1. Introduction
The atmosphere, land, and ocean, along with their interactions in various temporal and spatial scales, play a crucial role in shaping the quality and sustainability of human life and nature. Among these components, weather profoundly impacts numerous sectors such as civil protection, constructions, tourism, offshore energy applications (both renewable and not), insurance and reinsurance, shipping and transportation, natural inhabitant and cultural heritage security (Hinnenthal and Clauss 2010; Luo et al. 1994; Patlakas et al. 2017; Stathopoulos et al. 2013; Zeng 2000). Moreover, the interaction between weather and land use, particularly in agriculture, is of utmost importance for ensuring food security (Sheffield et al. 2014). At the same time, the oceanic and wave components have impacts in several fields ranging from ports, transportation, and offshore platforms to fisheries and natural reserves (Makris et al. 2021; Solari et al. 2012). Consequently, accurate forecasts of atmospheric, wave, and oceanographic parameters are essential for supporting socioeconomic activities, mitigating risks (Baruque et al. 2010), and securing the lives and livelihoods of habitants.
Recognizing the significance of weather forecasting, the authorities in Saudi Arabia have underscored its importance (Qahtani et al. 2014). Consequently, taking into account: 1) the key geographical location of Saudi Arabia, 2) the need for a complete meteorological forecasting system for the wider region, 3) the evolution of high-performance computers and 4) the progress of numerical analysis combined with new graphical representations of the computational results, an initiative was launched at the National Center for Meteorology (NCM) to develop and install an integrated numerical forecasting system (Papageorgiou and Al-omary 2018). The objective of this initiative was to design, develop, and support high-resolution operational procedures for atmospheric, wave, and ocean forecasts while continuously improving the system. Although efforts had been made in the past, they were based on systems with limited capabilities and lacked a focus on different components.
Since its establishment in 2020, this multiparametric system has been operational, bolstering the national capabilities in high-resolution numerical weather prediction and extending its coverage well beyond the physical boundaries of the peninsula. Currently, the system provides meteorological products and services to diverse clients and sectors. It is important to note, however, that the system is continuously evolving. Future plans include implementing a four-times-per-day forecast for all models, testing and employing alternative assimilation techniques and setting up an additional atmospheric modeling system with distinct characteristics. Regarding the latter, we acknowledge the limitations of deterministic forecasting. Therefore, alongside employing deterministic models with varying characteristics, our ultimate goal is to utilize the main modeling system for ensemble prediction, enabling us to quantify variability and uncertainty in weather forecasts.
It is well known that numerical weather prediction (NWP) systems heavily rely on observational data to evaluate their accuracy and reliability. Unfortunately, the scarcity of comprehensive observational networks in certain regions presents significant challenges in this regard. The local authorities are aware of this limitation and are making ongoing efforts to expand the station network throughout the region.
Taking all the above into consideration, in the subsequent sections, we provide a comprehensive presentation of the initial version of the weather/wave/ocean modeling system, including its setup in NCM’s supercomputer system (cf. the appendix). Moreover, we conduct an extensive evaluation of each component of the system, covering one year of operational forecasts.
2. A short description of the atmospheric and oceanic conditions of the Arabian Peninsula
The Arabian Peninsula is a region characterized by diverse climatic conditions (Patlakas et al. 2019). These can be attributed to large-scale atmospheric and ocean circulations, its location and geomorphological characteristics, and the constant presence of dust particles in the atmosphere. It is well known that desert dust has an impact on the weather and climate (Mahowald et al. 2014, 2010). It has a profound effect on the radiative budget and energy distribution of the atmosphere and highly affects the lives and economy in the region.
Focusing on the atmosphere, mean annual temperatures range between around 30°C in the southeast and 20°C in the northern regions. From June to September, mean monthly temperatures reach 35°–40°C, with daily maximum temperatures sometimes exceeding 50°C, even in populated areas (Patlakas et al. 2019). One characteristic wind pattern is the shamal winds, a result of the large-scale atmospheric systems of high pressure over the Mediterranean and Europe and the relatively lower pressure over the Indian Ocean (Yu et al. 2016). The winds, blowing over the southeast Arabian Peninsula, often cause dust and sand storms, restricting visibility and affecting health. The peninsula’s coastal orographic formations, along with physiographic characteristics, interact with regional wind conditions. A characteristic example is the Tokar Gap jet (Davis et al. 2015). Precipitation in the peninsula exhibits low annual amounts with seasonal and spatial variability (Patlakas et al. 2021b). It is highly affected by Mediterranean lows during the winter and monsoonal activity mainly during the summer months. Strong convection on smaller spatial scales is also something rather common. Extreme precipitation events are quite frequent, leading to floods and posing a threat to human life and activities on an annual basis.
The peninsula is surrounded by the Arabian marginal seas and gulfs (AMSG). The dominant geophysical processes affecting the oceanographic conditions of the area are monsoon driven (Hoteit et al. 2021), exhibiting high seasonal and interannual variability. Additional key factors influencing the region’s thermohaline properties are the net evaporation rates and the air–sea interaction processes (Sofianos and Johns 2015, 2002). The large contrast in the bathymetric features of the surrounding seas also yields different responses to external forcings. For example, tidal currents are important in the Arabian Gulf (Hyder et al. 2013), but less important in the Red Sea, which is mainly driven by air–sea fluxes (Sofianos et al. 2002). Both marginal seas are connected to the Indian Ocean along the coastal region of the southern Arabian Peninsula, which is well known for its intense upwelling events modulating the ecosystem properties and affecting biodiversity (Raitsos et al. 2017).
In the Red Sea and the Arabian Gulf, the wave field can be characterized as wind driven (Langodan et al. 2014, 2017b). Red Sea wind patterns are responsible for the north west waves formed over the Red Sea (Hoteit et al. 2021; Langodan et al. 2017a, 2014). During summer, the regional wind system weakens and the local wave conditions can be affected by the dominant sea/land breezes (Aboobacker et al. 2021). In the south, the Arabian Sea is an open area influenced by monsoonal activity and remote-generated south west swell from the southern Indian Ocean (Colosi et al. 2021; Francis et al. 2020). During the summer monsoon the region is characterized by strong southwest waves (Anoop et al. 2015). During winter, shamal winds can affect the wave characteristics of the northeast Arabian Sea and can create northeast swells traveling toward the coastline of India (Aboobacker et al. 2011).
Forecasting meteorological conditions in Saudi Arabia presents unique challenges due to its diverse geography and land–sea interaction. Limited availability of observational data in remote areas further complicates accurate forecasting. To address these challenges, a high-resolution operational forecasting system is being developed specifically for Saudi Arabia, encompassing atmospheric, wave, and ocean parameters. The aim is to improve forecasting capabilities and provide tailored forecasts for the region.
3. System and components configuration
The development of Saudi Arabia’s NCM forecasting system involves three components: atmosphere, ocean, and wave field. The atmospheric component plays a dual role as it serves not only for atmospheric forecasting but also as the driving force for the other components. Therefore, the focus was placed on optimizing the quality of atmospheric forecasts by refining the system’s setup and initial conditions through assimilating real-time measurements. A graphical summary of these complex interactions is provided in Fig. 1, with detailed explanations following in subsequent sections.
