from ERA-interim to ERA5 is striking, even though the differences between CHIRPS and ERA5 are
still substantial.
These results show that using reanalysis data for trend analysis is problematic, in particular for
ERA-interim. While this is frequently done in research [55,56], reanalysis is generally considered
unsuitable for identifying long-term trends [57–59]. This is because observational datasets are included
in the assimilation system at various points in time [28,33,45]. This can lead to jumps in the data and
explain the differences between observations and reanalysis with regards to long-term trends [57].

Figure 7. Trend of annual mean near-surface temperature in observations (CRU, (a)), ERA-interim (b)
and ERA5 (c) from 1981 to 2017. Hatching represents regions where trends are significant at the
1% level.
Atmosphere 2020, 11, 996 9 of 19

Figure 8. Trend of annually aggregated precipitation in observations (CHIRPS, (a)), ERA-interim (b)
and ERA5 (c) from 1981 to 2017. Hatching represents regions where trends are significant at the
1% level.

3.2. Regional Analysis

3.2.1. East African Precipitation Cycle

As the tropical regions exhibit a strong seasonality in precipitation, we investigated how well
the reanalysis products captured the seasonal cycle of precipitation in our East African focus region.
Figure 9 shows the time–latitude diagram of precipitation averaged over the longitude band between
25◦ E and 50◦ E for CHIRPS and the bias in the two reanalysis datasets. CHIRPS shows the north–south
migration of the East African rainfall following the movement of the inter-tropical convergence zone
(ITCZ). Maximum precipitation is located in the Southern hemisphere around 12◦ S with an average
value of more than 9 mm/day during northern hemisphere winter, which is the main rainy season
of Southern Tanzania. During northern hemisphere spring the precipitation maximum migrates
northward. During northern hemisphere summer the maximum precipitation rate remains in the
north of East Africa around 10◦ N with rates just below 5 mm/day representing the main rainy
season of Ethiopia (June to September). During northern hemisphere fall the rainbelt wanderes
south again resulting in a bimodal precipitation pattern of Uganda, Kenya, Southern Ethiopia,
and Northern Tanzania.
ERA-interim and ERA5 both capture the seasonality and position of the rain belt very well
(not shown), but total rainfall amounts do not agree everywhere with observations. ERA-interim
underestimates the precipitation maximum in the very south of the East Africa box by up to
2.25 mm/day. On the other hand, the rainy seasons of Uganda and Kenya in northern hemisphere
spring and fall (precipitation maximum around the equator) are overestimated by more than
2.25 mm/day and in particular the spring rains begin too early in the year. The timing of the main
rainy season of Ethiopia in northern hemisphere summer is captured well but also overestimated by
up to 1.75 mm/day.
The changes from ERA-interim to ERA5 are striking, in particular over the most northern and
southern positions of the precipitation belt: the dry bias in the northern hemisphere winter as well
as the wet bias during the Ethiopian rainy season are reduced to less than 0.75 mm/day. In addition,
the wet bias in the locations of bimodal rainfall regimes is reduced to below 1.25 mm/day. Consistently
with ERA-interim, ERA5 also shows a slightly smaller northward extent of the rainbelt compared to
the CHIRPS data.
Atmosphere 2020, 11, 996 10 of 19

Figure 9. Monthly mean precipitation averaged over longitude band 25◦ E–50◦ E (East Africa box) and
time period 1981–2017 from 13◦ S to 17◦ N in observations (CHIRPS, (a)), and the difference between
ERA-interim (b)/ERA5 (c) and observations.

