Assessment of ERA5 and ERA-Interim in Reproducing Mean and Extreme Climates over West Africa

West African countries are susceptible to global warming and its consequences because of their vulnerability to extreme weather conditions and poor adaptation potential compared to other parts of the world (Fitzpatrick et al., 2020). The climate in West Africa is anticipated to change due to rising temperatures and unpredictable rainfall patterns. Its population, on the other hand, is projected to grow faster than the global average (Mechiche-Alami and Abdi, 2020). Countries in the region have often contended with extreme weather and climate events, such as floods (Badou et al., 2019; Tramblay et al., 2020) and droughts (Rodríguez-Fonseca et al., 2015; Ekwezuo and Ezeh, 2020). Climate change has been identified as an increasing danger to food security and poor nutrition in these countries, where the main practice is subsistence agriculture (Sorgho et al., 2020). Therefore, understanding the climatic conditions is critical for identifying present and future vulnerabilities and developing climate change adaptation strategies (Araya-Osses et al., 2020).

Weather stations and associated observational datasets are sparse over Africa. Their numbers are frequently below the World Meteorological Organization’s recommended minimum and are declining (Le Coz and Van De Giesen, 2020). In addition, the number of stations in Africa in recent decades has decreased compared to the decade of the 1970s (Harris et al., 2020). Consequently, the highest error in the Climate Research Unit (CRU) dataset is observed over the edge of the Sahara and other dry regions. Even in areas where coverage is reasonably good, data may not be easily accessible (Quagraine et al., 2020). Therefore, the improved availability of gridded climate data obtained from reanalysis datasets is a possible alternate solution that will help compensate for the region’s sparse gauge network (Muthoni, 2020; Quagraine et al., 2020). Reanalysis datasets are an important component to climate studies and have thus been widely used (Dou et al., 2021; He et al., 2021). Reanalysis data are developed by integrating atmospheric models with available observations from several sources, using data assimilation techniques to determine the atmospheric conditions at different temporal and spatial scales (Alghamdi, 2020; Rakhmatova et al., 2021). Reanalyses data provide a physically consistent estimate of the atmospheric fields, allowing for process-based analysis of precipitation variability, temperature changes, and other related atmospheric drivers (Quagraine et al., 2020). Furthermore, they are open-access data, which makes them easily available and accessible for use.

However, existing literature has shown that the accuracy of reanalysis data differs significantly across regions and among variables (e.g., Rakhmatova et al., 2021). These datasets can have significant biases in regions with a paucity of observations or in complex remote terrain; thus, temperature data are usually more reliable than other data (e.g., precipitation)(Martins et al., 2017; Luo et al., 2019).

Several studies (e.g., Zhang et al., 2013; Lee and Biasutti, 2014; Zhan et al., 2016; Nkiaka et al., 2017; Berntell et al., 2018; Hua et al., 2019; Igbawua et al., 2019; Quagraine et al., 2020) have been conducted to assess the performance of reanalysis datasets from the Modern-Era Retrospective Analysis for Research and Applications (MERRA; Rienecker et al., 2011), ECMWF 20th century reanalysis (ERA20C; Poli et al., 2016), National Oceanic and Atmospheric Administration Twentieth Century reanalysis (NOAA-20CR; Slivinski et al., 2019), Japanese 55-year Reanalysis (JRA55; Kobayashi et al., 2015), and the European Center for Medium-Range Weather Forecasting (ECMWF) interim reanalysis (ERAI; Dee et al., 2011) over various parts of Africa and demonstrated the pros and cons in their representation of rainfall. While the reliability of such datasets in reproducing long-term trends has been questioned (Dee et al., 2011; Bloomfield et al., 2018), some reproduced key climatological features better than others, thereby justifying their extensive use in climate science. For instance, Hua et al. (2019) compared seven reanalysis datasets over central Africa and found that all reanalyses captured the seasonal evolution rainfall well, despite exhibiting biases in the intensity and other spatiotemporal characteristics. Similarly, Nkiaka et al. (2017) have shown that ERA-Interim and Climate Forecasting System Reanalysis (CFSR) reproduced the rainfall pattern over the Lake Chad basin in the Sahel region, although the ERA-Interim reanalysis represented climatology better than CFSR.

The most recent reanalysis product from the ECMWF is ERA5 (Hersbach et al., 2019), which succeeds the ERAI. Following ten years of model and data assimilation growth, ERA5 represents a significant upgrade (He et al., 2021). In comparison to ERAI, this latest reanalysis offers several enhancements, including an improved spatial and temporal resolution and a more accurate description of geophysical processes in the prediction model and added observational contributions to the data assimilation scheme (Hoffmann et al., 2019).

