Climatic features of the Red Sea from a regional assimilative model

Abstract

TheAdvanced Research version ofWeather Research and Forecasting (WRF-ARW) modelwas used to generate
a downscaled, 10-km resolution regional climate dataset over the Red Sea and adjacent region. The model simulations are performed based on two, two-way nested domains of 30- and 10-km resolutions assimilating all conventional observations using a cyclic three-dimensional variational approach over an initial 12-h period. The improved initial conditions are then used to generate regional climate products for the following 24 h. We combined the resulting daily 24-h datasets to construct a 15-year Red Sea atmospheric downscaled product from 2000 to 2014. This 15-year downscaled dataset is evaluated via comparisons with various in situ and gridded datasets. Our analysis indicates that the assimilated model successfully reproduced the spatial and temporal variability of temperature, wind, rainfall, relative humidity and sea level pressure over the Red Sea region. The model also efficiently simulated the seasonal and monthly variability of wind patterns, the Red Sea Convergence Zone and associated rainfall. Our results suggest that dynamical downscaling and assimilation of available observations improve the representation of regional atmospheric features over the Red Sea compared to global analysis data from the National Centers for Environmental Prediction. We use the dataset to describe the atmospheric climatic conditions over the Red Sea region.

Full text

INTERNATIONAL JOURNAL OF CLIMATOLOGY
Int. J. Climatol. 37: 2563– 2581 (2017)
Published online 16 August 2016 in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/joc.4865
Climatic features of the Red Sea from a regional assimilative
model
Yesubabu Viswanadhapalli,a,b Hari Prasad Dasari,aSabique Langodan,aVenkata Srinivas Challac
and Ibrahim Hoteita*
aPhysical Sciences and Engineering Division, King Abdullah University of Science and Technology (KAUST), Thuwal, Saudi Arabia
bNational Atmospheric Research Laboratory, Gadanki, India
cRadiological Safety Division, Radiological Safety & Environment Group, Indira Gandhi Centre for Atomic Research, Kalpakkam, India
ABSTRACT: The Advanced Research version of Weather Research and Forecasting (WRF-ARW) model was used to generate
a downscaled, 10-km resolution regional climate dataset over the Red Sea and adjacent region. The model simulations are
performed based on two, two-way nested domains of 30- and 10-km resolutions assimilating all conventional observations
using a cyclic three-dimensional variational approach over an initial 12-h period. The improved initial conditions are then
used to generate regional climate products for the following 24 h. We combined the resulting daily 24-h datasets to construct
a 15-year Red Sea atmospheric downscaled product from 2000 to 2014. This 15-year downscaled dataset is evaluated
via comparisons with various in situ and gridded datasets. Our analysis indicates that the assimilated model successfully
reproduced the spatial and temporal variability of temperature, wind, rainfall, relative humidity and sea level pressure over
the Red Sea region. The model also efciently simulated the seasonal and monthly variability of wind patterns, the Red
Sea Convergence Zone and associated rainfall. Our results suggest that dynamical downscaling and assimilation of available
observations improve the representation of regional atmospheric features over the Red Sea compared to global analysis data
from the National Centers for Environmental Prediction. We use the dataset to describe the atmospheric climatic conditions
over the Red Sea region.
KEY WORDS Red Sea reanalysis; WRF model; dynamical downscaling; data assimilation; 3DVAR
Received 19 January 2016; Revised 15 June 2016; Accepted 19 July 2016
1. Introduction
Accurate knowledge of surface weather conditions such as
temperature, wind, rainfall, cloud cover and solar radiation
over a region is essential to a broad range of activities from
policy making to science and engineering (Ansari et al.,
1986; Exell and Fook, 1986; Habali et al., 1987; Rehman,
2005). Climatology based on observations from a single
station is limited to a small area around the observed
location and does not provide adequate spatial and tem-
poral resolution to account for heterogeneity over an
extended domain. Nowadays, gridded atmospheric infor-
mation is generated using numerical models in the form of
global reanalysis (Kalnay et al., 1996; Dee et al., 2011).
However, the use of global atmospheric reanalysis is also
limited by various constraints, such as coarse horizontal
resolution, model-induced uncertainties (Weare, 1997;
Scott and Alexander, 1999; Yang et al., 1999; Ladd and
Bond, 2002), and lack of a sufcient amount of data
for assimilation (Waliser et al., 1999; Putman et al.,
2000). Dynamical downscaling, a method for generating
high-resolution, region-specic climatic information from
* Correspondence to: I. Hoteit, Physical Science and Engineering Divi-
sion, King Abdullah University of Science and Technology (KAUST),
Thuwal 23955-6900, Saudi Arabia. E-mail: ibrahim.hoteit@kaust.edu.sa
coarse grid climate conditions using regional models
(Giorgi and Mearns, 1991; Giorgi et al., 1993; Jones
et al., 1995; Wilby and Wigley, 1997) can ll these gaps
by providing higher spatial and temporal resolution while
enabling the assimilation of available regional datasets.
The Weather Research and Forecasting (WRF) model
has been proven to be a reliable tool for simulating regional
climatic features via downscaling of global reanalysis
elds (Lo et al., 2008; Caldwell et al., 2009; Jiang et al.,
2009; Zhang et al., 2009; Dasari et al., 2014a, 2014b;
Srinivas et al., 2014, 2015; Hari Prasad and Srinivas,
2015). For example, Heikkila et al. (2011) demonstrated
the efciency of a two-way nested domains WRF cong-
uration of 30- and 10-km for reducing bias in temperature
and rainfall data over Norway. Using a similar setup for
the Red Sea region, Jiang et al. (2009) reported that the
10-km WRF conguration described well the regional fea-
tures of mean seasonal wind patterns by downscaling the
nal (FNL) global analysis data from the National Cen-
ters for Environmental Prediction (NCEP). The efciency
of WRF has been also demonstrated by many studies
(Mukhopadhyay et al., 2010; Srinivas et al., 2012, 2015;
Raju et al., 2015a, 2015b) for downscaling and investi-
gating the regional rainfall characteristics during Indian
summer monsoon region.
© 2016 Royal Meteorological Society