3.2.2. Extreme Precipitation Years

The time series of annual precipitation aggregated within the East Africa box (13◦ S–17◦ N,
25◦ E–50◦ E) is shown in Figure 10a for observations and reanalysis products. Observations do
not show a trend in annual mean precipitation. The driest year of the time period is 2005, which
corresponds to a strong negative Indian Ocean Dipole (IOD) event during the June to September
period [60]. However, also the devastating drought of 1983/1984 in Ethiopia [61] is captured.
The wettest years of the time series are 1997 and 2006, which are associated with combined positive IOD
and ENSO (El Niño–Southern Oscillation) years [62]. According to Bahaga et al. [63], strong positive
IOD events contribute more to above-normal rainfall over Eastern Africa compared to positive ENSO
years. This is because a warm SST anomaly over the western Indian ocean favors strong moisture flux
over Eastern Africa. Although ERA-interim has a consistent strong overall wet bias and therefore fails
to capture the dry years, it captures the wet years as well as the lack of a trend. By contrast, ERA5 has
an overestimated drying trend. The wet bias is much reduced compared to ERA-interim, particularly
starting from the late 1990s, likely due in part to the increase in observational data flowing into the data
assimilation procedure. Due to the drying trend, the great drought of 1983/1984 is not particularly dry
in ERA5, but the mentioned wet years are well-captured and also the dry year 2005 is very dry in ERA5.
The time series shows that the similarity between ERA5 and observations increase over time, with the
drying trend in ERA5 reducing the differences from observations. As there is large uncertainty in the
trend of precipitation products in East Africa [64], we removed the time series’ trends in Figure 10b.
The detrended time series of ERA5 precipitation agree very well with observations, with a correlation
of 0.84. This highlights that the year-to-year variability is captured well. By contrast, the detrended
time series of ERA-interim precipitation only has a correlation of 0.48 with observations.
We selected the year 2005 as an example of an extremely dry year and 1997 as an example
of an extremely wet year for a more detailed look at the rainfall patterns during extreme years.
Hastenrath et al. [65] noted that the deficit rainfall in 2005 led to drought conditions over Equatorial
Africa and associated this extreme dry event with the development of a positive pressure anomaly over
the western Indian Ocean during the October and November, which resulted in anomalous subsidence
over the region. In 1997, on the other hand, the East African October–December rains were in many
areas more than 5–10 times the normal [66].
The observed precipitation anomalies in 2005 show that East Africa was affected by dry conditions
mostly in the Southern hemisphere (Figure 11a). All of Tanzania and most of Uganda and Kenya were
Atmosphere 2020, 11, 996 11 of 19

drier than usual. This is consistent with the anomalous precipitation pattern associated with negative
IOD years. Both ERA-interim and ERA5 capture a type of bimodal pattern with a normal to wet
north and a dry south (Figure 11b,c). This suggests that both reanalysis datasets capture the apparent
relationship between the negative rainfall anomalies over Equatorial and Southern East Africa and a
negative IOD correctly. However, ERA-interim strongly overestimates the wet conditions in large parts
of Ethiopia and most of South Sudan. While the dry conditions in South Somalia and Eastern Kenya
are captured, conditions in Tanzania are also too wet compared to observations. This explains why the
average precipitation of East Africa is not particularly dry in 2005 in ERA-interim. ERA5 shows an
intensified picture of the observed conditions with slightly too wet conditions in Ethiopia and slightly
too dry conditions in the southern hemisphere. For the year 1997, observations shows wet conditions
in most of East Africa, with the strongest anomalies in East Kenya and South Somalia, with anomalies
between 700 and 900 mm for the year (Figure 12a). This year corresponds to the strongest El-Niño
of the century and a positive IOD event. Both of these events led to above-normal rainfall over
Southern Ethiopia and the Equatorial and Southern East Africa regions in the winter and spring
seasons. ERA-interim and ERA5 reproduce these anomalously wet conditions over the coastal areas of
East Africa (Figure 12b,c). However, ERA-interim shows dry anomalies in West Ethiopia and Uganda,
and the strongest wet anomalies are located too far East and limited to the coastal areas of Somalia.
By contrast, ERA5 captures the observed wet pattern quite well, while slightly underestimating the
maximum precipitation rates.

Figure 10. Annual precipitation averaged over longitude band 25◦ E–50◦ E and 13 ◦ S–17 ◦ N (East
Africa box) in observations (CHIRPS, black), ERA-interim (red), and ERA5 (blue) from 1981 to 2017 as
annual sums (a) and detrended annual anomalies (b).
Atmosphere 2020, 11, 996 12 of 19

Figure 11. 2005 annual precipitation anomalies in East Africa in CHIRPS (a), ERA-interim (b) and
ERA5 (c).

Figure 12. 1997 annual precipitation anomalies in East Africa in CHIRPS (a), ERA-interim (b) and
ERA5 (c).

3.3. Country-Level Evaluation

We further narrowed down the analysis to the four focus countries Ethiopia, Kenya, Uganda,
and Tanzania. Figures 13 and 14 show scatter plots of reanalysis data versus observations to allow for
a more detailed investigation of the temporal temperature and precipitation distribution within the
countries and also present the seasonal cycle of temperature and precipitation at the country level.
The scatter plots of near-surface temperature in Figure 13 show that ERA5 is closer to observations
than ERA-interim in all four countries, as the values are located much closer to the lines of best
fit. While correlation ranges from 0.54 to 0.82 for ERA-interim, the range for ERA5 is from 0.78 to
0.93. The strongest change and best fit for near-surface temperature is found for Tanzania, with a
correlation coefficient of 0.93 (from 0.61 in ERA-interim). The least good fit for temperature of both
reanalysis products is found in Uganda, and even here ERA5 (0.78) is closer to observations than
ERA-interim (0.54).
Figure 13 also displays the annual cycle of near-surface temperature from CRU, ERA-interim
and ERA5. Observations show the highest temperatures during late winter/early spring for Ethiopia,
Kenya, and Uganda and during November for Tanzania. Conversely, the lowest temperature for
Kenya, Uganda, and Tanzania is seen during July, whereas for Ethiopia the coldest month is December.
Both reanalysis products reproduce the observed annual cycle well, even if the peak is shifted for
Ethiopia and Tanzania by a month. ERA-interim shows an overall cold bias in all four countries.
This bias is of a magnitude around 1–2 ◦ C, while the overall temperature range throughout the year is
2–3 ◦ C . While this bias is present all year round in Kenya and Uganda, it is only present in Ethiopia
after March and from October to June in Tanzania. In all four countries this bias is much reduced in
Atmosphere 2020, 11, 996 13 of 19