A graphical representation of the NCM operational forecasting system components and interoperability.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
a. Atmospheric data assimilation
The data assimilation procedure is based on the Limited Atmospheric Processing System (LAPS; Albers 1995). LAPS was developed and first operated by NOAA’s Earth System Research Laboratory in Boulder, Colorado. It is a meteorological assimilation tool with the capability of combining a wide array of observed meteorological datasets into a gridded atmospheric analysis (Hiemstra et al. 2006). More specifically, it can employ several types of available observations (meteorological networks, radar, satellite, soundings, and aircraft) to generate a spatially distributed, three-dimensional representation of atmospheric features and processes. These, enriched with atmosphere profile data, satellite and radar retrievals and measurements taken from ships or aircrafts can improve the accuracy of the assimilation output.
The assimilation system LAPS is used here for the initialization of all atmospheric components. It has a domain that covers a large area including the Middle East and large parts of Europe, Africa, and Asia with a resolution of 15 km (Fig. 2).
LAPS domain employed for the initialization of the atmospheric modeling systems. Black dots represent World Meteorological Organization (WMO) stations, and red triangles represent local stations used by LAPS and/or evaluation.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
Vertically, it is stretched from the surface up to the lower stratospheric layers through 41 pressure levels. Global geography datasets were used for topography, land fraction, land use (U.S. Geological Survey dataset with a resolution of 30 s), soil type, greenness fraction (Food and Agriculture Organization—FAO 1991), and albedo definition (Csiszar and Gutman 1999). The background used consists of GFS analysis (0.25°—all cycles) from the National Center for Environmental Prediction (NCEP). The system incorporates several observational data including meteorological aerodrome reports (METAR), surface synoptic observations (SYNOP) and radiosonde observations (RAOB). Moreover, ground observational data are provided by local authorities with a frequency of 30 min.
The configuration is set to run hourly and the outputs are used as initial conditions for all atmospheric models. At the same time the LAPS output is used for the display of the present conditions in the visualization system.
b. High-resolution atmospheric forecasting/wave forecasting
1) WRF (cycles 0000 and 1200 UTC)
For the atmospheric simulations, the Advanced Research version of WRF (WRF-ARW) version 4.1.3 (Powers et al. 2017; Skamarock et al. 2008, 2019) is used. WRF-ARW is a limited-area atmospheric model designed for operational forecasting and research activities (Otero-Casal et al. 2019; Patlakas et al. 2023). It is based on a fully compressible, nonhydrostatic dynamical core. On the vertical plane it has terrain-following, mass-based, hybrid sigma-pressure vertical coordinates based on dry hydrostatic pressure, with vertical grid stretching permitted, while for the horizontal grid, Arakawa C-grid staggering is employed.
The model was configured to run in two interactive grid nests (Fig. 3). The outer nest covers a large domain including large parts of Africa, Europe, and Asia with a spatial resolution of 4.8 km, while the inner one covers the Arabian Peninsula with a spatial resolution of 1.6 km. In the vertical the model has 48 levels, with 8 of them within the first 1 km, the first free level at 52.5 m, and the last at 20.92 km. The coarse grid is configured to provide a 10-day forecast while the fine one is set for 5-day forecasts (+12-h spinup time). Both run in two cycles: 0000 and 1200 UTC.
WRF (black line frame) and WAM (white line frame) model domains.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
The model utilizes the 0.25° global forecast (GFS) for its boundary conditions. The daily analysis field from NCEP is employed for sea surface temperature (SST). Terrain elevation data are obtained from the ASTER Global Digital Elevation Map (GDEM) by the United States Geological Survey (USGS; Slater et al. 2011) with a resolution of 30 m, and land use information from the Corine (Coordination of Information on the Environment) database (2010) with 250-m resolution. The main physics options and parameterizations used are summarized in Table 1.
WRF Model physical schemes and properties.
The inner nest operates at cloud-resolving scales, leading to improved representation of extreme precipitation events. These events often result in floods, even in urban areas, with streets inundated, vehicles swept away by fast-flowing floodwaters, disruptions to services like electricity and telecommunications, and unfortunately, loss of lives. A flooding event that affected cities like Mecca (Makkah) and Riyadh with high accumulated precipitation is depicted in Fig. 4. This adverse weather was forecasted between 14 and 15 January 2022 (Figs. 4a,c). Additionally, the description of the vertical structure of the atmosphere in critical locations assists operational meteorologists, air traffic controllers, and scientists, especially during extreme conditions (Fig. 4b).
(a) 3-h accumulated precipitation (mm) over the Arabian Peninsula (WRF inner domain—1800 UTC 14 Jan 2022). (b) Skew T diagram showing temperature (black line), dewpoint (blue line), and wind barbs for Jeddah (WRF inner domain—1800 UTC 14 Jan 2022), and (c) 24-h accumulated precipitation (mm) over the Arabian Peninsula (WRF inner domain—14 Jan 2022).
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
While the primary purpose of the coarse domain is to serve as an intermediate step for high-resolution forecasts, it also provides valuable information about the weather conditions in the surrounding regions (Fig. 5a). Additionally, it effectively resolves interactions with neighboring regions such as the Indian Ocean, Africa, and the eastern Mediterranean. A notable example is the presence of the Tokar Gap jet, which is captured by the coarse domain (Fig. 5b). This highlights the system’s ability to represent and account for important atmospheric features and dynamics in the broader geographical context.
(a) Temperature at 2 m (°C) and MSLP (hPa) during one of the hottest days of June 2021 (WRF outer domain—1200 UTC 2 Jun 2021). (b) Wind speed (m s−1) during a Tokar Gap jet event (red circle, WRF outer domain—0600 UTC 27 Jul 2021).
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
2) WAM (cycles 0000 and 1200 UTC)
The wave conditions are forecasted based on the third-generation spectral wave model WAM Cycle4 (Komen et al. 1994; WAMDI Group 1988). WAM solves the wave transport equation explicitly and it represents the physics of wave evolution using a two-dimensional wave spectrum. The rate of change of the wave spectrum is determined by the energy transfer from wind, the nonlinear wave–wave interactions and finally the dissipation by white capping. It can run for both deep and shallow waters since bottom dissipation source function and refraction terms are also included. It can run on a spherical latitude–longitude grid and can be used in any given regional or global grid with a prescribed bathymetric dataset. The version used incorporates state-of-the-art features, such as an advanced corner transport upstream advection scheme that provides a uniform propagation in all directions and a parameterization of shallow water effects that affects both the time evolution of the wave spectrum and the determination of the kurtosis of the wave field (Janssen and Onorato 2007).
WAM predicts the directional spectra and the spectral integrated wave parameters such as significant wave height, mean wave direction and frequency at each grid point at selected output times. It makes a distinction between wind–sea and swell and is able to provide the height, the mean direction and the mean frequency of each component. Finally, it describes the extreme conditions based on two extreme wave parameters, the average maximum wave height and the corresponding wave period (Mori and Janssen 2006).
The system has been configured for the meteorological conditions of the Arabian Peninsula and has been set up to cover three sea areas: the Arabian Sea, the Arabian Gulf, and the Red Sea. For each area a different configuration has been used based on their physiographic characteristics. To describe the wave conditions of the Arabian Sea a 0.05° resolution domain is nested within a 0.25° outer one that covers the Indian Ocean and southern seas (WAM-d01, Fig. 6). The latter is extended in an area that is adequate to model the remote-generated swell that reaches the Arabian Sea, particularly during the monsoonal season (Fig. 6). It feeds the inner grid with spectral boundary conditions every hour. For the Indian Ocean domain, the surface winds from the 0.25° GFS model configuration are utilized. The nested Arabian Sea domain (WAM-d02) is using winds from the outer WRF grid. This setup is able to resolve the impact of the dominant monsoonal activity, shamal winds, and tropical cyclones on the wave characteristics of Arabian Sea (Fig. 7).