2564 Y. VISWANADHAPALLI et al.
35°N
30°N
25°N
20°N
15°N
10°N
5°N
EQ
15°E 20°E 25°E 30°E 35°E 40°E 45°E 50°E 55°E 60°E
2400
2200
2000
1800
1600
1400
1200
1000
800
600
400
200
0
–200
–400
–600
–800
–1000
–1200
–1400
–1600
–1800
–2000
Terrain
(m)
Figure 1. Model domains along with topography. Domain 1 has 30 km
and domain 2 has 10 km horizontal resolution. [Colour gure can be
viewed at wileyonlinelibrary.com].
We use dynamical downscaling to simulate the climatic
conditions over the Red Sea and adjacent regions based
on a regional assimilative conguration of WRF at 10-km
resolution. The study region (shown in Figure 1) covers
the central part of the Middle East and North Africa, act-
ing as a bridge between the African and Asian continents.
Following the climate classication of Köppen-Geiger
(Kottek et al., 2006), the northern, and eastern parts of
the study region are characterized as ‘hot desert climate
type’ (BWh), with ‘hot steppes climate type’ (BSh) along
the coastal Red Sea, ‘cold desert climate type’ (BWk)
over south-eastern part, ‘equatorial savannah type with dry
winter’ (Aw) over mountain gaps. A warm temperature
climate, hot summer and dry winter climatic conditions
prevail over the mountain regions of North Eastern Africa.
Almazroui et al. (2012) investigated the seasonal variabil-
ity of temperature and rainfall over the Arabian Peninsula
using the Climatic Research Unit (CRU) analysis and syn-
optic data. The study reported that maximum temperatures
in winter are highest in the southeastern part of Sudan and
the west-central Saudi Arabia, and minimum temperatures
are lowest over the northern part of the Arabian Peninsula
and the southern mountainous region. Rainfall over the
western Saudi Arabia is largely conned to winter months
(October–January). In summer months, rainfall is mainly
concentrated over the southern parts of Sudan, associated
with the intrusion of the southwest India Monsoon winds
(Almazroui, 2011; De Vries et al., 2013).
The mountain ranges bordering the Red Sea inuence the
local dominant wind regimes and transform the Red Sea
into a virtual wind channel, where along-axis winds are
the dominant feature. Larger and smaller valleys cut across
the bordering mountain ridges, creating winds that are rel-
evant to local wind regimes. The most relevant of these
valleys is the Tokar Gap, a 110-km-wide valley at about
18∘N on the African side of the Red Sea (see Figure 1).
In winter, the convergence of tropical and extra-tropical
synoptic-scale systems forms the Red Sea Convergence
Zone (RSCZ) at around 18∘N, characterized by cloudy
skies and drizzle in contrast to the ubiquitous clear weather
typical of the area (Pedgley, 1966; Langodan et al., 2015).
A low-level, northeast Indian monsoon-induced trough (at
around 850 hPa) over equatorial Africa, referred to as the
Red Sea trough (RST), is another dominant synoptic fea-
ture of this region. It extends northward from the south-
ern Red Sea toward the eastern Mediterranean, inuencing
the surrounding weather and climatic conditions (Krichak
et al., 2012; De Vries et al., 2013). The position of the
RST and the two high-pressure systems that modulate its
behaviour are integral to the movement of the RSCZ and
the intensication of the surface conditions. The convec-
tive activity associated with the surface of the RSCZ and
the position of the RST regulates the rainfall over the Red
Sea and adjacent regions (De Vries et al., 2013).
Few studies have investigated the atmospheric climato-
logical conditions (Evans et al., 2004; Krichak et al., 2007,
2012; Kunstmann et al., 2007; Almazroui et al., 2012; De
Vries et al., 2013; Pal and Eltahir, 2015) over the Mid-
dle East. Abdullah and Al-Mazroui (1998) and Al-Ahmadi
and Al-Ahmadi (2013) showed that the distribution of
annual rainfall varies between 10 and 600 mm from north
to south and that its variation strongly depends on the
topography. Krichak et al. (2012) analysed various aspects
of the RST using NCEP reanalysis data and implemented
a methodology to identify active RSTs (ARSTs) using
convective available potential energy, precipitable water
vapour and geo-potential heights at 1000 and 500 hPa.
In a recent study, De Vries et al. (2013) investigated the
dynamics of the ARSTs and associated extreme precipi-
tation events using European Centre for Medium-Range
Weather Forecasts (ECMWF) Re-Analysis (ERA) and
Aphrodite rainfall data for the 1979–2010 period and
reported that the formation of ARSTs is favourable in
autumn months (October and November). However, both
studies pointed out the need for high-resolution datasets to
describe both mesoscale and local-scale characteristics of
the RST. Recent downscaling studies of the Arabian Penin-
sula further suggest that coarse-resolution regional climate
models do not accurately reproduce regional climate fea-
tures in the Red Sea region (Almazroui, 2011, 2013).
We used an ARW model to produce a dynamically down-
scaled atmospheric dataset for the 15-year period span-
ning 2000–2014 to study regional climate patterns over
the Red Sea and adjacent region. Our rst objective is to
generate high-resolution atmospheric forcing elds and to
evaluate the generated data sets against in situ observations
and gridded elds of temperature, wind, relative humid-
ity (RH) and rainfall. Second, we use the dataset to study
the main atmospheric features in the region. Once vali-
dated, these products will be used to drive ocean, wave
and hydrological models over the western Saudi Arabian
Peninsula and the Red Sea. The paper is organized in four
sections. Section 2 describes the model and experimen-
tal setup, along with the methodology and data used for
assimilation and validation. The results and discussion are
presented in Section 3. Section 4 summarizes main results
and conclusions.
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

ACCURATE KNOWLEDGE OF SURFACE WEATHER CONDITIONS SUCH AS TEMPERATURE 2565
2. Data and methodology
The Advanced Research WRF (WRF-ARW, Ska-
marock et al., 2008) model version 3.6.1 developed
by NCEP/NCAR was used in this study. A high-resolution
WRF conguration was implemented based on two,
two-way nested domains with respective horizontal reso-
lutions of 30 and 10 km (Figure 1), each with 35 vertical
levels. The outer domain extends from 5∘Sto39
∘Nand
from 16∘Eto64
∘E and the inner domain covers the entire
Red Sea basin from 9∘to 30∘N and from 31∘to 47∘E. The
model’s physics are the same as in Jiang et al. (2009) and
Langodan et al. (2014), employing Yonsei University’s
k-prole scheme (Hong et al., 2006) for boundary layer
turbulence diffusion, the Kain– Fritsch mass ux scheme
for cumulus convection (Kain and Fritsch, 1993; Kain,
2004), the WRF Single-Moment 3-class (WSM3) for
microphysical processes, the Noah land surface scheme
(Chen and Dudhia, 2001) for surface processes, the
Rapid Radiation Transfer Model for long-wave radiation
(Mlawer et al., 1997), the Dudhia scheme (1989) for
short-wave radiation and MM5 Monin– Obukhov sim-
ilarity theory for surface layer processes. The model’s
topography, such as terrain elevation, land-use and soil
types, are interpolated from United States Geological
Survey data at arc 5′and 2′for 30 and 10 km domains,
respectively. For better representation of the land–sea
contrast, the coarse-resolution lower boundary conditions
(sea surface temperature, SST) from NCEP FNL analysis
data were replaced with the time-varying high-resolution
the real-time global high-resolution SST (Gemmill et al.,
2007).
The model simulations were performed over the 15-year
period between 2000 and 2014 using a consecutive inte-
gration method with daily initialization. The advantage of
consecutive reinitialization has been demonstrated in many
downscaling studies (e.g. Pan et al., 1999; Qian et al.,
2003; Lo et al., 2008; Jiang et al., 2009; Lucas-Picher
et al., 2013). The model was initialized from NCEP FNL
data available at 1∘×1∘and the boundary conditions
were updated every 6 h. Simulations were conducted with
daily initializations at 1200 UTC and integrated up to
a 36-h lead-time. A cyclic three-dimensional variational
(3DVAR) assimilation system was used in the rst 12 h
to optimize the downscaled initial conditions with avail-
able observations. The cyclic data assimilation approach
uses the forecast as the background in the next assim-
ilation cycle, which is more consistent with the parent
model. The available observations were assimilated every
6 h (two cycles) to replace the WRF’s rst guess by the
3DVAR analysis. With these improved initial conditions,
the model was then run in a free forecast mode over the
subsequent 24 h. The methodology is applied for the entire
study period (2000–2014). The resulting 24-h free fore-
cast elds are combined and compared with the observa-
tions. These model products are then used for the anal-
ysis of climatic features. We assimilated the ‘prepared’
and quality-controlled observational data set (PrepBUFR)
available in Binary Universal Form for the Representation
(BUFR) format from the NCEP Atmospheric Data Project
(ADP), which includes conventional observations from
surface stations (synoptic stations, Metar, ship and buoy),
upper-air soundings (Rawinsonde and pilot balloon), and
satellite observations, such as wind vectors from the Quick
Scatterometer (QuikSCAT), Windsat and ASCAT scat-
terometers, and atmospheric motion vectors from geosta-
tionary satellites. The dataset also contains information
about observational errors specic to each source type.
Further information about this dataset can be obtained
from http://rda.ucar.edu/datasets/ds337.0.
To validate the downscaled products, we compared
them against different gridded as well as station obser-
vations. We used CRU data for temperature and rainfall,
Cross-Calibrated Multi-Platform (CCMP) data for wind,
tropical rainfall measuring mission (TRMM) and climate
prediction centre MORPHing (CMORPH) data for rain-
fall, and surface station (synoptic) observations from the
National Climate Data Center for wind speed and rainfall.
CRU TS3.10 is a global climate dataset available at a hor-
izontal grid resolution of 0.5∘×0.5∘from the University
of East Anglia (Harris et al., 2014). The CCMP wind data
are available in 6-h grids at 0.25∘×0.25∘resolution (Atlas
et al., 2011). Satellite merged rainfall estimates, TRMM
(Huffman et al., 2007) and CMORPH (Joyce et al., 2004)
available at 0.25∘×0.25∘grid resolution, were used to val-
idate WRF downscaled rainfall.
3. Results and discussion
We compare the atmospheric elds generated by the
regional model with the global FNL analysis as well as
with observations from several sources to assess the advan-
tages of downscaling and the improvements resulting from
the higher-resolution conguration. Surface temperature
(minimum, maximum and mean values of each day), RH,
wind speed and direction, sea level pressure (SLP) and
rainfall are analysed to validate the derived high-resolution
Red Sea downscaled product.
3.1. Methodology validation
An assimilative regional model is expected to perform
better than a global model, as it benets from increased
resolution and the incorporation of additional local obser-
vations. However, it may also degrade global predictions
over certain areas in certain situations by amplifying biases
in the global elds due to insufcient resolution and/or
errors from inappropriate physics (Srinivas et al., 2015).
To demonstrate the added values at the different stages
of the adopted methodology, we computed various error
statistics following Xu and Yang (2015) and presented
them in Table 1. The error statistics are computed between
the observations and NCEP FNL data; WRF 30 km
downscaled data; 10 km downscaled data and the 10 km
assimilative data for the year 2009. We investigated differ-
ent surface variables, including wind speed, temperature,
SLP and RH from 64 surface weather stations. The adopted
methodology reduces the cold bias for temperature, but
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