ERA5, which displays a bias around 0.5 ◦ C at the most. In particular in Tanzania the bias is so unevenly
distributed over the year that its reduction brings the ERA5 curve much closer to the observed curve.
The scatter plots for monthly mean precipitation rates in Figure 14 show a similar change for
temperature, which is closer to observations for ERA5 than for ERA-interim. As already seen in the
African correlation maps, in East Africa the correlation between observed monthly mean precipitation
and reanalysis is generally higher than the correlation between observed temperature and reanalysis.
For precipitation, the correlation changed from 0.68 to 0.93 for ERA-interim to a range of 0.9 to 0.98 for
ERA5. As for temperature, the best fit for monthly mean precipitation is found in Tanzania and the
worst fit is found in Uganda.
The annual cycles of precipitation from observations and the reanalysis datasets are displayed
in Figure 14 (lower panels) and show that Ethiopia and Tanzania get their maximum rains during
their respective summer seasons, whereas the countries around the equator (Kenya and Uganda)
receive the peak rains during spring and autumn following the movement of the ITCZ. Both reanalysis
products reproduce the observed seasonality of precipitation well despite the difference in amounts.
In Ethiopia, both reanalysis products have a wet bias, which is most pronounced in the rainy summer
season. This bias is not substantially reduced from ERA-interim to ERA5. In Kenya, both reanalysis
products have a wet bias in the dry seasons. However, there is a clear bias reduction in the short
rains (northern hemisphere winter), where ERA5 is much closer to observations than the too wet
ERA-interim data. In Uganda, despite the increased correlation with observations, the wet bias is more
pronounced in ERA5 than in ERA-interim throughout most of the year and in particular in the rainy
seasons. In Tanzania, there is little change from ERA-interim to ERA5. Both products overestimate
precipitation in the beginning of the year and are close to observations during the rest of the year.

Figure 13. Monthly mean near-surface temperature from 1981 to 2017 in observations (CRU) versus
ERA-interim (a–d) and ERA5 (e–h) in the four focus countries (columns). Annual cycle of monthly
mean near-surface temperature (i–l) averaged over the corresponding country as observed (black line),
in ERA-interim (blue line) and ERA5 (red line).
Atmosphere 2020, 11, 996 14 of 19

Figure 14. Monthly mean precipitation from 1981 to 2017 in observations (CHIRPS) versus ERA-interim
(a–d) and ERA5 (e–h) in the four focus countries (columns). Annual cycle of monthly mean precipitation
(i–l) averaged over the corresponding country as observed (black line), in ERA- interim (blue line) and
ERA5 (red line).

4. Summary and Conclusions

This study assessed the performance improvements of ERA5 over ERA-interim reanalysis over
Africa for the 1981–2017 period by comparing with gridded observations. Our analysis for near-surface
temperature and precipitation data in Africa showed substantial changes with the introduction of
the new reanalysis product. ERA5 data were generally closer to observations than ERA-interim,
which agrees with findings of studies on other aspects of the datasets [34–37].
Over most of Africa the bias in temperature as well as precipitation was clearly reduced and
correlation with observations increased in ERA5 reanalysis. It is interesting to note, that in ERA5
the temporal variability of observed precipitation was better captured than that of temperature in
large parts of Africa. This was reflected in higher temporal correlation values for the precipitation
than for the temperature field despite more complex processes involved in the generation of
precipitation. Further diagnostic study of this seemingly weaker performance of temperature compared
to precipitation would be an interesting topic for future study.
Despite a slightly better performance of ERA5 over ERA-interim in representing the observed
trend in precipitation, and to some extent that of temperature, this is an aspect where both reanalysis
fields performed less well. Both reanalysis products did not capture the observed precipitation trends
in most of Africa. This is not surprising, as precipitation trends are uncertain for most regions of the
globe, as noted in [64]. The observed temperature trend was much better captured in ERA5 than
in ERA-interim, which even showed cooling trends in several regions of Africa. These regions are
marked by strong wetting trends as well, which implies cooling due to increased cloud cover. Overall,
reanalysis data are generally considered unsuitable for trend analysis [58].
The representation of the observed precipitation cycle of East Africa has also improved
substantially from ERA-interim to ERA5, reducing most of the bias in the zonally averaged precipitation
data. In contrast to ERA-interim, ERA5 captured the variability well, but a not-observed drying trend
in the data contorts the occurrence of extreme years. However, this result has to be taken with a grain
of salt, as the precipitation trend might be underestimated in the CHIRPS data. In particular in the
Atmosphere 2020, 11, 996 15 of 19