To apply the data for climate studies and operational usage in West Africa, a thorough assessment of temperature and precipitation variability and long-term trends, as well as an adequate representation of climate extremes is needed (Rakhmatova et al., 2021). In particular, when new reanalysis data sets such as ERA5 are produced and become available, pointing out its advantage and improvement compared to its predecessor is crucial for building confidence in their application. For instance, a study conducted by Hersbach et al. (2019) found a better simulation of precipitation in ERA5 over ERAI for global average precipitation. Other recent studies have also shown that ERA-5 has a higher precision in simulating observations than ERAI for various parameters, periods, and regions (Beck et al., 2019; Wang et al., 2019; Zhang et al., 2019; Gleixner et al., 2020). Most of these studies have focused on the evaluation of the mean climate variables and provided less emphasis on assessing the representation of climate extremes. However, to the best of the author’s knowledge, an evaluation of ERA5 reanalysis in representing both the mean and extreme climate characteristics over West Africa has yet to be thoroughly examined. Therefore, the goal of this work is to evaluate the performance of ERAI and ERA5 in terms of representing the spatial pattern, and temporal variability of both mean and extreme climate conditions of West Africa for the 1981–2018 period. The comparison will be done relative to observational datasets from CRU and Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS), which are regarded as reference datasets.

The remainder of this paper is organized as follows. Section 2 gives a description of the location of the study area, datasets employed, and the methods applied in this study. Section 3 is dedicated to presenting and discussing the results. Finally, summary and conclusions are provided in section 4.

This research focuses on the entirety of West Africa and the three climatic zones within the region (Fig. 1). It is geographically located between longitudes 20°W and 20°E and latitudes 0° and 25°N. The study region is divided into three climatic zones (Fig. 1), the Guinea Coast (4°–8°N), Savannah (8°–11°N), and Sahel (11°–16°N), which well-represent the major climatic regimes of the West African region as described in Gbode et al. (2019). Analysis of the climatic zones is limited to the longitudinal bounds of 10°W–10°E. The Guinea Coast marks the southern part of the domain bordering with the Atlantic Ocean; the Sahel zone (Mauritania, Mali, and Niger) encompasses the northern borders, bordered on the west by the Atlantic Ocean and the east Republic of Cameroon, and the Republic of Chad; and the Savannah lies within the middle-belt of region bordered by Sahel to the north and Guinea Coast to the south. A few localized mountains influence West Africa's climate (e.g., the Guinea Highlands, Cameroon Mountains, and Jos Plateau). Additionally, the West African Monsoon (WAM), which accounts for more than 70% of annual rainfall, significantly dominates West Africa’s climate during the boreal summer season (Hagos and Cook, 2007).

Figure 1.  A map of West Africa showing the topography (m) and the three climatic zones: the Guinea Coast (4–8°N), Savannah (8–11°N), and Sahel (11–16°N).

We relied on gridded observational datasets since in situ data is sparse and sometimes unevenly distributed over Africa. Gridded observed monthly temperature and daily precipitation data were obtained from the CRU gridded Time Series (CRU TS) version 4.02 (Harris et al., 2020) and CHIRPS-v2 (Funk et al., 2015) databases, respectively. The CRU TS v4.02 dataset is the fourth edition of the gridded product produced by the University of East Anglia’s Climate Research Unit at a spatial resolution of 0.5° × 0.5°. CRU data is obtained from the World Meteorological Organization archives and covers the period from 1901 to 2017 (Harris et al., 2014, 2020). The CHIRPS product was developed by researchers at the US Geological Survey and the University of California, Santa Barbara. CHIRPS is a quasi-global (50°S–50°N), high spatial resolution (0.05°) precipitation dataset, available from 1981 to near real-time. It was created by integrating station data and a wide range of satellite measurements (Funk et al., 2015). CHIRPS precipitation estimates have been assessed through a comparison with station data and satellite precipitation products globally and regionally, and they have been found to perform relatively reasonably well (Prakash, 2019; Paredes-Trejo et al., 2021).

This study concentrates on two reanalyses, ERAI and ERA5; since they are produced with atmospheric models and state-of-the-art assimilation methods, they tend to yield better performances (e.g., Chen et al., 2014; Alghamdi, 2020). ERAI (Dee et al., 2011) is the 4th-generation global atmospheric reanalysis produced by the ECMWF. In comparison, ERA5 is the fifth-generation and latest global atmospheric reanalysis by the ECMWF (Hersbach et al., 2019). These two versions of reanalysis use four-dimensional variational data assimilation (4D-Var) and span the entire globe. ERA-5 provides several substantial improvements over ERA-Interim: 1) it employs an updated and improved version of the ECMWF Integrated Forecast System (IFS Cycle 41r2 versus Cycle 31r2); 2) it has a higher spatial resolution (a 0.25° grid against a 0.75° grid, and 137 levels against 60 vertical levels); 3) it has a higher temporal resolution; 4) it assimilates substantially more and newer versions of observational datasets; 5) it uses more contemporary and better forcing datasets, including sea surface temperature and sea-ice concentration; 6) ERA-5 covers a longer period (i.e., 1950 to present).

