A Reanalysis System for the Generation of Mesoscale...
Transcript of A Reanalysis System for the Generation of Mesoscale...
A Reanalysis System for the Generation of Mesoscale Climatographies
ANDREA N. HAHMANN,*,1 DORITA ROSTKIER-EDELSTEIN,*,# THOMAS T. WARNER,*,@
FRANCOIS VANDENBERGHE,* YUBAO LIU,* RICHARD BABARSKY,& AND SCOTT P. SWERDLIN*
* National Center for Atmospheric Research, Boulder, Colorado**1 Risø National Laboratory for Sustainable Energy, Technical University of Denmark, Roskilde, Denmark
@ Department of Atmospheric and Oceanic Sciences, University of Colorado, Boulder, Colorado& National Ground Intelligence Center, Charlottesville, Virginia
(Manuscript received 22 July 2009, in final form 4 November 2009)
ABSTRACT
The use of a mesoscale model–based four-dimensional data assimilation (FDDA) system for generating
mesoscale climatographies is demonstrated. This dynamical downscaling method utilizes the fifth-generation
Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5), wherein
Newtonian relaxation terms in the prognostic equations continually nudge the model solution toward surface
and upper-air observations. When applied to a mesoscale climatography, the system is called Climate-FDDA
(CFDDA). Here, the CFDDA system is used for downscaling eastern Mediterranean climatographies for
January and July. The downscaling method performance is verified by using independent observations of
monthly rainfall, Quick Scatterometer (QuikSCAT) ocean-surface winds, gauge rainfall, and hourly winds from
near-coastal towers. The focus is on the CFDDA system’s ability to represent the frequency distributions of
atmospheric states in addition to time means. The verification of the monthly rainfall climatography shows that
CFDDA captures most of the observed spatial and interannual variability, although the model tends to un-
derestimate rainfall amounts over the sea. The frequency distributions of daily rainfall are also accurately
diagnosed for various regions of the Levant, except that very light rainfall days and heavy precipitation amounts
are overestimated over Lebanon. The verification of the CFDDA against QuikSCAT ocean winds illustrates an
excellent general correspondence between observed and modeled winds, although the CFDDA speeds are
slightly lower than those observed. Over land, CFDDA- and the ECMWF-derived wind climatographies when
compared with mast observations show similar errors related to their inability to properly represent the local
orography and coastline. However, the diurnal variability of the winds is better estimated by CFDDA because
of its higher horizontal resolution.
1. Introduction
Atmospheric models have been used for decades for
filling space and time gaps among observations with var-
ious motivations, for example, long-term reanalyses have
been generated with global-model data assimilation
systems. The resulting model-assimilated datasets (MADS)
are used for analyzing trends, studying physical pro-
cesses, and identifying erroneous observational data. On
smaller scales, mesoscale models have been used in short
simulations for ‘‘case study’’ analyses, where good model
simulation skill at observation locations has been justi-
fication for believing the simulation in the space and
time gaps between the observations. In both cases, the
strength in using the models to fill the observation gaps
is that the fields are dynamically consistent and they are
defined on a regular grid. Additionally, the models re-
spond to local forcing that adds information beyond
what can be represented by the observations.
More recently, the availability of greater computing
power has allowed the generation of long-term mesoscale
reanalyses (Castro et al. 2007; Kanamitsu and Kanamaru
2007; Lo et al. 2008; among many others), where
the resulting gridded datasets are used for various
# Permanent affiliation: Israel Institute of Biological Research,
Ness-Ziona, Israel.
** The National Center for Atmospheric Research is sponsored
by the National Science Foundation.
Corresponding author address: Andrea N. Hahmann, Risø Na-
tional Laboratory for Sustainable Energy, Technical University of
Denmark, P.O. Box 49, 4000 Roskilde, Denmark.
E-mail: [email protected]
954 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 49
DOI: 10.1175/2009JAMC2351.1
� 2010 American Meteorological Society
applications. For example, mesoscale climatographies1
that are based on reanalyses can be used to define the
statistical distribution of wind speeds for wind power
assessment (Hagemann 2008). Boundary layer (BL) cli-
matographies can be used with transport and diffusion
(T&D) or air-quality models to define prevailing pat-
terns in source–receptor relationships. An example of the
latter application is the use of a multimonth mesoscale
reanalysis of BL properties in Iraq to calculate the ex-
posure of civilian and military personnel to potential
releases of Sarin nerve gas during the 1991 Gulf War
(Shi et al. 2004; Warner and Sheu 2000). In a similar
but more-contemporary application, the U.S. National
Ground Intelligence Center (NGIC) routinely performs
40-yr mesoscale reanalyses for various areas of the world
to estimate the T&D of hazardous material released into
the atmosphere for different seasons and various times-
of-day. In addition, such BL climatographies can allow an
assessment of the statistical risk to populations from the
release of hazardous material into the atmosphere from
chemical or nuclear power generation facilities. Last, the
automated interpretation of long-period reconstructions
of the atmosphere can be used to define physical pro-
cesses in ways that are much more robust than what is
obtainable from the use of a few case studies.
Other dynamical downscaling efforts have verified
climatographies of near-surface model winds over land.
Zagar et al. (2006) verified near-surface model winds
obtained by the dynamical downscaling of the European
Centre for Medium-Range Weather Forecasts (ECMWF)
40-yr Reanalysis (ERA40) with a mesoscale model using
a 10-km horizontal grid increment. Verification was per-
formed against wind observations at 10-m height AGL
for a period of 70 days. Kanamitsu and Kanamaru (2007)
have compared near-surface winds from the 57-yr
California reanalysis downscaling on a 10-km grid to the
North American regional reanalysis that used a 32-km
grid. Recently, dynamical downscaling has become a
well-established method for wind power resource as-
sessment purposes (e.g., Landberg et al. 2003), however,
very few objective verifications have been reported in
the literature. Jimenez et al. (2007) performed a verifi-
cation of downscaled near-surface winds from the Na-
tional Centers for Environmental Prediction–National
Center for Atmospheric Research (NCEP–NCAR) Re-
analysis Product by the Pennsylvania State University–
NCAR Mesoscale Model (MM5; their procedure does
not include the assimilation of observations as done in
our technique described in section 2) using measure-
ments taken on masts at several heights up to 130 m AGL.
However, the emphasis in that work was on offshore winds
rather than on winds over the land, where there is the
complication from orographic effects. Hagemann (2008)
conducted an extensive study of the sensitivity of
MM5-simulated winds to various combinations of land
surface and BL parameterizations with emphasis on the
wind power resource assessment in South Africa.
For a MADS to be used for a particular purpose, the
veracity of the analysis needs to be verified in the con-
text of the application. For example, for wind power
generation purposes the model must be able to reason-
ably reproduce the probability distribution function (PDF)
of the wind speed at the height of the generator. Because
the generation equipment is vulnerable to speeds above
a threshold, it is especially important that the model
capture the upper tail of the speed PDF. Similarly, for
air-quality or other T&D applications the BL depth
must be simulated accurately, as must the mean wind in
the BL, and for hydrologic purposes the PDF of rainfall
intensity is needed.
This paper describes a mesoscale reanalysis system
and demonstrates its ability to define unassimilated vari-
ables and fill in data voids. Section 2 describes the system’s
numerical basis, and section 3 summarizes the prevailing
meteorology for the eastern Mediterranean (EM) Sea
and nearby land areas. Section 4 describes the accuracy
of the mesoscale reanalysis of unassimilated variables,
and section 5 contains a summary and conclusions.
2. Numerical basis of the CFDDA system
Given the many existing and potential applications of
high-quality mesoscale MADS, NCAR has adapted its
real-time four-dimensional data assimilation (RTFDDA;
Liu et al. 2008) system for their creation. The RTFDDA
system uses data assimilation by Newtonian relaxation to
generate a mesoscale reanalysis that is consistent with
both observations and model dynamics. Even though it
was originally developed for dynamic initialization of
mesoscale forecasts with gridded datasets that contained
fully developed mesoscale processes, the continuous data
assimilation process in RTFDDA is also ideal for the
generation of mesoscale climatographies. The RTFDDA
technology applied in this way is called the climate-
FDDA (CFDDA) system. Versions of the CFDDA sys-
tem are based on both MM5 and its sequel, the community
Weather Research and Forecast model. One current
application of CFDDA is related to the previously
mentioned mission of the NGIC, which must assess the
potential risk to civilian and military personnel of the
release of hazardous material into the atmosphere for
1 The term climatography is used here to refer to a description of
the climate based on an interpretation of a series of observations in
contrast to the term climatology, which is the study of climate.
