The role of low-level clouds in the West African monsoon system

Realistically simulating the West African monsoon system still poses a substantial challenge to state-of-the-art weather and climate models. One particular issue is the representation of the extensive and persistent low-level clouds over southern West Africa (SWA) during boreal summer. These clouds are important in regulating the amount of solar radiation reaching the surface but their role in the local energy balance and the overall monsoon system has never been assessed. Based on sensitivity experiments using the ICON model for July 2006, we show for the first time that rainfall over SWA depends 5 logarithmically on the optical thickness of low clouds, as these control the diurnal evolution of the planetary boundary layer, vertical stability and finally convection. In our experiments, the increased precipitation over SWA has small direct effects on the downstream Sahel, as higher temperatures due to increased surface radiation are accompanied by decreases in low-level moisture due to changes in advection, leading to almost unchanged equivalent-potential temperatures in the Sahel. A systematic comparison of simulations with and without convective parameterisation reveals agreement in the direction of the precipitation 10 signal but larger sensitivity for explicit convection. For parametrized convection the main rainband is too far south and the diurnal cycle shows signs of unrealistic vertical mixing, leading to a positive feedback on low clouds. The results demonstrate that relatively minor errors, variations or trends in low-level cloudiness over SWA can have substantial impacts on precipitation. Similarly they suggest that the dimming likely associated with an increase in anthropogenic emissions in the future would lead to a decrease of summer rainfall in the densely populated Guinea Coastal area. Future work should investigate longer-term 15 effects of the misrepresentation of low clouds in climate models, e.g. moderated through effects on rainfall, soil moisture and evaporation. Copyright statement. TEXT


S1.1 Climatology and model evaluation
In addition to the general characterization of the meteorological conditions in southern West Africa for the variables precipitation and radiation, a brief discussion about the diurnal cycle of the vertical atmospheric structure is given for the wet monsoon season July, August and September 2006. Comparison of ICON CLIM with ERA-I and observations confirms the applicability 5 of the ICON model for the sensitivity experiments of the main article. Figure S1 shows average profiles of cloud cover (CLC), relative humidity (RH), horizontal wind speed v horiz as well as specific cloud water content q c for 00, 06, 12 and 18 UTC (corresponding to local time in our study region). At 00 UTC the NLLJ is already well established and the low-level cloud deck is beginning to form (Fig. S1a). ICON shows a considerably stronger jet than ERA-I reaching 7 ms −1 at 925-950 hPa and consistently lower values in CLC, RH and q c . In contrast to 10 ERA-I, ICON tends to concentrate cloud water in the upper parts of the cloud deck around 850 hP a. The relatively small differences in RH between the two datasets (more than 90% from 830 hP a downwards) in contrast to differences in CLC and q c illustrates a substantial sensitivity to the subgrid-scale cloud scheme or possibly differences in spatial variance of RH, as the dependance of CLC on RH is quadratic in ICON. The tendency of stronger NLLJ and less cloud was also found in many climate models (Hannak et al. 2017). At midlevels around 560 hPa ICON shows a secondary peak in v horiz , CLC, RH and 15 q c not found in the overall smoother and moister ERA-I profiles. At 06 UTC the NLLJ is very similar to 00 UTC but the low-level cloud deck increases markedly in cover and q c accompanied by an increase in RH to values well above 95 % below 900 hPa (Fig. S1b). Maximum CLC occurs at 950 hPa reaching 25% in ICON and about 45% in ERA-I, which is more realistic (cf. van der Linden et al., 2015). Overall the discrepancies between the two models are qualitatively similar to 00 UTC ( Fig. S1a). At midday (Fig. S1c), radiative heating lifts and dissolves the 20 low-level cloud deck shifting the maximum in CLC and RH to 850 hPa, where a pronounced peak in q c develops. Surface heating and turbulent mixing markedly slows down the low-level jet (e.g. 4.5 ms −1 in ICON) and decreases RH to under 90% below 900 hPa with ICON being substantially drier and less cloudy in that layer. Finally at 18 UTC (Fig. S1d) the low-level jet starts re-accelerating, keeping the generally higher values in ICON found at all times of day. The deep daytime mixing has reduced CLC and q c and created an almost vertically constant offset between the two modeling systems. RH is already 25 increasing at this time of day, particularly in ICON, where also the sharp gradient in v horiz suggests a beginning decoupling of the surface. Such an early evening transition is consistent with observations as documented in Fig. 3d in Schuster et al. (2013). The comparison between the two datasets shows considerable biases at all times of day with generally higher low-and midlevel wind maxima in ICON but moister and more cloudy low levels in ERA-I. Investigating the reasons for these discrepancies is beyond the scope of this paper but the overall agreement in vertical structure and diurnal cycle suggests that sensitivities tested 30 with ICON should be qualitatively meaningful.