This study presents an analysis of the total precipitation and average 2-m temperature (T2m) from 1981 to 2018. Because the datasets used in this study are of different spatial resolutions, all data were re-gridded from their original resolutions to the coarsest 0.75° × 0.75° horizontal resolution grid that ERA-I utilizes, using a first-order conservative remapping technique (Jones, 1999), to enable comparison at the same spatial scale. Also, since the CRU temperature is only usable as monthly means, the two-reanalysis and selected observed data were converted to monthly and annual time scales for comparison purposes. For both temperature and precipitation, annual averages and total, respectively, were used for correlation analysis. We calculated the trend by taking annual sums of precipitation and annual means of temperature. The precipitation cycle was estimated from datasets by also taking the monthly means. We used absolute means to compute averages at the zonal scale since the data’s overall performance, including the variability across the three climatic zones, is needed. A Student’s t-test (Mishra et al., 2019) is used to test the hypothesis based on the difference between spatial temperature and precipitation means. This is a parametric test that assumes a normal distribution of the population distribution from which the sample is drawn. The t-test helps to determine the probability that two populations are the same with respect to the variable tested and provides the level of statistical significance based on the probability. Also, the Mann-Kendall (Mann, 1945; Kendall, 1975) statistical test was used to test whether there is a statistically significant trend in the observed and reanalysis data. Mann-Kendall statistical trend test is a non-parametric technique used to determine whether a set of data values is increasing or decreasing over time and whether the trend is statistically significant. The magnitude of change is not evaluated by the Mann-Kendall test.

Six various extreme indices chosen from the list defined and suggested by the Expert Team on Climate Change Detection and Indices (ETCCDI) (Zhang et al., 2011; Xi et al., 2018) were computed from the daily precipitation series to assess the performance of reanalysis datasets in representing climate extremes. These extreme indices are consecutive dry days (cdd), consecutive wet days (cwd), annual maximum 5-day precipitation (Rx5day), number of heavy rain days (r10mm), contributions from very wet days when precipitation exceeding the 95th percentile threshold (r95ptot), and the total annual total wet day, when daily precipitation exceeds 1 mm (prcptot).

Figure 2 shows the spatial distribution of the annual climatology (1981–2018) of near-surface temperature over West Africa for CRU, ERAI, and ERA5, as well as the temperature biases of the reanalysis. Over West Africa, the CRU temperatures show a maximum temperature with zonal distribution, with high values (>30°C) centered near 15°N in the Sahel and low values (22°C) along the Guinea coast and in the complex terrain such as the Guinea highlands, Cameroon mountains, and Jos Plateau. Clearly, both the ERAI and ERA5 adequately capture this spatial pattern of annual mean temperature over the region. ERAI underestimated the near-surface temperature over the Guinea Coast region. It exhibits a cold bias distribution of approximately 1°C to 2°C (Fig. 2d). ERA5 also shows weak biases around the Guinea highlands. Nonetheless, it displays cold biases close to 2°C over the western Sahara (Fig. 2e). Average cooling of about 0.26°C and 0.3°C is found in ERAI and ERA5 (Table 1), respectively, over the entire West African domain (Fig. 2d-e). Similarly, cold biases of 0.26°C–0.84°C in ERAI and 0.18°C–0.52°C in ERA5 persists over the three climatic zones, with the Guinea Coast being the coldest zone (Table 1). The spatial distribution of temperature correlation values for the two reanalysis fields (against CRU) is shown in Figs. 2f-g. For both reanalyses, the highest correlation (>0.9) is located over the northern part of the domain and decreases slightly southward to the coastal region. The result clearly demonstrates that both reanalysis datasets have well-reproduced the spatial variability of temperature (Fig. 2f-g), with ERA5 performing much better than ERAI. This suggests that these datasets can be used to verify seasonal forecasts or for variability studies.