north of East Africa other datasets display a drying trend since the 1950s [64] that is not captured in
the CHIRPS data for the chosen time period.
Analysis at the country level showed the strongest increase in correlation with observations in
East Africa in Tanzania with regards to temperature as well as precipitation. The one major lack of
improvement we found was in the precipitation bias in Uganda, which has increased from ERA-interim
to ERA5, even though variability has improved.
Overall, this study mainly suffers from the uncertainty of observational products that has been
shown for precipitation as well as surface temperature data [64,67,68]. Ideally, this study would
compare the reanalysis data to station data, but the sparsity of meteorological observations in Africa
limits such an analysis approach.
While ERA5 is much closer to observations than its predecessor, there is still room for further
improvement. For example, the resolution of 0.25◦ is still considered too coarse for very regional
studies and impact modeling. One first step to address this problem is the recently launched land-only
version of ERA5, called ERA5-Land. This dataset is a dynamically down-scaled version of ERA5’s
land component bias-corrected precipitation at a 9 km resolution [47,69]. It is produced with the Tiled
ECMWF Scheme for Surface Exchanges over Land incorporating land surface hydrology (H-TESSEL)
and produces surface and sub-subsurface processes on a very local scale. A first analysis of ECMWF
showed some improvements of ERA5-Land over ERA5 [69]. Another major improvement of reanalysis
data would be assimilation for precipitation.
From our results, we conclude that precipitation and temperature reanalysis in ERA5 are much
improved compared to ERA-interim in Africa, and we therefore discourage the continued use of
ERA-interim data. In contrast to previous reanalysis datasets, even the precipitation field in ERA5
displays a high agreement with observations. This implies that, though observational precipitation
datasets can provide higher resolution and accuracy, reanalysis precipitation can be used when a
coherent set of multiple climate variables is needed, like in the investigation of the natural variability
of coupled systems or the provision of initial conditions for land surface, hydrological, or vector
transmission models [58].
While the results of this study highlight the strengths as well as the limitations of the two
reanalysis products over East Africa in terms of monthly mean values of temperature and precipitation,
further in-depth analysis of the performance of reanalysis from daily to sub-seasonal time scales
targeting extreme events would be required to better assess the robustness and applicability of these
dataset for impact studies, such as agricultural and hydrological studies.

Author Contributions: Conceptualization, S.G. and T.D.; methodology, S.G. and T.D.; software, S.G.; validation,
S.G. and T.D. and G.T.D.; formal analysis, S.G.; investigation, S.G.; resources, S.G.; data curation, S.G.;
writing—original draft preparation, S.G. and T.D. and G.T.D.; writing—review and editing, S.G. and T.D. and
G.T.D.; visualization, S.G.; supervision, T.D. and G.T.D. All authors have read and agreed to the published version
of the manuscript.
Funding: This research received no external funding.
Acknowledgments: We thank CCAFS East Africa Director Dawit Solomon and the Ethiopian office of CCAFS
for hosting Stephanie Gleixner and therefore providing the opportunity for this study. The research was also
supported by the EPICC project, funded by IKI and supported by BMU, and the project "Adapting crop and
livestock systems in Mali to climate change" funded by the Norwegian Ministry of Foreign Affairs. G.T.D. is
currently affiliated with the Canadian Meteorological Centre, Environment and Climate Change Canada.
Conflicts of Interest: The authors declare no conflict of interest.
Atmosphere 2020, 11, 996 16 of 19