The summer monsoonal activity, as depicted by the operational wave modeling system WAM (0000 UTC 21 Jun 2021) showing (a) the significant wave height, (b) the swell height, and (c) the mean wave period. This coarser domain that covers the entire Indian Ocean and southern seas provides the information needed regarding the long-period remote-generated swells.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
Typical wave fields for the Red Sea and Arabian Gulf, as depicted by the operational wave modeling system WAM, are presented. (a) The wave field of Red Sea under the influence of the Tokar Gap winds (27 Jul 2021) and (b) a shamal event in the Arabian Gulf (0600 UTC 13 Jun 2021). The cyclonic storm Shaheen as depicted by the Arabian Sea domain of WAM (0000 UTC 3 Oct 2021) in (c) web and (d) zoomed view. The wave fields in all cases are described by significant wave height.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
For the Red Sea (WAM-d03) and Arabian Gulf (WAM-d04) domains, a higher resolution configuration has been adopted since both areas are closed basins governed by local atmospheric circulations that directly affect the wave field and need to be adequately resolved (Fig. 7). Both domains are set up with a 1/60° resolution. They utilize high resolution winds from the fine WRF domain for the first 5 days while for the remaining 5 days the winds are interpolated from the coarse one. The WAM domains are illustrated in Fig. 3 with white frames.
The bathymetry for these model configurations is derived from the ETOPO1 dataset (Amante and Eakins 2009). All the domains run in two cycles: 0000 and 1200 UTC.
3) RAMS-ICLAMS (cycles 0600 and 1800 UTC)
Due to the large spatial extend of the Arabian Peninsula and the different prevailing weather patterns triggered from multivariate factors, forecasting information from various sources is necessary. To counterbalance the uncertainty in weather forecasting and support decision-making purposes, a second NWP model was selected, namely, the Regional Atmospheric Modeling System–Integrated Community Limited Area Modeling System (RAMS-ICLAMS) model. Model characteristics together with some distinct capabilities, such as the online treatment of natural aerosols and an explicit cloud droplet activation scheme, can offer a useful supplementary role toward the forecasting variability in areas of specific interest.
The RAMS-ICLAMS system (Solomos et al. 2011), developed within the Atmospheric Modeling and Weather Forecasting Group of the University of Athens is an advanced version of the RAMS model (Cotton et al. 2003). It has been tested and used in various academic and operational initiatives (Patlakas et al. 2021a; Stathopoulos et al. 2020a,b). Aerosol parameterization schemes are employed for the implementation of mineral dust and sea salt emissions. In particular, the core RAMS model is fully coupled with the SKIRON dust submodel (Astitha et al. 2010; Kallos et al. 2009; Kushta et al. 2014; Nickovic et al. 2001; Solomos et al. 2012, 2011; Spyrou et al. 2010), using eight particle size bins as proposed by Pérez et al. (2006). A sea salt cycle model is also implemented (Gong 2003) and emitted sea salt particles are estimated with the use of white cap fraction as a function of surface wind speed and a semiempirical relationship for the description of size distribution. Additionally, an updated cloud droplet activation scheme is available, offering a microphysical communication between aerosols and clouds (Fountoukis and Nenes 2005). Aerosol characteristics such as concentration, size distribution and composition are in constant interaction with cloud dynamics properties. The radiation scheme, Rapid Radiative Transfer Model (RRTM; Iacono et al. 2008), also considers the effects of natural aerosols depending on their size and water content. Altogether, natural aerosols contribute to model calculations through feedback mechanisms including direct, semidirect and indirect effects in the radiation scheme and the ice nuclei (IN), cloud condensation nuclei (CCN), and giant cloud condensation nuclei (GCCN) estimations.
The RAMS-ICLAMS model has been configured to run in three different setups. The inner grids are focused on three different areas of Saudi Arabia, while the outer ones (RAMS-ICLAMS-d01) overlap and cover almost the entire peninsula (Fig. 8). The selected areas exhibit particular interest to the NCM operational center, and the configurations are according to characteristic scales of the landscape variability and the prevailing local phenomena. The nesting ratio between the coarse and the fine grids is equal to 4. The northwest domain (RAMS-ICLAMS-d03) that includes Jeddah, Mecca, and Medina has a horizontal grid resolution of 2 km. The east province domain (RAMS-ICLAMS-d04) that includes Riyadh, Dammam–Dhahran–Al Khobar, Jubail, and up to the Kuwait border is set up to 2.5 km and the southwest domain (RAMS-ICLAMS-d02) that includes Abha, Jizan, and other major cities is configured at 3 km. In the vertical, the model has the necessary number of vertical layers to cover the deep convection. The model is set up to have 40 vertically stretched levels, extended to 22 000 m, with the first 25 levels reaching up to 10 000 m. The first level is set to be at 100 m and the vertical grid stretch ratio is 1.10. The maximum level distance is set to 1000 m. The forecasting horizon is limited to 72 h.
The model uses Shuttle Radar Topography Mission (SRTM) (90 m) elevation data. The soil texture and properties are retrieved from FAO while for the vegetation and land cover Olson Global Ecosystem (OGE) categorization (30 arc s) is employed. The SST used is the daily analysis field from NCEP with a resolution of 0.083°. Further physical schemes and properties employed are listed in Table 2.
RAMS-ICLAMS model physical schemes and properties.
The current configurations aim to cover in detail the prevailing meteorological conditions in specific locations of interest such as densely populated and industrial areas. This implementation is aiming beyond the traditional operational runs, focusing on a better understanding of the dust and sea salt cycle in the atmosphere alongside their direct, indirect and semidirect interactions. Particular emphasis is given to extreme events where more targeted analysis is performed when needed (Figs. 9a,b). The northwest and southwest model domains enclose wide arid and extra-arid regions together with extended mountainous areas. The interaction of this complex landscape with the Red Sea leads to significant variations in temperature and humidity. These features serve as key triggering mechanisms for local-scale phenomena such as sea and mountain breezes, as well as more intense events like dust storms, heat waves, and rainfall events. The last model domain focuses on the eastern province with the intention to capture the local climate and especially the northerly and northwesterly winds (shamal winds) and the moist sea breeze systems. The physical schemes and parameterizations of RAMS-ICLAMS model have the potential to represent all these kinds of events. Indicative examples from the simulations in each subdomain are given in Fig. 9. An extreme dust event affecting mainly the western Arabian Peninsula and causing several issues in Riyadh took place in March 2021. The model forecast for the event is presented in Fig. 9a. In Fig. 9b a further analysis of the event is presented. A typical mountain breeze during the day depicted in Fig. 9c, showcases the domain’s (d03) ability to capture the dynamics of the local atmospheric conditions. Furthermore, Fig. 9d, illustrates a flood event that took place in the eastern province in July 2021.
(a) Dust concentration at the surface (μg m−3); (b) vertical cross section of dust concentration (μg m−3), equivalent potential temperature (K), and streamlines at 49°E (domain d02—1900 UTC 12 Mar 2021); (c) 10-m wind speed intensity (m s−1) and direction (vectors) (domain d03—1500 UTC 11 Jul 2021); and (d) last 3-h accumulated precipitation (mm) (domain d04—23 1500 UTC Jul 2021).
