3.1 Seasonal cycle of the TJ in CMIP6 model climatology
Presence of the TJ throughout the year is investigated by masking wind vectors for winds speeds below 5 m/s at 850 mb level. The pressure level is chosen since the wind here flows above the terrain and therefore, it is the lowest level of the TJ which is free from obstruction by the terrain barrier. Figure 1 (panels a to c) shows the mean wind vectors in ERA5 versus wind from 2 CMIP6 models chosen to illustrate two contrasting models. High wind speeds of at least 5 m/s in both ERA5 and the HadGEM-GC31-LL are constrained within the Turkana channel throughout the year. The BCC-ESM1 model has the winds spread out over Kenya.
Through the year, the prevailing wind direction shifts from southerly during May to September, to northerly during October to April. The southerly flow during May to September splits over northern Kenya, into two streams: south-easterly flow (the TJ), and the larger south-westerly flow (East Africa Low Level Jet- EALLJ). This splitting is first seen in May, as evident in figure 1a and 1b (May panel in ERA5 and the HadGEM-GC31-LL). The north-easterly stream forms the TJ with a clear axis from southeast to northwest first seen in April and October in both ERA5 and the HadGEM-GC31-LL model.
The mean wind speeds for 1981 to 2014 are shown for a point at mid Turkana channel (Figure 1d). In ERA5, they remain relatively high throughout the year with mean values ranging between 10 and about 13 m/s. While the CMIP6 ensemble mean similarly has a 2 m/s range between maximum to minimum mean values, the magnitude of the model values is lower throughout the year. The difference is largest from May to October, when the EALLJ is also evident in both the ERA5 and the HadGEM-GC31-LL model. During the May to October period, the spread of individual model values about the ensemble (standard deviation) is minimal, indicating that the majority of the models show stronger departure in values from ERA5. During the same climatological period, the BCC-ESM1 model exhibited climatological mean wind speeds ranging between 5 and 9 m/s with particularly low values in presence of the EALLJ, from April to September. In the BCC-ESM1 model, the EALLJ is merged with the TJ, hence the TJ appears to terminate within this Turkana region in this BCC-ESM1 model. This shortfall appears to be characteristic of most of the models that contribute to the ensemble. Further analysis in subsequent sections will set up thresholds that will isolate the models that exhibit the shortfall.
The southerly winds carry moisture into the region during northern summer time compared to other months of the year (illustrated by red shading in figures 1a to 1c) and coincides with a period in which the regions bordering the northwest of Kenya receives extended rainfall (Chamberlin, 2018). Considering the TJ is a major transporting agent for moisture into northwest of East Africa (Vizy and Cook 2019; Munday et al., 2020), simulated impacts of the TJ to rainfall are likely to be different in the reanalysis and in the CMIP6 models that show a weaker TJ during the season.
3.2 Spatial Structure in climatology of the TJ in CMIP6
The mean vertical structure of the lower atmosphere (surface to 500 mb pressure level) winds is analyzed along the Turkana channel in CMIP6 models (figure 2). In ERA5 as well as in the models, the relatively high-speed winds are located above the surface; wind speeds increase between the surface and 850 mb and then decrease with altitude. The center of intensity appears between 925 mb and 850 mb pressure levels near the middle of the Turkana channel following acceleration at this point (Patwardhan and Asnani, 1999; Indeje et al., 2001).
In JJAS, at the mid-Turkana area, ERA5 and both versions of the CNRM model exhibit the highest change in wind speeds from maximum (approximately 12 m/s) at 850mb to minimum (approximately 7 m/s) before 500 mb level is reached (figure 2a). Other models with relatively large changes with height, of more than 5 m/s, include HadGEM3-GC31-MM, MRI-ESM2-0, CESM2, FGOALS-f3-L, SAM0-UNICON, ACCESS-CM2, HadGEM3-GC31-LL, IPSL-CM6A-LR and CanESM5. The rest of the models show lower wind shears as mean winds are slower at the mid-Turkana channel. They show the center of intensity further away from the middle area of the Turkana channel. The models are namely BCC-ESM1, CAMS-CSM1-0, SAM0-UNICON, TaiESM1, MIROC6, NESM3, INM-CM4-8, and INM-CM5-0. In these models, high wind speeds fail to extend up to mid-Turkana Channel, an apparent consequence of the TJ merging with the EALLJ during the JJAS season.