References
1. IPCC. Impacts, Adaptation, and Vulnerability. Part B: Regional Aspects. Contribution of Working Group
II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. In Climate Change
2014; Barros, V., Field, C., Dokken, D., Mastrandrea, M., Mach, K., Bilir,T.E., Ebi, K., Estrada, Y., Genova, R.,
Girma, B., et al., Eds.; Cambridge University Press: Cambridge, UK; New York, NY, USA, 2014; p. 688.
2. Ahmed, S.A.; Diffenbaugh, N.S.; Hertel, T.W.; Lobell, D.B.; Ramankutty, N.; Rios, A.R.; Rowhani, P.
Climate Volatility and Poverty Vulnerability in Tanzania; The World Bank: Washington, DC, USA 2009.
3. Hertel, T.W.; Burke, M.B.; Lobell, D.B. The poverty implications of climate-induced crop yield changes by
2030. Glob. Environ. Chang. 2010, 20, 577–585. [CrossRef]
4. World Bank. Agriculture for Development; World Bank: Washington, DC, USA 2008.
5. World Bank. Development and Climate Change. World Development Report 2010; Technical Report; World Bank:
Washington, DC, USA 2010.
6. Dinku, T. Challenges with availability and quality of climate data in Africa. In Extreme Hydrology and
Climate Variability; Melesse, A.M., Abtew, W., Senay, G.B.T.E.H., Variability, C., Eds.; Elsevier: Amsterdam,
The Netherlands, 2019; Chapter 7, pp. 71–80.
7. Harris, I.; Jones, P.D.; Osborn, T.J.; Lister, D.H. Updated high-resolution grids of monthly climatic
observations—The CRU TS3.10 Dataset. Int. J. Climatol. 2014, 34, 623–642. [CrossRef]
8. Maidment, R.I.; Grimes, D.; Black, E.; Tarnavsky, E.; Young, M.; Greatrex, H.; Allan, R.P.; Stein, T.; Nkonde, E.;
Senkunda, S.; et al. A new, long-term daily satellite-based rainfall dataset for operational monitoring in
Africa. Sci. Data 2017, 4, 170063. [CrossRef] [PubMed]
9. Novella, N.S.; Thiaw, W.M. African rainfall climatology version 2 for famine early warning systems. J. Appl.
Meteorol. Climatol. 2013, 52, 588–606. [CrossRef]
10. Dinku, T.; Funk, C.; Peterson, P.; Maidment, R.I.; Tadesse, T.; Gadain, H.; Ceccato, P. Validation of the
CHIRPS satellite rainfall estimates over eastern Africa. Q. J. R. Meteorol. Soc. 2018, 144, 292–312. [CrossRef]
11. Funk, C.; Peterson, P.; Landsfeld, M.; Pedreros, D.; Verdin, J.; Rowland, J.; Romero, B.; Husak, G.;
Michaelsen, J.; Verdin, A. A quasi-global precipitation time series for drought monitoring. U.S. Geol.
Surv. Data Ser. 2014, 832, 1–12. [CrossRef]
12. Harris, I.C.; Osborne, T.; Jones, P.; Lister, D. Version 4 of the CRU TS monthly high-resolution gridded
multivariate climate dataset. Sci. Data 2020, 7, 2052–4463. [CrossRef]
13. Osborn, T.J.; Jones, P.D. The CRUTEM4 land-surface air temperature data set: Construction, previous
versions and dissemination via Google earth. Earth Syst. Sci. Data 2014, 6, 61–68. [CrossRef]
14. Lenssen, N.; Schmidt, G.; Hansen, J.; Menne, M.; Persin, A.; Ruedy, R.; Zyss, D. Improvements in the
GISTEMP uncertainty model. J. Geophys. Res. Atmospheres. Geophys. Res. Atmos. 2019, 124, 6307–6326.
[CrossRef]
15. Smith, T.M.; Reynolds, R.W.; Peterson, T.C.; Lawrimore, J. Improvements to NOAA’s historical merged
land-ocean surface temperature analysis (1880–2006). J. Clim. 2008, 21, 2283–2296. [CrossRef]
16. Lawrimore, J.H.; Menne, M.J.; Gleason, B.E.; Williams, C.N.; Wuertz, D.B.; Vose, R.S.; Rennie, J. An overview
of the Global Historical Climatology Network monthly mean temperature data set, version 3. J. Geophys.
Res. Atmos. 2011, 116, 1–18. [CrossRef]
17. Decker, M.; Brunke, M.A.; Wang, Z.; Sakaguchi, K.; Zeng, X.; Bosilovich, M.G. Evaluation of the reanalysis
products from GSFC, NCEP, and ECMWF using flux tower observations. J. Clim. 2012, 25, 1916–1944.
[CrossRef]
18. Simmons, A.J.; Jones, P.D.; da Costa Bechtold, V.; Beljaars, A.C.; Kållberg, P.W.; Saarinen, S.; Uppala, S.M.;
Viterbo, P.; Wedi, N. Comparison of trends and low-frequency variability in CRU, ERA-40, and NCEP/NCAR
analyses of surface air temperature. J. Geophys. Res. Atmos. 2004, 109, 1–18. [CrossRef]
19. Lindsay, R.; Wensnahan, M.; Schweiger, A.; Zhang, J. Evaluation of Seven Different Atmospheric Reanalysis
Products in the Arctic. J. Clim. 2014, 27, 2588–2606. [CrossRef]
20. Luo, H.; Ge, F.; Yang, K.; Zhu, S.; Peng, T.; Cai, W.; Liu, X.; Tang, W. Assessment of ECMWF reanalysis
data in complex terrain: Can the CERA-20C and ERA-interim data sets replicate the variation in surface air
temperatures over Sichuan, China? Int. J. Climatol. 2019, 39, 5619–5634. [CrossRef]
21. Sahlu, D.; Moges, S.A.; Nikolopoulos, E.I.; Anagnostou, E.N.; Hailu, D. Evaluation of High-Resolution
Multisatellite and Reanalysis Rainfall Products over East Africa. Adv. Meteorol. 2017. [CrossRef]
Atmosphere 2020, 11, 996 17 of 19