Figure 2.  Spatial distribution of annual climatological mean 2-m surface temperature (T2m; °C) from (a) CRU, (b) ERAI, and (c) ERA5 for the periods 1981–2018, and the temperature biases for (d) ERAI – CRU (e) ERA5 – CRU, and correlation of annual average T2m (f and g) relative to CRU. Stippling denotes areas where differences or correlations are statistically significant at the 0.01 level

The spatial distribution of mean precipitation over West Africa from 1981 to 2018 is shown in Fig. 3 for both CHIRPS observations and the reanalyses. The CHIRPS shows a zonal pattern, and the precipitation maxima associated with the Fouta Djallon highlands, the Cameroonian highlands, and Jos Plateau are also highlighted (Fig. 3a). Both the reanalysis simulations exhibit similar spatial patterns to CHIRPS, with the precipitation increasing from north to south. Although the reanalysis reproduces the zonal bands and localisations of precipitation maxima fairly well, their intensities vary (Figs. 3b-c). For example, ERAI reproduced the precipitation pattern with an underestimate over the two peak locations, one over Liberia and Sierra Leone and the other over the southern part of the Nigeria-Cameroon border. In addition, ERAI shows a significant precipitation bias and a widespread underestimation in northern West Africa (Fig. 3d). About a 109.85 mm difference between ERAI and CHIRPS exists over the entire West African domain (Table 2). It can be deduced that ERAI is excessively dry in the region’s northern belt, implying a lack of northward migration of the monsoon rainband and low rainfall intensities. ERA5, on the other hand, shows a significant reduction in dry bias over a large part of West Africa (i.e., an average of 56.08 mm dryness; Fig. 3e), even though the underestimation over the Sahel persisted. For example, the bias over Guinea highlands is reduced from 80% in ERAI to less than 20% in ERA5. In the sub-regions, drier conditions dominate the Savannah and Sahel, especially in the latter region, where a 252.33 mm and 117.02 mm reduction in precipitation is found in ERAI and ERA5, respectively. At the same time, precipitation overestimates of 13.83–87.33 mm is obtained in Guinea Coast (Table 2). The difference between CHIRPS and ERAI is greater when compared with that of ERA5.

Figure 3.  Spatial distribution of mean annual total precipitation (mm yr^–1) from (a) CHIRPS, (b) ERAI, and (c) ERA5, and the precipitation biases (d) ERAI – CHIRPS, (e) ERA5 – CHIRPS, and the correlation of annual total precipitation values (f and g) for the 1981–2018 period. Stippling denotes areas where the differences or correlations are statistically significant at the 0.01 level.

The temporal correlation between both precipitation reanalyses and CHIRPS observations from 1981 to 2018 displays spatial differences (Figs. 3f-g). ERA5 is significantly and positively correlated to CHIRPS across West Africa, with correlation coefficients of up to 0.9. Both negative and positive values exist in the correlation between ERAI and CRU, with an average of 0.1 over the entire region. The correlation in ERA5 improved significantly, attaining an average of 0.34. Values reaching 0.6 are obtained between longitudes 5°W and 5°E. This suggests that ERA5 is better at reproducing the variability of precipitation. Gleixner et al. (2020) obtained a similar result, suggesting that model physics, resolution, and data assimilation method advancements in ERA5 continue to improve precipitation simulations. A closer physical relationship between ERA5 and CHIRPS observations can be inferred based on the results.

The spatial pattern of the annual mean T2m temperature trend is depicted in Fig. 4a according to observation and reanalysis datasets. In the CRU observations, the most significant positive trends are observed across most parts of West Africa. Despite all three datasets agreeing on a warming trend across the region, ERAI shows a few patches of decreasing/cooling temperature along the coast from Sierra Leone to Senegal. This cooling trend of ERAI agrees with the results of Simmons et al. (2010) and Gleixner et al. (2020), which may be partially explained by a change in the source of analysis. Both studies showed a drop in temperature when comparing the 1998–2008 period to the 1989–1998 period and when analyzing temperature in the 1981–2017 period in ERAI (Gleixner et al., 2020). ERA5 exhibits a widespread significant positive trend across the West African region. This trend value in ERA5 is closer to the CRU observation except that the stronger positive trend extends further to the northwest part of the region. There are no areas in ERA5 that have a negative temperature trend. Overall, the slope of the trend is substantially closer to observations in ERA5 than ERAI, particularly across Guinea and the Savannah. Nonetheless, the warming trend in ERA5 is greater than in CRU.

Figure 4.  Spatial patterns of trends in annual mean T2m [°C (10 yr)^–1] and precipitation (mm yr^–1) from observations (CRU and CHIRPS) and reanalysis (ERAI and ERA5). Stippling denotes regions where trends are statistically significant at the 0.05 level.