During the October to May months, the EALLJ is absent, the center of intensity in the wind speeds shifts northwards to mid-channel area, it is elevated, and has a vertical elongation up to about 825 mb (figure 2b). As the winds flow through the channel in absence of the EALLJ, they are displaced upwards by the gently rising slope indicated in shown topography below each panel. The vertical extension of the TJ core is accompanied by weaker vertical wind shear. The weakening shear with height appears pronounced in a set of models which exhibited the center of intensity further away from the middle area of the Turkana channel during JJAS season (BCC-ESM1, CAMS-CSM1-0, SAM0-UNICON, TaiESM1, MIROC6, NESM3, INM-CM4-8, and INM-CM5-0). This phenomenon indicates that while the winds begin to accelerate at mid-Turkana channel, the gain in wind speed is lesser in these models compared to ERA5.
Those CMIP6 models which have lower wind speeds in the lower-atmosphere are of lower resolution. Considering that a TJ characterized by high wind speeds that decrease aloft result from acceleration within the Turkana channel, the disparities between the ERA5 and some of the CMIP6 models are likely to result from TJ interaction with topography within the channel.
3.3 Mean vertical flow in the Turkana channel
Figure 3 shows the mean omega through the study period, plotted for the northwest to southeast axis parallel to the Turkana channel (for models with the omega data available and the same transect is used as in Figure 2). Other than in the HadGEM-GC31-LL and the two versions of the CNRM model, descending motions dominate the entrance of the Turkana channel during the JJAS season and the mid-channel area during the months of October to May in CMIP6 models. Relating these results with those in Figure 2 shows that descending motions dominate lower atmospheric levels where the intensity of the TJ is strong. The ERA5, HadGEM-GC31-LL and the two versions of CNRM model have relatively faster winds at mid-Turkana area during JJAS season following intense TJ core in the season.
Additionally, the influence of the elevated terrain is noticeable. In both the JJAS and October to May seasons, ascending motion emanates between the entrance and the mid-channel area where the floor of Turkana channel begins depicting an elevation, due to frictional interaction of the wind with the sloping floor of the Turkana channel. During JJAS season the TJ core accompanied by subsidence aloft is further to the southeast of the mid-Turkana channel. This allows low-level ascents manifesting at the TJ entrance to incline towards the middle and exit area of the channel with height as generation of the ascents diminish as surface elevation slopes downwards at Turkana channel exit. This further strengthens uplifts at the exit area in the deep troposphere. During October to May months, the TJ core accompanied by subsidence aloft is at the mid-Turkana channel area. Thus, ascents from the TJ entrance area are inhibited from lateral movement towards the mid-Turkana area. At TJ exit, vertical motions are generated by deceleration of winds that leads to build up of pressure at lower atmosphere.
Other than the ERA5 and the two CNRM models, other CMIP6 models show stronger ascending motion, potentially from enhanced blocked lower atmospheric flow at the mid-Turkana channel. The ERA5, HadGEM-GC31 and CNRM models appear to depict unblocked flow through the channel with pockets of near surface descending motion associated with faster wind speeds at the mid-Turkana channel area. During the months of October to May, the mid-channel descents are stronger. This suggests that although strong TJ intensity induces the subsidence in the Turkana channel, the interaction between the TJ and the Turkana channel influences the intensity of the TJ core.
