Abstract. Heavy precipitation is one of the most devastating weather extremes in the
western Mediterranean region. Our capacity to prevent negative impacts from
such extreme events requires advancements in numerical weather prediction,
data assimilation, and new observation techniques. In this paper we investigate the impact of two state-of-the-art data sets with very high
resolution, Global Positioning System (GPS)-derived zenith total delays (GPS-ZTD) with a
10 min temporal resolution and radiosondes with ∼ 700 levels,
on the representation of convective precipitation in nudging experiments.
Specifically, we investigate whether the high temporal resolution, quality,
and coverage of GPS-ZTDs can outweigh their lack of vertical information or
if radiosonde profiles are more valuable despite their scarce coverage and
low temporal resolution (24 to 6 h). The study focuses on the Intensive
Observation Period 6 (IOP6) of the Hydrological cycle in the Mediterranean
eXperiment (HyMeX; 24 September 2012). This event is selected due to its
severity (100 mm/12 h), the availability of observations for nudging and
validation, and the large observation impact found in preliminary
sensitivity experiments. We systematically compare simulations performed
with the Consortium for Small-scale Modeling (COSMO) model assimilating
GPS, high- and low-vertical-resolution radiosoundings in model resolutions
of 7 km, 2.8 km, and 500 m. The results show that the additional GPS and
radiosonde observations cannot compensate for errors in the model dynamics and
physics. In this regard the reference COSMO runs have an atmospheric
moisture wet bias prior to precipitation onset but a negative bias in
rainfall, indicative of deficiencies in the numerics and physics, unable to
convert the moisture excess into sufficient precipitation. Nudging GPS and
high-resolution soundings corrects atmospheric humidity but even further
reduces total precipitation. This case study also demonstrates the potential
impact of individual observations in highly unstable environments. We show
that assimilating a low-resolution sounding from Nîmes (southern France)
while precipitation is taking place induces a 40 % increase in
precipitation during the subsequent 3 h. This precipitation increase
is brought about by the moistening of the 700 hPa level (7.5 g kg−1)
upstream of the main precipitating systems, reducing the entrainment of dry
air above the boundary layer. The moist layer was missed by GPS observations
and high-resolution soundings alike, pointing to the importance of profile
information and timing. However, assimilating GPS was beneficial for
simulating the temporal evolution of precipitation. Finally, regarding the
scale dependency, no resolution is particularly sensitive to a specific
observation type; however, the 2.8 km run has overall better scores, possibly
as this is the optimally tuned operational version of COSMO. Future work
will aim at a generalization of these conclusions, investigating further
cases of the autumn 2012, and the Icosahedral Nonhydrostatic Model (ICON)
will be investigated for this case study to assert whether its updates are
able to improve the quality of the simulations.