Sensitivity of precipitation forecasts to cumulus

Document Sample
Sensitivity of precipitation forecasts to cumulus Powered By Docstoc
					           Sensitivity of precipitation forecasts to cumulus
             parameterizations in Catalonia (NE Spain)
                             Jordi Mercader (1), Bernat Codina (1),
                           Abdelmalik Sairouni (2), Jordi Cunillera (2)

        (1) Dept. of Astronomy and Meteorology, University of Barcelona, Barcelona, Spain

                     (2) Meteorological Service of Catalonia, Barcelona, Spain


1. Introduction

        Precipitation forecasts in Catalonia
(Northeastern Iberian Peninsula) have always
been a challenge due to the complexity of its
orography and its situation beside the
Northwestern Mediterranean Sea (Figures 1a
and 1b). Heavy rainfall events are frequent and
24-hour precipitation amounts exceeding 200
mm are reached almost every year. Despite             a)                          b)
being a small region (32,000 km2), contrasts          Figure 1. a) Situation of Catalonia (orange square), in
between precipitation amounts recorded in a           NE Iberian Peninsula. b) Enlarged area displaying the
particular event can be very large.                   most important mountain ranges. In both maps, the
        In order to evaluate the sensitivity of       green square indicates the situation of Barcelona.
precipitation forecasts to the cumulus
parameterizations and some of the cloud               2. Selected events
microphysical schemes provided by the WRF
model, as well as the ability of this model to               The first event considered as a case took
simulate the most important features of rain          place on August 25, 2006 and was
events in Catalonia, several simulations for          characterized by cold and moist air in mid-
selected episodes are being performed. This           levels of the troposphere. According to the
work is included in a larger scope project            Barcelona sounding, at 12Z the temperature at
carried out by the Meteorological Service of          500 hPa was close to -13ºC and the CAPE 649
Catalonia (SMC) and the University of                 J. The main event of that day was an isolated
Barcelona. This project is aimed to set up a          storm that originated a flash-flood in a quite
suitable operational configuration of WRF-            populated touristic area of the coast (Figures
ARW to be used as a forecasting tool in SMC.          3a,c,e,g). More than 35 cars parked in a usually
        In the poster, preliminary results            dry stream bed were swept to the sea. The
obtained for two convective precipitation             economical losses were noticeable. Also, the
events have been selected to get an                   effects of a tornado were observed.
appreciation of the performance of WRF                       The synoptic situation on September 13,
model: August 25 and September 13, during             2006 was characterized by a through with cold
later summer of 2006.                                 and moist air that crossed the Iberian
                                                      Peninsula from NW to SE. Precipitations
                                                      affected all Catalonian areas, many stations
recorded 24-hour precipitation amounts            3.2. Initial and boundary conditions
exceeding 100 mm, with a maximum record
reaching 238 mm, while other nearby                       To generate both initial and boundary
observatories recorded total daily amounts        conditions, GFS forecasts are used, with a
between 10 and 20 mm. Most of the rainfalls       horizontal resolution of 1º and a 6-hour time
were concentrated in two rain bands (Figures 4    interval between two consecutive forecast
a-k) that crossed the region following a SW-      outputs.
NE direction. The second one became                       Specifically, to initialize a WRF
stationary for almost 3 hours when reaching       simulation at 00 or 12Z, GFS forecasts
the NE of Catalonia and caused the largest        initialized 12 hours earlier have been used.
amounts of precipitation of that event (Figure    Consequently, the first guess is actually
4k).                                              generated from a GFS 12-hour forecast and
                                                  subsequent 6-hour forecasts are used to
3. Model simulation configuration                 provide boundary conditions for the coarsest
      In this work, the WRF-ARW model                     For simulations in nested domains, both
version 2.2 has been used, with the               initial and boundary conditions are supplied
configuration described below.                    from their parent domain’s run. Both 12 and 4-
                                                  km simulations are initialized at the same time
3.1. Model domains                                as the coarsest domain run and boundary
                                                  conditions are updated every 3 hours.
      Simulations were performed using a set
of 3-nested domains with horizontal grid-point    3.3. Observational data assimilation
resolutions of 36-12-4 km. The outer domain is
a grid of 94x102 points which covers the SW              Conventional surface and upper-air data
of Europe; the second domain has a grid size      are assimilated, using WRF-3DVar applications
of 70x70 points, and the finest mesh covers       (Barker, D.M, et al., 2004), in order to improve
Catalonia and its surrounding area (Figure 3)     the first guess and the boundary conditions in
with a grid of 88x88 points.                      the coarsest domain.
      Vertical levels used in these simulations          This configuration described above has
are the default 31 levels defined in WRF          been made in order to compare WRF with
(Wang, W., et al., 2007).                         MM5 forecasts generated in SMC for
                                                  operational purposes (RAM-SMC, 2005).