2566 Y. VISWANADHAPALLI et al.
Table 1. Different Statistical indices between Model and Observations for different surface variables. Total 64 station observations
are considered for the year 2009.
OBS mean Model mean OBS std Model std Slope Bias RMSE SI CC
T2
FNL 25.56 25.23 7.39 7.31 0.9−0.33 3.08 0.12 0.91
WRF30 25.53 23.78 7.43 7.26 0.91 −1.75 3.2 0.13 0.93
WRF10 25.53 23.81 7.43 7.31 0.92 −1.72 3.1 0.12 0.94
WRFVAR 25.54 25.69 7.42 7.30.94 0.15 2.14 0.08 0.96
RH
FNL 47.43 41.88 21.85 21.35 0.79 −5.56 14.36 0.3 0.81
WRF30 53.44 59.55 19.96 19.39 0.74 6.114.85 0.28 0.76
WRF10 53.09 59.420.07 19.99 0.78 6.31 14.67 0.28 0.78
WRFVAR 51.71 55.45 19.83 16.08 0.66 3.74 12.21 0.24 0.81
SLP
FNL 1010.34 1010.42 5.63 4.40.68 0.08 2.86 0 0.87
WRF30 1010.34 1011.15 5.63 4.43 0.68 0.81 2.94 0 0.87
WRF10 1010.34 1011.14 5.63 4.50.70.82.88 0 0.87
WRFVAR 1010.34 1010.58 5.63 4.77 0.75 0.24 2.63 0 0.89
WS
FNL 3.54 3.16 2.52 1.67 0.42 −0.38 1.99 0.56 0.63
WRF30 3.74.09 2.65 2.14 0.52 0.39 2.1 0.57 0.65
WRF10 3.67 4.01 2.64 2.23 0.58 0.34 1.99 0.54 0.69
WRFVAR 3.65 3.94 2.61 2.09 0.55 0.28 1.92 0.53 0.69
OBS, Observations; SI, Scatter Index; WS, Wind Speed.
slightly increases moist bias compared to FNL. It further
produces slightly lower pressures and stronger winds in
the study region. Successive improvements are noticeable
at the different stages of the adopted methodology in
terms of reduction in root-mean-square-error (RMSE) and
increase in correlations for all the considered parame-
ters. The statistical metrics indicate that our assimilative
product has less bias, higher correlation and best t to
the observed data than the global as well as dynami-
cal downscaled products. The advantages of this model
conguration and adopted methodology have been also
recently demonstrated in Langodan et al. (2016).
3.2. Temperature patterns at 2-m height
The 15-year mean summer (May–August) daily max-
imum, minimum and mean temperatures from the
observed CRU (top panel), FNL (middle panel) and WRF
(bottom panel) are plotted in Figure 2. FNL shows
relatively higher maximum, minimum and mean daily
temperatures distributed along the southwest– northeast
direction across the Red Sea, but with less intensity com-
pared to the CRU data, suggesting that the global model
underestimates regional temperatures. WRF’s downscaled
temperature elds further show improvements over FNL,
accurately reproducing the mean patterns of daily max-
imum temperatures as observed in CRU. Unlike FNL,
WRF’s simulated north– south and east–west tempera-
ture gradients are also in good agreement with the CRU
data. Since CRU elds are masked over water bodies,
comparisons could not be performed over the Red Sea
and the Gulf of Aden. Relatively warmer (30– 36 ∘C)
temperatures are simulated over the Red Sea and adjacent
areas compared with FNL data. Although the general
patterns of maximum temperatures are better simulated
compared with FNL and CRU data, the peak temperatures
(>42 ∘C) are conned to a smaller area on the western
side of the Red Sea and extend to a larger area along the
north–south direction on the eastern side of the Red Sea,
suggesting the inuence of the local topography on the
temperature. Unlike FNL, the high-resolution congura-
tion of WRF enables reproducing realistic local variations
of temperature. FNL indicates a cold bias (<2∘C) over
the region between 18∘–21∘N and 30∘–36∘E relative to
CRU data, whereas the WRF simulation substantially
reduces this bias. Similar differences between CRU and
FNL and WRF simulations are also noticeable for the
mean and minimum temperatures. In simulating minimum
temperatures, the model exhibits a wider area of higher
temperature over the mountain belt and valley regions of
the southern Red Sea than both FNL and CRU.
Figure 3 shows that the regional model also improves
the distribution of maximum, minimum and mean temper-
atures in winter compared with FNL, in close agreement
with CRU data. The WRF simulation substantially reduces
the cold bias in minimum temperatures found in FNL data
over the entire model domain. The Ethiopian mountain
ranges are exception from the overall good performance
of the model, which underestimates the maximum temper-
atures of 3–5 ∘C. The model produces a slight warm bias
(∼3∘C) over the southern mountain ranges on both sides
of the Red Sea. Analysis of the simulated temperature thus
suggests that the model exhibits a slight warming bias in
minimum temperatures and cold bias in maximum temper-
atures over mountainous areas. Moreover, the WRF simu-
lated winter temperatures over the Red Sea are warmer by
∼2∘C compared with the observations.
During summer, strong winds associated with the south-
westerlies of the Indian summer monsoon cross over the
African coast and then reach the Red Sea through the Tokar
gap (Pedgley, 1966, 1974; Jiang et al., 2009). These winds
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

ACCURATE KNOWLEDGE OF SURFACE WEATHER CONDITIONS SUCH AS TEMPERATURE 2567
Maximum tMean t Minimum t
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
(a) (b) (c)
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
42
39
36
33
30
27
24
21
18
15
12
9
°C
(d) (e) (f)
(g) (h) (i)
Figure 2. The mean of minimum, maximum and mean temperatures (∘C) in summer from CRU data (a–c), FNL (d –f) and WRF (g –i). [Colour
gure can be viewed at wileyonlinelibrary.com].
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

2568 Y. VISWANADHAPALLI et al.
Maximum tMean t Minimum t
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
(a) (b) (c)
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
(d) (e) (f)
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E
(g) (h) (i)
39
36
33
30
27
24
21
18
15
12
9
6
3
°C
Figure 3. The mean of minimum, maximum and mean temperatures (∘C) in winter from CRU data (a– c), FNL (d– f) and WRF (g –i). [Colour gure
can be viewed at wileyonlinelibrary.com].
merge with the northwest winds blowing from the north of
the Red Sea toward the Gulf of Aden. The cross-equatorial
ow transports moisture from the equatorial Arabian Sea
to the western Red Sea and modulates air temperatures by
adverting cold moist air. During winter, the southeast and
northwest wind systems transport moisture from north and
south of the Red Sea. Once these opposite systems con-
verge over the Red Sea, the moisture is transported to the
west through the Tokar gap, and occasionally to the east
under some favourable conditions (De Vries et al., 2013).
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