22. Wang, G.; Zhang, X.; Zhang, S. Performance of three reanalysis precipitation datasets over the qinling-daba
mountains, eastern fringe of tibetan plateau, China. Adv. Meteorol. 2019. [CrossRef]
23. Kalnay, E.; Kanamitsu, M.; Kistler, R.; Collins, W.; Deaven, D.; Gandin, L.; Iredell, M.; Saha, S.; White, G.;
Wollen, J.; et al. The NCEP/NCAR 40-year reanalysis project. Bull. Am. Meteorol. Soc. 1996, 77, 437–470.
[CrossRef]
24. Uppala, S.M.; Kållberg, P.W.; Simmons, A.J.; Andrae, U.; da Costa Bechtold, V.; Fiorino, M.; Gibson, J.K.;
Haseler, J.; Hernandez, A.; Kelly, G.A.; et al. The ERA-40 re-analysis. Q. J. R. Meteorol. Soc. 2005,
131, 2961–3012. [CrossRef]
25. Hollmann, R. Precipitation—Essential Climate Variable (ECV) Factsheet; Technical Report; GCOS: Geneva,
Switzerland 2020.
26. Bauer, P.; Thorpe, A.; Brunet, G. The quiet revolution of numerical weather prediction. Nature 2015,
525, 47–55. [CrossRef]
27. Legates, D.R. Climate Models and their Simulation of Precipitation. Energy Environ. 2014, 25, 1163–1175.
[CrossRef]
28. Poccard, I.; Janicot, S.; Camberlin, P. Comparison of rainfall structures between NCEP/NCAR reanalyses
and observed data over tropical Africa. Clim. Dyn. 2000, 16, 897–915. [CrossRef]
29. Diro, G.; Grimes, D.I.F.; Black, E.; O’Neill, A.; Pardo-Iguzquiza, E. Evaluation of reanalysis rainfall estimates
over Ethiopia. Int. J. Climatol. 2009, 29, 67–78. [CrossRef]
30. Diro, G.; Toniazzo, T.; Shaffrey, L. Ethiopian rainfall in climate models. In African Climate and Climate Change;
Springer: Dordrecht, The Netherlands, 2011; pp. 51–69.
31. Lemma, E.; Upadhyaya, S.; Ramsankaran, R. Investigating the performance of satellite and reanalysis
rainfall products at monthly timescales across different rainfall regimes of Ethiopia. Int. J. Remote Sens. 2019,
40, 4019–4042. [CrossRef]
32. Koutsouris, A.J.; Chen, D.; Lyon, S.W. Comparing global precipitation data sets in eastern Africa: A case
study of Kilombero Valley, Tanzania. Int. J. Climatol. 2016, 36, 2000–2014. [CrossRef]
33. Hersbach, H.; Bell, B.; Berrisford, P.; Horányi, A.; Sabater, J.M.; Nicolas, J.; Radu, R.; Schepers, D.;
Simmons, A.; Cornel Soci, D.D. Global reanalysis: Goodbye ERA-interim, hello ERA5. ECMWF Newsl. 2019,
159, 17–24.
34. Martens, B.; Schumacher, D.L.; Wouters, H.; Muñoz-Sabater, J.; Verhoest, N.E.C.; Miralles, D.G. Evaluating
the surface energy partitioning in ERA5. Geosci. Model Dev. Discuss. 2020, 1–35. [CrossRef]
35. Urraca, R.; Huld, T.; Gracia-Amillo, A.; Martinez-de Pison, F.J.; Kaspar, F.; Sanz-Garcia, A. Evaluation
of global horizontal irradiance estimates from ERA5 and COSMO-REA6 reanalyses using ground and
satellite-based data. Sol. Energy 2018, 164, 339–354. [CrossRef]
36. Betts, A.K.; Chan, D.Z.; Desjardins, R.L. Near-Surface Biases in ERA5 Over the Canadian Prairies.
Front. Environ. Sci. 2019, 7, 129. [CrossRef]
37. Tarek, M.; Brissette, F.P.; Arsenault, R.; De, É.; West, N.D. Evaluation of the ERA5 reanalysis as a potential
reference dataset for hydrological modelling over North America. Hydrol. Earth Syst. Sci. 2020, 24, 2527–2544.
[CrossRef]
38. Beck, H.E.; Zimmermann, N.E.; McVicar, T.R.; Vergopolan, N.; Berg, A.; Wood, E.F. Present and future
köppen-geiger climate classification maps at 1-km resolution. Sci. Data 2018, 5, 1–12. [CrossRef] [PubMed]
39. Nicholson, S. The Turkana low-level jet: Mean climatology and association with regional aridity.
Int. J. Climatol. 2016, 36, 2598–2614. [CrossRef]
40. Yang, W.; Seager, R.; Cane, M.A.; Lyon, B. The annual cycle of East African precipitation. J. Clim. 2015,
28, 2385–2404. [CrossRef]
41. Holmes, T.R.; Hain, C.R.; Anderson, M.C.; Crow, W.T. Cloud tolerance of remote-sensing technologies to
measure land surface temperature. Hydrol. Earth Syst. Sci. 2016, 20, 3263–3275. [CrossRef]
42. Awange, J.L.; Ferreira, V.G.; Forootan, E.; Andam-Akorful, S.A.; Agutu, N.O.; He, X.F. Uncertainties in
remotely sensed precipitation data over Africa. Int. J. Climatol. 2016, 36, 303–323. [CrossRef]
43. Dinku, T.; Ceccato, P.; Grover-Kopec, E.; Lemma, M.; Connor, S.J.; Ropelewski, C.F. Validation of satellite
rainfall products over East Africa’s complex topography. Int. J. Remote Sens. 2007, 28, 1503–1526. [CrossRef]
44. Dinku, T.; Ceccato, P.; Connor, S.J. Challenges of satellite rainfall estimation over mountainous and arid
parts of east Africa. Int. J. Remote Sens. 2011, 32, 5965–5979. [CrossRef]
Atmosphere 2020, 11, 996 18 of 19