Figure 5.  Time series of (a) mean annual temperature (°C) and (b) annual total precipitation (mm) over the three sub-regions from observation (CRU; black), ERAI (dashed blue), and ERA5 (solid blue) in the Guinea Coast, Savannah, and the Sahel for the period 1981−2018. S represents the slope, and pV is the p-value computed at the 0.05 significance level.

Trends in annual total precipitation reveal considerably more significant disparities between reanalysis and observations (Figs. 4d–f). CHIRPS indicates a weak dipole signal with an insignificant decreasing trend across the coastal regions, except for some regions in Cameroon and Sierra Leone. In contrast, a wetting trend is found throughout most of the Sahel region north of 10°N. On the other hand, the reanalyses display a significant drying trend in this region, particularly the ERAI. It is also observed that ERAI has an opposing wetting trend over Sierra Leone, where the observations indicated a significant drying trend. ERA5 reproduces the reduction in rainfall over the central Guinea coast, but significant drying is seen extending further east and west over the region. This drying tendency in reanalysis datasets is consistent with previous studies (Lin et al., 2014; Gleixner et al., 2020; Quagraine et al., 2020). Gleixner et al. (2020) found extreme drying in ERA5 and ERAI across west Africa, and Quagraine et al. (2020) noted that ERAI and ERA5 do not reproduce the wetting trends found in observational datasets (CRU, CHIRPS, and GPCC) across the Sahel for the study period 1981–2016. According to these findings, utilizing reanalysis products for trend analysis, particularly the ERA-interim product, is subject to doubt. While reanalysis is often used in research (Collins, 2011; Jury, 2013), it is typically regarded as unsuitable for discovering long-term trends (Bengtsson et al., 2004; Thorne and Vose, 2010). The reason concerns the incorporation of observational datasets into the assimilation system at distinct moments in time (Poccard et al., 2000; Hersbach et al., 2019). This can result in data jumps and account for the discrepancies between observations and reanalysis regarding long-term trends (Bengtsson et al., 2004).

The time series of annual mean temperature and precipitation aggregated over the Guinea Coast, Sahel, and Savannah region are shown in Figs. 5a and 5b, respectively. Figure 5a shows that both ERAI and ERA5 reanalysis products could capture inter-annual temperature variability over the three regions, especially in ERA5, whose trends are similar to that of CRU. The CRU has a gradual increase in the mean annual intensity of temperature beginning in 2000, which is well-captured by both reanalyses, although with cold biases, particularly over the Guinea domain (Fig. 5a; top panel). ERA5 is in closer agreement with the CRU temperature over the Guinea and Savannah domains than ERAI. It is interesting to note that the spikes in the temperature time series, namely, 1987, 1998, 2010, and 2016 are related to the mature phase of ENSO years (i.e., the El Nino), and both reanalyses are able to capture this teleconnection between regional temperature variations with ENSO.

Figure 5b presents the time series of area-averaged annual total precipitation of CHIRPS and reanalyses datasets for the three sub-regions. ERA5 is in relatively good agreement with the CHIRPS observations regarding the magnitude of error over the Savannah and Sahel. By contrast, precipitation in ERAI shows a decreasing trend starting from around 1995, possibly due to the change in the satellite and other observational datasets assimilated to the system. The pronounced drought of 1983/84 is well-captured by both products; however, for this drought year (1983/84), ERA5 overestimated precipitation in the Guinea area. The time series shows that the similarity between ERA5 and CHIRPS increases over time and illustrates that the new reanalysis system (ERA5) manages to eliminate the artificial drying trend that was present in ERAI. This temporal trend and inter-annual correlation/variation improvement in ERA5 over ERAI in precipitation can be attributed to the improvement in the input data sources as well as the improvements in the data assimilation methods, model physics, and increased model resolution.

Figure 6 presents a Hovmöller diagram of precipitation averaged over 10°W to 10°E for the observation (CHIRPS) and the reanalyses (ERAI and ERA5). Over West Africa, the monsoon system is inextricably linked to the ITCZ’s northward migration during boreal spring and summer (Janicot et al., 2010; Akinsanola and Zhou, 2019). The three distinct phases of the West African monsoon annual cycle are shown by CHIRPS observations: the first phase is the monsoon onset occurring in May−June over the Guinea coast subregion, located in the south, with the core around latitude 6°N, the second phase involves a northward jump of the rainfall band towards the Sahel, resulting in the maximum rainfall delivery over this region in August, and the third, a southward retreat of the rain band towards the Guinea Coast, resulting in the end of rainfall and onset of dry season for the larger parts of the West African region. Relative to CHIRPS, the core of the first phase is displaced southward at 6°N. Although ERAI reproduces the observed monsoon jump, there is a reduction in precipitation intensity compared to observations. For example, ERAI did not accurately capture the nonlinear northward transition of the monsoon rains between June−July and the maximum precipitation in the Sahel around June–August. This suggests that ERAI fails to migrate the precipitation north enough, and probably precipitates all the water prematurely.