ACCURATE KNOWLEDGE OF SURFACE WEATHER CONDITIONS SUCH AS TEMPERATURE 2569
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E 33°E 36°E 39°E 42°E 45°E 33°E 36°E 39°E 42°E 45°E 33°E 36°E 39°E 42°E 45°E
42
39
36
33
30
27
24
21
18
15
12
9
6
°C
January
(a) (b) (c) (d)
(e) (f) (g) (h)
(i) (j) (k) (l)
April JulyOctober
Figure 4. The mean of minimum, maximum and mean temperatures (∘C) from CRU data (a– d), FNL (e–h) and WRF (i– l) for January, July, April
and October. [Colour gure can be viewed at wileyonlinelibrary.com].
The presence of cold and moist air transported from the
north of the Red Sea reduces the temperatures in winter
compared to summer months.
In addition to the seasonal temperature analysis, we also
evaluated the model’s performance with respect to indi-
vidual monthly mean temperatures. Figure 4 plots the
composite means of January, July, April and October rep-
resenting the peak months of winter and summer and the
transition periods between the two seasons, respectively.
The FNL mean temperature exhibits a cold bias (∼1∘C)
in all these months with small differences in spatial vari-
ation. WRF improves the simulation of mean tempera-
tures across the domain compared with FNL, in close
agreement with CRU data. In January, colder temperatures
are noticeable in the northern parts and over mountain
regions in the southern parts, and warmer temperatures
are found over the Red Sea and to the southwest of the
model’s domain. While FNL exhibits a larger area of north-
ern cold temperatures extending up to 24∘N, the CRU
data suggest that such temperatures are conned to 26∘N.
WRF shows improved temperature simulation compared
with FNL, accurately simulating areas of cold tempera-
tures as compared to CRU. In summer months, the con-
trasting features of higher temperatures along the central
Arabian Peninsula and lower temperatures over mountain
regions are well reproduced by the model, as seen in FNL
and CRU. The model reproduces the high temperature
in the western Red Sea during the transition months; but
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

2570 Y. VISWANADHAPALLI et al.
30
25
20
15
10
30
25
20
15
10
30
25
20
15
10
32 34 36 3840 42 44 46 32 34 36 3840 42 44 46 32 34 36 3840 42 44 46 32 34 36 3840 42 44 46
–3 –2 –1 0
Bias(m s–1)
1 2 3 0.5 1.5 2.5
RMSE(m s–1)
0.1 0.3 0.5
Scatter index
0.6 0.8
Correlation
(a) (b) (c) (d)
(e) (f) (g) (h)
(i) (j) (k) (l)
Figure 5. Different statistical indices (bias, RMSE, SI and CCs) between daily mean temperatures from surface observations (59 stations) and nearest
or corresponding grid points from the model. The top panel is for the entire 15-year period (a– d); the middle panel is for summer (e– h); the bottom
panel is for winter (i–l). [Colour gure can be viewed at wileyonlinelibrary.com].
exhibits a cold bias of 1–2 ∘C over the eastern parts of the
Red Sea.
To assess the downscaled product, we computed sta-
tistical indices such as bias, RMSE, scatter index and
correlation coefcients (CC) for daily maximum,
minimum and mean temperatures at 59 synoptic obser-
vation locations as presented in Figure 5. The top panel
of Figure 5 shows that, at all stations and over the whole
study period, the model reproduces accurate temperatures
at a signicance level of 95%. In most locations, the bias
between the model and observations falls in the range of
±2∘C, with a RMSE of about 2.5 ∘C and CC values greater
than 0.9. The model error due to topography leading to
increased bias and RMSE and lower CC for mountain
regions. These few points are located in the southern
and bordering region of the Red Sea where land and sea
interactions play a major role in modulating temperature
and rainfall. Likewise, the local circulation has signicant
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

ACCURATE KNOWLEDGE OF SURFACE WEATHER CONDITIONS SUCH AS TEMPERATURE 2571
inuence on the diurnal cycle of air temperature in
those regions. During summer months, the errors signi-
cantly increase (Figure 5(e)–(h)), particularly during the
summer–autumn transition period, with variations in CC
from 0.6 to 0.9 and an increased warm bias and higher
RMSE (up to 2 ∘C over most of the region). The summer
temperatures suggest that the model has a warm bias over
most of the region except for a small area over the south-
west corner of the domain. The Indian summer monsoon
current that passes over the African sub-continent plays
a major role in summer temperatures. The transported
Arabian Sea moisture to this region during the summer
monsoon lowers the summer temperatures compared with
those in other parts of the Red Sea region. The model
reproduces low temperatures over the southern Red Sea as
observed in FNL and CRU with good correlations and low
RMSE errors. Similarly in winter (Figure 5(i)–(l)), sig-
nicant correlations are observed over the entire domain,
except at a few locations around the mountains that are
distributed across the southern Red Sea region where the
model exhibits a warm bias of about 2 ∘C.TheCCvalues
are greater than 0.8 over the entire region, except for the
coastal regions where the values decrease to 0.7 with a
90% signicance level.
Comparison of the collocated pairs of values with the
corresponding best-t slope provides a good criterion to
evaluate the model’s outputs (shown in Fig. S1a, Support-
ing information). As observed in the spatial representation
of the statistics in Figure 5, the simulated temperature
is underestimated with a slope of 0.96. The modelled
temperature correlates well with the observations
(CC =0.98), with a low correlation for minimum temper-
atures (not shown), suggesting poor model performance
on night-time temperatures. The data points around the
best-t line with less scattering indicate that the model’s
values are in good agreement with the observations.
In addition, we also compared the simulated monthly
temperatures (maximum, mean and minimum) with syn-
optic observations at four locations representing different
geographic/topographic patterns in the model’s domain
(Figure 6). The maximum temperatures at the Asyut sta-
tion (27.05∘N, 31.017∘E), located in the plains of Egypt,
show a gradual increase in temperatures from winter to
summer, with an average variability (standard deviation)
of ±8∘C in winter months and ±4∘C in summer months.
This annual trend is reproduced by the model, but with
an average underestimation of 1 ∘C throughout the year,
particularly in summer. Similar patterns are noticed in the
mean and minimum temperatures with slightly increased
errors in summer months. In summer months, the model’s
cold bias of about 3 ∘C suggests that the night-time min-
imum temperatures are under predicted. Similar results
are observed at the Al-Wejh station (26.2∘N, 36.483∘E)
located in the coastal region of northwestern Saudi Arabia.
The temperature at the remaining two stations, i.e. Kassala
(15.467∘N, 36.400∘E), located in the Tokar valley, and
Lekemte (9.083∘N, 36.450∘E), located in the southwest
mountainous region, exhibits a bi-modal distribution
(two peaks) during the year, which is well simulated by
the model. Although the northern locations follow the
seasonal trend of both winter and summer with maxima
during June to September, the Tokar gap and southwest
mountainous region show two maxima for April and
October months.
Overall, the seasonal and monthly variability of max-
imum, mean and minimum temperatures over the entire
domain are well reproduced by WRF outputs. The spa-
tial distribution of temperatures in the model is in good
agreement with the CRU and FNL data as well as with
synoptic station observations. The errors are within rea-
sonable limits (<10%), but they remain relatively high for
the maximum temperatures during summer months. The
errors in simulated temperature may arise from diurnal
variability, land –sea inuences, local desert effects and
hydrological processes, which need to be accurately
represented in the model with appropriate model physics.
The relatively poor performance of the model over the
mountainous regions also needs to be addressed very
carefully. The steep mountainous lands associated with
sharp land–water interfaces on either side of the Red Sea
cause complex interactions, such as mountain-valley and
sea–land diurnal ows, that affect the local temperature.
The 10-km conguration may not be adequate to properly
capture orographic effects, such as fast radiative cooling,
solar heating and orographic convection.
3.3. Analysis of simulated surface winds
During winter months, the mean seasonal surface winds
(at 10 m height) from CCMP, FNL and WRF model data
exhibit (Figure 7) distinguishing features of wind ow with
southeasterly winds from the Arabian Sea over the south-
ern Red Sea and northwesterly winds from mid-latitudes or
the Mediterranean region over the northern Red Sea. The
WRF simulation suggests improvement over FNL data in
terms of stronger northwest winds over the northern Red
Sea, stronger northerly winds over the western areas as
well as stronger southeast winds over the south, in close
agreement with CCMP data. During summer months, the
northwest winds cover the entire Red Sea basin, turn north-
east under the inuence of orography and ow toward the
Arabian Sea.
Also, the southwesterly winds of the East African
monsoon enter the Red Sea through the Tokar gap and
merge with the northwest winds coming from the northern
Red Sea. The advantage of using a high-resolution model
(WRF) over FNL is evident in simulating the stronger
southwesterly monsoon winds as observed in CCMP.
However, WRF slightly under predicts the wind strength
over the Gulf of Aden in the southeast corner of the
domain.
The location and intensity of RSCZ depend on the
strength of the southeast and northwest winds. The amount
of moisture orographically lifted over the coastal and
adjacent mountain regions governs the occurrence of
extreme rainfall over the central Red Sea region (Pedg-
ley, 1974). The narrow branch of southeast winds pro-
vides an important source of moisture for the southern
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