MAY 2010 H A H M A N N E T A L . 955
times that are months or years in the future. For this
purpose, the CFDDA climatography is generated for
a region of interest in the world and typical BL condi-
tions are used to define likely directions and speeds of
hazardous material transport for different seasons and
times of day. In addition, by quantifying the variance in
the wind fields over the historical period and inputting
this data to the probabilistic T&D model one can rep-
resent the uncertainty in consequence assessment prod-
ucts that is attributable to historical weather variability.
a. The basic CFDDA system configuration
The CFDDA system is currently based on the MM5,
version 3.6 (Grell et al. 1995). The general configuration
of MM5, as used in the CFDDA system, is summarized
in Table 1—details about specific model characteristics
can be found in the NCAR technical note on MM5
(Grell et al. 1995). In the simulations presented here,
default values have been used for the various parame-
ters in the parameterizations and no attempt has been
made to tune them to better represent the climate of the
region.
b. The data assimilation technique
Data assimilation by Newtonian relaxation is accom-
plished by adding nonphysical nudging terms to the
model predictive equations. These terms force the model
solution at each grid point to observations, or analyses of
observations, in proportion to the difference between
the model solution and the observations or analysis. This
approach is used because it is relatively efficient com-
putationally; it is robust; it allows the model to ingest
data continuously rather than intermittently; the full
model dynamics are part of the assimilation system so
that analyses contain all locally forced mesoscale fea-
tures; and it does not unduly complicate the structure
of the model code. The implementation of Newtonian
relaxation in the CFDDA system forces the model so-
lution toward observations (Stauffer and Seaman 1994)
rather than toward a gridded analysis of the data. This
approach was chosen because observations on the me-
soscale are sometimes sparse and typically are not very
uniformly distributed in space, making objective analy-
sis difficult. With observation nudging, each observation
is ingested into the model at its observed time and loca-
tion, with proper space and time weights, and the model
spreads the information in time and space according to
the model dynamics. In addition, the observational values
are quality controlled against the large-scale analysis
value as described by Liu et al. (2004).
Studies using Newtonian relaxation include Stauffer
et al. (1991), Stauffer and Seaman (1994), Seaman et al.
(1995), and Fast (1995). A common finding in these studies
is that analysis nudging works better than intermittent
assimilation2 on synoptic scales. Stauffer and Seaman
(1994) and Seaman et al. (1995) showed that nudging
toward observations was more successful on the meso-
scale than nudging toward analyses.
c. Model configuration and setup
The model uses 36 computational levels, with approx-
imately 12 levels within the lowest 1 km and with the
model top at 50 hPa. Figure 1 displays the geographic
location of the model grids and the surface elevation on
each domain. The model has a horizontal grid increment
of 45 km on the outer grid (D1) that covers most of the
Mediterranean Sea and extends eastward to cover the
Black and Caspian Seas. The first nested grid (D2),
which has a grid increment of 15 km, covers the EM and
Middle East regions. A third nested domain (D3), with
a grid increment of 5 km, is located along the Levant
(Lebanon, Syria, and Israel). One-way nesting is used to
preserve the horizontal continuity of the results on each
nest.
TABLE 1. Summary of setup and parameterizations used in the CFDDA system.
Nonhydrostatic dynamics
One-way interactive nesting procedure
Radiative upper-boundary condition that mitigates noise resulting from the reflection of vertically propagating waves
Coarse-domain time-dependent lateral-boundary conditions, relaxed toward large-scale reanalyses
Grell (1993) cumulus parameterization on 10-km grid increment, or coarser, grids
Reisner et al. (1998) mixed-phase microphysics parameterization that includes explicit prognostic equations
for cloud water, rainwater, ice particles and snow processes
Modified medium-range forecast model (Hong and Pan 1996; Liu et al. 2006) boundary layer parameterization
Cloud effects on radiative transfer (Dudhia 1989) for shortwave and rapid radiative transfer model (Mlawer et al. 1997) for longwave
Noah land surface model (Chen and Dudhia 2001) with four soil layers
2 Intermittent data assimilation involves restarting the model at
regular intervals, where the initial conditions are typically based on
data within a certain time window being used to correct a first-guess
field that is the forecast from the previous cycle. Restart intervals
may typically be 1–6 h.
956 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 49
The CFDDA simulations cover 10 Januarys and
10 Julys for the period 1998–2007. Individual simulations
are reinitialized every 8.5 days, using the large-scale reanal-
ysis (2.58 3 2.58 grid) of the NCEP–U.S. Department of
Energy reanalysis II (R2; Kanamitsu et al. 2002). These
same analyses are used to calculate the lateral boundary
conditions. Even though the rationale for the cold starts
is that they prevent the accumulation of error in data-
sparse areas, there has been no evidence of this problem.
Each simulation overlaps the previous one by 12 h, to
avoid using the time during which the model is spinning
up mesoscale processes. The choice of 8.5-day in-
tegration segments is due to restrictions on the size of
the lateral boundary condition file and the need to
minimize the error growth in areas with sparse observa-
tions. Qian et al. (2003) and Lo et al. (2008) recommend
this approach. An added advantage of segmenting the
simulation is that it allows for simultaneous integrations
for various time periods.
Initial land surface conditions are based on the Na-
tional Aeronautics and Space Administration (NASA)
Global Land Data Assimilation System (GLDAS; Rodell
et al. 2004) dataset that is defined on a 18 3 18 grid. The
GLDAS fields used here are substrate soil moisture and
temperature at depths of 10 and 200 cm, ground skin
temperature, and snow water equivalent. These fields
are interpolated to all land grid points on each model
domain and replace the standard values defined in the
R2 analysis. Because the same land surface scheme is
used in both GLDAS and MM5, no soil moisture con-
version is required. Sea surface temperatures (SSTs) are
interpolated to CFDDA ocean grid points on each model
domain using the NCEP Version 2.0 global SST dataset
(Reynolds et al. 2002), which is defined on a 18 3 18 grid
and updated daily.
d. Data sources for assimilated observations
The type, spatial density, and frequency of observa-
tions used by the CFDDA system vary greatly in different
parts of the world. The Advanced Data Processing
datasets, available from the NCAR Computational and
Information Systems Laboratory’s (CISL) Data Support
Section, are employed. The data represent a global syn-
optic set of surface hourly and 6-hourly data reports,
operationally collected by NCEP. The surface dataset
includes mostly surface synoptic observations (SYNOP)
and Meteorological Aerodrome Report (METAR) land
reports but a few ship observations also exist. The upper-
air data consist mainly of radiosonde soundings.
Figure 2 displays the geographic data distribution and
the approximate frequency of data availability in January
2000 for the model’s outermost domain. The map rep-
resenting surface data (Fig. 2a) shows a very inho-
mogeneous geographic distribution, with most of the
observations having greater than 3-h time resolution
located over western and central Europe—the lowest
concentration of observations is found over northern
Africa away from the coast. Both the Mediterranean
and Red Seas show a few locations with oceanic data
from ships, but their time resolution is poor (less than
1 observation per day on average). During January 2000
in this region, there were over 615 000 observations at
FIG. 1. CFDDA system configuration and terrain elevation of
(a) outermost domain (Dx 5 45 km; D1) and (b) innermost domain
(Dx 5 5 km; D3). The outer edges of D2 and D3 are shown by the
two rectangles in (a). The dots in (b) indicate the approximate
location of rain gauges used in the verification section of this paper.
Black squares at A and B mark the location of the wind towers.