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
Overall, these examples highlight the crucial role of RAMS-ICLAMS as a key element in the forecasting system, particularly when investigating and understanding extreme weather phenomena and localized events.
c. Regional atmospheric and dust forecasting/regional oceanic forecasting
1) WRF-Dust (cycles 0600 and 1800 UTC)
To improve the forecasting of the dust cycle in the region, an additional version of the WRF model, known as WRF-chem, has been employed alongside RAMS-ICLAMS. WRF-chem is a full online atmospheric chemistry model with the ability to interact with physics via aerosols affecting radiation and microphysics. For the dust emission calculations in WRF-chem, the model utilizes the Chemistry Aerosol Radiation and Transport (GOCART) model (Chin et al. 2000; Ginoux et al. 2001), which employs a five-bin particles size distribution (0.2–2, 2–3.6, 3.6–6, 6–12, 12–20 μm). Several approaches have been introduced resulting in three dust emission schemes compatible with GOCART in WRF-chem. The first one the original GOCART WRF scheme (dust_opt = 1) (Belly 1964; Gillette and Passi 1988), the second is the Air Force Weather Agency (AFWA) scheme (dust_opt = 3) (Marticorena and Bergametti 1995; Su and Fung 2015), and the last one, the University of Cologne (UoC) scheme (dust_opt = 4) (Shao 2001, 2004; Shao et al. 2011). More details on the schemes can be found in LeGrand et al. (2019) and Ukhov et al. (2021).
WRF-Dust component is configured to cover a substantial portion of Africa, Europe and Asia through a single domain with spatial resolution of 4.5 km (Fig. 10). In the vertical dimension it has 48 layers. It is designed to provide a 10-day forecast (+12-h spinup time), running in two cycles: 0600 and 1800 UTC. The initial and boundary conditions, the terrain elevation, the land use data, the main physics options and the employed parameterizations are the same as the other WRF runs. For the dust production parameterization, the GOCART dust emissions are employed (LeGrand et al. 2019).
WRF-Dust and NEMO model domains. The red triangles correspond to the Aeronet stations used for the evaluation.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
The selection of the model domain was specifically done to capture the long-range transport of dust including the transport from North Africa toward the Middle East, from East Africa toward the western part of Saudi Arabia, and mainly from Iraq and Iran toward the eastern province of Saudi Arabia (Kallos et al. 2020). This domain allows for a comprehensive examination of dust conditions affecting neighboring regions, as depicted in Fig. 11a. Additionally, such a large domain offers the opportunity to explore other atmospheric phenomena and features, such as the characteristics of the jet stream (Fig. 11b), monsoonal activity, and more.
(a) Dust load (mg m−2) as calculated for 0600 UTC 21 Jul 2021and (b) jet stream at 0000 UTC 31 Jan 2022 (wind speed at 200 hPa; m s−1).
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
Using large domains with high resolution offers also the advantage of forecasting and analyzing extreme regional events, such as cyclones, in greater detail. In cases of interest, from a forecasting or a scientific perspective, these events can be further analyzed (Fig. 12).
Wind speed at 10 m shown (a) in knots (kt; 1 kt ≈ 0.51 m s−1) and (b) in meters per second (m s−1), (c) vertical cross section of wind speed (m s−1) and equivalent potential temperature (K) at 58.95°E, and (d) 24-h accumulated precipitation (mm) during the passage of the Severe Cyclonic Storm Shaheen (0000 UTC 3–4 Oct 2021).
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
2) NEMO (cycle 0000 UTC)
The ocean circulation model is based on NEMOv3.6 (Nucleus for European Modeling of the Ocean; http://www.nemo-ocean.eu/; NEMO 2023; Madec et al. 2016), which is a state-of-the-art modeling framework for oceanographic research and operational oceanography. The NEMO platform is used by several monitoring forecasting services worldwide (OceanPredict 2023; https://oceanpredict.org/science/operational-ocean-forecasting-systems, ETOOFS 2022). It solves the primitive equations in spherical coordinates. The NEMO architecture includes five prognostic variables relative to the ocean circulation: three-dimensional velocities, nonlinear sea surface height, conservative temperature, and absolute salinity, projected onto an Arakawa C-grid type.
A regional high-resolution configuration of NEMO is used covering the Arabian Sea, the Red Sea, and the Arabian Gulf (32.1°–70.9°E, 5°–30.6°N) as shown in Fig. 10. The primitive equations are discretized on a 1/36° curvilinear Arakawa C-grid, based on the so-called “ORCA” tripolar grid (Bernard et al. 2006), assuming hydrostatic and Boussinesq approximations. The model domain is configured as an exact 3:1 refinement of the global operational system PSY4V3R1 (Copernicus 2023; http://marine.copernicus.eu/). The vertical grid has 50 unevenly spaced geopotential levels, with 22 levels in the upper 100 m and a resolution that decreases from ∼1 m in the upper 10 m to more than 400 m in the deep ocean. The bathymetry was constructed using the General Bathymetric Chart of the Oceans (GEBCO) 08 dataset (resolution of 30 arc s; GEBCO 2023; http://www.gebco.net), which was filtered (using a Shapiro filter) and modified manually in critical areas such as the Strait of Bab el Mandeb and the Strait of Hormuz.
Meteorological fields are provided at 1-hourly intervals from the WRF-Dust modeling configuration. Evaporation, latent and sensible heat fluxes, as well as wind stresses are computed within the model using the CORE bulk forcing algorithm (Large and Yeager 2004). As a river source, we consider the Shatt al-Arab in the northern Arabian Gulf formed by the confluence of the Tigris, Euphrates, and Karun Rivers. River runoff is imposed on the model as a surface freshwater flux of constant salinity 0.1 with a minimum value of 350 m3 s−1 in October and a maximum of 650 m3 s−1 in April (Kämpf and Sadrinasab 2006; Vasou et al. 2020).
The initial and open boundary conditions for the ocean state are provided by the Copernicus Marine Service (CMEMS; http://marine.copernicus.eu/) global analyses at 1/12° horizontal resolution and daily frequency. The CMEMS analyses consist of the full state vector (i.e., potential temperature, salinity, ocean currents and sea surface height). Tidal forcing is also applied at the open boundaries as the sum of 13 tidal constituents (Q1, O1, P1, K1, N2, M2, S2, K2, Mf, Mm, M4, Ms4, and Mn4). The amplitude and phase of each tidal constituent are provided by the TPXO 7.1 global ocean tide model (Egbert and Erofeeva 2002). The amplitude and phase of the major semidiurnal constituent in the domain is presented in Fig. 13. The M2 propagation shows high amplitudes along the Strait of Hormuz and the Arabian Gulf, indicating the importance of including tidal forcing in the region. The two amphidromic points of M2 (zero-amplitude) located in the Arabian Gulf and the one in the central Red Sea are displayed as well. Finally, an inverse barometer signal is also added at the open boundaries to consider the response of the sea level to atmospheric pressure.
Amplitude (cm; shaded in colors) overlaid with phase contours (black lines) of modeled M2 tidal constituent. The black arrows indicate the location of the Strait of Hormuz and the Strait of Bab El Mandeb.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
In terms of physics, vertical mixing is based on the generic length scale (GLS) turbulent closure scheme, using k-ϵ parameterization and Canuto type-A stability functions (Umlauf and Burchard 2003, 2005). The horizontal tracers’ advection is computed with the QUICKEST scheme (Leonard 1979). The bottom roughness is based on a quadratic drag coefficient applied with a logarithmic formulation, being important over shallow shelf areas (Karagiorgos et al. 2020).
