Before analysing climate projections, it is paramount to first study the performance of the models in simulating current climate, as model projections cannot be credible if models are unable to reproduce skilfully the current climate. In this section we focus on the results of simulations under current climate, corresponding to the orange section of the flowchart shown in Fig. 2. The period (1981–2010) will be considered as the reference or control period for the climate-change projections.
Seasonal mean climatology
As most of the African continent lies within the tropics, the seasonal migration of the tropical rainbelt, which regulates the alternation of wet and dry seasons, is an important characteristic of the climate. In boreal winter, the tropical rainbelt is at its southernmost position and the precipitation maximum lies mostly south of the equator. To the north, the continent is under the influence of the dry northeasterly winds. In boreal summer, the tropical rainbelt is at its northernmost location and the precipitation in confined essentially in a band north of the equator, reaching up to 15°N in the Sahel; in this season, the humid southwesterly monsoonal winds penetrate further north into the continent. These changes in circulation and precipitation are induced by the seasonal variations of the surface temperature contrast between land and ocean. In boreal summer the Sahara is very hot and the east equatorial Atlantic Ocean is colder than in boreal winter. The large-scale pressure gradient between the hot Sahara and the colder equatorial Atlantic Ocean brings the southerly flow from the ocean to the land and the West African Monsoon (WAM) precipitation into the Sahel. In fact, in boreal summer the two key players of the northern migration of the WAM rainfall are the thermal low located over the Sahara (the Saharan Heat Low, SHL), and the cold tongue present in the east equatorial Atlantic Ocean (e.g., Thorncroft et al. 2010; Lafore et al. 2010; Nguyen et al. 2011).
Figure 3 shows the SST bias of the two GCMs for current climate (1989–2008), in boreal winter (JFM) and boreal summer (JAS). The most striking feature is the warm bias in the Guinea Gulf and the tropical Atlantic Ocean along the West Coast of the continent southward of the equator in JAS; in fact in both GCMs this bias extends quite far off the African coast (not shown), which certainly introduces biases in the atmospheric circulation serving as LBC to CRCM5. Both models fail to represent properly the cold tongue that develops in boreal spring (AMJ, not shown) and reaches its maximum intensity in JAS. As it is the development of the cold tongue in spring and its reinforcement in summer that, combined with the evolution of the
SHL and its associated Shallow Meridional Circulation (SMC), brings the WAM rainfall from the equator to the Guinea Coast region in spring and later over the Sahel region in summer, this SST bias will have repercussions on the GCM-driven CRCM5 simulations of the WAM, as will be shown later.
In order to capture at a glance the performances of the models in current climate we show results for the JFM and JAS seasonal-mean 2-m air temperature for the 20-year CORDEX period 1989–2008 (Fig. 4); for precipitation (Fig. 5), a shorter period 1997–2008 is used due to limitations in the availability of observational datasets. In both figures, top left panels show one source of observation used as reference; all other panels except rows three and five show biases as departure from this reference. Top right panels show the structural bias of the ERA-driven CRCM5 simulation (RCM_SB) that was extensively analysed in HD12. The second row shows the GCM-driven RCM bias resulting from the combined effects of CRCM5 structural bias and the lateral and lower boundary conditions (RCM_SB_BC) when driven by CanESM2 (left) and MPI (right). The third row shows difference between the GCM-driven and ERA-driven CRCM5 simulations, reflecting the impact of lateral and lower boundary errors on the CRCM5 simulation (BC_E) when driven by CanESM2 (left) and MPI (right); this field is in fact equal to the difference between RCM_SB_BC and RCM_SB. The fourth row shows the bias of the driving GCMs (GCM_SB), CanESM2 on the left and MPI on the right. Finally the fifth row shows the difference between the GCM-driven CRCM5 and the driving GCM simulations (RCM_AV_p), which kind of constitutes the CRCM5 added value under current climate, again when driven by CanESM2 (left) and MPI (right); this field is in fact equal to the difference between RCM_SB_BC and GCM_SB. For the temperature (Fig. 4) only the third and fifth rows show the differences over the oceans.