MAY 2010 H A H M A N N E T A L . 957
over 2000 stations. The geographic distribution of upper-
air observations in Fig. 2b displays a similar pattern, with
the most frequently reporting (2–4 times a day) stations
over western Europe. Over the region covered by D3,
only four radiosonde sites are available, with an aver-
aged frequency of 1–2 observations per day.
3. Prevailing meteorological processes in the EM
The EM, located within the transition area between
the subtropical high-pressure belt and the midlatitude
westerlies, is very interesting and meteorologically com-
plex. Superimposed on the large-scale processes are
mesoscale effects related to the complex coastline and
nearby mountains (see terrain elevation in Fig. 1b). The
cold-season precipitation in this area is associated with
Mediterranean cyclones, for which there are many cli-
matological descriptions (e.g., Alpert et al. 1990; Trigo
et al. 1999; Genoves et al. 2006). There are strong near-
coastal gradients in observed precipitation, the positions
of which have huge hydrologic consequences. These
gradients are due to the preferred tracks followed by the
cyclones, their intensity, and their interaction with the
local topography (e.g., Saaroni et al. 2010). The north-
eastern and eastern coast of the Mediterranean basin is
rich in cyclogenetic regions, resulting in considerable
precipitation amounts. The opposite is true in the south-
eastern coast, leading to scarce precipitations and desert
climate characteristics. Some adjacent lands to the north-
ern and eastern Mediterranean coastal areas are rich in
cyclonic activity as well, leading to significant precipi-
tation amounts during the cold season—that is, the south
coast of the Black sea and the Fertile Crescent extend-
ing throughout north Levant to the Persian Gulf. Thus,
proper reproduction of the winter climate over this region
compels correct simulation of the synoptic flow domi-
nated by extratropical cyclones as well as the mesoscale
flow dictated by the complex coast and terrain forcing.
Throughout the summer, the EM is dominated by a
subtropical ridge extending from the north-African coast
to the east and by a Persian trough extending from the
monsoonal low through the Persian Gulf to the north-
east Mediterranean and Turkey (Alpert et al. 1992). The
synoptic-scale near-surface wind direction is generally
westerly and is modulated by the sea breeze. This results
in a westerly to northwesterly wind flow of nearly 7 m s21
during the daytime. In addition to the synoptic flow, the
westerly daytime sea breeze is accelerated by anabatic
winds forced by the inland mountainous regions (Alpert
and Rabinovich-Hadar 2003; Doron and Neumann 1977;
Berkovic and Feliks 2005). The easterly land breeze
develops along the coast, because of the sea–land dif-
ferential cooling after sunset. As the night progresses,
katabatic flow originating on the western slopes of the
mountain ridges (;30 km inland in the south of Israel
and adjacent to the coast in the north) reaches the coast,
increasing the wind intensity.
4. Verification of CFDDA for the easternMediterranean
This paper assesses the success of the downscaling sys-
tem by comparing its climatographies to various datasets
that have not been assimilated during the model inte-
gration. The verification focuses on the ability of the
downscaling system to represent the frequency distri-
bution of atmospheric states in addition to the simple
time means.
Because precipitation data are not assimilated and
many aspects of model physics typically need to operate
properly for precipitation to be correctly simulated,
FIG. 2. Geographic location and time frequency of (a) surface
and (b) upper-air observations assimilated in the CFDDA system
within the domain during January 2000. The various symbols rep-
resent the approximate temporal frequency of the observations.
958 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 49
verifying the simulated precipitation fields is a good test
of the CFDDA system’s ability to define unobserved or
poorly observed fields. We verify the spatial distribution
of monthly multiyear totals, the interannual variability
of regional precipitation totals, and the frequency distri-
bution of daily rainfall totals against gridded and gauge
rainfall totals. Also evaluated are the CFDDA clima-
tography of the 10-m winds over the Mediterranean Sea
using NASA Quick Scatterometer (QuikSCAT) mea-
surements (OBS). Coastal wind tower observations are
also used for verification.
a. Large-scale circulation
Before focusing on the details of the validation of
the CFDDA high-resolution simulations, it is impor-
tant to assess the ability of the modeling system to
correctly represent the large-scale features of the cir-
culation over this particular region. To accomplish this,
the 6-hourly real-time analyses from the ECMWF have
been used to provide an estimate of the large-scale
circulation. This ECMWF operational analysis is avail-
able at 1.1258 3 1.1258 resolution. Figure 3 displays the
anomaly correlation between the ECMWF analysis and
the corresponding CFDDA-D1 fields interpolated to
the ECMWF grid. Correlations are plotted every 6 h for
each January and July for 1998–2007 for the 700-hPa
temperature and relative-humidity fields.
The large-scale thermal and moisture patterns simu-
lated by the CFDDA modeling system closely correlate
with those defined by the ECMWF analysis. The anomaly
correlation values for 700-hPa temperature (Figs. 3a,b)
average 0.985, with higher overall values in winter than
in summer. The same seasonal trend is true for the
anomaly correlations for 700-hPa relative humidity (RH;
Figs. 3c,d), which has an average of 0.83. Higher in the
troposphere, anomaly correlation values (not shown)
are even larger. The anomaly correlations for each year
of the simulation are displayed separately, and it can be
seen that there is a tendency for higher correlations to
exist for later years. This is probably a consequence of
the increased number and quality of observations as-
similated in both the ECMWF and CFDDA analysis
and improvements in the ECMWF analysis cycle. For
January 2006, a CFDDA simulation with no assimila-
tion of point observations was conducted. The anomaly
correlation values decreased to 0.85–0.9 for the 700-hPa
temperature and to 0.55–0.7 for the RH. This result
emphasizes the importance of the FDDA portion of
FIG. 3. Time series of anomaly correlation between upper-air fields from the ECMWF analysis and those simulated by the CFDDA-D1
system for the period 1998–2007. Correlations are for 700-hPa temperature for (a) January and (b) July, and 700-hPa relative humidity for
(c) January and (d) July.
MAY 2010 H A H M A N N E T A L . 959
the system for constraining the simulation of the large
scales.
b. Monthly precipitation climatography
The model-derived monthly rainfall totals are com-
pared to three different gridded estimates of monthly
rainfall. For gauge-only estimates, we employ the Global
Precipitation Climatology Centre (GPCC; Beck et al.
2005) dataset, which provides monthly precipitation to-
tals from 1901 to the present on a 0.58 3 0.58 global grid
but only over land. Because a considerable fraction of
the CFDDA domain is located over the sea, we also
utilize the merged Tropical Rainfall Measuring Mission
(TRMM) product (3B43; Huffman et al. 2007). These
data combine precipitation estimates from multiple sat-
ellites (retrievals from measurements in the microwave
and infrared regions of the spectrum) as well as gauge-
based analyses on a 0.258 3 0.258 grid that extends from
508N–508S for the period from 1998 to the present. Fi-
nally, following the work of Ebert et al. (2007), we also
use precipitation derived from the short-range forecasts
(i.e., the accumulated precipitation during the forecast
period between 12 and 36 h) from the ECMWF model,
available daily on a 1.1258 3 1.1258 global grid. These
daily estimates were aggregated to produce a monthly
rainfall total and have an accuracy that is comparable to
that of satellite estimates, especially in winter and at
higher latitudes (Ebert et al. 2007). The focus here is
mainly in the geographic distribution of rainfall; there-
fore, we have chosen not to interpolate the data from the
various sources to a common grid. A more quantitative
analysis will be done later.
Figures 4a–c show the average (1998–2007) January
precipitation amount based on the three observational
datasets described above. Large precipitation maxima
over land are seen along the coastlines of the Levant,
FIG. 4. Total monthly rainfall (mm) averaged during 10 Januarys (1998–2007) for (a) 0.58 3 0.58 GPCC, (b) 0.258 3 0.258 3B43 merged
TRMM, (c) 12–36-h forecast rainfall from ECMWF model, and (d) CFDDA-D1 simulations. The two black boxes in (a) outline the areas
averaged in Fig. 7.
960 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 49
Turkey, and along the west coast of the Balkan states.