3.4 Topography in CMIP6
Figure 4a shows that the Turkana channel is the largest topographic feature below the 850 mb level within the study area bounded by longitudes 33°E and 41°E and latitudes 2°S and 9°N. The floor of the channel at the entrance is roughly 980 mb in ERA5 and the CMIP6 models. The channel floor gently rises to between 945 and 960 mb at mid-channel area with walls of the channel rising to about 915 mb in ERA5 and the models with less than 1° longitudinal grid step size. In coarser resolution models, the floor of the channel rises to the level that matches or approximates that of the wall and therefore, the channel appears flattened. This phenomenon is more apparent in the CanESM5 and the BCC-ESM1 models.
Figure 4b shows the cross-section of the Turkana channel at each longitude. The channel narrows from about 500 to 100 square kilometers in both the ERA5 and the CMIP6 model datasets as one approaches the longitude 37°E from the channel entrance situated at around 41°E, within latitudes 2°S and 9°N. The INM, NorCPM1 and NorESM2-LM models attain this constriction at 38°E. Exiting at the 37°E longitude; the channel widens to about 200 square kilometers further to the west of this longitude according to the ERA5.
While the narrowing results in the formation of a constriction at the 37°E longitude, the changes between adjacent cross sections are varied in CMIP6 models compared to ERA5, consistent with results in Vizy and Cook (2019). Other than the HadGEM3-GC31-MM and MPI-ESM1-HR models, cross-sectional areas at some adjacent longitudes are maintained or least changed compared to the ERA5. Strikingly, CanESM5 exhibits higher cross-sectional area changes whenever a difference is observed at its relatively coarse spatial grid resolution. The results from ERA5 and the HadGEM3-GC31-MM and MPI-ESM1-HR which are two high-resolution CMIP6 models show that reducing the grid-step size increases topographic details being resolved and forms a narrowing Turkana channel.
The difference between the cross-sectional area of the channel at the entrance and exit of the jet is computed. Figure 5 shows the narrowing index for CMIP6 models as well as the horizontal resolution (longitudinal grid step size) of the model. Since the cross-sectional area entrance of the channel (42°E) in the ERA5 and all the CMIP6 models is about 800 square kilometers and reduces to about an eighth of this at the mid-channel, models with zero changes between adjacent longitudes have higher gradient whenever shape changes. Thus, a low index (Figure 5, y-axis), which is characteristic of low-resolution models with higher grid step sizes, characterizes steep change in terrain followed by blocks of wall between the adjacent longitudes with constant cross-section area between them.
The ERA5 has the highest narrowing index for the Turkana channel. It is followed by CMIP6 models of less than 1.5° resolution in which the index is either maintained or weakens in models with higher grid-step sizes. At greater than 1.5° resolution, the CMIP6 models scattered values for the index and show inconsistency in how the narrowing of the channel is lost with increasing coarseness in the topography. The CanESM5 model is an outlier, returning a high value of the quantity due to high rates of change associated with very large grid-step sizes and a smaller number of points in the channel (Figure 4b). Therefore, linking this result with those from Figure 3 indicates that CMIP6 models with resolution greater than 1.5° have steep walls along the floor and height of the channel. This is a likely source of the impediment on the smooth flow near the surface along the Turkana channel.
3.5 Topographic influence on the TJ in CMIP6
The association between the strength of the TJ and the shape of the Turkana channel represented by the narrowing index, is explored in this section by adjusting the thresholds for JJAS season maximum wind speed and vertical shear (between the point of maximum and minimum wind speed before reaching 500 mb). Figure 6 shows TJ counts for wind speeds at 850 mb in range 6 to 10 m/s and wind shear above the TJ core in range 2 to 6 m/s, for each model and its narrowing index.
Most models return a TJ with maximum wind speeds exceeding 6 m/s and vertical wind shear above the TJ core exceeding 2 m/s. Only the CMIP6 models INM and ACCESS-ESM have the TJ structure weaker than this category within the 35-year study period. Higher thresholds of 9 m/s at 850 mb and vertical wind shear of 5 m/s TJs are still common in the majority (17) of the 26 CMIP6 models. Nearly half of the CMIP6 models return zero TJ counts at stricter criteria of higher than 9 m/s maximum wind speeds and greater than 5 m/s vertical shears above the TJ core. Therefore, we consider those thresholds as optimal.