2572 Y. VISWANADHAPALLI et al.
Figure 6. Time series of monthly means of maximum (a, d, g, j), mean (b, e, h, k) and minimum (c, f, I, l) temperatures from four stations
along with model outputs. The standard deviations are shown in bars for both observations and model values. [Colour gure can be viewed at
wileyonlinelibrary.com].
Red Sea and neighbouring regions during winter, whereas
the branch of northwest winds brings moist cold air. The
moisture along with temperature gradients plays a major
role in generating rainfall over the RSCZ. The mixing of
warm, dry air with cold, moist air blowing from opposite
directions leads to upward frontal lifting that triggers con-
vection and subsequently leads to heavy rainfall episodes
over the Red Sea and adjoining regions (Pedgley, 1974).
The RSCZ is clearly observed in CCMP winds, and is
better simulated in WRF than in FNL analysis in terms
of location and intensity of the two opposite branches of
winds.
To understand directional changes in the wind and to
determine the months favoured for forming the RSCZ,
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

ACCURATE KNOWLEDGE OF SURFACE WEATHER CONDITIONS SUCH AS TEMPERATURE 2573
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E 33°E 36°E 39°E 42°E 45°E 33°E 36°E 39°E 42°E 45°E
33°E 36°E 39°E 42°E 45°E 33°E 36°E 39°E 42°E 45°E 33°E 36°E 39°E 42°E 45°E
(a) (b) (c)
(d) (e) (f)
10
9.0
8.0
7.0
6.0
5.0
4.0
3.0
2.5
2.0
1.5
1.0
0.5
m s–1
10
Figure 7. The mean seasonal 10 m height winds for different seasons. The top panel is for summer (a– c) and the bottom panel is for winter (d–f).
Winds are from CCMP (a and d), FNL (b and e) and WRF (c and f). [Colour gure can be viewed at wileyonlinelibrary.com].
we analysed the mean composite monthly winds over
the 15-year study period as shown in Figures 8 and
9. The monthly surface wind ow patterns at 10 m
clearly show that the southeasterly ow from the Arabian
Sea and RSCZ begins to develop in October and then
intensies through the winter months until December.
During January to April, the intensity of the convergence
gradually decreases. The progress of the northwesterly
ow of the Mediterranean branch begins in October and
reaches 19–20∘N. With the onset of winter, the north-
westerly wind moves further south to 18∘N and forms
a convergence zone with strong southeasterly ow from
the Arabian Sea. This phenomenon is clearly seen in the
model’s outputs as well as CCMP. During summer, the
northwesterly winds gain strength from May to June over
the Red Sea and weaken slightly when the Tokar jet picks
up strength in July. Once the Tokar jet starts to weaken
in early September, the northwesterly winds spread
widely over the Red Sea and reach their full strength.
FNL does not capture the strength of the winds in all
these months. As net convergence and rainfall are closely
associated with the intensity of the wind, the WRF sim-
ulation clearly improves the representation of the RSCZ
compared with FNL, in better agreement with CCMP.
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

2574 Y. VISWANADHAPALLI et al.
30°N (a) (b) (c) (d) (e) (f)
(g) (h) (i) (j) (k) (l)
(m) (n) (o) (p) (q) (r)
27°N
24°N
21°N
18°N
15°N
12°N
9°N
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E
36°E
39°E
42°E
45°E
33°E
36°E
39°E
42°E
45°E
33°E
36°E
39°E
42°E
45°E
33°E
36°E
39°E
42°E
45°E
33°E
36°E
39°E
42°E
45°E
33°E
36°E
39°E
42°E
45°E
JanuaryFebruaryMarch April May June
m s–1
10
9
8
7
6
5
4
3
2.5
2
1.5
1
0.5
10 10 10 10 10 10
Figure 8. The mean monthly 10 m height winds from January to June. (a– f) are from CCMP, (g– l) are from FNL and (m–r) are from WRF. [Colour
gure can be viewed at wileyonlinelibrary.com].
The strengthening of the southeast winds over the Red
Sea from October to December/January, its weakening
from January to April and the strength of northwesterly
winds and Tokar gap winds are also well captured by
the model.
The joint frequency distribution of winds (the distribu-
tion of winds of different strength in different direction
sectors), also called a ‘wind rose’, describes the variability
of winds at any given location over a given period of
time. To complement the above analysis, we plot the mean
seasonal wind roses in Figure 10 at four stations to analyse
the performance of the model in terms of wind direction.
Clearly, the Asyut station located in the northern plains
of Egypt exhibits northwesterlies in both summer and
winter, with a small inuence of winds (∼5%) from other
directions. The wind rose for the Kassala station located
north of RSCZ shows that the winds in summer are basi-
cally driven by south-southwesterlies (associated with the
Indian summer monsoon winds blowing from East Africa)
and the winds in winter are due to north-northeasterlies. At
Hodeida, a coastal Yemen city located in the southeastern
Red Sea region, the model-generated summer and win-
ter wind ows agree well with observations. The wind
roses from these three stations indicate good agreement
between the model and observed wind elds in both
seasons. The observations at Borama, a station located
south of Gulf of Aden, shows that the maximum fre-
quency of winds is from northwest whereas the model
identies the predominant direction as southwest. The
performance of the model in terms of simulating surface
winds is also evaluated using synoptic observations.
A scatter plot between the WRF outputs and observa-
tions from 59 synoptic observation locations (Figure
S1b) shows that the simulated wind speeds are overesti-
mated and the errors are more in simulating lower wind
speeds.
3.4. Relative humidity and sea level pressure analysis
Figures 11 and 12 depict the time series of monthly means
of RH at 2 m and surface SLP at six locations over 15
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

ACCURATE KNOWLEDGE OF SURFACE WEATHER CONDITIONS SUCH AS TEMPERATURE 2575
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E
36°E
39°E
42°E
45°E
33°E
36°E
39°E
42°E
45°E
33°E
36°E
39°E
42°E
45°E
33°E
36°E
39°E
42°E
45°E
33°E
36°E
39°E
42°E
45°E
33°E
36°E
39°E
42°E
45°E
JulyAugust September October November December
m s–1
10
9
8
7
6
5
4
3
2.5
2
1.5
1
0.5
10 10 10 10 10 10
(a) (b) (c) (d) (e) (f)
(g) (h) (i) (j) (k) (l)
(m) (n) (o) (p) (q) (r)
Figure 9. The mean monthly 10 m height winds from July to December. (a–f) are from CCMP, (g –l) are from FNL and (m –r) are from WRF.
[Colour gure can be viewed at wileyonlinelibrary.com].
years. The results suggest that the model simulates the
monthly variations in RH at all stations except Gizan,
where the model fails to produce the minimum values
during summer months. The small differences in humidity
between the model and observations may be attributed
to the differences in the moisture advection. This obser-
vation is reinforced by the simulated and observed wind
speed differences. The model’s simulated SLP shows
good agreement with the observations at all stations. In
particular, the minima in summer and maxima in winter
are well simulated by the model. Scatter plot (Figure S1c)
of WRF’s simulated RH values with their corresponding
observations from 59 synoptic locations shows a wet bias
with a RMSE of 13%. Although the comparison suggests
a greater spread, the model highly correlates (CC =0.83)
with observations. The cold bias in temperature and wet
bias in humidity is probably due to excessive mixing
produced by the model’s boundary layer physics (Shin
and Hong, 2011). Also, from the scatter plot, the simulated
SLP (Figure S1d) is in good agreement with observations
(CC =0.92), with a RMSE of 2.38 hPa.
The mean spatial distribution of the RH and SLP sim-
ulated by WRF and FNL for summer and winter seasons
are presented in Figure 13. Though WRF and FNL exhibit
almost similar spatial patterns for RH, the high-resolution
grid allows WRF to produce the variability in greater
detail over complex terrain. In summer, the RH distribution
exhibits higher values along the Red Sea and southwest of
the study domain. The lower RH values over the western
and eastern sides of the Red Sea are associated with the
dry weather of the desert. In winter, RH values are higher
over the land compared with summer months, but lower
over the Red Sea. The spatial distribution of SLP for both
summer and winter indicate that the WRF model resolves
the pressure in greater detail compared with FNL. This
suggests that a proper representation of local effects (i.e.
the land–sea contrast and mountain effects) is important to
effectively resolve local atmospheric circulation patterns.
3.5. Rainfall patterns
The spatial patterns of the seasonal mean rainfall for the
winter and summer seasons are presented in Figure 14.
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

