Wind velocity is typically reported at 10 meters above the ocean surface (referred to as U10) for marine applications. Numerous researches studied the correlation between the horizontal wind velocity over aquatic environment and adjacent terrestrial land in proximity to estuaries and bays. Assessment of time series of U10 across 15 monitoring stations along the Mississippi Delta and Louisiana coastlines revealed a distortion in wind measurements taken at boundaries of land and sea. Close to such features, the inland measurements cannot be used as a representative of the prevailing wind conditions across estuarine waters (Mariotti et al. 2018). Measurements taken close to the coastline underestimate the wind speed for seaward winds because even a small stretch of land (< 1 km) with rough surfaces can considerably decrease U10. The wind direction in such condition tends to align with the lengthwise direction of the water body.
As wind speed increases, the ocean surface becomes progressively more irregular. To effectively account for the interplay between wind and surface roughness, an adapted version of the Charnock relationship, established by Charnock in 1955, has been widely utilized in the literature (Edson et al. 2013); however, in shallow waters, the surface roughness surpasses the corresponding values recommended over the open ocean (Jiménez and Dudhia 2018). Hence, during the process of executing wind simulations in coastal areas, the bias error increases due to the sudden fluctuations in the surface texture and the intricate geographical arrangements that are encountered. The precise simulation of wind components along the coastline is important, particularly in the context of energy harvesting facilities; e.g. Santos-Alamillos et al. (2013) extensively examined the potential for local wind energy production along the southern shores of Spain by optimizing the most suitable physical models within the Weather Research & Forecasting Model (WRF hereafter).
In state-of-the-art meteorological models, like the WRF model, the optimum values for parameters in the model depend on the specific local circumstances. As a result, there exist no universally adaptable physical frameworks capable of precisely ascertaining meteorological variables for broad-range of applications. Fundamentally, the most appropriate combination of parameters for a given region might not present the optimum combination for another study area (Krieger et al. 2009); e.g. Borge et al. (2008) conducted a meticulous sensitivity analysis of the WRF model in the Iberian Peninsula, encompassing 23 distinct model configurations. These configurations involved various components such as Planetary Boundary Layer (PBL) schemes, microphysics, surface land models, radiation schemes, sea surface temperature (SST), and four-dimensional data assimilation. After performing the model assessment, the optimal configuration was determined as using the Yonsei University PBL, the Noah land-surface model, the single-moment 6-class microphysics scheme, the eta geophysical fluid dynamics laboratory longwave radiation scheme, and the MM5 shortwave radiation scheme. This configuration, in conjunction with other pertinent user settings such as time-varying SST and combined grid-observational nudging, exhibited the superior performance in estimating temperature, wind speed components, and humidity fields at the surface level.
The performance of WRF model in estimating wind vectors was evaluated under different numerical and physical options for an area of the Portugal, located in a complex terrain and characterized by its significant wind energy resource. The results showed the utilization of grid nudging with an integration time of less than 2 days was the best numerical configuration. Also, the physical schemes of MM5 and Yonsei University-Noah outperformed the other options for this case. The results were less accurate in sites with higher terrain complexity, mainly due to limitations of the model to include the small scale changes in the topography (Carvalho et al. 2012).
A series of thirty-six multi-physics ensembles were generated to model wind vectors in southeastern Australia, with each ensemble member featuring distinct physics parameterizations. However, no single ensemble member consistently outperformed others across all events, variables, and metrics, indicating sensitivity of model tuning parameters to varying environmental conditions. Nevertheless, the combination of the Mellor-Yamada-Janjic PBL and the Betts-Miller-Janjic cumulus scheme emerged as the most successful pairing. Furthermore, it is advised against using the Yonsei University PBL, Kain-Fritsch cumulus scheme, and RRTMG radiation scheme together in this region. Notably, the model's accuracy was compromised during periods of intense rainfall events. (Evans, Ekström, and Ji 2012).
In their study, Santos-Alamillos et al. (2013) assessed different parameterizations and terrain representations within a WRF mesoscale model, employing 32 configurations and two microphysics, cumulus, PBL, shortwave, and longwave radiation schemes. They then evaluated wind estimates from various experiments with spatial resolutions of 1, 3, and 9 km against data from four stations in southern Spain, providing hourly wind speed and direction at 40 meters above ground level. The results indicated that the choice of PBL parameterization predominantly affected the standard deviation (STD) of wind speed and bias values. Although spatial resolution had a minor impact on STD, it significantly influenced bias, with higher absolute values observed at coarser resolutions. However, wind direction sensitivity to the physical configuration was minimal.
Based on a high-resolution (3 km) WRF simulations, Gevorgyan (2018) reproduced Investigating the dynamics of topographically-driven Low-level jets (LLJs) and mountain-valley winds in the vicinity of Yerevan, utilizing a combination of detailed observational data and sophisticated modeling techniques. Both PBL configuration and the lateral boundary conditions highly affected Examining the attributes of surface-level winds and low-level jet (LLJ) features documented on July 4th, 2015. Among the nine tested PBL parameterizations, The PBL schemes MYJ, QNSE, and TEMF models demonstrated better proficiency in simulating valley winds near the surface compared to other schemes, which notably underestimated wind speeds. The majority of PBL schemes effectively simulated the distinct low-level jets (LLJs) observed during evening valley winds in Yerevan. These simulated jet cores typically occurred at altitudes ranging from 150 to 250 meters above ground level, with velocities varying from 12 to 21 meters per second. However, most configurations notably underestimated the observed intensity of nocturnal low-level jets (LLJs) in Yerevan, which typically occurs at 110 meters above the surface level. Conversely, the Shin and Hong schemes and the YSU PBL approach tended to overestimate the speed of the nocturnal LLJ. Notably, simulations conducted with the ECMWF ERA-5 dataset as the initial condition showed improvements in both near-surface winds and nighttime potential temperatures in Yerevan compared to those driven by the Global Forecast System fields. It is clear that the simulation of wind speed and direction in atmospheric models like WRF depends greatly on the parameterizations used for the boundary layer, including Land Surface Models (LSM), the Surface Layer (SL), and PBL. These parameters have a substantial impact on both global and regional scales (Stensrud 2009; Carvalho et al. 2012). As mentioned above, the WRF model offers various physical options for components such as cumulus convection, radiation, and microphysics. However, including all possible model configuration options in sensitivity analysis would be impractical and unnecessary (Nossent, Elsen, and Bauwens 2011).