There is also precipitation over the Mediterranean Sea
but in various degrees from TRMM (Fig. 4b) or ECMWF
(Fig. 4c). A distinct line of maximum rainfall is also
apparent along the Fertile Crescent. The precipitation
gradient is very large perpendicular to the North Afri-
can coastline, with values of about 50 mm along the
Libyan coast, decreasing to less than 5 mm across the
interior just 100 km inland. The gradient is somewhat
less along the Levant and Turkey coasts, but it is still
large. Overall, the precipitation estimated by CFDDA
(Fig. 4d) represents most of these spatial patterns, ex-
cept for the observed coastal maximum over Algeria
and France (upper-left edge of the domain). There is less
precipitation estimated by CFDDA over the central and
southern Mediterranean than is estimated by TRMM
(more tan color, Fig. 4b). However, verification of pre-
cipitation over the Mediterranean Sea is problematic
because the TRMM dataset has a known negative bias
to low-precipitation rates over the oceans (Huffman
et al. 2007).
During July (Fig. 5), observations show areas of en-
hanced precipitation along the eastern Black Sea coasts
of Turkey and Georgia. A secondary region is seen over
Romania, extending northward and southwestward to
the Balkan countries. The largest maximum is located
over the Alps, along the northern Italian border. All
observational estimates show no significant rainfall be-
low about 358N and considerably more rainfall over land
than over oceanic regions. Overall, a very similar pattern
is seen in the CFDDA rainfall climatography (Fig. 5d).
However, the model rainfall has more small-scale vari-
ability than that observed by the blended TRMM prod-
uct that is shown at a similar spatial resolution (Fig. 5b).
For example, in the CFDDA-derived rainfall climato-
graphy, a series of northwest to southeast oriented areas
of maximum precipitation are seen over Romania and
southern and western Ukraine. Similar maxima, but more
FIG. 5. Total monthly rainfall (mm) averaged during 10 Julys (1998–2007) for (a) 0.58 3 0.58 GPCC, (b) 0.258 3 0.258 3B43 merged TRMM,
(c) 12–36-h forecast rainfall from ECMWF model, and (d) CFDDA-D1 simulations.
MAY 2010 H A H M A N N E T A L . 961
widespread in character, are seen in the GPCC and
TRMM climatographies (Figs. 5a,b, respectively) and a
similar pattern is seen in the ECMWF-forecasted rain-
fall (Fig. 5c).
c. Variability of daily rainfall
Daily rainfall totals simulated by the CFDDA system
are verified against a database of daily precipitation to-
tals from rain gauge stations for the geographic region
covered by D3. The locations of the rainfall stations are
shown in Fig. 1b. The data for a majority of these rain
gauge observations were obtained from the Israeli Min-
istry of Agriculture. The dataset was augmented (in
particular for the stations over Lebanon) with daily
rainfall totals from the National Oceanic and Atmo-
spheric Administration global surface summary of the
day (GSOD). The analysis is conducted with only sta-
tions that have data for at least 70% of the days in the
10 Januarys covering the period from 1998 to 2007.
Rainfall in this region during the summer months is vir-
tually nonexistent; thus, we have limited our analysis to
the winter season.
During winter, a marked north–south gradient in rain-
fall exists in D3, with the larger totals in the northern
Levant and lesser amounts toward the south. This gra-
dient is mainly a consequence of the preferred path of
extratropical cyclones (Alpert et al. 1990; Trigo et al.
1999; Schadler and Sasse 2006). The rainfall stations have
been classified into three groups based on this north–
south contrast. A summary of the rainfall statistics for
each group of stations is presented in Table 2, with
the lowest observed monthly rainfall totals in southern
Israel (stations 1–6) and the largest observed values in
northern Israel (stations 7–12). The CFDDA-D3 and
the ECMWF operational 12–36-h forecast daily rainfall
totals have been interpolated to the station locations
for those days with observed values—only nonzero pre-
cipitation days in both models and observations are used
in the statistics and subsequent frequency distributions.
The simulated CFDDA monthly rainfall totals compare
well with those observed over the two regions in Israel, but
they overestimate the total over Lebanon. The ECMWF
forecasts underestimate rainfall totals in all three re-
gions. The mean daily total rainfall for rainy days is
underestimated by CFDDA for Israel but values are
better than those from ECMWF in all three regions.
Table 2 also presents the percentage of ‘‘drizzle’’ days
with precipitation of less than 0.1 mm but not including
days with zero precipitation. This varies from 13% in
southern Israel to almost zero over Lebanon. This value
is reasonably well simulated for Israel by both CFDDA
and ECMWF, but both models severely overestimated
this percentage for Lebanon. Similar behavior is seen
in other weather and climate models (Chen et al. 1996;
Knievel et al. 2004).
The frequency distribution of observed daily rainfall
totals was compared for each group of stations to the
distributions from the CFDDA simulations and ECMWF
forecasts. For southern Israel (Fig. 6a) the observed
distribution shows that most of the rainfall events result
in daily totals of 4–8 mm; for the northern Israel stations
(Fig. 6b) and the stations in Lebanon (Fig. 6c) the dis-
tributions are shifted toward 8–16 mm of daily rainfall.
The northern Israel stations show a few events (;10)
with daily rainfall totals above 64 mm, with fewer such
events in southern Israel and Lebanon. The CFDDA
distribution of daily rainfall totals in categories above
1–2 mm is similar to the observed distribution; however,
CFDDA consistently overestimates the number of light
rainfall events (0.1–1 mm) in all regions, perhaps by a
similar mechanism to that responsible for the overpre-
diction of drizzle days over Lebanon. The consistency
between CFDDA and the observed distribution is par-
ticularly good for the northern Israel stations (Fig. 6b),
with errors of less than 15 events in all classes above
2 mm. In Lebanon, the CFDDA system overestimates
the values in the higher-intensity classes (.16 mm). This
may be due to a lack of representativeness of the ob-
served rainfall totals in stations located in complex terrain
(Chen et al. 2002). Furthermore, rainfall totals for the
Lebanon stations were derived from the GSOD archive,
which has only minimal quality control of the measure-
ments. Similar error statistics have been found in other
MM5 studies at similar resolutions over complex terrain
(Colle and Mass 2000; Mass et al. 2002).
TABLE 2. Rainfall statistics for January 1998–2007 for the three subregions in Fig. 1b.
Statistic Southern Israel Northern Israel Lebanon
OBS CFDDA ECMWF OBS CFDDA ECMWF OBS CFDDA ECMWF
Total monthly rainfall (mm) 108.6 121.7 70.7 169.3 167.6 103.5 152.6 269.4 124.3
Mean daily rainfall
intensity (mm day21)
7.0 4.9 3.5 10.8 7.0 5.0 10.9 10.4 5.6
Fraction of days with
0 , p , 0.1 mm (%)
12.9 13.2 10.0 6.3 10.0 7.7 0.1 10.1 8.0
962 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 49
The rainfall from the ECMWF forecasts consistently
has excessively high frequencies in the low-rainfall classes
and too low frequencies for the higher precipitation
amounts for all station groups. Daily amounts above
32 mm, which are not rare in the observed or CFDDA
distributions, are almost nonexistent in the ECMWF
daily rainfall totals. Therefore, although the monthly
rainfall values forecast by ECMWF are reasonably close
to those observed, the rainfall occurs too often and with
a lower-than-observed intensity.
d. Interannual variability of rainfall totals
The gridded January rainfall data from observations
and models are compared in terms of their interannual
variability for two geographic regions: the Levant (318–
368N, 368–378E) and southwest Turkey (368–388N, 258–
328E), which are delineated by black boxes in Fig. 4a.