The optimum thresholds are surpassed in CMIP6 models having a high index for the shape of the Turkana channel such as HadGEM3-GC31-MM, CNRM and MRI. This is similar to the ERA5 which has the highest index (Figure 5) considering TJs whose core is characterized by at least 9 m/s maximum wind speeds which decrease by at least 5 m/s above. Referring to Figure 2, the mean TJ in ERA5 reaches higher values for maximum wind speeds at 12 m/s, decreasing by 7 m/s above the core. But CMIP6 models with low index for shape of the Turkana channel exhibit mixed TJ characteristics. For instance, the CNRM and FGOALS-f3-L models exhibit TJs when NorESM2-LM and INM models of comparable index for the Turkana channel, return zero TJ count considering the optimum thresholds for the TJ in CMIP6 models. There were 9 CMIP6 models with zero TJs and 4 models with thirty-five TJs reaching the optimum threshold in 35 JJAS seasons. The remaining 13 CMIP6 models occasionally exhibited TJs reaching the optimum threshold for a strong jet (maximum wind speeds of 9 m/s and decreasing aloft the core by at least 5 m/s).
The different CMIP6 models (coarser than 1.5° longitudinal resolution) with low narrowing index show a lower consistency in narrowing cross sectional area. Considering that the narrowing supports channeling of the winds and acceleration of the winds (Indeje et al., 2001), models with coarser than 1.5° longitudinal grid step size which occasionally exhibit weak TJ structure, are here found to be related to poor resolution of the Turkana channel. These irregularities in the TJ strength could also influence rainfall at TJ exit, as the TJ is considered a moisture transport mechanism (Munday et al., 2020).
3.6 Relationship between TJ and East African Precipitation in CMIP6
The influence of the TJ on the climate of East Africa is shown using the composite anomalies of precipitation for the JJAS season (Figure 7). Mean precipitation anomalies are shown for strong TJs in the CMIP6 models as they best match TJ strength in the ERA5 (at least 9 m/s at the 850 mb and vertical wind shear of at least 5 m/s above TJ core) in the “jet” panel and also when the strength was weaker “no jet cases” panel.
JJAS is a major season with rainfall accumulation of up to 1,500 mm over the northwest regions of East Africa while other parts of the region remain generally dry (Camberlin 2018). This is the average condition, exhibited when the TJ strength remains regular. Fluctuations in the TJ strength which is common in nearly half of the CMIP6 models used, induce spatial distribution of dry and wet anomalies.
In Figure 7 (left panel), when a perennially weak TJ strengthens in model climatology, dry anomalies at the border region of the exit the Turkana channel are exhibited. Specifically, in the whole JJAS season (122 days), rainfall is suppressed by up to 122 mm below the climatological mean (~ 8% of the 1,500 mm), observed in CAMS-CSM1-0, BCC-CSM2-MR GFDL-CM4 and FGOALS-g3. The dryness extends up to the west of South Sudan, east of Central African Republic, northern Congo and to the west of Ethiopian Highlands. The increasing dry anomalies at the TJ exit area is associated with stronger inflow of air with low moisture content during this season through lower atmosphere. Furthermore, during a strong TJ, easterly flow is so strong that it inhibits southwesterly advection of moist air from equatorial Congo.
Conversely, wet anomalies characterize the immediate exit of Turkana channel and Ethiopian highlands in the CMIP6 models when perennially strong TJ weakens below the thresholds (Figure 6; right panel). Precipitation is enhanced by up to 1 mm more than the climatological mean in the models. Further to the west of South Sudan, in the Central African Republic and northern Congo, dry anomalies below the climatological mean are exhibited. This is likely associated with enhanced uplift of moist air that is transported from the Indian Ocean through the Turkana channel at TJ exit. In this case, pressure piles up at the immediate exit due to deceleration of winds and promotes zonal advection of moisture-laden air to the western regions of South Sudan (Vizy and Cook 2019). Furthermore, during a weak TJ, southwesterly advection of moist air from equatorial Congo is likely being promoted.