Seasonal mean climatology
The seasonal mean 2-m temperature from CRU gridded analysis, as well as ERA-Interim reanalyses and CRCM5 simulation during the periods January–February–March (JFM) and July–August-September (JAS) are shown in Fig. 2, along with the simulation bias (CRCM5 minus Observations). The periods JFM and JAS will henceforth be referred to as the austral and boreal summers, since these periods match best with the migration of the tropical rainbelt over tropical Africa. The simulation reproduces the overall features of geographical and seasonal temperature distribution. Compared to CRU, there is however a cold bias in the CRCM5 simulation in JFM (Fig. 2a) over most of the domain south of the equator, as well over the West Africa region and the Saudi Arabia peninsula. At the same time, there is a warm bias near the border between the Central African Republic and the Democratic Republic (DR) of Congo. The cold bias appears larger compared to UDEL dataset but is smaller when compared to ERA. The warm bias in Democratic Republic (DR) of Congo is on the contrary larger when compared to ERA-Interim.
In JAS (Fig. 2b), CRCM5 exhibit a cold bias over the East Africa Highlands, the elevated terrains of Ethiopia and Sudan, eastern Madagascar, the NW of Saudi Arabia, and over the coastal countries of the Gulf of Guinea when compared to CRU and UDEL datasets. On the other hand, the model is warmer than CRU in regions such as the Sahara, the apparent warm bias reaching up to 2° in regions such as the southern Congo basin and Oman. In fact, as can be seen when comparing with ERA-Interim reanalyses or UDEL dataset, the CRCM5 simulation has a weak cold bias over the Sahara rather than a warm bias as would imply the CRU data. We hypothesise that due to the paucity of observations over the Sahara the CRU gridded analysis misses the hottest pool, with temperatures exceeding 36° on average over northern Mali and southern Algeria according to ERA-Interim and UDEL.
In conclusion, albeit with some biases, the CRCM5 simulation succeeds in reproducing the detailed geographical distribution and seasonal variations of temperatures. As a matter of fact Sylla et al. (2010), in their simulation of the African climate using a regional climate model, found that a temperature bias of ±2° is in the range of state-of-the-art regional climate models. What constitutes the best reference dataset for simulation validation remains an open issue in regions with poor observational coverage.
CRCM5-simulated precipitation fields for JFM and JAS are compared to those from GPCP, TRMM, CRU and UDEL datasets in Fig. 3. As most of the African continent lies within the tropics, the seasonal migration of the tropical rainbelt that regulates the alternation of wet and dry seasons, is the principal characteristic of precipitation over the continent. Furthermore, small shifts in the position of the rainbelt can result in large local changes in precipitation; this has a direct impact upon water resources and agriculture in the semiarid regions of the continent, such as the Sahel. There are also regions on the northern and southern limits of the continent with winter rainfall regimes that are governed by the passage of mid-latitude fronts. The wettest regions in Africa are those of the equatorial, tropical rainforest climate type, where there is rain throughout the year, with two peak periods corresponding to the double passage of the tropical rainbelt. The driest regions are those of the desert climate type, such as those of the Sahara, Kalahari and Somali deserts, where there is very little precipitation.
During the period of high sun in austral Africa (JFM), the tropical rainbelt is at its southernmost position (Fig. 3a). The bulk of precipitation over the continent is thus mainly circumscribed to the regions south of the equator. Rainfall peaks of 8–12 mm/day are found in Mozambique, Malawi, Zambia and Gabon, but the most intense are found in the east and northwest coasts of Madagascar with more than 20 mm/day. The model simulation compares well with observational datasets. In fact for some regions it can be argued that the model-simulated precipitation may be more realistic than the 1° GPCP and even the 0.5° gridded datasets; this is the case for the Gabon, DR of Congo and Madagascar regions. This may be the result of the CRCM5 higher resolution that allows a fairly realistic description of physiographic effects such as lowlands, escarpments and plateaus that characterise these regions. For example, according to the Historical and Meteorological Data Center (HMDC) for Madagascar (NOAA Central Library, http://docs.lib.noaa.gov/rescue/data_rescue_madagascar.html), the lowlands along the southern and western coasts of Madagascar receive 400–800 mm of annual rainfall, while most of the mountainous east coast, as well as a small region in the northwest, have an annual rainfall over 2,000 mm. Precipitation in the central plateau has intermediate values: 1,000–1,500 mm according to HMDC. Being exposed to the moisture-laden trade winds, the east coast is wet for much of the year, while the rest of the island receives most of the rainfall between November and March (HMDC). CRCM5 and TRMM reproduce this pattern while the coarser mesh GPCP analysis misses it. GPCP also misses the rainfall peak in the highlands of Gabon and the DR of Congo. It must be mentioned also that in the southeast part of the domain, near the boundary, there is an overestimation of rainfall over the ocean, especially in this season. The reasons for this bias are unknown at the moment, but we hypothesize as does Nikulin et al. (2012) that this may be related to the representation of convection in the model boundary relaxation zone.