Figure 6.  Hovmöller diagram of monthly means of daily precipitation (mm d^–1) for the period 1981–2018 for (a) CHIRPS, (b) ERAI, and (c) ERA5.

In contrast, ERA5 is better at reproducing the characteristics of West African monsoon precipitation. The timing of the onset season in Guinea, the heavy precipitation in the Sahel, and the eventual southward retreat in October are well-captured. These results indicate that ERA5 can reproduce the basic climatological features of West African summer precipitation.

Figures 7 and 8 present scatter plots of the ERA5 and ERAI reanalyses against observations to illustrate a more thorough examination of the distribution of temporal temperature and precipitation in the three sub-regions. The seasonal temperature and precipitation cycles in each of the sub-regions are also presented to gain further insights into the capability of the reanalysis products to capture phases and peak precipitation in the specific sub-regions in West Africa. In all three locations, the scatter plots of the average monthly precipitation in Fig. 7 shows that ERA5 is more closely related to observations than ERAI (maximum R is 0.95 in the Savannah) since the values cluster closer to the best fit lines with a correlation coefficient of 0.98. The precipitation clustering around the best fit is higher in Guinea Coast and Savannah, suggesting a reduced error in magnitude relative to CHIRPS. The worst fit for precipitation of both reanalysis products is in the Sahel. However, even here, the ERA5 is closer to observations than ERAI, noting that both products underestimate the observed monthly values. The annual cycles of precipitation from observations and the reanalysis datasets show that the maximum rainfall during their wet season coincides in the three sub-regions (third row; Fig. 7). Both reanalysis products well-reproduce the observed seasonality of precipitation, despite the differences in amounts. In Guinea, both reanalysis products displayed two peaks of precipitation, the first peak in June, a secondary one in September, and a relative minimum in August. However, the two reanalyses produce an early onset and late cessation in the Guinea Coast. This disparity is substantially reduced in ERA5. Over the Savannah, the ERAI reanalysis product has a dry bias in the peak of the rainy seasons, while ERA5 is closely related to the CHIRPS observation. Over the Sahel, an underestimation is also pronounced in ERAI than in ERA5 throughout most of the month and in the rainy seasons.

Figure 7.  Mean monthly precipitation (mm; CHIRPS) compared to ERAI (a–c) and ERA5 (d–f) over the three sub-regions between 1981 and 2018. Annual cycle of mean monthly precipitation (g–i) averaged over the respective sub-regions according to observations (black line), ERAI (dashed blue line), and ERA5 (solid blue line).

Figure 8.  Mean monthly 2-meter temperature (in °C; CRU) compared to ERAI (a–d) and ERA5 (d–f) over the three sub-regions between 1981 and 2018. (columns). Annual cycle of mean monthly 2-meter temperature (g–i) averaged over the respective sub-regions according to observations (black line), ERAI (dashed red line), and ERA5 (solid red line)

The scatter plots for monthly mean T2m for the three sub-regions in Fig. 8 show a similar pattern as precipitation, closer to observations for ERA5 than ERAI. The best fit for the monthly mean T2m is seen over the Sahel. Also shown in Fig. 8 (third row) is the annual cycle of T2m from CRU, ERA5, and ERAI. CRU has the maximum temperatures occurring around pre-monsoon (March–May) and in the winter months (October–November) and the lowest temperatures during the peak of the monsoon (July–August). The lower temperatures during July/August are attributed to the reduction of solar radiation due to an increase in cloud cover coupled with convective activities. Both reanalysis products reproduce the observed annual cycle well, even though they show cold biases over the Guinea and Savannah domains. In these two regions, this bias is reduced more in ERA5 as compared to ERAI.