In a similar way with the WRF-Dust domain, the ocean model configuration was selected to cover the extended area of the Arabian marginal seas and gulfs, being of paramount importance for maritime activities in the region. Figure 14 shows an example of the ocean circulation and SST patterns during the passage of the Shaheen cyclone nearing landfall in the Gulf of Oman on 3 October 2021. An intense SST cooling can be observed in the region occupied by the storm (black arrow in Fig. 14a) as a result of ocean heat loss by air–sea fluxes and Ekman pumping due to wind stress curl and surface current divergence. Other prominent ocean circulation features in the model domain are the meanders of the Great Whirl anticyclonic vortex and the Socotra Gyre, observed off the east coast of Africa (black box in Fig. 14b) and associated with the end of the monsoon season. The seasonal variability of the Great Whirl also affects coastal upwelling and primary production, depending on the location, intensity and shape of the eddies in the region (Liao et al. 2016; Wiggert et al. 2005).
(a) NEMO sea surface temperature and (b) currents during the passage of the Severe Cyclonic Storm Shaheen (0000 UTC 3 Oct 2021). The black arrow in (a) indicates the SST cooling due to the passage of the storm. The black box in (b) highlights the location of the Great Whirl–Socotra Gyre system.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
Figures 15a and 15d show the sea surface height and circulation patterns focusing on the Gulf of Aden and the Gulf of Oman on the same date. The vigorous mesoscale eddy field in these Gulfs dominates, influencing the pathway and the fate of the water masses exchanged between the Red Sea and the Arabian Gulf, and the Indian Ocean through the Strait of Bab El Mandeb and the Strait of Hormuz, respectively (Bower et al. 2002; L’Hégaret et al. 2016). The vertical cross sections in Figs. 15b and 15c denote the signature of the cold and saline water mass near the bottom called the Red Sea Outflow Water (RSOW). The formation of RSOW is associated with open-ocean convection in the northern part of the Red Sea during the winter, as suggested by simulation studies by Sofianos and Johns (2003). In Fig. 15f, high salinities are observed at depths up to 40 m in agreement with observation data in the Strait of Hormuz by Johns et al. (2003), which found salinities greater than 38 psu in the surface layer during the fall period. In the deep southern part of the Strait of Hormuz, there is also a confined cold and saline water (Figs. 15e,f) that is associated with the Arabian Gulf outflow (Chao et al. 1992; Johns et al. 2003; Vasou et al. 2020).
Snapshots of sea surface height (m) and circulation streamlines surrounding (a) the Gulf of Aden and (d) the Gulf of Oman, respectively, on 3 Oct 2021. Hydrographic sections of temperature and salinity in the (b),(c) Strait of Bab El Mandeb and (e),(f) Strait of Hormuz are shown for the same date.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
4. System evaluation
a. Atmospheric component
To assess the quality and performance of the numerical weather prediction systems, an evaluation procedure is performed. This is useful not only for monitoring the system but also for improving the model setup and thus the final outcome with respect to the requirements. For the needs of this study, the evaluation covers a period from June 2020 to May 2021.
The WRF model is utilized to assess weather conditions in Saudi Arabia and the surrounding region. The high-resolution domain, which covers the entire Arabian Peninsula, is considered the core of the system, and its performance is crucial. Surface wind speed and temperature are among the standard meteorological parameters. Hourly wind speed, estimated at a height of 10 m AGL and temperature estimated at 2 m AGL are evaluated against measurements from surface meteorological stations [the evaluation metrics used can be found in Nascimento et al. (2021)]. The model data are extracted employing the “nearest model gridpoint” methodology. This is considered to be a commonly accepted approach for the extraction of time series in specific locations when working with model results. However, it might be less accurate in areas with significant topographic variations or near the coastlines.
Beginning with temperature, the model performs well and efficiently. The correlation is around 0.93, slightly declining for later forecasting times (Fig. 16). A lower correlation is evident during the summer season. RMSE ranges between 2.6° and 3.3°C with higher values anticipated during warm periods due to intraday variability. The model exhibits a slight underprediction, indicated by a bias ranging between −0.3 and −1.2 for spring (March–April–May), summer (June–July–August), and autumn (September–October–November). Higher correlation and slightly positive values are found during winter (December–January–February). In all cases the model performs worse for greater forecasting horizons. This is associated with the initial errors induced by the boundary conditions. These errors are accumulated and multiplied through the numerical analysis leading to growing biases. It should be noted that bias values derived from the initial conditions were larger before applying the assimilation procedure. The standard deviation is smaller during the winter period, something expected due to the lower values and daily variability.
Taylor diagrams (Taylor 2001) for temperature at 2 m (°C) (WRF). Dot colors correspond to the forecast lead time, and the red star corresponds to the observations. Radial distance represents the standard deviations of the datasets (simulated and observed) (°C), and azimuthal angle represents model–data correlation. Gray arcs represent centered root-mean-square difference (°C) between model and data.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
Modeled and measured values of wind speed are in a general good agreement, presenting an increased level of model simulation accuracy (Fig. 17). The metrics confirm that the model forecast skill declines for later forecasting times. Correlation drops by more than 0.1 for the fifth day of forecast, something expected due to the errors growing for larger simulation periods. It should be noted that the inherent variability of wind speed leads to smaller correlation values as compared to other meteorological parameters, such as temperature. The RMSE ranges between 2.1 and 2.5 approximately. The derived outcome suggests a small variability of error with slightly higher values during the summer. Bias on the other hand, is lower during summer and spring followed by winter and autumn with values around 0.35, 0.39, 0.57, and 0.69, respectively. This outcome is related to the lower wind speed values observed during the dry period.
Taylor diagrams for wind speed at 10 m (m s−1) (WRF). Diagram features are as in Fig. 16, but for wind speed.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
Although the region is typically characterized as arid or semiarid, the southeast receives significant amounts of precipitation due to factors such as topography, warm seas, and monsoonal activity. Moreover, the occurrence of severe precipitation often leads to floods, posing a threat to human life and activities. As a result, accurate forecasting of precipitation becomes of utmost importance, particularly for this component. Toward this way we analyze the behavior of WRF by comparing its predictions against observations and satellite data. Regarding the first, the model’s forecasts are compared against in situ observations obtained from a wide network of WMO stations and data provided by local authorities (see Fig. 2). Despite the abundance of data available, only a fraction could be utilized due to common issues with observations, such as missing data and lack of quality. Additionally, and due to the relatively limited precipitation occurrence across most seasons, our analysis covers the entire year.
The evaluation metrics were based on daily accumulated precipitation, revealing the model’s reliability with correlation coefficients ranging between 0.71 and 0.59 (Fig. 18). The weakest forecast performance was observed on day 4, which could be attributed to specific events where the forecast for day 5 might have been more accurate. Apart from that, we encountered a few cases of model failures caused by technical issues, potentially impacting the sample in some instances. The standard deviation ranges between 6.23 and 8.32 while the RMSD is between 6.19 and 8.59. In both cases, the metrics generally decreased as the forecasting horizon extended, except for the fourth day where there was an exception. For the first forecasting day, we found a negative bias of −0.27 that reaches values of −0.69 for the fifth one.
Taylor diagrams for daily accumulated precipitation (mm) (WRF). Diagram features are as in Fig. 16, but for precipitation.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
It is worth noting that extracting data from the nearest grid point may lead to inferior statistics, as the spatiotemporal variability of precipitation is greater than that of wind speed or temperature. Although other approaches, such as incorporating weighted factors, might offer better results, they are beyond the scope of this work.