In AMJ (not shown), the tropical rainbelt begins its northward displacement and is located near the equator. This is the beginning of the rainy season for the Guinean coastal region and the first rainfall onset of the West African Monsoon (WAM). Rainfall peaks are still present over the east coast of Madagascar and the highlands regions of Gabon. The CRCM5 simulates rainfall peaks of 8–12 mm/day in the Guinea coastal region, which do not appear in GPCP.
During the boreal summer (JAS), the tropical rainbelt moves to its northernmost location (Fig. 3b) and so does the rainfall peak of the WAM, which is located around 10°N. This is the second rainfall onset of the WAM and the beginning of the rainy season for the Sahel. Precipitation is mostly confined between the equator and around 15°N; the width of the rain band is slightly narrower in the model than in all of the observation datasets, resulting in a dry bias in the Sahel and the DR of Congo as well as in the Gulf of Guinea. CRCM5 and GPCP as well as CRU and UDEL show 3 maxima of precipitation (from 8 to 12 mm/day and for some up to 24 mm/day) over the highlands of Guinea, near the Nigeria—Cameroon border and Ethiopia; the CRCM5 has higher values than the gridded analyses in those regions. TRMM shows an underestimation of rainfall in the Ethiopian Highlands region when compared to the other datasets as well as to CRCM5. In the southern hemisphere, away from the equator, the only African regions to receive non-negligible rain during this season are the east coast of Madagascar and some parts of Mozambique; in Madagascar the CRCM5 simulates a sharp line of precipitation with peak values higher than in GPCP and the other datasets, and in Mozambique all datasets fail to show any substantial precipitation at all.
In OND (not shown), the tropical rainbelt is on its journey southward and so does the spatial pattern of precipitation over the continent.
Annual cycle of precipitation
Figure 4 shows the mean annual cycle of precipitation for some of the African-CORDEX regions as displayed in Fig. 1b. For each region are shown the CRCM5 simulation, as well as the CRU and GPCP data sets for the period 1997–2008, and the TRMM dataset for the period 1998 to 2008. In the regions near the equator (Fig. 4b, d, e) there is double peak of rainfall due to the double passage of the tropical rainbelt. The simulations and the observations are very close for the Central Africa—North (Fig. 4d) and South (Fig. 4e) regions, although the bimodal character of precipitation is more visible in the simulation than in the observations in the case of the Central Africa—North region (Fig. 4d). Over the West Africa—South region (Fig. 4b), the observations exhibit two pronounced rainfall peaks: the coastal monsoon onset at the end of May, which is the first rainy season for this region, and the second one when the tropical rainbelt is in its southern migration towards its boreal winter location. The CRCM5 simulation agrees better with the observations in the magnitude and timing of the second peak of precipitation. For the first rainfall peak, the simulation overestimates the magnitude and the monsoon onset is somewhat too early.
The West Africa—North region (Fig. 4a), which includes the Sahel, is characterised by a single peak of precipitation occurring in August when the monsoon reaches its farthest northward position. The peak precipitation intensity is somewhat deficient in the simulation, although simulated precipitation is somewhat excessive in the April–May period.
Over the Ethiopian Highlands region (Fig. 4c), maximum precipitation occurs around July–August. The simulation is able to reproduce accurately the gradual rise of precipitation from January to July, but the simulated peak is overestimated and precipitation is excessive till September. In the East Africa Highlands (Fig. 4f) maximum precipitation occurs around January-March; in the simulation there is a little overestimation of rainfall during the dry season and before the time of the peak. Over the South Africa—East region (Fig. 4g), the annual cycle of precipitation is similar to that of the East Africa Highlands, but weaker; the simulation successfully captures the timing and magnitude of the precipitation cycle.