These areas were chosen because of the large prevail-
ing rainfall amounts in January (Figs. 4b,c). The area-
averaged rainfall for these two regions during all Januarys
for the period 1998–2007 is displayed in Fig. 7. Because
the CPCC rainfall estimate is available for land only,
and TRMM data are available for land and water, for
CFDDA we display both the values averaged over all
grid points and over land-only grid points. In most years,
these two values are very similar over coastal Turkey but
the averages over the Levant using D2 rainfall data differ
significantly. Over the Levant, the CFDDA-D2 and
ECMWF rainfall totals are consistently larger than the
TRMM estimates for all years. Over southwest Turkey,
the TRMM estimates are often larger than the CFDDA
and ECMWF values. The CFDDA-D2 precipitation
values are uniformly larger than those for D1, for both
regions and all years. Considerable interannual variabil-
ity is seen in the observed precipitation in both regions,
and the variation from one year to the next in both re-
gions is well represented in the CFDDA and ECMWF
rainfall.
e. Near-surface winds over the ocean
To verify the CFDDA-simulated low-level winds over
the ocean, we use data from the NASA QuikSCAT
satellite. The SeaWinds instrument on QuikSCAT is an
active microwave radar that measures the backscatter
from surface ocean waves, and winds can be obtained in
FIG. 6. Distribution of daily rainfall totals for observations and
model simulations from CFDDA-D3 and 12–36-h forecast from
the ECMWF model for the stations shown in Fig. 1b located in (a)
southern Israel (stations 1–6), (b) northern Israel (stations 7–12),
and (c) Lebanon (stations 13–15).
MAY 2010 H A H M A N N E T A L . 963
all conditions except for moderate to heavy rain (Hoffman
and Leidner 2005). A function is used to relate the
measured backscatter to the 10 m MSL neutral stability
equivalent winds. The rms differences between quality-
controlled ship winds and QuikSCAT winds are typically
about 1 m s21 in speed and 158 in direction (Bourassa
et al. 2003). Very high- and very low-speed winds have
problematic retrievals. In particular, low speeds result in
very poor directional skill, while winds above 25 m s21
are often underestimated (Hoffman and Leidner 2005).
Also, because of the previous assumption of neutral sta-
bility, winds of less than 6 m s21 are underestimated by
20%–30% under unstable conditions, which are often
found over the Mediterranean during winter (Zecchetto
and De Biasio 2007).
In the validation presented here, we use the QuikSCAT
level-3 daily, gridded ocean wind vectors. These data are
provided on an approximate 0.58 3 0.58 global grid with
separate maps for the ascending (0600 LT equator cross-
ing) and descending (1800 LT equator crossing) passes.
The data are available for the period 2000–present. The
CFDDA 10-m winds for each domain are linearly in-
terpolated to each individual QuikSCAT wind vector
location, and matched to the closest valid time of the
observation. Figures 8 and 9 compare the 10-m wind
speed composites from QuikSCAT and CFDDA-D2 for
January and July 2000–07, respectively. During January,
FIG. 7. Area-averaged January precipitation (mm) as a function
of the year (1998–2007) for (a) Levant (318–368N, 348–378E; right
rectangle in Fig. 4a) and (b) SW Turkey (368–388N, 258–328E; left
rectangle in Fig. 4a) for CFDDA system simulations over domains
D1 and D2, and for all points vs land-only (L) points for observa-
tional estimates from GCPP and TRMM and the 12–36-h forecasts
from the ECMWF model.
FIG. 8. January (2000–07) composite of 10-m winds (m s21) from
(a) QuikSCAT, (b) CFDDA-D2 system simulations, and (c) the
difference between the CFDDA-D2 and QuikSCAT composites.
The black rectangles in (c) mark the locations of the data used in
the wind roses in Figs. 10 and 11.
964 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 49
the observed wind climatography is dominated by the
northwesterly Etesian winds (Ziv et al. 2004) in the
Levantine Basin and the mainly northerly flow in the Black
Sea area. During winter, the winds in the entire region are
characterized by low steadiness (a measure of the con-
stancy of the wind direction) resulting from the frequent
passage of fronts (Zecchetto and De Biasio 2007). The
observed mean 10-m wind speed is 6–10 m s21 across the
Mediterranean Sea (Fig. 8a), with speeds decreasing
toward the southeast. Over the Black Sea, wind speeds
range from up to 10 m s21 in the west to 4 m s21 in the
east. The CFDDA 10-m winds (Fig. 8b) display a similar
geographic distribution but the speed is underestimated
by 1–2 m s21. In summer (Fig. 9), when conditions over
the Mediterranean Sea are dominated by steady local
winds, the CFDDA system is able to capture very well
the wind distribution observed by QuikSCAT. The area
of wind funneling through the strait separating the islands
of Crete and Rhodes (Zecchetto and De Biasio 2007) is
clearly seen in the CFDDA-derived analysis. Mean ab-
solute differences between QuikSCAT and the model are
mostly below 1 m s21 over more than 90% of the ocean
grid points in D2.
To further examine the model-simulated winds as
they relate to those observed by QuikSCAT, we show
wind roses of model-derived 10-m winds and observed
winds for two 58 3 58 regions in the EM. Solid lines in
Fig. 8c show the two regions. In Fig. 10, the wind roses
for January 2000–07 (combining both QuikSCAT ascent
and descent times) for the area 328–378N and 258–308E
(left box in Fig. 8c) are shown for QuikSCAT and
CFDDA-D2. Observed winds are mostly from the west
and northwest, with the most frequent high speeds (av-
eraging 9 m s21) being from the northwest. The roses
generated from the CFDDA-D2 winds show very simi-
lar directional frequency distributions but the model un-
derestimates the frequency of the strongest wind class.
Almost identical results are seen in the CFDDA-D1
winds (not shown). In Fig. 11, the wind roses generated
with data from July 2000–2007 for a region closer to the
EM coast (328–378N, 308–358E, right box in Fig. 8c) are
shown. This time, separate wind roses are shown for the
morning (ascent) and evening (descent) overpasses to
isolate diurnal effects. During this time of the year,
westerlies dominate with stronger winds more likely in
the evening than in the morning. The CFDDA-derived
winds reproduce this diurnal variation quite well, espe-
cially at higher resolution (Figs. 11c,f). As in winter, the
model underestimates the speeds. However, the differ-
ences are more evident in the morning than in the
evening. Because very few observations are available
over the sea (Fig. 2a), only a limited impact of data as-
similation is expected in these regions.
f. Boundary layer winds over land
Here, the CFDDA wind speed distributions for January
and July are compared to observations on two towers.
Data from these towers were not assimilated in the model
integration. In addition, we verify the winds from the
6-hourly real-time ECMWF analyses at a resolution of
1.1258. The aim of this comparison is to assess the ca-
pability of CFDDA to reproduce wind speed statistics
that are relevant to wind power generation applications.
The observations used in the verification consist of
winds measured at the two locations depicted in Fig. 1b.
Hourly observations at 65 m AGL were available for
January and July 1999–2007 at station A, and half-hourly
observations at 60 m AGL were available for January
and July 1998–2007 at station B. Both stations are situ-
ated at a distance of about 8 km inland from the coast,
FIG. 9. As in Fig. 8, but for July.
MAY 2010 H A H M A N N E T A L . 965
in areas characterized by a mixed canopy consisting of
urban buildings, shrubs, and trees, with a roughness length
of approximately 0.3 m. Station A is located in an area of
rather flat terrain—the tower base is 62 m MSL and the
coastline is relatively straight. Station B is situated in
a region of complex terrain and complex coastline con-
figuration and the tower base is at 60 m MSL. The latter
area has been described in Rostkier-Edelstein and Givati
(2003). Although neither of these observations is located
in open rural areas suitable for wind power production,
we opted to verify CFDDA against them because they
provide a high-quality multiyear dataset at heights rele-
vant to wind power applications.
The CFDDA-derived winds from D3 were interpolated
to the location of the observations. The CFDDA terrain
heights at the location of the stations are 36 m and 51 m
MSL, for stations A and B, respectively. The lowest
three CFDDA levels are located at approximately 15,
57, and 113 m AGL at both locations, and winds were
vertically interpolated with respect to height using log-
arithmic weights, between the latter two levels, to the
height of the masts. Therefore, the height AGL of the
observed and interpolated wind is the same despite
the elevation difference of the model terrain and the real
terrain. Land grid squares (mainly urban type, with as-
signed roughness length of 1 m) surround the location of
station A in D3. In contrast, station B is surrounded on
three sides by land grid squares, except for the northwest
quadrant, where over 40% of all modeled winds originate,
which is an ocean grid square (with an assigned back-
ground roughness of 0.001 m).
Similarly, ECMWF analysis winds archived on the
1.1258 grid were interpolated to the stations locations.