To further quantify the model performance, an additional evaluation employing threat scores is performed. Toward this way, several metrics are calculated such as accuracy (ACC), probability of detection (POD), success ratio (SR), false alarm ratio (FAR), critical success index (CSI) and bias (Nascimento et al. 2021). The dataset used for this is the Global Precipitation Measurement Integrated Multi-satellitE Retrievals for GPM (GPM-IMERG) product (Final Run). GPM-IMERG (Huffman et al. 2014) comes from intercalibration and interpolation of many satellite microwave precipitation estimates, together with microwave-calibrated infrared (IR) satellite estimates, precipitation gauge analyses, and potentially other precipitation estimators at fine time and space scales over the entire globe.
The results show reasonable performance for all forecasting days with model accuracy ranging from 93.69% to 94.82% within the forecasting horizon (Table 3). The POD and SR show a decline for longer range forecasts, as uncertainties in numerical prediction become dominant. The FAR variable is complementary to SR. The CSI is the most demanding index, because it includes both misses (false negatives) and false alarms (false positives). A declining trend from 30.26% on the first day to 19.35% on the fifth day is depicted as well. Focusing on the bias index, the model shows a slight underestimation of rain events, as its values lie between 0.641 (fifth day) and 0.719 (second day).
Evaluation matrix (WRF, fine domain).
WRF-Dust component is primarily used to forecast dust conditions covering a large domain that includes almost the entire Middle East and North Africa. Its evaluation here will be limited to the basic meteorological variables and dust. The correlation regarding the temperature forecasts is around 0.95, slightly declining for later forecasting times (Fig. 19). A higher correlation is evident during autumn and spring while RMSE ranges between 1.6° and 2°C. Bias is slightly negative ranging between −0.4 and −0.9 for spring, summer and autumn (not shown). Higher correlation and lower values are found during winter reaching −0.2°C for the 10th forecasting day. In all cases the model performs worse for later forecasting times.
Taylor diagrams for temperature at 2 m (°C) (WRF-Dust). Diagram features are as in Fig. 16, but for temperature.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
Wind speed modeled and measured values are in general good agreement (Fig. 20). Correlation drops by 0.1–0.2 for the 10th day of forecast, something expected due to the errors growing for larger simulation periods. The RMSE ranges between 1.6 and 1.9 approximately (not shown). The derived outcome suggests a small variability of error with slightly higher values during the summer. Bias on the other hand, is lower during the warm season.
Taylor diagrams for wind speed at 10 m (m s−1) (WRF-Dust). Diagram features are as in Fig. 16, but for wind speed.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
Focusing on dust emissions, the evaluation was based on the aerosol optical depth (AOD). Modeled AOD at 550 nm (AOD550) was compared against estimates retrieved from Aerosol Robotic Network (AERONET 2023; https://aeronet.gsfc.nasa.gov) stations. The model seems to perform quite well following the distribution of AOD550. In summer, autumn and winter there is a slight underestimation especially in the distribution tail. On the other hand, there is an obvious overestimation during spring (Fig. 21).
Q-Q plots of modeled (WRF-Dust) and observed AOD550 for (a) summer, (b) autumn, (c) winter, and (d) spring.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
The metrics confirm that the model forecast skill is acceptable with the correlation ranging between 0.53 and 0.71 depending on the season (Table 4). The RMSE ranges between 0.11 and 0.17 nm approximately. The analysis suggests a small variability of error with slightly higher values during the summer. Bias on the other hand, is minimal during the entire year. It should be noted that dust production and transport are quite complex processes that involve various uncertainties. These uncertainties stem from several factors, including wind speed, precipitation, and other meteorological conditions. It is crucial to acknowledge that the skill of dust aerosol forecasting is significantly influenced by these uncertainties.
Evaluation metrics for AOD550 (WRF-Dust).
The last part of the atmospheric component evaluated here is RAMS-ICLAMS, which is a forecast system mainly used to target areas of interest. Therefore, its evaluation will cover the entire spectrum described above. To examine the performance, wind speed, temperature and precipitation are compared for a horizon of two forecasting days while dust for one.
For all seasons, an increased correlation (to the observations) is depicted in the predicted temperatures with values close to 0.9, while the RMSE differences lie in the range between 2.5 and 3.5 (Fig. 22). Standard deviations appear higher in both modeled and observed temperatures in autumn and spring seasons when diurnal variation is more considerable. In regard to the relation of the level of accuracy with the forecasting horizon, the marginal differences presented in the statistical indicators showcase insignificant variations in the performance within the first 48 h.
Taylor diagrams for temperature at 2 m (°C) (RAMS-ICLAMS). Diagram features are as in Fig. 16, but for temperature.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
For the wind speed, the statistical evaluation does not manifest important seasonal variability in the model accuracy (Fig. 23). Moreover, as in the evaluation of temperature, there is an absence of important differences in forecasting performance between the first and second day of simulation. Correlation values are ranging between 0.4 and 0.6 with the lower values emerging in the summer months and the higher in spring. Similarly, the standard deviation of the modeled values is generally higher. During the summer season, however, the values are similar to the outcome of the observations. The lowest root-mean-square differences emerge in summer, while the highest are found in spring.
Taylor diagrams for wind speed at 10 m (m s−1) (RAMS-ICLAMS). Diagram features are as in Fig. 16, but for wind speed.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
On an annual basis, the precipitation exhibits a satisfactory correlation, specifically on the first forecasting day, considering the limited number of rain stations available (Fig. 24). The results of the mean bias error indicate a slight tendency for underestimation in both forecasting horizons (not shown). There are also noticeable differences in the centered RMSD. The last is slightly lower on the first day of the forecast compared to the second one.
Taylor diagrams for daily accumulated precipitation (mm) (RAMS-ICLAMS). Diagram features are as in Fig. 16, but for precipitation.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
In relation to the aerosol optical depth, an underestimation is overall evident in most of the quantiles, apart from the summer months where the modeled values slightly overestimate the measured ones in quantiles above 0.25 (Fig. 25). Among the four seasons, in autumn and winter observed and modeled distributions match better in most of the quantiles, diverging, however, significantly in the distribution tail.
Q-Q plots of modeled (RAMS-ICLAMS) and observed AOD550 for (a) summer, (b) autumn, (c) winter, and (d) spring.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
b. Wave component
This section provides an evaluation analysis of the wave model WAM. The satellite data SARAL/AltiKa GDR-T is used to assess the quality of the significant wave height derived from the wave model (Verron et al. 2021). Utilizing such data poses several difficulties. Shallow waters, particularly in the Arabian Sea, complex coastlines, the presence of small islets (especially in the southwest Arabian Gulf and Red Sea), and various objects such as ships and platforms contribute to the challenges. Two other important factors should also be considered: the repetition time of the satellite in the Red Sea and Arabian Gulf region is very coarse in time and the accuracy near the land is bad. Moreover, erroneous wave height estimates can also result from the sensitivity of the Ka-band backscatter coefficient to areas with very low waves (surface reflections) and environmental factors like clouds and rain.