Diurnal cycle of precipitation
The diurnal cycle of precipitation is another good metric for assessing the skill of climate model simulations because of the multiple dynamical, thermal and radiative processes that are associated with it (Ploshay and Lau 2010). As stated by Yang and Slingo (2001): “The simulation of the amplitude and phase of the diurnal cycle provides an ideal test bed for model parameterizations and for the representation of the interactions between the surface, the boundary layer, and the free atmosphere.”
The main feature of the diurnal cycle of precipitation in the tropics is a late afternoon—evening maxima over land (Dai et al. 2007, Yang and Slingo 2001). Using the 3 hourly TRMM observational dataset as reference, the mean diurnal cycle of simulated precipitation for the period 1998–2008 were plotted for some of the Africa CORDEX regions shown in Fig. 1b, during the season of their maximum precipitation. For the regions with two rainy seasons, only one is shown but the other is commented upon if necessary. As can be seen from the TRMM data in Fig. 5 most regions have a late afternoon maximum of precipitation, around 18 GMT, while the West Africa ones (Fig. 5a, b) exhibit a later occurrence of their maximum, around 21 h GMT. This would be related to the fact that in West Africa most of the precipitation comes from mesoscale convective systems triggered by the orography and elevated daytime heating (Hodges and Thorncroft 1997; Yang and Slingo 2001) which are typically initiated between 17 and 18 h (Local Time) but for which the maximum of precipitation occurs at the mature stage, later in the night (McGarry and Reed 1978; Hodges and Thorncroft 1997; Yang and Slingo 2001; Nikulin et al. 2012).
Figure 5 shows that overall CRCM5-simulated diurnal cycle of precipitation agrees with TRMM observations in phase and amplitude, albeit with some differences from one region to another. The model does particularly well in the Central Africa North (Fig. 5d) and South (Fig. 5e) regions, although the timing of maximum rainfall in CRCM5 is somewhat earlier than observed. This difference in the time of maximum rainfall is still more pronounced in the first rainy season (AMJ) for the Central Africa—North region (not shown), in which case the simulation also gives higher rates of precipitation. Concerning the Central Africa—South region the diurnal cycle of both seasons (FMA, not shown and OND) is very similar. In the Ethiopian Highlands region (Fig. 5c) there is an overestimation of the rate of precipitation in the simulation. On the other hand, in the East Africa Highlands (Fig. 5f), there is a little overestimation of the rate of rainfall before the time of the maxima but an underestimation afterwards. This is also the case of the South Africa—East region (Fig. 5g). The timing of the simulated rainfall peak is too early for these two regions. In the West Africa South (Fig. 5b) region the peak of simulated rainfall occurs earlier than observed for the two rainy seasons (AMJ, not shown, and ASO), but the model is better in reproducing the diurnal cycle of the ASO peak although the magnitude of the rainfall peak is somewhat underestimated. On the other hand, the results for the West Africa North (Fig. 5a) region show a very good amplitude of the simulated diurnal cycle, but its timing is in advance by two to three hours.
Although the simulated diurnal cycle is not perfect, being always in advance for the timing of rainfall peak, we can say that CRCM5 reproduces the pattern of the observed diurnal cycle most of the time. In their comparison of several regional models over Africa, Nikulin et al. (2012) remarked that CRCM5 best reproduced the diurnal cycle of precipitation.
West African Monsoon climatology
A detailed analysis of the West Africa—North region is presented in this section. Figure 4a showed that CRCM5 captures the timing of the monsoon onset for the region but underestimates the magnitude of precipitation. This is clearly reflected in the Hovmöller-type diagrams presented in Fig. 6. This figure shows time-latitude cross-sections of daily precipitation from GPCP as well as from the CRCM5 simulation, averaged over 10°W–10°E for the period 1997–2008. A 31-day moving average has been applied to remove high-frequency variability. The simulation captures to some extent the evolution of the monsoonal rainfall. However in the simulation the first peak occurring around day 150 is excessive compared to that of GPCP and the northern limit of precipitation (as defined by the 1 mm/day contour) is located around 15°N while it is found at almost 20°N in GPCP; CRCM5 has thus a dry bias in the Sahel. Despite the model dry bias in the Sahel that is still present, the intensity of the rainfall peaks in the Hovmöller diagram matches better with those of the TRMM dataset shown in Fig. 6 of Nikulin et al. (2012).
