The deployment of special instrumentation for the Hydrological Cycle in the Mediterranean Experiment (HyMeX) provides a valuable opportunity to investigate the spatio‐temporal variability of atmospheric water vapour across scales in relationship with the occurrence of Heavy Precipitation Systems (HPSs) in the north Western Mediterranean (WMed) during the Intensive Observation Period (IOP12), which is the focus of this investigation. High‐resolution convection‐permitting COSMO simulations complement the observational network and allow the calculation of on‐line trajectories. In addition to the presence of a favourable large‐scale situation and low‐level convergence, atmospheric moisture changes resulting in conditionally unstable air are identified as responsible for convective initiation (CI). All HPSs within the north‐WMed form in periods/areas of maximum integrated water vapour (IWV; 35–45 kg/m2) after an increase of about 10–20 kg/m2. The most intense events receive moisture from different sources simultaneously and show a sudden increase of about 10 kg/m2 between 6 and 12 h prior to the event, whereas in the less intense events the increase is larger, about 20 kg/m2, over a period of at least 24–36 h. Changes in the lower (∼900 hPa) and mid‐troposphere (∼700 hPa) control the evolution of the atmospheric moisture and the instability increase prior to CI. Spatial inhomogeneities in the lower boundary layer determine the timing and location of deep convection, whereas enhanced moisture in the mid‐troposphere favours intensification. Moister and deeper boundary layers, with updraughts reaching up to 2 km are identified in those pre‐convective environments leading to HPS, whereas dry, shallow boundary layers are found everywhere else. The build‐up time and vertical distribution of the moisture changes are found to be crucial for the evolution and severity of the HPSs rather than the amount of total column atmospheric moisture.

<p>In this study we investigate the impact of several selected sources of uncertainty on convective precipitation prediction. For this purpose, we conduct numerical simulations with the ICOsahedral Non-hydrostatic (ICON) model for two consecutive days in June, 2021, on which deep moist convection triggered by different synoptic forcing occurred over southwestern Germany. We use single- and double-moment microphysics schemes and vary the initial soil moisture, grid spacing, and cloud condensation nuclei (CCN) concentration. We compare the results with measurements conducted on the same two days during the Swabian MOSES (Modular Observation Solutions for Earth Systems) field campaign. We find that the applied dry bias (initial soil moisture in the model reduced by 25%) much better represents the actual soil moisture conditions and leads to an improved quantitative precipitation forecast when compared to radar-derived precipitation amounts. Furthermore, the model resolution impacts the precipitation amount, intensity, and the timing of convection initiation: while 1-km runs show the least root mean square error for 24-hour precipitation sums, the onset of convective precipitation in 2-km resolution runs matches better the observations. However, the overall impact of this factor is not always systematic. The comparison of several radiosounding-derived convective indices (e.g. lifted index, convective available potential energy, convective inhibition) with model data yield many non-systematic results. For instance, CCN concentrations do not seem to have any significant impact on any of the calculated indices. At the same time, runs with coarser resolution (2-km) often better depict the temporal development of CAPE but overestimate its amount.</p>