45. Dee, D.P.; Uppala, S.M.; Simmons, A.J.; Berrisford, P.; Poli, P.; Kobayashi, S.; Andrae, U.; Balmaseda, M.A.;
Balsamo, G.; Bauer, P.; et al. The ERA-interim reanalysis: Configuration and performance of the data
assimilation system. Q. J. R. Meteorol. Soc. 2011, 137, 553–597. [CrossRef]
46. Simmons, A.J.; Uppala, S.; Dee, D.; Kobayashi, S. ERA-interim: New ECMWF reanalysis products from 1989
onwards. ECMWF Newsl. 2007, 110, 25–36. [CrossRef]
47. Hennermann, K.; Giusti, M. ERA5: Data Documentation; Technical Report; ECMWF: Reading, UK 2020.
48. Weedon, G.P.; Balsamo, G.; Bellouin, N.; Gomes, S.; Best, M.J.; Viterbo, P. The WFDEI meteorological forcing
data set: WATCH Forcing Data methodology applied to ERA-interim reanalysis data. Water Resour. Res.
2014, 50, 7505–7514. [CrossRef]
49. Muthoni, F.K.; Odongo, V.O.; Ochieng, J.; Mugalavai, E.M.; Mourice, S.K.; Hoesche-Zeledon, I.; Mwila, M.;
Bekunda, M. Long-term spatial-temporal trends and variability of rainfall over Eastern and Southern Africa.
Theor. Appl. Climatol. 2019, 137, 1869–1882. [CrossRef]
50. Hay, S.I.; Lennon, J.J. Deriving meteorological variables across Africa for the study and control of vector-borne
disease: A comparison of remote sensing and spatial interpolation of climate. Trop. Med. Int. Health 1999,
4, 58–71. [CrossRef] [PubMed]
51. Sylla, M.B.; Giorgi, F.; Coppola, E.; Mariotti, L. Uncertainties in daily rainfall over Africa: Assessment of
gridded observation products and evaluation of a regional climate model simulation. Int. J. Climatol. 2013,
33, 1805–1817. [CrossRef]
52. Tesfaye, T.W.; Dhanya, C.; Gosain, A. Evaluation of ERA-interim, MERRA, NCEP-DOE R2 and
CFSR Reanalysis precipitation Data using Gauge Observation over Ethiopia for a period of 33 years.
AIMS Environ. Sci. 2017, 4, 596–620. [CrossRef]
53. Simmons, A.J.; Willett, K.M.; Jones, P.D.; Thorne, P.W.; Dee, D.P. Low-frequency variations in surface
atmospheric humidity, temperature, and precipitation: Inferences from reanalyses and monthly gridded
observational data sets. J. Geophys. Res. Atmos. 2010, 115. [CrossRef]
54. Lin, R.; Zhou, T.; Qian, Y. Evaluation of Global Monsoon Precipitation Changes based on Five Reanalysis
Datasets. J. Clim. 2014, 27, 1271–1289. [CrossRef]
55. Jury, M.R. Climate trends in southern Africa. S. Afr. J. Sci. 2013, 109, 1–11. [CrossRef]
56. Collins, J.M. Temperature Variability over Africa. J. Clim. 2011, 24, 3649–3666. [CrossRef]
57. Bengtsson, L.; Hagemann, S.; Hodges, K.I. Can climate trends be calculated from reanalysis data? J. Geophys.
Res. Atmos. 2004, 109. [CrossRef]
58. Trenberth, K.E.; Koike, T.; Onogi, K. Progress and Prospects for Reanalysis for Weather and Climate.