Compared to CRU analysed 2-m temperatures, there is a slight cold bias in the ERA-driven CRCM5 simulation in JFM (Fig. 4a: RCM_SB) over most of the domain south of the equator, as well over a part of the West Africa region. On the other hand, there is a warm bias near the border between the Central African Republic and the Democratic Republic (DR) of Congo. The MPI-driven CRCM5 simulation exhibits a stronger cold bias that occupies a larger region (RCM_SB_BC: MPI), as confirmed by the panel BC_E: MPI showing the difference between the MPI-driven and the ERA-driven CRCM5 simulations. The MPI GCM exhibits a strong cold bias over the Guinea Coast states and a warm bias in a narrow band along the southwest coast of Africa (GCM_SB: MPI); the CRCM5 is almost neutral at correcting the cold bias, but succeeds at correcting part of the warm bias, as can be seen in RCM_AV_p: MPI. On the other hand, the ERA-driven CRCM5 weak cold bias over the Sahara combines with the equally weak cold bias of the MPI GCM there, and results somehow in a stronger cold bias in MPI-driven CRCM5. Compared to the ERA-driven simulation the CanESM2-driven CRCM5 simulation exhibits stronger cold bias over most of West Africa but a reduced cold bias south of the equator, and in fact almost no bias over Botswana, Zimbabwe and Zambia, which is better than ERA-driven simulation, possibly because BC_E: CanESM2 cancels the RCM_SB.
In JAS (Fig. 4b), ERA-driven CRCM5 exhibits a cold bias over the East Africa Highlands, the elevated terrains of Ethiopia and Sudan, eastern Madagascar, and over the coastal countries of the Gulf of Guinea when compared to CRU (and UDEL, not shown). On the other hand, the model is slightly warmer than CRU in regions such as the Sahara, the apparent warm bias reaching up to 2 °C in regions such as the southern Congo basin and Oman. It is worth mentioning however that 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 noted in the comparison with CRU data (see the related discussion in HD12). Both GCMs exhibit a generalised warm bias over the continent (GCM_SB: Can and GCM_SB: MPI). With the exception of the cold bias in the Guinea Gulf Coast region which is peculiar to the CRCM5 and that is still present when driven by the two GCMs, everywhere else the GCMs warm bias is partly corrected by the CRCM5, as can be seen in RCM_AV_p.
The skill of the models in simulating seasonal mean precipitation, using GPCP as reference, is presented in Fig. 5 (a: JFM; b: JAS). In JFM the ERA-driven CRCM5 simulation has a rather small bias. Precipitation over the equatorial Atlantic Ocean is overestimated by both GCMs and the maximum is displaced southward, as seen in GCM_SB. When driven by GCMs, CRCM5 improves the tropical rainbelt compared to the driving GCMs, as seen in RCM_AV_p that has overall the opposite sign of GCM_SB. On the other hand, the ERA-driven CRCM5 dry bias in the Gulf of Guinea coast remains in all GCM-driven CRCM5 simulations, and is in fact amplified although only MPI has also a dry bias there. There is a wet bias east of Madagascar in the ERA-driven CRCM5 simulation (RCM_SB) and a general dry bias over Madagascar in both GCMs; the wet bias is in fact intensified in the GCM-driven CRCM5 simulations, as confirmed in BC_E.
In JAS, the ERA-driven CRCM5 has a dry bias in the Sahel region and in the DR of Congo, as well as in a small region of the Guinea Gulf (RCM_SB). A dry bias in the Sahel is also present to some extent in the MPI simulation (GCM_SB: MPI). The CRCM5 dry bias remains when driven by MPI (RCM_SB_BC), and is amplified when driven by CanESM2 as confirmed by the difference field BC_E: CanESM2. Both GCMs exhibit a wet bias over the Guinea Gulf (GCM_SB), larger in MPI. This wet bias is associated to the GCM-simulated SST biases. In both GCM-driven CRCM5 simulations, the GCM wet bias is further propagated to the Guinea Gulf Coast states. On the other hand the CRCM5 reduces the wet bias over the ocean, as seen in RCM_AV_p that has the opposite sign to GCM_SB. CRCM5 also corrects the MPI dry bias in the South Sudan (GCM_SB: MPI), as can be seen from RCM_AV_p: MPI and RCM_SB_BC: MPI. Interesting, the CanESM2-driven CRCM5 has a dry bias in this region (RCM_SB_BC: Can) while its driving GCM has a wet bias (GCM_SB: Can).