Figure 7 (panels a and b) shows maps with a superposition of the 2-m temperature (colours), mean sea level pressure (MSLP, contours) and low-level wind vectors (arrows) during the boreal summer, for the ERA-Interim data (Fig. 7a) and CRCM5 simulation (Fig. 7b). The contrast between the hot Sahara and the colder equatorial Atlantic Ocean in this season can be clearly seen. The large-scale pressure gradient between the two regions gives rise to a southerly flow from the ocean to the land. The zone where the humid and cooler southwesterly (monsoonal) flow meets the dry and hot northeasterly (Harmattan) flow lies in the southern fringe of the West African Heat Low (WAHL), or Saharan Heat Low (SHL) as it is often named for its summer position over the Sahara region (e.g., Lavaysse et al. 2009). The surface position of this lower tropospheric convergence is called the Inter-Tropical Front (ITF) or Inter-Tropical Discontinuity (ITD; Lavaysse et al. 2009). In the reanalyses, it is located as far inland as 20–22°N.
The CRCM5 simulation reproduces the overall features present in the ERA-Interim analyses, but with some deficiencies. In the CRCM5 simulation, the maximum temperatures located in the Sahara (northern Mali and southern Algeria) are not high enough, but the secondary maximum located further southeast in Niger and Chad is somewhat too high. The resulting southward shift of the position of maximum surface temperature induces a southward bias in the position of the SHL, as can be seen in Fig. 7c, where the bias of mean sea-level pressure (MSLP) is shown. As a consequence, the monsoonal flow does not extend far enough to the north, and this could be the reason why the summer monsoonal rainfall in the Sahel region is much too weak in the simulation, as we shall see below.
In order to better analyse the WAM, Fig. 8 zooms on the region between 10°W–10°E and 10°S–35°N, over the period 1997–2008, as represented by ERA-Interim (top) and the CRCM5 simulation (bottom). The left panels present maps with a superposition of the fields of 2 m-temperature (colours), 925-hPa horizontal wind vectors (arrows) and precipitation (contours); the right panels present vertical cross-sections with a superposition of vertical velocity (mm/s) plotted in colours and arrows displaying vectors combining vertical (mm/s) and meridional (m/s) velocity components, and a line curve showing the GPCP precipitation intensity. Such composite diagram allows grasping at a glance some of the key dynamical processes associated with the WAM and to evaluate the skill of the model in simulating the WAM.
The most striking feature when looking to the left panels of Fig. 8 is the difference in surface temperature between the model and reanalysis in this region. Compared to the analyses the simulation shows a weaker surface temperature gradient and a narrower precipitation belt that is located further south. Also noticeable is the fact that the monsoonal winds barely reach 15°N in the simulation whereas in the reanalysis they reach up to 20°N; not surprisingly, the northern precipitation isoline of 1 mm/day has a similar southward displacement bias in the simulation.
As noted before, in the simulation the Sahara is not warm enough and the warmest part is located too far south and east. As a consequence, the heat low that develops in summer over this region is displaced to the south. This implies a southward displacement of the zone of convergence of the humid and cooler south-westerly monsoonal flow with the hot and dry north-easterly Harmattan flow. As a result, the vertical ascent associated with the low-level convergence of the Harmattan and monsoonal winds is located too far south in the CRCM5 simulation with respect to the reanalysis, as can be seen in the right panel of Fig. 8. This region of shallow ascent corresponds to the shallow meridional circulation (SMC) cell from the surface to the mid-troposphere (Nicholson 2009; Thorncroft et al. 2011) associated to the Saharan Heat Low. As can be seen in Fig. 8, the maximum ascent is located north of 20°N in the reanalysis while in the simulation it is found farther south (between 15°N and 20°N).
