A comparison of the daily evolution of OLR-cluster sizes with SEVIRI observations shows that simulations with parameterized convection initiate deep convection several hours before it is observed and these simulations fail to develop larger, organized clusters. Simulations with explicit convection, even with 12-km horizontal grid length, have an improved timing of the diurnal cycle and allow for organized convection to develop, similar to observations, although the cloud amounts appear too high.
The vertical cloud structure in the simulations is evaluated against CloudSat observations, allowing for a comparison of different cloud types. The simulations with parameterized convection have cumulonimbus and anvil occurring too far south and at the wrong time compared to observations. These simulations, as well as those with explicit convection, generate too much low-level cloud across West Africa, including far north into the Sahara, where it is not frequently observed. The model produces the correct amount of mid-level cloud in both the parameterized and the explicit simulations, but in all simulations this cloud is positioned directly at the freezing level, several kilometers lower than observed.
The errors apparent from the evaluation with satellite observations can be directly associated with errors in the diurnal cycle. We show how the simulations with explicit convection have a more realistic diurnal cycle in convection. This changes the relative timing of pressure-gradients induced by convective heating and synoptic-scale flows, which are maximized at night when boundary-layer mixing is weak. This generates a weaker and more realistic monsoon flow from the Sahel to the Sahara in explicit runs. Cold-pool outflows from convection form a significant component of the WAM in explicit runs, consistent with observed errors in a global model from cold pools in this region. Errors in the diurnal cycle in simulations with parameterized convection are shown to establish within the first 24 hours of the run and are reinforced by the inability to trigger convection far enough north. This combination of process-based analysis of the various simulations and model cloud verification using complementary satellite observations assists in understanding the West African monsoon water cycle and highlights critical areas for model improvement.