Understanding climate extremes is very important because they clearly show the inherent unique characteristics of nonlinear systems and the extent of the potential impact on socioeconomic activities. Figure 9 shows the spatial pattern of the climatology of extreme indices together with the corresponding biases in the two reanalysis datasets over West Africa. Observations show higher extreme precipitation (e.g., prcptot) and a smaller number of cdd along the Guinea Coast and less extreme precipitation and a greater number of cdd in the Sahel. The reanalysis reproduced this pattern, albeit with biases. Figure 9 (second and third row) suggests that the reanalysis datasets are better at reproducing cdd compared to cwd. Both reanalyses systematically overestimate the climatology of cwd and this is associated with the overestimation of light rain, known to be a common feature of weather and climate models (e.g., Dai and Trenberth, 2004; Stephens et al., 2010; Sillmann et al., 2013). The overestimation of the frequency of consecutive wet days over Savannah and most of the Guinea Coast suggests that the underestimation of extreme precipitation does not occur because of having a smaller number of rainy days; rather, it is due to an underestimation of the intensity. Over the southeastern part of West Africa, the biases of ERA5 are smaller than those of ERAI regarding prcptot, rx5day, and r10mm, especially along the regions located in the southeast belt. Nevertheless, ERA5 depicts less of a dry bias in cdd. Notably, ERA5 overestimates the fraction of the 95th percentile to annual total extreme precipitation (r95ptot). Notwithstanding, the dry bias in ERA5 over most of West Africa is significantly reduced compared with ERAI. For instance, the dry bias for r10mm in ERAI was as high as 120%, but it was less than 60% in ERA5 over West Africa.

Figure 9.  Climatology of selected precipitation-based climate indices for CHIRPS (column one) and spatial bias for ERAI (column two), and ERA5 (column three) relative to CHIRPS. Stippling denotes areas where the differences are significant at the 0.01 level.

Analysis of the linear relationship between observation and reanalysis for the six chosen precipitation-based indices (prcptot, cdd, cwd, rx5day, r10mm, and r95ptot) are shown in Fig. 10, where the correlation of the frequency indices over West Africa was computed over a 38-year period (1981–2018). In general, the relationship is stronger in ERA5 compared to ERAI. For example, rainfall contributions recorded from rain days in the ERA5 dataset are in strong agreement with the observations, especially around parts of the Guinea Coast and across the middle belt of the region covering parts of the Savannah and Sahel, while ERAI shows a negative correlation around the eastern parts of the region (Fig. 10; first row). The frequency of dry and wet days (cdd and cwd) also shows good agreement in both reanalysis products. Other indices do not show a clear relationship pattern between the reanalyses and observation, although noticeable improvements in heavy rainfall events exist in ERA5 (Fig. 10; fifth row, R10mm) for larger areas to the south of the region.

Figure 10.  Correlation of selected precipitation-based climate extreme indices for ERAI (column one) and ERA5 (column two) relative to CHIRPS. Stippling denotes areas where correlation coefficients are significant at the 0.01 level.

Figure 11 illustrates the spatial pattern of the linear trend for extreme precipitation indices from observation and reanalyses in West Africa based on the Z and Sen tests. Shaded areas represent the Sen's slope, and stipplings correspond to points where trends are statistically significant at the 95% level. The spatial distribution for the trend of the observed prcptot (Fig. 11; top row) showed a general, slight, non-significant, increasing trend across West Africa, with a mean magnitude of 1.08 mm yr^–1. However, statistically significant negative trends are found for the reanalyses based prcptot over the eastern and western regions. Relative to CHIRPS with trend of 1.08 mm yr^–1, the mean trend in prcptot decreases for both ERAI (–4.24 mm yr^–1) and ERA5 (–0.73 mm yr^–1), although at less magnitude in ERA5 (Figs. 11b-c; top row).

Figure 11.  Spatial pattern of trend in prcptot (first row), cdd (second row), cwd (third row), rx5d (fourth row), r10mm (fifth row) and r95ptot (sixth row) from CHIRPS (column a), ERA-Interim (column b) and ERA5 (column c) for the period 1981–2018. Dots overlaid in the plots represent grid points where trends are statistically significant at the 0.05 level. Also shown is the domain-averaged value of the trend in change yr^–1.

The CHIRPS cdd shows increasing trends over West Africa from 1981 to 2018 (Fig. 11; second row). Both ERAI and ERA5 display mixed signal trends and agree less with the observation. Over the study region, the number of cwds decreases slightly for the CHIRPS observation with a mean value of –0.02 days yr^–1. Both ERAI and ERA5 overestimated the declining trend, at more significant mean values of –0.15 for ERA5 and –0.33 days yr^–1 for ERAI, respectively (Figs. 11b-c; third row). The declining trend in cwd, and prcptot in reanalysis datasets is in line with the decreasing trend of the seasonal mean precipitation discussed above.

The results for the r$ x $5day index show a similar increasing spatial trend for CHIRPS observation and ERA5 reanalysis (Figs. 11a, c; fourth row). ERAI did not capture this increasing trend but rather showed a strong and significant drying trend across the region (Fig. 11b; fourth row).

For the extreme precipitation index, r95ptot, the CHIRPS observation showed no trend. In contrast, there is no agreement between observations (CHIRPS) and reanalyses for the r10mm index (Fig. 11; fifth row). While ERA5 had a positive significant trend with a mean value of 0.27% yr^–1 (Fig. 11c; fifth row), ERAI had a negative significant trend with a mean value of –0.15% yr^–1 (Fig. 11b; fifth row).