Because of changes in model resolution, the terrain heights
of the ECMWF analyses at the observation locations
changed during the study period. The ECMWF terrain
height values were appreciably higher than the actual
ones (159–214 m MSL and 171–216 m MSL for stations
A and B, respectively). The ECMWF wind values used
for comparison with the observations are based on log–
height interpolation from the pressure levels of 1000
and 925 hPa. The model that generated the operational
ECMWF analysis system used here underwent changes
in horizontal and vertical resolution, model physics, and
data assimilation during the period 1998–2007. There-
fore, no temporal consistency is expected, but the wind
fields provide the ‘‘best’’ baseline analysis at the time.
Figure 12 displays the wind speed distributions (for all
6 hourly data) based on CFDDA-D3, ECMWF, and
observations for both locations during January and July.
Both models reproduce the observed differences in the
wind speed distributions between January and July. The
results for January (Figs. 12a,b) are characterized by
FIG. 10. Wind roses generated from January 2000–07 winds from
(a) QuikSCAT retrievals (OBS), and (b) CFDDA-D2 (15 km).
The winds are restricted to the area between 328–378N and
258–308E (left square in Fig. 8c) and include both ascent and descent
times. The wind rose radius represents the wind frequency (%) for
each sector and speed range as presented in the grayscale bar below.
966 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 49
a wide distribution of wind speeds (relative to those for
July) with a high-speed tail showing a few observed
events that exceed 10 m s21. These events are associated
with strong pressure gradients within extratropical cy-
clones. The periods between the cyclonic activity, which
are mainly dominated by Red Sea troughs and winter
highs, lead to weaker winds. The observed distributions
for this month show no significant differences between
the two sites, in spite of the differences in the orography
and coastline characteristics. In winter, in the absence of
strong solar heating, no pronounced land–sea tempera-
ture gradients develop and local thermal circulations are
likely weak. Near-surface winds are thus dominated by
the synoptic flow. Both models are able to reproduce the
broad observed distributions. However, there is some
model-dependent disagreement between the modeled
and observed distributions, and some differences exist
between the model solutions for the two sites. The
ECMWF analysis tends to overestimate the frequency
for the low-speed part of the January distributions at
both sites and underestimates the frequency at higher
speeds. The PDFs for CFDDA show underestimation of
the speeds in the middle of the distribution and over-
estimation for winds above 7 m s21 at both sites. The
CFDDA terrain and coastline (not shown, but partly
described previously) is not able to correctly represent
the ridge and bay in the vicinity of station B, as well as
the correct distance to the coast. As a result, the site is
located much closer to the coast in the model than in
reality and is thus affected by lower roughness upstream
during periods of strong winds.
The observed wind distributions for July (Figs. 12c,d)
are narrower than for January, with speeds that do not
exceed 11 m s21. This is a result of the summer weather
FIG. 11. Wind roses generated from July 2000–07 winds derived from (a) QuikSCAT ascent retrievals (OBS), (b) CFDDA-D1 (45 km)
ascent times, (c) CFDDA-D2 (15 km) ascent times, (d) QuikSCAT descent retrievals (OBS), (e) CFDDA-D1 (45 km) descent times, and
(f) CFDDA-D2 (15 km) descent times. The winds are restricted to the area between 328 and 378N and between 308 and 358E (right box on
Fig. 8c). The wind rose radius represents the wind frequency (%) for each sector and speed range in the grayscale bar.
MAY 2010 H A H M A N N E T A L . 967
regime that is characterized by weak westerly to north-
westerly large-scale flow (Section 3), upon which sea–land
breeze and terrain-induced circulations are superimposed.
The observed distributions show some differences be-
tween the two nearby sites, presumably because of the
local orography and coastline. For example, while sta-
tion A (Fig. 12c) shows a maximum frequency for speeds
of about 3 m s21, station B (Fig. 12d) shows a bi-
modal behavior with maximum frequencies for speeds
of 1–2 m s21 and 4–6 m s21. For summer, the ECMWF
product overestimates some of the low- and medium-
speed categories and underestimates higher speeds. The
CFDDA wind distribution for station A is too flat, greatly
underestimating the speeds in the 3–5 m s21 range. For
station B, the modeled distribution is insufficiently flat.
The model does produce a bimodal distribution but the
minimum is at too high a speed.
To better understand the role of the land–sea breeze
in generating the observed wind distributions, and to
assess the ability of the models to reproduce it, Fig. 13
displays the observed and model histograms for wind
speeds at station A for 0000, 0600, and 1200 UTC for
July. These times are not necessarily ideal for showing the
phases of the land–sea breeze but rather are dictated by
the availability of the ECMWF data. Figure 13a (0300 LT)
shows the wind speed distributions at a time dominated by
the land breeze, which is normally weaker than the sea
breeze. The observed wind distribution shows a significant
frequency of calm winds (,1 m s21) and speeds never
exceed 6 m s21. The ECMWF-derived winds fail to re-
produce the calm speeds and most of the events fall in the
2–4 m s21 category. In contrast, CFDDA-derived winds
better capture the distribution. The coarse resolution of
the ECWMF analyses most likely masks the land breeze
FIG. 12. Frequency distributions of 6-hourly wind speeds from observations, and simulated by the CFDDA-D3
(5 km) and ECMWF operational forecast model for (a) station A and (b) station B during January 1999–2007 and
(c) station A and (d) station B during July 1998–2007.
968 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 49
and terrain-induced flow at this time. In the absence of
these, no opposing flow weakens the large-scale westerly
to northwesterly flow. At 0900 LT (Fig. 13b), the sea
breeze begins to develop, with higher observed wind
speeds. The ECMWF winds remain rather similar to
those at 0300 LT. In contrast, CFDDA winds correctly
shift toward higher speeds but the peak is not sufficiently
large. At 1500 LT (Fig. 13c), all three distributions show
a narrow profile, with speeds in the range 3–9 m s21.
The winds derived from the ECMWF analysis still un-
derestimate the observed frequency peak and the higher
speeds.
5. Summary and discussion
We present a method for generating high-resolution
regional climatographies through the use of a mesoscale
model driven by R2 and the assimilation of surface and
upper-air observations by Newtonian relaxation. The re-
sults of 10 yr of simulations for January and July are
verified in terms of the horizontal and seasonal distri-
bution of rainfall and the near-surface wind field, both
over the ocean and at two land sites.
The verification of monthly rainfall shows that CFDDA
is able to correctly capture most of the geographic fea-
tures and interannual variability of the precipitation
field in this region. Over the oceans, CFDDA appears to
underrepresent the monthly rainfall; however, the error
in the CFDDA estimates as compared to those from
TRMM is within the uncertainty of these satellite esti-
mates over the oceans. Daily rainfall totals from the
CFDDA system are verified against rain gauge obser-
vations, and the results show remarkable agreement
both in terms of the mean amount but also in terms of
their temporal distribution for three distinct regions. As
seen in many other models, the CFDDA system appears
to overestimate the amount of precipitation over higher
elevations. It is unclear what the origins of this problem
in the model simulation are, or whether the observations
underestimate the true amount of rainfall of this region
by undercatchement errors (Colle and Mass 2000) or
preferential placement of the verifying gauges in lower
elevation valley locations. Nevertheless, the high quality
of the depiction of the rainfall distribution in most places
by the CFDDA system demonstrates its possible use-
fulness in coupling to hydrological models.
FIG. 13. Frequency distributions of 6-hourly wind speeds
from observation and simulated by the CFDDA-D3 (5 km) and
ECMWF forecast model from station A during July 1999–2007 for
(a) 0000, (b) 0600, and (c) 1200 UTC.
MAY 2010 H A H M A N N E T A L . 969
Wind roses constructed with wind vectors from CFDDA
show considerable accuracy when compared with winds
derived from QuikSCAT observations over the EM Sea.
Although the directional distribution of winds is well
represented, wind speeds simulated by the CFDDA
system over the oceans are consistently slightly lower
than those observed by QuikSCAT. The surface-layer
formulation within the CFDDA system uses Charnock’s
formula as written by Delsol et al. (1971) to compute the
surface roughness over the ocean as a function of the
wind speed. However, the constant used in MM5 is twice
as large as that based on oceanic data (Wu 1969). Colle
et al. (2003) reported similar biases in their verification
of wind speeds over the Gulf of Mexico and the eastern
Atlantic.