Taking all these under consideration, the wave system has been evaluated, and the results indicate that the model is capable of accurately simulating the wave fields over the three basins of interest, Red Sea, Arabian Gulf, and Arabian Sea, during all seasons (Fig. 26). Overall, the lowest RMSE values are found during the winter, slightly higher (between 0.3 and 0.5) during spring and autumn, while the highest ones are found in summer. The model performs best in terms of correlation during summer. It drops by around 0.2 for the 10th day of forecast, for all seasons, something expected due to the errors growing for larger simulation periods. It should be noted that the variability of wind speed is inherited in waves as this is their driving factor. It is obvious that the forecast skill of the model declines for later forecasting times, following the behavior of the atmospheric model. However, the performance is still acceptable during the whole 10-day forecast period.
Taylor diagrams comparing modeled significant wave height (m) against the significant wave height as measured by the SARAL/Altika satellite over the three basins Red Sea, Arabian Gulf, and Arabian Sea for all seasons. Diagram features are as in Fig. 16, but for significant wave height.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
c. Ocean component
This section provides an evaluation analysis of the NEMO ocean forecasting system. The modeled SST is compared with available satellite data. The observational dataset used is a daily analysis (“Level 4”) of the Operational Sea Surface Temperature and Ice Analysis (OSTIA) SST product (Donlon et al. 2012) on a global 0.05° grid.
The quantitative metrics confirm that the model forecasts were skillful in estimating the SST, shown to be well correlated with the observations exceeding 95% for all forecast lead times (Fig. 27). The highest values of standard deviation and RMSD were observed during spring, associated with mixed layer restratification processes in the model during the spring shoaling of the thermocline. On the other hand, late-autumn and early-winter SST shows good statistical properties with increasing forecast horizon (i.e., less than 0.2°C RMSD and around 0.4°–0.5°C standard deviation), where the upper ocean is controlled by intense cooling in the northern subregions.
Taylor diagrams comparing model SST (°C) against OSTIA data over the modeled domain for all seasons. Diagram features are as in Fig. 16, but for SST.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
5. Visualization
Visualizing and providing the model output in an informative and clear context is of great importance both for forecasters and for several authorities that use the system. The system management and nowcasting is performed through the Advanced Weather Interactive Processing System (AWIPS) (UCAR 2023; https://www.unidata.ucar.edu/software/awips2/). AWIPS is a software package for the processing, visualization and analysis of all types of meteorological data. All the model outputs, the selected surface and upper-air observations and satellite images are incorporated into AWIPS for each cycle. In addition to the running models and observations, the GFS global forecasts are included. Meteorologists monitor the weather conditions on a 24–7 basis, processing modeled parameters as well as in situ, radar and satellite data (Fig. 28).
(a) An example of the display of the data processing and visualization system AWIPS for GFS (including the tailor-made utilities). (b) Vertical cross section of wind speed (kt), potential temperature (K), and relative humidity (%) during an extreme precipitation event that took place on 1 Jan 2022 (WRF, fine domain).
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
At the same time, and in order to serve the public and private sector, a mixture of easy-to-follow and more sophisticated outputs are provided through a web page specifically designed during this collaboration. All model outputs and special features are easily accessible via the homepage (Fig. 29).
Website (1800 UTC 15 Sep 2021).
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
The user can navigate and choose from among different models and the last three cycles according to their preference. The atmospheric component includes surface and upper air parameters such as temperature, wind speed and direction, geopotential height, atmospheric pressure, cloud cover, relative humidity, and dust concentration, depicted for all available domains. A similar approach is taken for the oceanic and wave components. Apart from its industrial and operational applications, all data are also preserved for scientific purposes.
To facilitate the extraction and display of local meteorological forecasting variables, an interactive system has been developed. This system incorporates a database where all model forecasting data are stored. It includes a window displaying an ESRI geomorphological mapping system and provides the necessary interface for selecting the desired model (coarse and high-resolution weather forecasts, wave and ocean forecasts for the peninsula and the surrounding seas) and the area to be covered. Users can select the desired location either by clicking on a point, typing the location name (e.g., city name), or entering the geographic coordinates. An example is shown in Fig. 30.
Interactive forecast example (15 Sep 2021).
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
6. Conclusions
Weather forecasting plays a crucial role in various economic and social sectors while providing early warnings can save lives and mitigate risks. Moreover, the broad range of applications associated with numerical weather prediction demonstrates the significant impact accurate forecasts can have on the economy and society as a whole. To address these needs, the National Center for Meteorology of Saudi Arabia has developed a comprehensive, multifaceted forecasting system. The system incorporates atmospheric, wave and ocean model components, along with software applications for forecasting, monitoring and analyzing the meteorological conditions across the Arabian Peninsula. The system has been configured to efficiently capture the diverse weather patterns and local characteristics of the region. Notably, a methodology blending different models and systems has been employed, as opposed to a single model approach. Furthermore, a data assimilation system and advanced analysis tools enhance the system’s capabilities.
Since its deployment in 2020, the system has undergone extensive evaluation. The atmospheric component shows good performance for short-term forecasts (1–3 or 1–5 days, depending on the variable), with some degradation observed for longer forecasting horizons (5–10 days). The ocean and wave components exhibit even better performance due to the lower variability of the ocean and the sea state as compared to the atmosphere. Throughout its operations, the system has proven to be a valuable tool for collaborating organizations and authorities working with the National Center for Meteorology.
Looking ahead, future plans include several key enhancements. First, we aim to incorporate a hydrologic component into the system, allowing for a more comprehensive understanding and prediction of hydrological processes. Additionally, alternative assimilation techniques are actively being explored in an effort to improve the accuracy and reliability of the initial conditions used by models. Moreover, the implementation of an additional atmospheric and ocean modeling system with distinct characteristics and the employment of ensemble prediction techniques (Vervatis et al. 2021a,b; Ferrarin et al. 2023), will help us better quantify the uncertainty associated with the forecasts. These advancements will provide users with valuable information about the range of possible outcomes, enabling more informed decision-making and improving predictability.
Acknowledgments.
This work was supported by the Statement of Work OPP-0000350118, which was carried out under the Master Agreement for Services MSA102021/WeMet/v1.0.
Data availability statement.
Due to confidentiality agreements, supporting data can only be made available to researchers upon agreement. More details are available from the National Center for Meteorology of the Kingdom of Saudi Arabia (https://ncm.gov.sa). Software for this research is available in the in-text data citation references.
APPENDIX
Description of the Supercomputer System at NCM
NCM required a robust computer system to handle computationally expensive models used for product generation. These algorithms often needed capability-oriented systems to be executed as single or small jobs. Ancillary processing further increased capacity demands. Additionally, NCM’s workload encompassed data assimilation and analysis for a more accurate description of past atmospheric and oceanic states. This portion of the work can strain capacity resources more than capability. All these tasks require very high speed input/output (I/O) for large amounts of data and extremely high data storage capacities to save and analyze the past atmosphere states. I/O requirements (transaction rate, bandwidth, and capacities) will scale near linearly with computational capabilities and the supercomputer system must satisfy all three I/O requirements. The I/O requirements are driven both by the increasing power of the simulation algorithms and by rapid increases in data available for use by these algorithms.
In view of these challenges, NCM has installed a Lenovo Supercomputer consisting of 200 computational nodes interconnected with the Intel Omnipath Fabric. Each node contains two Intel XEON “skylake” 6148 CPUs, for a total of 40 compute cores per node. Each core operates at 2.4 GHz. The nodes were outfitted with 192 GB of memory operating at 2666 MHz. All the nodes run the CentOS 7 operating system and have access to specialized storage devices, serving Luster file systems to the nodes via the Intel Omnipath interconnect. The total available storage is 5 PB. There are login nodes and other service nodes available for users. Fig. A1 displays the NCM system.