2576 Y. VISWANADHAPALLI et al.
Figure 10. Wind rose plots between observations and simulations for summer and winter at four locations. [Colour gure can be viewed at
wileyonlinelibrary.com].
It is particularly challenging to predict rainfall quantita-
tively over this arid region because of its low amount,
short duration and the limited small number of rainy
days. As rainfall observations in the Middle East are not
consistently available with reasonable accuracy (Benes-
tad et al., 2012), we chose to make use of various derived
precipitation products, such as TRMM and CMORPH,
in addition to CRU to evaluate the model’s ability to
simulate rainfall. The results indicate singularized spatial
variability in winter and summer seasons. In winter, the
maximum amount of rainfall is simulated in the south-
ern parts of the domain with maxima over the south-
west region. The winds from the northwest and south-
east converge around 18∘–20∘N, leading to a westward
wind ow through the Tokar gap, which ultimately results
in cyclonic circulation over the southwest corner of the
domain. The orographic uplift of winds triggers the con-
vection. In summer, more rainfall is observed over moun-
tain regions (southwest side of the Red Sea) in accor-
dance with the progress of the southwest monsoon winds
and the associated strengthening of surface-level conver-
gence. Isolated rainfall is also observed in the north-
east sector. The model simulates these features very well,
in good agreement with the TRMM rainfall data. Sim-
ilarly, the simulated strong cyclonic circulation with a
higher amount of rainfall over the southwest corner (or
mostly southern part) in winter and the weak cyclonic
circulation with a lower amount of rainfall over the
northeast sector in summer are in good agreement with
observations.
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

ACCURATE KNOWLEDGE OF SURFACE WEATHER CONDITIONS SUCH AS TEMPERATURE 2577
Figure 11. Time series of monthly means of RH (%) from six stations along with model outputs. The standard deviations are shown in bars for both
observations and model values. [Colour gure can be viewed at wileyonlinelibrary.com].
Figure 12. Time series of monthly means of SLP (hPa) from six stations along with model outputs. The standard deviations are shown in bars for
both observations and model values. [Colour gure can be viewed at wileyonlinelibrary.com].
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

2578 Y. VISWANADHAPALLI et al.
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
30°N
(a) (b)
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E 33°E 36°E 39°E 42°E 45°E
FNL
RH(%)
WRF
90
80
79
60
50
40
30
20
10
5
Figure 13. Mean of seasonal means of RH (shaded, %) and SLP (contours, hPa) for (a) summer and (b) winter for both FNL (top panel) and the
model (bottom panel). [Colour gure can be viewed at wileyonlinelibrary.com].
4. Summary and conclusions
The resolution of atmospheric datasets is crucial to study
the regional climatic features over complex regions. The
Red Sea and adjacent regions are an extreme example in
this respect, especially because of the complicated orog-
raphy. We generated a 10-km resolution regional climate
dataset for the Red Sea and adjacent regions using the
WRF-ARW regional model. The model was initialized on
a daily basis using FNL data and NCEP ADP observations
are assimilated in the rst 12 h using 3DVAR to improve
the initial conditions. The model was then integrated in
a free forecasting mode for the subsequent 24 h. Follow-
ing this methodology, a continuous dataset was generated
over a period of 15 years (2000– 2014) from which daily,
monthly and seasonal means of different atmospheric vari-
ables were estimated. Maximum, minimum and mean tem-
peratures, winds at 10 m, RH, SLP and rainfall resulting
from the simulations were validated against all possible
sources of observations. We specically focused on the
validation of the surface variables for different time scales
and their regional climatic characteristics. The results sug-
gest that the model performs well in reproducing differ-
ent regional climatic features, especially the evolution and
decay of the RSCZ, an important factor in modulating
weather and climatic conditions over the Middle East.
The temperature analysis shows that the maximum tem-
peratures in summer months are observed over north
Sudan and Central Arabian Peninsula, where as in win-
ter months the high maximum temperatures are spotted
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

ACCURATE KNOWLEDGE OF SURFACE WEATHER CONDITIONS SUCH AS TEMPERATURE 2579
(a) (b) (c) (d)
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
30°N
27°N
24°N
21°N
18°N
15°N
12°N
9°N
33°E 36°E 39°E 42°E 45°E 33°E 36°E 39°E 42°E 45°E 33°E 36°E 39°E 42°E 45°E 33°E 36°E 39°E 42°E 45°E
400
300
250
200
150
100
80
60
40
20
10
1
(e) (f) (g) (h)
Figure 14. The seasonal total rainfall (mm) for summer (top panel) and winter (bottom panel) seasons from CRU (a, e); TRMM (d, f); CMORPH (e,
g) and WRF (d, h) data. [Colour gure can be viewed at wileyonlinelibrary.com].
in Tokar region. From available stations observations, it
is also noticed that the northern part follows the seasonal
trend of winter and summer with maxima during June to
September. The Tokar gap and the mountainous region in
the southern parts exhibit two maxima during April and
October months, respectively. The analysis of mean sea-
sonal and monthly winds in winter clearly shows the for-
mation of RSCZ around Tokar gap. In winter months, the
northeast monsoon-induced wind ows through the Gulf
of Aden (from Arabian Sea) to the Red Sea and converges
with the northwest winds blowing from the northern Red
Sea. Furthermore, the generated dataset well demonstrate
the impact of southwest Indian monsoon ow during sum-
mer months over Tokar gap and more generally the Red
Sea region. The southwest wind ows through the Tokar
gap enters into the Red Sea from June and reaches maxi-
mum strength during July and August months.
The evolution and duration of RSCZ and favourable
conditions for its intensication and associated rainfall
variability were well simulated by the model, and are in
good agreement with the corresponding observations. The
model also simulated the higher amount of rainfall associ-
ated with active phases of the RSCZ in the winter season
(November– February) over the south–southwest region
of the Red Sea below 18∘N and the north and northeast-
ern parts of the Saudi Arabian region. The regional model
reproduced the seasonal shift in the high amount of rain-
fall over the northeastern parts of Saudi Arabia and the
scanty rainfall over the southwest or south of 18∘N lati-
tude in the summer seasons. Although localized, the win-
ter rainfall amount in the southern part is much higher
than the summer rainfall over the northeastern parts of
the domain. This can be attributed to the development
of intense low-pressure/cyclonic circulation systems over
the southern parts in winter compared with the cyclonic
circulation systems over the northeastern sectors in sum-
mer. The relatively higher meridional temperature gradi-
ents in winter compared with summer may also lead to
more active convection over low-pressure areas in win-
ter. These localized rainfall features are well represented
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