S1.2 Temporal stability of opacitiy-induced effects
An additional aspect to be discussed is the response time of the atmospheric system to the imposed cloud modifications. To investigate this we use EXPL experiments, in which f op = 0.1 is applied for the first 4 days but then switched off for 6 more 35 days of simulation time. Control runs with f op = 1.0 for all times were produced for comparison. As in EXPL, simulations were started every 4th day but run out to 10 days and two starting dates in August were added (3rd and 7th of August) to give better statistics for the time evolution. Figure S2 shows box-averaged 10-day time-series of RR, cover of low clouds CLC low and that of high clouds CLC high (below 800 hPa and above 400 hPa, respectively) for the DACCIWA region (Figs. S2a-c). The corresponding differences between the two sets of simulations are provided in the right-hand side panels of Fig. S2.

40
After the switch-off at 12 UTC on the fifth simulation day (i.e. after 96 hours), the differences in SSI and low-level T are reduced almost immediately (not shown), but for other variables the response is slower. RR shows the enhancement of afternoon and evening precipitation for f op = 0.1 as in EXPL (Figs. S2a and f). The enhancement is still fully visible for the first 14 hours after switch-off, indicating that the influence of the forcing during the morning hours is already enough to generate more instability and trigger more convection later in the day. After that, differences between the two runs become negligible. Before the switch in f op , CLC low shows the familiar afternoon decrease and nighttime increase (Figs. S2b and g). On the day of the change, some signal remains until the morning of the following day, similar to RR. The small but on average slightly positive differences after that may be a reflection of increased surface fluxes after the strongly enhanced rainfall of the first five days. CLC high (Figs. S2c and h) shows a considerably slower response. Differences between the two runs need one full diurnal cycle to establish and are then positive for the next three days. After the switch, there is a marked decrease in differences but then an overall tendency for relatively large values for two more days. This is consistent with Raymond et al. (2011), who show a considerably longer response time in the tropical upper troposphere than at low levels for a given perturbation.
Another interesting question is the impact on regions to the north, i.e. downstream of the DACCIWA box with respect to the monsoon flow. Figures S2d, e, i and j show corresponding plots for RR over the Sahelian regions 10-15 • N and 15-20 • N, 5 both averaged over 8 • W-8 • E. For the former, again an initial response time of about one day is observed followed by a period of small positive differences. During the last five days of the simulation there is then no clear net difference between the two sets of experiments but much larger fluctuations. As these do not follow a strict diurnal cycle, we speculate that this is mostly a reflection of the overall chaotic nature of the atmosphere growing with leadtime. This conclusion is consistent with the similar behavior found for the 15-20 • N band.

10
So in summary, this experiment shows that low-level variables such as SSI and T react almost immediately to changes in low cloud during the day. Low-level cloud cover and rainfall respond after one full diurnal cycle, while upper-level variables and neighboring regions show even longer responses, but also increasingly chaotic behavior. The latter is reflected in the growing shaded areas denoting the envelope of all runs in the bias in the right-hand side panels of Fig. S2. S1.3 Table of acronyms   Table S1.