Schematic representation of the NCM supercomputing system architecture and its components.
Citation: Weather and Forecasting 38, 11; 10.1175/WAF-D-23-0034.1
REFERENCES
Aboobacker, V. M., P. Vethamony, and R. Rashmi, 2011: “Shamal” swells in the Arabian Sea and their influence along the west coast of India. Geophys. Res. Lett., 38, L03608, https://doi.org/10.1029/2010GL045736.
Aboobacker, V. M., S. V. Samiksha, S. Veerasingam, E. M. A. S. Al-Aansari, and P. Vethamony, 2021: Role of Shamal and easterly winds on the wave characteristics off Qatar, central Arabian Gulf. Ocean Eng., 236, 109457, https://doi.org/10.1016/j.oceaneng.2021.109457.
AERONET, 2023: Aerosol Robotic Network Program. NASA Goddard Space Flight Center, accessed 10 July 2023, https://aeronet.gsfc.nasa.gov/.
Albers, S. C., 1995: The LAPS wind analysis. Wea. Forecasting, 10, 342–352, https://doi.org/10.1175/1520-0434(1995)010<0342:TLWA>2.0.CO;2.
Amante, C., and B. W. Eakins, 2009: ETOPO1 Global Relief Model converted to PanMap layer format. PANGAEA, NOAA/National Geophysical Data Center, accessed 10 July 2023, https://doi.org/10.1594/PANGAEA.769615.
Anoop, T. R., V. S. Kumar, P. R. Shanas, and G. Johnson, 2015: Surface wave climatology and its variability in the North Indian Ocean based on ERA-Interim reanalysis. J. Atmos. Oceanic Technol., 32, 1372–1385, https://doi.org/10.1175/JTECH-D-14-00212.1.
Astitha, M., G. Kallos, C. Spyrou, W. O’Hirok, J. Lelieveld, and H. A. C. Denier van der Gon, 2010: Modelling the chemically aged and mixed aerosols over the eastern central Atlantic Ocean—Potential impacts. Atmos. Chem. Phys., 10, 5797–5822, https://doi.org/10.5194/acp-10-5797-2010.
Bernard, B., and Coauthors, 2006: Impact of partial steps and momentum advection schemes in a global ocean circulation model at eddy-permitting resolution. Ocean Dyn., 56, 543–567, https://doi.org/10.1007/s10236-006-0082-1.
Baruque, B., E. Corchado, A. Mata, and J. M. Corchado, 2010: A forecasting solution to the oil spill problem based on a hybrid intelligent system. Info. Sci., 180, 2029–2043, https://doi.org/10.1016/j.ins.2009.12.032.
Beljaars, A. C. M., 1995: The parametrization of surface fluxes in large-scale models under free convection. Quart. J. Roy. Meteor. Soc., 121, 255–270, https://doi.org/10.1002/qj.49712152203.
Belly, P.-Y., 1964: Sand movement by wind. USACE Tech. Memo. 01, 84 pp., https://repository.tudelft.nl/islandora/object/uuid:f5051163-7a83-440d-9d02-92683a442549/datastream/OBJ/download.
Bower, A. S., D. M. Fratantoni, W. E. Johns, and H. Peters, 2002: Gulf of Aden eddies and their impact on Red Sea Water. Geophys. Res. Lett., 29, 2025, https://doi.org/10.1029/2002GL015342.
Chao, S.-Y., T. W. Kao, and K. R. Al-Hajri, 1992: A numerical investigation of circulation in the Arabian Gulf. J. Geophys. Res., 97, 11 219–11 236, https://doi.org/10.1029/92JC00841.
Chin, M., and Coauthors, 2000: Atmospheric sulfur cycle simulated in the global model GOCART: Comparison with field observations and regional budgets. J. Geophys. Res., 105, 24 689–24 712, https://doi.org/10.1029/2000JD900385.
Colosi, L. V., A. B. Villas Bôas, and S. T. Gille, 2021: The seasonal cycle of significant wave height in the ocean: Local versus remote forcing. J. Geophys. Res. Oceans, 126, e2021JC017198, https://doi.org/10.1029/2021JC017198.
Copernicus, 2023: Copernicus Marine Service. Accessed 1 January 2023, http://marine.copernicus.eu/.
Cotton, W. R., and Coauthors, 2003: RAMS 2001: Current status and future directions. Meteor. Atmos. Phys., 82, 5–29, https://doi.org/10.1007/s00703-001-0584-9.
Csiszar, I., and G. Gutman, 1999: Mapping global land surface albedo from NOAA AVHRR. J. Geophys. Res., 104, 6215–6228, https://doi.org/10.1029/1998JD200090.
Davis, S. R., L. J. Pratt, and H. Jiang, 2015: The Tokar gap jet: Regional circulation, diurnal variability, and moisture transport based on numerical simulations. J. Climate, 28, 5885–5907, https://doi.org/10.1175/JCLI-D-14-00635.1.
Donlon, C. J., M. Martin, J. Stark, J. Roberts-Jones, E. Fiedler, and W. Wimmer, 2012: The Operational Sea Surface Temperature and Sea Ice Analysis (OSTIA) system. Remote Sens. Environ., 116, 140–158, https://doi.org/10.1016/j.rse.2010.10.017.
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, https://doi.org/10.1175/1520-0469(1989)046<3077:NSOCOD>2.0.CO;2.
Egbert, G. D., and S. Y. Erofeeva, 2002: Efficient inverse modeling of barotropic ocean tides. J. Atmos. Oceanic Technol., 19, 183–204, https://doi.org/10.1175/1520-0426(2002)019<0183:EIMOBO>2.0.CO;2.
ETOOFS, 2022: Guide on implementing operational ocean monitoring and forecasting systems. IOC-UNESCO, GOOS-275, 392 pp., https://doi.org/10.48670/ETOOFS.
FAO, 1991: The State of Food and Agriculture 1991. United Nations, 244 pp., https://doi.org/10.18356/e3f71a7b-en.
Ferrarin, C., and Coauthors, 2023: Assessing the coastal hazard of Medicane Ianos through ensemble modelling. Nat. Hazards Earth Syst. Sci., 23, 2273–2287, https://doi.org/10.5194/nhess-23-2273-2023.
Fountoukis, C., and A. Nenes, 2005: Continued development of a cloud droplet formation parameterization for global climate models. J. Geophys. Res., 110, D11212, https://doi.org/10.1029/2004JD005591.
Francis, P. A., and Coauthors, 2020: High-resolution operational ocean forecast and reanalysis system for the Indian Ocean. Bull. Amer. Meteor. Soc., 101, E1340–E1356, https://doi.org/10.1175/BAMS-D-19-0083.1.
GEBCO, 2023: General bathymetric chart of the oceans. Accessed 1 January 2023, http://www.gebco.net.
Gillette, D. A., and R. Passi, 1988: Modeling dust emission caused by wind erosion. J. Geophys. Res., 93, 14 233–14 242, https://doi.org/10.1029/JD093iD11p14233.
Ginoux, P., M. Chin, I. Tegen, J. M. Prospero, B. Holben, O. Dubovik, and S.-J. Lin, 2001: Sources and distributions of dust aerosols simulated with the GOCART model. J. Geophys. Res., 106, 20 255–20 273, https://doi.org/10.1029/2000JD000053.
Gong, S. L., 2003: A parameterization of sea-salt aerosol source function for sub- and super-micron particles. Global Biogeochem. Cycles, 17, 1097,