2580 Y. VISWANADHAPALLI et al.
in the produced regional Red Sea datasets. The model
also successfully simulated the convective rainfall over the
northeast region in winter, which is associated with large
temperature gradients due to cold and warm air moisture
transport along the Red Sea. The temperature patterns dur-
ing different months and seasons were well reproduced by
the model and in good agreement with the CRU and FNL.
The simulated RH and SLP are in better agreement with
the observations than the simulated winds.
This study emphasizes the advantages of the application
of a downscaling assimilation methodology for generating
high-resolution climate data for the Red Sea region. This
is important for various applications, including studying
circulation in the Red Sea, ocean surface processes and
marine biodiversity in a region where atmospheric feed-
back is important to realistically simulating oceanic phe-
nomena. A few applications have already demonstrated the
relevance of this produced dataset [e.g. Langodan et al.
(2014, 2016) on Red Sea waves and Zhan et al. (2015)
on far-eld dispersion of concentrate discharges along the
Saudi coast of the Red Sea]. We also intend to use the pro-
duced dataset to improve our understanding of the atmo-
spheric circulation and variability in the region and to
study the effects of global warming.
Acknowledgements
This research work was supported by King Abdullah
University of Science and Technology (KAUST), Saudi
Arabia and the Saudi ARAMCO-KAUST Marine Envi-
ronmental Research Center (SAKMERC). This research
made use of the resources of the Supercomputing Lab-
oratory and/or computer clusters at KAUST. The NCEP
FNL, prepbufr global observational datasets were obtained
from http://rda.ucar.edu. TRMM-3B42 rainfall estimates
were downloaded from the TRMM NASA GFSC server.
CCMP datasets were downloaded from podaac.jpl.nasa
.gov. CMORPH rainfall estimates were downloaded from
the Climate Prediction Center.
Supporting information
The following supporting information is available as part
of the online article:
Figure S1. Scatter plots of all available station observa-
tions and corresponding nearest model values along with
different statistical indices. (a) Temperature (∘C); (b) wind
speed (m s−1), (c) RH (%) and (d) sea level pressure (hPa).
References
Abdullah MA, Al-Mazroui MA. 1998. Climatological study of the
southwestern region of Saudi Arabia – I. Rainfall analysis. Clim. Res.
9(3): 213– 223.
Al-Ahmadi K, Al-Ahmadi S. 2013. Rainfall-Altitude Relation-
ship in Saudi Arabia. Adv. Meteorol. 2013: 363029, doi:
10.1155/2013/363029.
Almazroui M. 2011. Sensitivity of a regional climate model on the simu-
lation of high intensity rainfall events over the Arabian Peninsula and
around Jeddah (Saudi Arabia). Theor. Appl. Climatol. 104: 261 –276.
Almazroui M. 2013. Simulation of present and future climate of Saudi
Arabia using a regional climate model (PRECIS). Int. J. Climatol. 33:
2247– 2259.
Almazroui M, Nazrul Islam M, Athar H, Jones PD, Rahman MA. 2012.
Recent climate change in the Arabian Peninsula: annual rainfall and
temperature analysis of Saudi Arabia for 1978– 2009. Int. J. Climatol.
32: 953– 966, doi: 10.1002/joc.3446.
Ansari J, Madni IK, Bakhsh H. 1986. Saudi Arabian Wind Energy Atlas.
KACST: Riyadh.
Atlas R, Hoffman RN, Ardizzone J, Leidner SM, Jusem JC, Smith DK,
Gombos D. 2011. A cross-calibrated, multiplatform ocean surface
wind velocity product for meteorological and oceanographic applica-
tions. Bull. Am. Meteorol. Soc. 92: 157– 174.
Benestad RE, Nychka D, Mearns LO. 2012. Spatially and temporally
consistent prediction of heavy precipitation from mean values. Nat.
Clim. Change 2: 544– 547.
Caldwell PM, Chin H-NS, Bader DC, Bala G. 2009. Evaluation of a
WRF based dynamical downscaling simulation over California. Clim.
Change 95: 499– 521.
Chen F, Dudhia J. 2001. Coupling an advanced land-surface/hydrology
model with the Penn State/NCAR MM5 modeling system. Part I:
Model description and implementation. Mon. Weather Rev. 129:
569– 585.
Dasari HP, Pozo I, Ferri-Yáñez F, Araújo MB. 2014a. A Regional
Climate Study of Heat Waves over the Iberian Peninsula. Atmospheric
and Climate Sciences 4: 841–853, doi: 10.4236/acs.2014.4574.
Dasari HP, Salgado R, Perdigao J, Srinivas CV. 2014b. A Regional
Climate Simulation Study Using WRFARW Model over Europe and
Evaluation for Extreme Temperature Weather Events. Int. J. Atmos.
Sci. 2014: 704079, doi: 10.1155/2014/704079.
De Vries AJ, Tyrlis E, Edry D, Krichak SO, Steil B, Lelieveld J. 2013.
Extreme precipitation events in the Middle East: dynamics of the
active Red Sea trough. J. Geophys. Res. Atmos. 118: 7087– 7108, doi:
10.1002/jgrd.50569.
Dee DP, Uppala SM, Simmons AJ, Berrisford P, Poli P, Kobayashi S,
Andrae U, Balmaseda MA, Balsamo G, Bauer P, Bechtold P, Beljaars
ACM, van de Berg L, Bidlot J, Bormann N, Delsol C, Dragani R,
Fuentes M, Geer AJ, Haimberger L, Healy SB, Hersbach H, Hólm
EV, Isaksen L, Kållberg P, Köhler M, Matricardi M, McNally AP,
Monge-Sanz BM, Morcrette J-J, Park B-K, Peubey C, de Rosnay P,
Tavolato C, Thépaut J-N, Vitart F. 2011. The ERA-Interim reanalysis:
conguration and performance of the data assimilation system. Q. J.
R. Meteorol. Soc. 137(656): 553– 597, doi: 10.1002/qj.828.
Dudhia J. 1989. Numerical study of convection observed during the win-
ter monsoon experiment using a mesoscale two-dimensional model.
J. Atmos. Sci. 46: 3077 –3107.
Evans JP, Smith RB, Oglesby RJ. 2004. Middle East climate simulation
and dominant precipitation processes. Int. J. Climatol. 24: 1671– 1694.
Exell RGB, Fook CT. 1986. The wind power potential of Malaysia. Solar
Energy 36(3): 281– 289.
Gemmill W, Katz B, Xu L. 2007. Daily real-time, global sea surface
temperature – high-resolution analysis. NCEP/EMC Off. Note 260,
Enviro. Modell. Cent., Camp Springs, MD.
Giorgi F, Mearns LO. 1991. Approaches to the simulation of regional
climate change: a review. Rev. Geophys. 29: 191–216.
Giorgi F, Bates GT, Nieman S. 1993. The multiyear surface climatology
of a regional atmospheric model over the western United States.
J. Clim. 6: 75– 95.
Habali SM, Hamdan MAS, Jubran BA, Zaid AIO. 1987. Wind speed and
wind energy potential of Jordan. Solar Energy 38(1): 59– 70.
Hari Prasad D, Srinivas CV. 2015. A study of precipitation climatology
and its variability over Europe using an advanced regional model
(WRF). Am. J. Clim. Change 4: 22– 39.
Harris I, Jones PD, Osborn TJ, Lister DH. 2014. Updated high-resolution
grids of monthly climatic observations – the CRU TS3.10 dataset. Int.
J. Climatol. 34: 623– 642, doi: 10.1002/joc.3711.
Heikkilä U, Sandvik A, Sorteberg A. 2011. Dynamical downscaling of
ERA-40 in complex terrain using the WRF regional climate model
Clim Dyn 37: 1551, doi: 10.1007/s00382-010-0928-6.
Hong SY, Noh Y, Dudhia J. 2006. A new vertical diffusion package with
explicit treatment of entrainment processes. Mon. Weather Rev. 134:
2318– 2341.
Huffman GJ, Adler RF, Bolvin DT, Gu G, Nelkin EJ, Bowman KP, Hong
Y, Stocker EF, Wolff DB. 2007. The TRMM Multisatellite Precip-
itation Analysis (TMPA): quasi-global, multiyear, combined-sensor
precipitation estimates at ne scales. J. Hydrometeorol. 8: 38 –55.
Jiang H, Farrar JT, Beardsley RC, Chen R, Chen C. 2009. Zonal surface
wind jets across the Red Sea due to mountain gap forcing along
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)