Eos Trans. Am. Geophys. Union 2008, 89, 234–235. [CrossRef]
59. Thorne, P.W.; Vose, R.S. Reanalyses Suitable for Characterizing Long-Term Trends. Bull. Am. Meteorol. Soc.
2010, 91, 353–362. [CrossRef]
60. Lim, E.P.; Hendon, H.H. Causes and Predictability of the Negative Indian Ocean Dipole and Its Impact on
La Niña During 2016. Sci. Rep. 2017, 7, 12619. [CrossRef] [PubMed]
61. Henricksen, B.L. Reflections on drought: Ethiopia 1983–1984. Int. J. Remote Sens. 1986, 7, 1447–1451.
[CrossRef]
62. Aparna, S.G.; McCreary, J.P.; Shankar, D.; Vinayachandran, P.N. Signatures of Indian Ocean Dipole and
El Niño–Southern Oscillation events in sea level variations in the Bay of Bengal. J. Geophys. Res. Ocean.
2012, 117. [CrossRef]
63. Bahaga, T.; Mengistu Tsidu, G.; Kucharski, F.; Diro, G. Potential predictability of the sea-surface temperature
forced equatorial East African short rains interannual variability in the 20th century. Q. J. R. Meteorol. Soc.
2015, 141, 16–26. [CrossRef]
64. Hartmann, D.; Tank, A.K.; Rusticucci, M.; Alexander, L.; Brönnimann, S.; Charabi, Y.; Dentener, F.;
Dlugokencky, E.; Easterling, D.; Kaplan, A.; et al. Observations: Atmosphere and surface. In Climate
Change 2013 the Physical Science Basis: Working Group I Contribution to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change; Hartmann, D.L., Klein Tank, A.M., Rusticucci, M., Alexander, L.V.,
Brönnimann, S., Charabi, Y.A.R., Dentener, F.J., Dlugokencky, E.J., Easterling, D.R., Kaplan, A., et al. Eds.;
Cambridge University Press: Cambridge, UK, 2013; pp. 159–254.
65. Hastenrath, S.; Polzin, D.; Mutai, C. Diagnosing the 2005 Drought in Equatorial East Africa. J. Clim. 2007,
20, 4628–4637. [CrossRef]
66. Bell, G.D.; Halpert, M.S. Climate Assessment for 1997. Bull. Am. Meteorol. Soc. 1998, 79, S1–S50. [CrossRef]
Atmosphere 2020, 11, 996 19 of 19

67. Beck, H.E.; Vergopolan, N.; Pan, M.; Levizzani, V.; van Dijk, A.I.J.M.; Weedon, G.P.; Brocca, L.;
Pappenberger, F.; Huffman, G.J.; Wood, E.F. Global-scale evaluation of 22 precipitation datasets using
gauge observations and hydrological modeling. Hydrol. Earth Syst. Sci. 2017, 21, 6201–6217. [CrossRef]
68. Xu, W.H.; Li, Q.X.; Yang, S.; Xu, Y. Overview of global monthly surface temperature data in the past century
and preliminary integration. Adv. Clim. Chang. Res. 2014, 5, 111–117. [CrossRef]
69. Muñoz-Sabater, J.; Dutra, E.; Balsamo, G.; Boussetta, S.; Zsoter, E.; Albergel, C.; Agusti-Panareda, A.
ERA5-Land: An improved version of the ERA5 reanalysis land component. In Proceedings of the 8th
Workshop-Joint ISWG and LSA-SAF Workshop, Lisbon, Portugal, 26–28 June 2018; pp. 26–28.

c 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).