Annual cycle of precipitation
Figure 6 displays the mean annual cycle of precipitation for some of the African-CORDEX regions shown in Fig. 1b. The observed annual cycle from three different datasets (CRU, GPCP and
as well as from the ERA-driven CRCM5 simulation, are shown in addition to the CanESM2- and MPI-driven CRCM5 simulations, as well as CanESM2 and MPI simulations. The equatorial regions (CA-NH, CA-SH and WA-S) have two rainy seasons because of the double passage of the tropical rainbelt. This is well accounted for by all models in CA-SH and CA-NH, although with some errors in timing and intensity; for these regions, the GCM-driven CRCM5 simulations are better than the driving GCM. For the WA-S region, only the ERA-driven CRCM5 simulation succeeds in representing the bimodal distribution of precipitation with two maxima, in AMJ and ASO. Both GCMs, as well as the GCM-driven CRCM5 simulations, miss this feature and show instead a single maximum of precipitation in JAS. Both GCMs overestimate the rainfall amounts in the WA-S region, and this wet bias is amplified in the GCM-driven CRCM5 simulations. In all the other regions however the GCM-driven CRCM5 simulations improve upon the GCM simulated annual cycle. An exception is the region of the Sahel (WA-N) where MPI and CanESM2 show a better representation of the annual cycle compared to CRCM5; this however appears to be due to a compensation of errors resulting from SST biases and coarse resolution.
Diurnal cycle of precipitation
The diurnal cycles in CRCM5 simulations driven by ERA, CanESM2 and MPI, are compared to that of the TRMM dataset for the period (1998–2008) in the same regions (Fig. 7). It is well known that maximum precipitation in tropical climates tends to occur in the evening or overnight rather than in the afternoon as in mid-latitudes. The proper simulation of the diurnal cycle over regions of the African continent is one of the most difficult tasks for the climate models (e.g., Nikulin et al. 2012). The CRCM5 simulates fairly well the diurnal cycle of precipitation in most of the regions, with some tendency for precipitation maximum somewhat early rather independently of the driving data except for two equatorial regions: CA-NH and CA-SH. The MPI-driven CRCM5 simulation does particularly well in these two equatorial regions, CA-NH and CA-SH, but overestimates the precipitation amounts in WA-S. In the WA-N region, the MPI-driven CRCM5 simulation does better than the CanESM2-driven one, and this is also true for the SA-E region. In the EH region the CanESM2-driven CRCM5 simulation performs better, although the shape of the cycle is displaced to an earlier rainfall peak. Finally, in the EA region the simulated diurnal cycles are very close.
Daily precipitation intensity distributions
Figure 8 shows the distribution, by range of intensities, of the contributions to daily precipitation amounts (hereinafter called the daily precipitation intensity distribution, DPID), from the three CRCM5 simulations and the two GCMs simulations, as well as from the TRMM observational dataset for the period (2001–2008) for the same regions of Figs. 6 and 7. These figures in fact should be drawn as histograms, but are shown as curves for ease of comparison between several datasets. These figures are designed such that the sum of all the bins gives the daily average precipitation. For example, in WA-N in JAS (Fig. 8a) the MPI model (in dashed cyan) has a value of 2 mm/day (Y axis) in the range of daily precipitation intensities from 8 to 16 mm/day (X axis); this corresponds to 184 mm, i.e. 2 mm/day multiplied by 92 days for JAS. The JAS mean precipitation amount is 487.6 mm, that is the product of the number of days (92) times the average daily precipitation (5.3 mm/day) shown in the upper left corner of the graphic. In other words, the range of daily precipitation intensities from 8 to 16 mm/day contributes for a fraction of 2/5.3 to the total precipitation.
For the WA-N region (Fig. 8a) although CanESM2 and MPI succeed in reproducing the average precipitation of the observations (5.3 mm/day), the DPID are very different, CanESM2 even exhibiting a bimodal distribution that is inexistent in the observations. In MPI and all the CRCM5 simulations, there is displacement of the maximum contribution toward smaller daily intensities (8–16 mm/day) with respect to observations (16–32 mm/day). Overall MPI shows a better distribution than CanESM2. The CRCM5 simulations have similar shapes to the observed distribution although with smaller average precipitation amounts. We recall that we saw in Fig. 6a that CanESM2 and MPI showed the best match with the observed annual cycle of precipitation in WA-N; their DPID however differ very much from the observations.
For the WA-S region (Fig. 8b) the ERA- and CanESM2-driven CRCM5 simulations have the best distributions. In fact, the DPID of CRCM5 is better than that of its driving model. The improvement is less remarkable in the case of the MPI-driven CRCM5 with respect to MPI. For the other two equatorial regions (CA-NH and CA-SH) (Fig. 8d, e) all the CRCM5 simulations (ERA- and GCM-driven) have similar DPID and best match the observed DPID. In these regions also the CRCM5 considerably improves upon the corresponding driving GCMs DPID. The same is true for the SA-E region (Fig. 8g). In this case however the CanESM2-driven CRCM5 shows a somewhat flattened distribution.