ACCURATE KNOWLEDGE OF SURFACE WEATHER CONDITIONS SUCH AS TEMPERATURE 2581
both sides of the Red Sea. Geophys. Res. Lett. 36: L19605, doi:
10.1029/2009GL040008.
Jones RG, Murphy JM, Noguer M. 1995. Simulation of climate change
over Europe using a nested regional climate model. I: Assessment of
control climate, including sensitivity to location of lateral boundaries.
Q. J. R. Meteorol. Soc. 121: 1413–1449.
Joyce RJ, Janowiak JE, Arkin PA, Xie P. 2004. CMORPH: a method that
produces global precipitation estimates from passive microwave and
infrared data at high spatial and temporal resolution. J. Hydrometeorol.
5: 487– 503.
Kain JS. 2004. The Kain-Fritsch convective parameterization: an update.
J. Appl. Meteorol. 43: 170–181.
Kain JS, Fritsch JM. 1993. Convective parameterization for mesoscale
models: the Kain-Fritcsh scheme. In The Representation of Cumulus
Convection in Numerical Models, Emanuel KA, Raymond DJ (eds).
Meteor Monogr No. 24. American Meteorological Society: Boston,
MA; 246 pp.
Kalnay E et al. 1996. The NMC/NCAR reanalysis project. Bull. Am.
Meteorol. Soc. 77: 437– 471.
Kottek M, Grieser J, Beck C, Rudolf B, Rubel F. 2006. World map of
the Koppen-Geiger climate classication updated. Meteorol. Z. 15(3):
259– 263.
Krichak S, Alpert P, Bassat K, Kunin P. 2007. The surface climatology of
the eastern Mediterranean region obtained in a three-member ensem-
ble climate change simulation experiment. Adv. Geosci. 12: 67 –80.
Krichak S, Joseph SB, Feldstein SB. 2012. A conceptual model for the
identication of active Red Sea trough synoptic events over the South-
eastern Mediterranean. J. Appl. Meteorol. Climatol. 51: 962– 971, doi:
10.1175/JAMC-D-11-0223.1.
Kunstmann H, Suppan P, Heckl A, Rimmer A, 2007. Regional
climate change in the Middle East and impact on hydrology in
the Upper Jordan catchment. Quantication and Reduction of
Predictive Uncertainty for Sustainable Water Resources Manage-
ment. In The Proceedings of Symposium HS2004 at IUGG2007,
Perugia, July 2007. IAHS Publ. 313, http://iahs.info/uploads/dms/
13976.21-141-149-313-18-Kunstmann.pdf (accessed 31 July 2016).
Ladd C, Bond NA. 2002. Evaluation of the NCEP/NCAR reanalysis in
the NE Pacic and the Bering Sea. J. Geophys. Res. 107(C10): 3158,
doi: 10.1029/2001JC001157.
Langodan S, Cavaleri L, Yesubabu V, Hoteit I. 2014. The Red Sea: a
natural laboratory for wind and wave modeling. J. Phys. Oceanogr.
44: 3139– 3159, doi: 10.1175/JPO-D-13-0242.1.
Langodan S, Cavaleri L, Viswanadhapalli Y, Hoteit I. 2015. Wind-wave
source functions in opposing seas. J. Geophys. Res. Oceans 120:
6751– 6768, doi: 10.1002/2015JC010816.
Langodan S, Viswanadhapalli Y, Hoteit I. 2016. The impact of atmo-
spheric data assimilation on wave simulations in the Red Sea. Ocean
Eng. 116: 200– 215, doi: 10.1016/j.oceaneng.2016.02.020.
Lo JCF, Yang ZL, Pielke RA Sr. 2008. Assessment of three dynam-
ical climate downscaling methods using the weather research and
forecasting (WRF) model. J. Geophys. Res. 113: D09112, doi:
10.1029/2007JD009216.
Lucas-Picher P, Boberg F, Christensen JH, Berg P. 2013. Dynamical
downscaling with reinitializations: a method to generate nescale
climate datasets suitable for impact studies. J. Hydrometeorol. 14:
1159– 1174, doi: 10.1175/JHM-D-12-063.1.
Mlawer EJ, Taubman SJ, Brown PD, Iacono MJ, Clough SA. 1997.
Radiative transfer for inhomogeneous atmosphere: RRTM, a validated
correlated-k model for the longwave. J. Geophys. Res. 102(D14):
16663– 16682.
Mukhopadhyay P, Taraphdar S, Goswami BN, Krishnakumar K. 2010.
Indian summer monsoon precipitation climatology in a high-regional
climate model: impacts of convective parameterization on systematic
biases. Weather Forecast. 25: 369–387.
Pal JS, Eltahir EAB. 2015. Future temperature in southwest Asia pro-
jected to exceed a threshold for human adaptability. Nat. Clim. Change
6: 197– 200, doi: 10.1038/NCLIMATE2833.
Pan Z, Takle E, Gutowski W, Turner R. 1999. Long simulation of
regional climate as a sequence of short segments. Mon. Weather Rev.
127: 308– 321.
Pedgley DE. 1966. The Red Sea convergence zone. Weather 21:
394– 406, doi: 10.1002/j.1477-8696.1966.tb02788.x.
Pedgley DE. 1974. An outline of the weather and climate of the
Red Sea. Publ. Cent. Natl. Exploit. Oceans Actes Colloq. 2:
9– 21.
Putman WM, Legler DM, O’Brien JJ. 2000. Inter-annual variability of
synthesized FSU and NCEP-NCAR reanalysis pseudostress products
over the Pacic Ocean. J. Clim. 13: 3003–3016.
Qian J, Seth A, Zebiak S. 2003. Reinitialized versus continuous simu-
lations for regional climate downscaling. Mon. Weather Rev. 131(11):
2857– 2874.
Raju A, Parekh A, Kumar P, Gnanaseelan C. 2015a. Evaluation of the
impact of AIRS proles on prediction of Indian summer monsoon
using WRF variational data assimilation system. J. Geophys. Res.
120(16): 8112– 8131, doi: 10.1002/2014JD023024.
Raju A, Parekh A, Sreenivas P, Chowdary JS, Gnanaseelan C. 2015b.
Estimation of improvement in Indian summer monsoon circulation by
assimilation of satellite retrieved temperature proles in WRF model.
IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 8(4): 1591 –1600,
doi: 10.1109/JSTARS.2015.2410338.
Rehman S. 2005. Prospects of wind farm development in Saudi Arabia.
Renew. Energy 30: 447–463.
Scott JD, Alexander MA. 1999. Net shortwave uxes over the ocean.
J. Phys. Oceanogr. 29: 3167– 3174.
Shin HH, Hong SY. 2011. Intercomparison of planetary boundary-layer
parameterizations in the WRF model for a single day from CASES-99.
Bound. Layer Meteorol. 139: 261– 281.
Skamarock WC, Klemp JB, Dudhia J, Gill DO, Barker DM, Dudha MG,
Huang X, Wang W, Powers Y. 2008. A description of the Advanced
Research WRF Ver.30. NCAR Technical Note, NCAR/TN-475CSTR,
Mesocale and Microscale Meteorology Davison, National Centre for
Atmospheric Research, Boulder, CO, 113 pp.
Srinivas CV, Hariprasad D, Bhaskar Rao DV, Anjaneyulu Y,
Baskaran B, Venkatrman B. 2012. Simulation of the Indian
summer monsoon regional climate using advanced research
WRF model. Int. J. Climatol. 33(5): 1195– 1210, doi: 10.1002/
joc.3505.
Srinivas CV, Bhaskar Rao DV, Hari Prasad D, Hari Prasad KBRR,
Baskaran R, Venkatraman B. 2014. A study on the inuence of the
Land Surface Processes on the Southwest Monsoon simulations using
a Regional Climate model. Pure Appl. Geophys. 172: 2791– 2811, doi:
10.1007/s00024-014-0905-9.
Srinivas CV, Hariprasad D, Bhaskar Rao DV, Baskaran B, Venkatrman
B. 2015. Simulation of the Indian summer monsoon onset-phase
rainfall using a regional model. Ann. Geophys. 33: 1097– 1115, doi:
10.5194/angeo-33-1097-2015.
Waliser DE, Shi Z, Lanzante JR, Oort AH. 1999. The Hadley circulation:
assessing NCEP/NCAR reanalysis and sparse in situ estimates. Clim.
Dyn. 15: 719– 735.
Weare BC. 1997. Comparison of NCEP-NCAR cloud radiative forcing
reanalyses with observations. J. Clim. 10: 2200– 2209.
Wilby RL, Wigley TML. 1997. Downscaling general circulation model
output: a review of methods and limitations. Prog. Phys. Geogr. 21:
530– 548.
Xu Z, Yang Z-L. 2015. A new dynamical downscaling approach with
GCM bias corrections and spectral nudging. J. Geophys. Res. Atmos.
120: 3063– 3084, doi: 10.1002/2014JD022958.
Yang SK, Hou YT, Miller AJ, Campana KA. 1999. Evaluation of
the Earth radiation budget in NCEP-NCAR reanalysis with ERBE.
J. Clim. 12: 477– 493.
Zhan P, Yao F, Kartadikaria AR, Yesubabu V, Ganesh G, Ibrahim
H. 2015. Far-eld ocean conditions and concentrate discharges
modeling along the Saudi Coast of the Red Sea. In Intakes and Outfalls
for Seawater Reverse-Osmosis Desalination Facilities, Innovations
and Environmental Impacts. Part II, ISBN: 978-3-319-13202-0,
Environmental Science and Engineering, Springer International,
501– 520, doi: 10.1007/978-3-319-13203-7_21.
Zhang Y, Duliere V, Mote P, Salathe EP Jr. 2009. Evaluation of WRF and
HadRM mesoscale climate simulations over the United States Pacic
Northwest. J. Clim. 22: 5511– 5526.
© 2016 Royal Meteorological Society Int. J. Climatol. 37: 2563– 2581 (2017)