                                                  3.4. Parameterization settings

                                                        These simulations have been carried out
                                                  using the NOAH (4 subsoil layers) land surface
                                                  model, the Yonsei University (YSU) planetary
                                                  boundary layer scheme, the Monin-Obukhov
                                                  scheme for surface layer physics, along with
                                                  the Dudhia and RRTM parameterizations for
                                                  short-wave and long-wave radiation processes.
                                                  (Skamarock, W.C., et al., 2005). These
                                                  parameterization settings have been kept
                                                  without change for all simulations.
                                                        On the other hand, in order to evaluate
Figure 2. The 3-nested domains with horizontal    the sensitivity of precipitation forecasts to each
resolutions of 36-12-4 km.
                                                  of the cumulus parameterizations supplied by
                                                  the WRF model and some of the available
                                                  clouds microphysical schemes, several
simulations have been performed for each               (not shown). Specifically, the Thomson scheme
selected event trying different combinations of        gives, in general, higher intensities than the
these physical parameterizations (see table 1).        WSM5 or WSM6 schemes.
                                                              In addition, the sensitivity of the
 Domain/          Cumulus               Clouds         precipitation field of a nested run to the
resolution        schemes            microphysical
             - Kain -Fritsch         - WSM3
                                                       cumulus parameterization used in its parent
36 km        - Betts–Miller–Janjic   - WSM5            domain’s run has been also evaluated. In fact,
             - Grell-Devenyi         - Thompson        many simulations in this event had shown that
             - Kain-Fritsch          - WSM3            sensitivity to this factor can be greater than the
             - Betts-Miller-Janjic   - WSM5
12 km
             - Grell-Devenyi         - WSM6            sensitivity to the clouds microphysics schemes
             - NO convection         - Thompson        (not shown).
             - Kain-Fritsch
             - Betts-Miller-Janjic   - WSM6
4 km
             - Grell-Devenyi         - Thompson
             - NO convection
Table 1. For each domain, cumulus and cloud
microphysical schemes that had been used to evaluate
the sensitivity of precipitation forecasts to them.

4. Results
                                                       a) RADAR: 18Z                b) WRF: 18Z
4.1. August 25, 2006 event

       All runs were initialized at 12Z. The 12-
km runs with the cumulus parameterizations of
Betts–Miller-Janjic       and       Grell-Devenyi
forecasted the convective rainfalls at the
mountainous area of Pyrenees, but they were            c) RADAR: 19Z                d) WRF: 19Z
not able to reproduce the isolated storm that
developed in the Pre-litoral mountain range
and later reached the coastline and produced a
flash-flood. In fact, both simulations
forecasted little precipitation on the sea when
the storm had already passed over the
coastline, where that storm actually finally went
(results not shown).                                   e) RADAR: 20Z                f) WRF: 20Z
       In contrast, the results of the 12-km run
with the Kain-Fritsch convection scheme (not
shown) displayed a rainfall between 18Z and
22Z that can be associated to the radar
signature of the storm.
       Although cumulus parameterizations
should not been used in simulations with
                                                       g) RADAR: 21Z                h) WRF: 21Z
horizontal grid-point resolutions of 4-km, one
of these schemes is needed to forecast this            Figures 3a,c,e,g. 1-hour accumulated precipitation from the
storm (Figures 3b,d,f,h). Otherwise, this event        radar imagery of the August 25, 2006 storm. From 18Z to
is missed.                                             21Z it followed a NW-SE trajectory, crossed the coastline
       The results provided with the runs that         about 20Z and produced a flash-flood. Figures 3b,d,f,h. 1-
                                                       hour accumulated precipitation output for the WRF 4-km
had used different clouds microphysical                run with the Kain-Fritsch cumulus parameterization (and the
schemes showed more differences in rainfall            Betts-Miller-Janjic cumulus parameterization on its parent
intensity than in any other feature of the             domain’s simulation). Despite the storm was forecasted in
precipitation field, such as timing or location        this work, precipitation intensity was underestimated.
4.2. September 13, 2006 event