On the other hand, the deep meridional circulation associated with the Hadley cell (Nicholson 2009; Thorncroft et al. 2011) is represented by a column of rising air centred around 10°N, but it is narrower in the simulation than in the reanalysis. This region of strong ascent is responsible for the bulk of precipitation; this is the tropical rainbelt (Nicholson 2008) and maximum rainfall coincides with maximum ascent. The dry bias of the CRCM5 simulations in the Sahel is in accordance with the southward bias in the position of the northern edge of this column of rising air, as well as the southward bias in the location of the SMC of the WAHL. However, the region of maximum ascent and rainfall in CRCM5 is collocated with that in ERA-Interim.
Finally, there is also a region of shallow ascent located around 5°N which corresponds to the coastal meridional sea-breeze circulation. This feature is less developed in the simulation than in the reanalyses, and consistently with this, the precipitation amounts simulated by the model are somewhat lower than those from GPCP at this latitude.
As we saw earlier the summer cold bias of CRCM5 over the Sahara and the southern bias of the maximum surface temperature have important implications for the simulated circulations of the WAM. This also has an impact upon the vertical distribution of winds over the West Africa region, as we shall see next. Figure 9 shows a vertical cross-section of the mean zonal wind in boreal summer (JAS) for the region 10°S–35°N and 10°W–10°E, top panel for ERA-Interim and bottom panel for the CRCM5 simulation. Between the surface and 850 hPa, the monsoonal and Harmattan flows appear as westerly and easterly components, respectively. At the surface, the transition occurs at 20°N in the reanalyses but further south, around 17°N, in the simulation.
Above this, in the mid-troposphere and centred around 600 hPa and 14°N, there is the African Easterly Jet (AEJ). The AEJ is a key element of the WAM circulation. It results from the strong baroclinicity between the hot and dry Sahara and the colder and humid Guinea coast, through the thermal wind relationship (Burpee 1972; Cook 1999; Thorncroft and Blackburn 1999; Parker et al. 2005a, b). There is a direct link between the divergent circulation around 600–700 hPa associated with the SMC of the SHL and the existence of the AEJ (Thorncroft and Blackburn 1999; Parker et al. 2005a). When compared to ERA-Interim, the core of the AEJ is of the right strength and almost the right height, but it is displayed southward by about 2°. At the same time, the shape of the zone of monsoonal winds is not exactly the same, with the maximum westerlies being overestimated in the model by about 2 m/s and being displaced northward: it is located at 6°N in ERA and between 5°N and 11°N in the model.
Further above, in the upper troposphere, lies the Tropical Easterly Jet (TEJ) at 200 hPa and 6°N. Compared to ERA-Interim, the core of the TEJ is slightly underestimated, by about 2 m/s; its position is correct but the shape is not exactly the same as in the reanalyses. Although in general there is an acceptable agreement between the model simulation and ERA-Interim, there are some differences in the configuration of wind distribution.
Nicholson (2008, 2009) noted that the column of rising motion associated with the tropical rainbelt is bounded to the south by the axis of the upper troposphere TEJ and to the north by the axis of the mid-troposphere AEJ. They also noted that the TEJ, which is the outflow of the Asian Monsoon, exhibits little latitudinal variations from year to year over Africa. The AEJ on the other hand can migrate south or north; a northern position of the AEJ means wet summers in the Sahel (wider tropical rainbelt) and a southern position of the AEJ implies dry summers in this region (narrower tropical rainbelt). The CRCM5 simulation appears to follow the empirical relationship established by Nicholson (2009) between the column of rising motion and the axis of the two jets. In the simulation, the axis of the AEJ is positioned close to the northern limit of the column of maximum rising motion. The model southward bias for the location of the AEJ (as is also the SHL), combined to the weakness of the TEJ, is in accordance with a deficit of rainfall in the Sahel, as noted by Nicholson (2008, 2009) in her analysis of observations. The climatology of the WAM as simulated by CRCM5 resembles the “dry Sahel” mode of the interannual variability of rainfall identified by Nicholson (2009), with a narrower column of rising motion and a southern position of the AEJ.
Interestingly enough the simulation performed by Sylla et al. (2010) using another RCM over a similar domain and at similar resolution, produced biases of opposite sign to our results. In their simulation the strength of the monsoon flow was overestimated and the location of the AEJ core was too far north compared with the reanalysis. They associated the stronger monsoon flow and northward shift of the AEJ to an overestimation of the low-level temperature gradients.