The DPID of the two GCM-driven and the ERA-driven CRCM5 simulations are very similar in the EA region (Fig. 8f), improving upon that of the corresponding driving GCMs. The situation is somehow different in the EH region (Fig. 8c) where although CanESM2-driven CRCM5 still improves upon CanESM2, the peak of the distribution is overestimated with
respect to that of the observations, but less than those of the very similar ERA-driven and MPI-driven CRCM5 simulations. It is worth to note that in this region the TRMM showed smaller amounts of precipitation than the other observational datasets (GPCP and CRU) as was seen in Fig. 6c.
Finally a recurrent feature of the CanESM2 in all regions is a bias of the precipitation distribution towards lower intensities and even occasionally a bimodal distribution that is not found in the observations; the CanESM2-driven CRCM5 simulation is able to correct this deficiency.
West African monsoon
As the WAM is one of the most important elements of the African climate, this section is dedicated to the analysis of the performance of the climate models employed in this study in simulating the WAM. Hovmöller-type diagrams presented in Fig. 9 show time-latitude cross-sections of daily precipitation, averaged over 10°W–10°E for the period 1997–2008, from GPCP data, as well as from the CRCM5 simulations driven by ERA, CanESM2 and MPI, and from the CanESM2 and MPI simulations. A 31-day moving average has been applied to remove high-frequency variability. It can be noted that only the ERA-driven CRCM5 simulation succeeds in reproducing the seasonal migration of the WAM precipitation, with a first peak in the Guinea Coast in May (around 5°N) and a second one in the Sahel region (around 10°N) in August. The ERA-driven CRCM5 simulation also reproduces the second rainy season of the Guinea Coast region in September. The two GCMs fail to reproduce this pattern as they show essentially only the maximum around 10°N. The two GCM-driven CRCM5 simulations suffer from the same handicap as their driving GCM, producing only one rainy season from the coast up to 10°N and with higher intensities, the MPI-driven simulation being the most intense. The GCM precipitation peak at 5°N occurs much later in the season than in the observations; given the coarse mesh of the GCM, this peak is probably simply an extension of nearby ocean precipitation rather than the beginning of the continental rainy season. In other words, the bimodality of the rainy season in the Gulf of Guinea coast, present in the ERA-driven CRCM5 simulation, disappears when driven by the GCMs and is also absent in the GCMs. This is related to the misrepresentation of the seasonal cycle of the GCM-simulated SSTs in the equatorial Atlantic Ocean (see Fig. 3). Furthermore CanESM2 shows precipitation too far north, beyond 20°N, while MPI gives a better representation closer to the observed values (between 15°N and 20°N). On the other hand, all of the CRCM5 simulations barely go beyond 15°N. The dry bias of reanalyses-driven CRCM5 in the Sahel region (as noted in HD12) remains when driven by CanESM2 and MPI, as can be seen also in Figs. 5b, and 6a.
The seasonal-mean vertical cross-sections of the mean zonal wind from 10°S to 35°N, averaged between 10°W and 10°E, is presented in Fig. 10; the left column shows the CRCM5 driven by ERA, CanESM2 and MPI, and the right column the corresponding driving data. The main features of the WAM circulation are represented in all simulations, although with more or less success; this includes the cooler and humid southwesterly monsoonal winds and the hot and dry easterly winds in the low levels, the African Easterly Jet (AEJ) in the mid levels, and the Tropical Easterly Jet (TEJ) in the upper troposphere. The monsoonal winds reach 20°N in the reanalyses while in CanESM2 and MPI they only reach 18°N. The shape of the dome they form are also different, MPI being the nearest to the reanalyses. The CRCM5 simulations also show a somewhat different shape of monsoonal winds when compared to their driving data. The surface convergence between the monsoonal and easterly winds is located further south (by almost 2°) in the CRCM5 simulations compared to the driving data, and the same is valid for the position of the AEJ. The core of the AEJ is nearly at the right height and intensity in almost all of the CRCM5 simulations, but it is smaller in the CanESM2-driven CRCM5 simulation. On the other hand, it is oversized in the MPI model, but better represented in the MPI-driven CRCM5 simulation. The southward bias in the position of the AEJ is thus present not only in the ERA-driven but also in the CanESM2- and MPI-driven simulations. This feature seems particular to CRCM5 and is in accordance with the dry bias in the Sahel region (Figs. 5b, 6a, 9) as discussed in HD12. The intensity and the shape of the TEJ are better represented in both CRCM5 simulations than in the driver models.