       Both the 12-km and 4-km runs, which
were initialized at 00Z of September 13, were
not able to capture the occurrence of the first
rain band (Figures 4a,4c), probably because of
the spin-up of the model and the wind
                                                       a) RADAR: 02 Z               b) WRF: 02 Z
direction simulated in low-levels, which was
different from the actual wind direction in
       However, the second rain band was quite
well forecasted in terms of the shape and
movement, but only when no convective
schemes had been used, both in the 4-km runs
and, surprisingly, in the 12-km runs.                  c) RADAR: 05 Z               d) WRF: 05Z
       In the 12-km run with explicitly-resolved
convection, the second rain band was actually
advanced in time, because it can be seen in the
model output (Figure 4b) such early as 02Z,
while the first signals of this rain band did not
appear in the radar signature until 07Z (Figure
                                                       e) RADAR: 08 Z               f) WRF: 08Z
       After its formation, this rain band
remained stationary for two hours and finally it
began to move towards NE (Figure 4g). This
behavior was forecasted by simulations too,
despite a little advance in time (Figures 4f,4h).
       Finally, the rain band became smaller,
changed its orientation and remained stationary
near the coastline in the NE of Catalonia              g) RADAR: 11 Z               h) WRF: 11 Z
(Figure 4k), producing the largest amounts of
precipitation of that event. As it can be seen in
figures 4j and 4l, this movement and the final
placement of the rain band were also well
simulated with a little displacement.
       The best agreements with the 4-km runs
were     achieved      with     explicitly-resolved
                                                       i) RADAR: 14 Z               j) WRF: 14 Z
convection too (not shown). In comparison
with the simulations provided by the 12-km
runs, they showed the second rain band
moving faster.
       On the other hand, the experiments
designed to evaluate the sensitivity of the
forecasted precipitation field to the cloud
microphysics        schemes      revealed       that   k) RADAR: 16 Z               l) WRF: 16 Z
simulations with the Thompson scheme
                                                       Figures 4a-k. In the left-hand side, 1-hour accumulated
produced higher intensities of precipitation           precipitation from the radar imagery of September 13, 2006.
than the simulations with other cloud                  In the right hand side, 1-hour accumulated precipitation
microphysics parameterizations, such as the            output for the WRF 12-km run, with explicitly-resolved
WSM5 or WSM6 (results not shown), as it has            convection (and the Kain-Fritsch cumulus parameterization
been seen for August 25 event yet.                     on its parent domain’s simulation).
       In addition, the influence of the cumulus   precipitation, while the simulations with
scheme used in the parent domain’s run was         explicitly-resolved convection showed the best
evaluated (results not shown). Such influence      agreement with observations after the first 6
increases with time and it implies differences     hours.
both in the position of the rain bands and the
intensity of the rainfall. Consequently, the
maximum 24-hours amounts of the forecasted
rain fields have shown to depend on the
cumulus parameterization scheme used in the
parent domain’s run. For instance, the nested
4-km run from a 12-km run with explicitly
resolved convection showed a maximum
amount of 24-hour forecasted precipitation up
to 228 mm (good agreement with observed
records) while the simulation whose parent was
a 12-km run with the Betts-Miller-Janjic
parameterization showed a maximum amount
of 137 mm.
       Finally, the rain field from the
operational output of the MM5 12-km run
(with the Grell convection parameterization
and the Schultz microphysics scheme) has been
compared with the rainfall forecasts from the
WRF simulations. At initial times, MM5
showed precipitation over the Northern areas
of Catalonia, as it was observed by the radar
imagery, while the WRF 12-km run (with
explicitly-resolved convection) concentrated       Figures 5a-f. On the left hand side, at the top, 1-hour
                                                   accumulated precipitation from the radar imagery of September
the rainfall in SW Catalonia (Figures 5a-c).       13, 2006 at 04Z; in the middle, 1-hour accumulated
However, after the first 6 hours, the WRF          precipitation output for the MM5 12-km run at the same time,
output displayed a rain band that began to         (with the Grell convection scheme) and in the bottom, 1-hour
move towards NE, while MM5 showed                  accumulated precipitation output for the WRF 12-km run
precipitations moving with this direction but      (explicitly-resolved convection). On the right hand side, the
                                                   same but at 10Z.
reproducing some rainfall cores rather than a
clearly band-shaped rainfall.
                                                          Such results suggest the need for a new
                                                   cumulus scheme at intermediate scales
5. Conclusions                                     (between 3-4 and 10 km) because the actual
                                                   parameterizations were designed for horizontal
       Two convective rainfall events in           grid-point resolutions coarser than 10 km.
Catalonia have been presented to get an                   It has been noted also that the most
appreciation of the sensitivity of precipitation   influence of the clouds microphysics schemes
forecasts to both cumulus parameterizations        on precipitation forecasts is done in the
and cloud microphysics schemes.                    precipitation intensity. In these events, the
       In the August 25 event, a cumulus           Thompson scheme produced higher intensities
parameterization in both 12-km and 4-km runs       than the WSM5 or WSM6 parameterizations.
is needed to reproduce the storm that caused a            Moreover, the sensitivity of nested runs
flash-flood in a coastal area of Catalonia.        to the cumulus parameterization used in their
However, for the September 13 event, neither       parent domain’s run has been shown. This
of the simulations using cumulus schemes in        sensitivity increases with time and exerts
the 12-km and 4-km domains were able to            influence on the location, timing and intensity
reproduce     the    observed     pattern     of   of the rain bands.
      Finally, a comparison between the
operational MM5 12-km run and the WRF 12-
km simulations has been shown. The MM5
forecasts agreed observations better than WRF
during the first hours, but the WRF 12-km run
with explicitly-resolved convection performed
better after these first 6 hours, specially
concerning the rain band structure.
      More       comparisons     between   the
performance of the WRF model and the other
models used operationally at SMC, specifically
MM5 and MASS, have to be done.