In regional modeling, the parameterizations of SL, PBL, and LSM play a vital role in controlling The reciprocal transfer of heat, moisture, and momentum between the Earth's surface and the atmosphere. Such exchanges are crucial for the accuracy of weather forecasts (Gilliam and Pleim 2010).
The planetary boundary layer (PBL) approach used in a model is particularly important in momentum and thermodynamic equations. They determine how wind components evolve and influence the model's performance significantly; e.g. the MYJ scheme uses a unique approach to align the viscous sublayer and SL, while the MRF and YSU schemes rely on logarithmic similarity formulas for flux calculations. These differences can lead to underestimation or overestimation of heat fluxes, depending on factors like surface roughness (Pagowski 2004). Parameters like thermal stability, PBL height, and the entrainment of air from the free atmosphere into the PBL all influence the wind speed within the PBL (Arya 2001). Numerous PBL schemes have been developed over the years to address these complex interactions (Hong, Noh, and Dudhia 2006; Janjić 1994; Nakanishi and Niino 2006, 2009; Olson et al. 2019; Pleim 2007a, 2007b; Angevine, Jiang, and Mauritsen 2010; Hong and Pan 1996).
Efficient heat transfer between the Earth's surface and the atmosphere is challenging, primarily because the temperature contrast between these two components is much larger than what similarity theory predicts. While all PBL schemes maintain logarithmic wind profiles in neutral atmospheric conditions, they differ in modeling potential temperature profiles. The MYJ scheme tends to have a steeper potential temperature gradient than predicted by similarity theory, especially in convective conditions. The MRF scheme struggles to replicate the curvature of potential temperature profiles, resulting in log-linear profiles. Both the MYJ and YSU schemes perform well in modeling momentum transfer under neutral conditions (Pagowski 2004).
The MYJ scheme is distinctive for its pronounced decoupling from the surface regarding heat transfer, characterized by excessively steep temperature gradients. Bulk transfer coefficients for heat and momentum, approximated using similarity functions in the WRF model, might lead to inaccuracies. An iterative approach has been proposed to remedy this problem. All implemented schemes in WRF face challenges in efficiently transferring heat exchange between the surface and the atmosphere within the SL, potentially resulting in weaker or delayed diurnal variations; the MYJ scheme exhibiting the most pronounced discrepancies though (Pagowski 2004).
In convective conditions with weak winds, the MRF and YSU schemes produce boundary layers (BL) that are twice as deep as those generated by the MYJ scheme. Under stronger wind conditions, the YSU scheme tends to yield shallower BLs, possibly addressing known issues of MRF overestimation of BL height. The MYJ scheme experiences numerical instability under convective conditions and high diffusivities, which hinders its ability to accurately capture atmospheric dynamics. In neutral conditions, the YSU scheme exhibits the least mixing, while it exhibits more mixing in stable BL conditions.(Pagowski 2004).
Additionally, the ACM2 (Asymptotic Convective Mixing) boundary-layer scheme offers a practical solution to accurately model turbulent transport within convective boundary layers, ensuring realistic profiles for heat flux and potential temperature. The YSU scheme outlines turbulent momentum flux calculations using K-theory, with considerations for stable conditions and critical Richardson numbers (Hong 2010). The MYNN (Mellor-Yamada-Nakanishi and Niino) scheme, rooted in the Mellor-Yamada prognostic turbulent kinetic energy (TKE) framework, incorporates K-theory and related parameters (Nakanishi and Niino 2006, 2009). In essence, the choice of PBL schemes in atmospheric modeling significantly affects the simulation of wind behavior, heat transfer, and profile curvature under different atmospheric conditions. Careful consideration of these factors is essential for successful simulation using atmospheric models.
The surface layer (SL) refers to the atmospheric layer closest to the Earth's surface, reaching approximately one-tenth of the height of the planetary boundary layer (PBL) (Carvalho et al. 2012). In this layer, the reciprocal transfer of heat, moisture, and momentum with the Earth's surface remains relatively constant, and fluxes of these properties exhibit minimal variations. The SL thickness typically varies, being less than 50 m during the day and less than 20 m at night (Stensrud 2009). The LSM schemes integrate atmospheric data from SL schemes with land surface characteristics, which are dependent on the land use. This integration allows for the assessment of vertical transport within PBL schemes, directly affecting the estimation of the PBL height (Han, Ueda, and An 2008).
In this research, the several WRF model configurations are employed to find the best option for simulating the wind component over the southern Caspian Sea, the largest lake on the Earth. Different ensembles were evaluated for the east, west and middle parts of the study area. Beside ASCAT satellite data, several coastal synoptic observations provided data used for model assessment along with offshore data from a buoy.