The comparison between CFDDA, ECMWF, and
observed wind speed distributions at two mast locations
is accomplished through the construction of wind speed
histograms for observed and simulated data. In this
comparison, representativeness error is not accounted for.
Clearly, both the ECMWF and CFDDA analyses fail to
truthfully represent the terrain (mainly at station B), the
coastline (mainly at station B), and the roughness
characteristics of the areas surrounding the masts ob-
servations. Wind measurements may be exposed to local
influences limiting their spatial representativeness to a
radius of only a few hundreds of meters. Some estimates
of representativeness error for near-surface winds com-
puted with a model grid increment of about 1 km lead
to a value of about 1 m s21 for areas of complex terrain
(e.g., Rife et al. 2004; Steinacker et al. 2000). Landberg
et al. (2003) discussed the nature of these local effects and
described methods to overcome them for wind resource
estimation. These methods rely on dynamical downscal-
ing using a mesoscale model and further fine tuning of the
wind climatographies with a microscale model (Frank
and Landberg 1997).
The verification results, despite the discussed caveats,
confirm the potential of using CFDDA reanalyses as a
tool for wind power resource estimation. Assimilated
surface and upper-air observations are sparse in the vi-
cinity of the masts, thus, showing the capability of the
mesoscale reanalysis to reproduce near-surface wind
statistics in poorly observed areas. To the best of our
knowledge, this is one of the few published verifications
of dynamical downscaled winds over land at turbine
height, in the context of wind resource assessment. It
demonstrates the benefit gained by high-resolution meso-
scale reanalysis as compared with coarse-resolution global
analyses, in particular when the circulation is dictated
by mesoscale mechanisms. Some of the limitations
encountered may be overcome by optimization of the
CFDDA model parameters for the specific domain, for
example, terrain, coastline, surface canopy, and specific
parameters in the physical process schemes. Others
limitations are imposed by the model horizontal grid
size, that is, the representativeness errors due to local
effects. These can be overcome by increasing the hor-
izontal resolution further or by postprocessing the
CFDDA reanalysis with statistical corrections or mi-
croscale modeling.
Many factors play a role in the quality of the meso-
scale reanalysis system: the quality of the mesoscale
model and its physical parameterizations, the quality of
lateral and lower boundary conditions, the quality and
spatial density of observations used in the data assimi-
lation, the increased horizontal and temporal resolution,
etc. It is beyond the scope of the paper to examine the
individual contribution of each of these, but we believe
all are necessary components of a credible system.
In the present study, no attempt has been made to
optimize the model parameterizations for the weather
conditions and climate of the EM. The choice of these is
based on our experience from real-time mesogamma-
scale weather forecasting in the United States (Liu et al.
2008). That type of optimization process, which is be-
yond the scope of this work, would very likely improve
the accuracy of the climatography derived from the
CFDDA system.
Acknowledgments. Andrew Negri of NASA provided
assistance relative to the use of TRMM data. Mark Perel
from the Israeli Ministry of Agriculture is thanked for
providing the data from the rain-gauge stations. Dr. Ilan
Setter from the Israeli Meteorological Service is thanked
for providing the masts wind data. Gregory Roux and
Ming Ge are thanked for their help with data processing
and Jason Knievel is thanked for his help editing the
manuscript. The suggestions of three anonymous re-
viewers greatly improved the quality of the manuscript.
We thank the Army Testing and Evaluation Command
(ATEC) and the NGIC for providing the financial support
for this study. Dorita Rostkier-Edelstein acknowledges
the NCAR Research Applications Laboratory visitor
funds. The CFDDA system simulations were performed
using NCAR/CISL supercomputers.
REFERENCES
Alpert, P., and M. Rabinovich-Hadar, 2003: Pre- and post-sea-
breeze frontal lines–A meso-g-scale analysis over South Is-
rael. J. Atmos. Sci., 60, 2994–3008.
——, B. U. Neeman, and Y. Shay-El, 1990: Climatological analysis
of Mediterranean cyclones using ECMWF data. Tellus, 42A,
65–77.
——, R. Abramsky, and B. U. Neeman, 1992: The prevailing
summer synoptic system in Israel—Subtropical high, not Per-
sian trough. Isr. J. Earth Sci., 39, 93–102.
970 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 49
Beck, C., J. Grieser, and B. Rudolf, 2005: A new monthly pre-
cipitation climatology for the global land areas for the period
1951–2000. Klimastatusbericht 2005, DWD, 181–190.
Berkovic, S., and Y. Feliks, 2005: The land breeze characteristics in
Israel during the summer by the MM5 model. World Weather
Research Program Symp. on Nowcasting and Very Short Fore-
casting, Tolouse, France, World Weather Research Program.
[Available online at http://www.meteo.fr/cic/wsn05/.]
Bourassa, M. A., D. M. Legler, J. J. O’Brien, and S. R. Smith, 2003:
SeaWinds validation with research vessels. J. Geophys. Res.,
108, 3019, doi:10.1029/2001JC001028.
Castro, C. L., R. A. Pielke Sr., and J. O. Adegoke, 2007: Inves-
tigation of the summer climate of the contiguous United States
and Mexico using the Regional Atmospheric Modeling system
(RAMS). Part 1: Model climatology (1950–2002). J. Climate,
20, 3844–3865.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–
hydrology model with the Penn State–NCAR MM5 modeling
system. Part I: Model implementation and sensitivity. Mon.
Wea. Rev., 129, 569–585.
Chen, M. X., R. E. Dickinson, X. B. Zeng, and A. N. Hahmann,
1996: Comparison of precipitation observed over the conti-
nental United States to that simulated by a climate model.
J. Climate, 9, 2233–2249.
Chen, M. Y., P. P. Xie, J. E. Janowiak, and P. A. Arkin, 2002:
Global land precipitation: A 50-yr monthly analysis based on
gauge observations. J. Hydrometeor., 3, 249–266.
Colle, B. A., and C. F. Mass, 2000: The 5–9 February 1996 flooding
event over the Pacific Northwest: Sensitivity studies and eval-
uation of the mm5 precipitation forecasts. Mon. Wea. Rev., 128,
593–617.
——, J. B. Olson, and J. S. Tongue, 2003: Multiseason verification
of the MM5. Part I: Comparison with the Eta model over the
central and eastern United States and impact of MM5 reso-
lution. Wea. Forecasting, 18, 431–457.
Delsol, F., K. Miyakoda, and R. H. Clarke, 1971: Parameterized
processes in surface boundary layer of an atmospheric circu-
lation model. Quart. J. Roy. Meteor. Soc., 97, 181–208.
Doron, E., and J. Neumann, 1977: Land and mountain breezes with
special attention to israel’s mediterranean coastal plain. Israel
Meteor. Res. Pap., 1, 109–122.
Dudhia, J., 1989: Numerical study of convection observed during the
winter monsoon experiment using a mesoscale two-dimensional
model. J. Atmos. Sci., 46, 3077–3107.
Ebert, E. E., J. E. Janowiak, and C. Kidd, 2007: Comparison of
near-real-time precipitation estimates from satellite obser-
vations and numerical models. Bull. Amer. Meteor. Soc., 88,
47–64.
Fast, J. D., 1995: Mesoscale modeling and four-dimensional data
assimilation in areas of highly complex terrain. J. Appl. Me-
teor., 34, 2762–2782.
Frank, H. P., and L. Landberg, 1997: Modelling the wind climate of
Ireland. Bound.-Layer Meteor., 85, 359–378.
Genoves, A., J. Campins, and A. Jansa, 2006: Intense storms in the
Mediterranean: A first description from the era-40 perspective.
Adv. Geosci., 7, 163–168.
Grell, G. A., 1993: Prognostic evaluation of assumptions used by
cumulus parameterizations. Mon. Wea. Rev., 121, 764–787.