The radar imagery has been supplied by
Remote Sensing team of SMC. The
information about domain configuration of
MM5 in SMC was given by Jordi Moré. We
would thank the help and advice given by J.R.
Miró, J. Toda, M. Aran and M. Bravo, from the
Applied Research and Modeling Area of SMC.

Applied Research and Modeling Area,
Meteorological Service of Catalonia (RAM-
SMC), 2005. “Verificació dels models
operatius de mesoescala al SMC”
(Verification of mesoscale operational
models in SMC) (Internal Technical Note)

Barker, D. M., W. Huang, Y. R. Guo, and
Q. N. Xiao., 2004. A Three-Dimensional
(3DVAR) Data Assimilation System For Use
With MM5: Implementation and Initial
Results. Mon. Wea. Rev., 132, 897-914.

Skamarock, W. C., Klemp, J. B., Dudhia, J.,
Gill, D. O., Barker, D. M., Wang, W., and
Powers, J. G., 2005 (revised 2007): A
description of the Advanced Research WRF
Version 2. NCAR Tech Notes-468+STR.

Wang, W., Barker, D., Bray, J., Bruyère, C.,
Duda, M., Dudhia, J., Gill, D., Michalakes,
J., 2007. User’s Guide for Advanced Research
WRF (ARW) Modeling System Version 2.2.
Mesoscale and Microscale Meteorology
Division - National Center for Atmospheric
Research (MMM-NCAR).