The purpose of the study is to evaluate the performance of the ERA5 and ERAI reanalyses in terms of representing the spatial patterns, and temporal variability of both mean and extreme climate conditions of West Africa for the period of 1981−2018. Our findings show that ERA5 yields a much better representation of temperature and precipitation than ERAI over Western Africa and its sub-regions (i.e., the Guinea Coast, Savannah, and Sahel). The ERA5 datasets were typically closer to the observations than ERAI, especially in reproducing precipitation in the Savannah and Guinea Coast and the temperature in the Sahel, which supports the results of investigations of other aspects of the datasets (Betts et al., 2019; Gleixner et al., 2020; Martens et al., 2020; Quagraine et al., 2020; Tarek et al., 2020). For domain mean precipitation and temperature, the ERA5 reanalysis dataset performs better due to its relatively small bias and improved correlation with observation data.

ERA5 provided a considerably better representation of the observed temperature trend than ERAI, which even indicated cooling tendencies in certain areas in West Africa. These locations are also characterized by wetting tendencies, which implies that the cooling resulted from increased cloud cover. In general, reanalysis data are deemed inappropriate for trend analysis (Trenberth et al., 2008; Gleixner et al., 2020). Analysis of precipitation trends indicated that, regardless of ERA5’s somewhat superior performance in capturing the observed trend, there are some regions where both reanalysis fields underperformed. Both reanalyses were unable to reproduce observed precipitation trends throughout the majority of West Africa. This was not unexpected because precipitation patterns are uncertain for most parts of the globe. These results highlight that reanalysis precipitation fields are not appropriate for applications where an accurate representation of linear trends in the precipitation field is required.

Moving from ERAI to ERA5, the reproduction of the observed precipitation cycle considerably improves in the three sub-regions of western Africa by eliminating much of the precipitation bias in the regionally averaged rainfall data. Unlike ERAI, the variability and magnitude were well-captured by the ERA5. Analysis at the sub-region level showed the strongest match between ERA5 and observations in West Africa was in the Sahel for T2m temperature and the Savannah for precipitation. In an ideal environment, a comparison would have been made between reanalysis data and station data; however, this is challenging in Africa due to the scarcity and inaccessibility of meteorological stations as well as the strict data-sharing policies among the National Meteorological Agencies of the respective countries; therefore, the observational products used are inherently uncertain. Although ERA5 is considerably closer than ERAI to observations, there is still room for improvement. For example, more effort could be put into increasing the resolution, as 0.25° is still considered too coarse for highly localized research and impact modeling.

Several extreme precipitation indices describing the intensity and frequency of heavy precipitation events have been computed to assess their representation of spatial and temporal variability in the two reanalysis datasets. It is found that both reanalyses underestimate the intensity and frequency of extreme heavy precipitation events. This occurs over most of the domain except over patches of the coastal areas of West Africa and the northeastern part of the domain, as reflected by the wet day precipitation total and frequency of precipitation greater than 10 mm. It is interesting to note that the frequency of consecutive wet days is actually overestimated over the Savannah and most of the Guinea coast, suggesting that the underestimation of extreme precipitation is not due to a smaller number of rainy days; rather, it is due to an underestimation of the intensity.

Results from trend analysis on selected extreme precipitation demonstrated that ERA5 has improved upon some of the ERA-Interim reanalysis weaknesses. This is more apparent in r5xday and prcptot. In these indices, while observations show no significant trend over the Sahel region, a significant drying trend is noted in ERAI dataset over central and eastern parts of West Africa. However, ERA5 shows either a neutral or slightly increasing trend.

The overall remarkable performance of ERA5 in representing key climatic features over West Africa, together with its accessibility at a higher temporal and spatial resolution, makes it a suitable dataset for near-real-time climate monitoring, particularly for the temperature over the region. It is also attractive for climate impact applications such as agricultural management and hydrology, where reliable high temporal and spatial resolution climate information is needed.

Our results show the strengths and shortcomings of ERAI and ERA5 over West Africa in terms of mean and extreme characteristics. Assessing the link between the variability of extreme climate with local and large-scale modes of variability would be an interesting aspect for future work to better understand causal relationships.

Acknowledgements. The ERA5 and ERA-Interim datasets were obtained from the European Center for Medium-Range Weather Forecast. The CRU and CHIRPS datasets were provided by the University of East Anglia’s Climate Research Unit and the US Geological Survey and the University of California, Santa Barbara, respectively. GTD is currently affiliated with Environment and Climate Change Canada.