——, J. Dudhia, and D. R. Stauffer, 1995: A description of the fifth-
generation Penn State/NCAR mesoscale model (MM5).
NCAR Tech. Rep. NCAR/TN-398, 122 pp.
Hagemann, K., 2008: Mesoscale wind atlas of South Africa. Ph.D.
thesis, University of Cape Town, 227 pp.
Hoffman, R. N., and S. M. Leidner, 2005: An introduction to
the near-real-time QuikSCAT data. Wea. Forecasting, 20,
476–493.
Hong, S. Y., and H. L. Pan, 1996: Nonlocal boundary layer vertical
diffusion in a medium-range forecast model. Mon. Wea. Rev.,
124, 2322–2339.
Huffman, G. J., and Coauthors, 2007: The TRMM multisatellite pre-
cipitation analysis (TMPA): Quasi-global, multiyear, combined-
sensor precipitation estimates at fine scales. J. Hydrometeor.,
8, 38–55.
Jimenez, B., F. Durante, B. Lange, T. Kreutzer, and J. Tambke,
2007: Offshore wind resource assessment with WAsP and MM5:
Comparative study for the German Bight. Wind Energy, 10,
121–134.
Kanamitsu, M., and H. Kanamaru, 2007: Fifty-seven-year Cal-
ifornia Reanalysis Downscaling at 10 km (CaRD10). Part I:
System detail and validation with observations. J. Climate, 20,
5553–5571.
——, W. Ebisuzaki, J. Woollen, S. K. Yang, J. J. Hnilo, M. Fiorino,
and G. L. Potter, 2002: NCEP-DOE AMIP-II reanalysis (R-2).
Bull. Amer. Meteor. Soc., 83, 1631–1643.
Knievel, J. C., D. A. Ahijevych, and K. W. Manning, 2004: Using
temporal modes of rainfall to evaluate the performance of a
numerical weather prediction model. Mon. Wea. Rev., 132,
2995–3009.
Landberg, L., L. Myllerup, O. Rathmann, E. L. Petersen, B. H.
Jorgensen, J. Badger, and N. G. Mortensen, 2003: Wind re-
source estimation—An overview. Wind Energy, 6, 261–271.
Liu, Y., F. Vandenberghe, S. Low-Nam, T. T. Warner, and
S. Swerdlin, 2004: Observation-quality estimation and its ap-
plication in the NCAR/ATEC real-time FDDA and forecast
(RTFDDA) system. Preprints, 20th Conf. on Weather Anal-
ysis and Forecasting and 16th Conf. on Numerical Weather
Prediction, Seattle, WA, Amer. Meteor. Soc., J1.7. [Available
online at http://ams.confex.com/ams/84Annual/techprogram/
paper_71934.htm.]
——, F. Chen, T. Warner, and J. Basara, 2006: Verification of
a mesoscale data assimilation and forecasting system for the
Oklahoma City area during the Joint Urban 2003 field project.
J. Appl. Meteor. Climatol., 45, 912–929.
——, and Coauthors, 2008: The operational mesogamma-scale
analysis and forecast system of the U.S. Army test and eval-
uation command. Part I: Overview of the modeling system, the
forecast products, and how the products are used. J. Appl.
Meteor. Climatol., 47, 1077–1092.
Lo, J. C.-F., Z.-L. Yang, and R. A. Pielke Sr., 2008: Assessment of
three dynamical climate downscaling methods using the
Weather Research and Forecasting (WRF) model. J. Geophys.
Res., 113, D09112, doi:10.1029/2007JD009216.
Mass, C. F., D. Ovens, K. Westrick, and B. A. Colle, 2002: Does
increasing horizontal resolution produce more skillful fore-
casts? Bull. Amer. Meteor. Soc., 83, 407–430.
Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and
S. A. Clough, 1997: Radiative transfer for inhomogeneous
atmospheres: RRTM, a validated correlated-k model for the
longwave. J. Geophys. Res., 102, 16 663–16 682.
Qian, J. H., A. Seth, and S. Zebiak, 2003: Reinitialized versus
continuous simulations for regional climate downscaling.
Mon. Wea. Rev., 131, 2857–2874.
Reisner, J., R. M. Rasmussen, and R. T. Bruintjes, 1998: Explicit
forecasting of supercooled liquid water in winter storms using
the MM5 mesoscale model. Quart. J. Roy. Meteor. Soc., 124,
1071–1107.
MAY 2010 H A H M A N N E T A L . 971
Reynolds, R. W., N. A. Rayner, T. M. Smith, D. C. Stokes, and
W. Q. Wang, 2002: An improved in situ and satellite SST
analysis for climate. J. Climate, 15, 1609–1625.
Rife, D. L., C. A. Davis, Y. B. Liu, and T. T. Warner, 2004: Pre-
dictability of low-level winds by mesoscale meteorological
models. Mon. Wea. Rev., 132, 2553–2569.
Rodell, M., and Coauthors, 2004: The global land data assimilation
system. Bull. Amer. Meteor. Soc., 85, 381–394.
Rostkier-Edelstein, S. B. D., and R. Givati, 2003: Statistical eval-
uation of high resolution mm5 surface winds and temperatures
over complex terrain. 13th PSU/NCAR Users Workshop,
Boulder, CO, National Center for Atmospheric Research.
[Available online at http://www.mmm.ucar.edu/mm5/workshop/
ws03/session5/Rostkier-Edelstein.pdf.]
Saaroni, H., N. Halfon, B. Ziv, P. Alpert, and H. Kutiel, 2010:
Links between the rainfall regime in Israel and location
and intensity of Cyprus lows. Int. J. Climatol., 30, 1014–
1025.
Schadler, G., and R. Sasse, 2006: Analysis of the connection be-
tween precipitation and synoptic-scale processes in the East-
ern Mediterranean using self-organizing maps. Meteor. Z., 15,
273–278.
Seaman, N. L., D. R. Stauffer, and A. M. Lariogibbs, 1995: A
multiscale four-dimensional data assimilation system applied
in the San Joaquin Valley during SARMAP. 1. Modeling de-
sign and basic performance-characteristics. J. Appl. Meteor.,
34, 1739–1761.
Shi, J. J., S. W. Chang, T. R. Holt, T. F. Hogan, and D. L. Westphal,
2004: A meteorological reanalysis for the 1991 Gulf War. Mon.
Wea. Rev., 132, 623–640.
Stauffer, D. R., and N. L. Seaman, 1994: Multiscale four-dimensional
data assimilation. J. Appl. Meteor., 33, 416–434.
——, ——, and F. S. Binkowski, 1991: Use of four-dimensional
data assimilation in a limited-area mesoscale model. Part 2.
Effects of data assimilation within the planetary boundary
layer. Mon. Wea. Rev., 119, 734–754.
Steinacker, R., C. Haberli, and W. Pottschacher, 2000: A trans-
parent method for the analysis and quality evaluation of ir-
regularly distributed and noisy observational data. Mon. Wea.
Rev., 128, 2303–2316.
Trigo, I. F., T. D. Davies, and G. R. Bigg, 1999: Objective clima-
tology of cyclones in the Mediterranean region. J. Climate,
12, 1685–1696.
Warner, T. T., and R. S. Sheu, 2000: Multiscale local forcing of the
Arabian Desert daytime boundary layer and implications for
the dispersion of surface-released contaminants. J. Appl. Me-
teor., 39, 686–707.
Wu, J., 1969: Froude number scaling of wind stress coefficients.
J. Atmos. Sci., 26, 408–413.
Zagar, N., M. Zagar, J. Cedilnik, G. Gregoric, and J. Rakovec, 2006:
Validation of mesoscale low-level winds obtained by dynami-
cal downscaling of ERA40 over complex terrain. Tellus, 58A,
445–455.
Zecchetto, S., and F. De Biasio, 2007: Sea surface winds over the
Mediterranean basin from satellite data (2000–04): Meso- and
local-scale features on annual and seasonal time scales. J. Appl.
Meteor. Climatol., 46, 814–827.
Ziv, B., H. Saaroni, and P. Alpert, 2004: The factors governing the
summer regime of the eastern Mediterranean. Int. J. Climatol.,
24, 1859–1871.
972 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 49