Detecting and Characterizing Sea Breezes Over the U.S. Northeast Coast with Implication for Offshore Wind Energy

In this study, a new two-step identification method is proposed to detect and characterize three types of sea breezes (pure, corkscrew and backdoor) over the U.S. Northeast coast from a year-long WRF (Weather Research and Forecasting) simulation. The results suggest that the proposed detection method can identify the three different types of sea 15 breezes in the model simulation. Key sea breeze features, such as the calm zone associated with pure sea breezes and coastal jets associated with corkscrew sea breezes, are evident in the sea breeze composite imagery. In addition, the simulated sea breeze events indicate a seasonal transition from pure to corkscrew sea breeze between March and August as the land-sea thermal contrast increases. Furthermore, the location and extension of the sea breeze front are different for each type of sea breeze, suggesting that the coastal impact of sea breeze varies with sea breeze type. From the wind energy perspective, the 20 power production associated with a 10 megawatts offshore wind turbine would produce approximately 3 to 4 times more electrical power during a corkscrew sea breeze event than the other two types of sea breezes. This highlights the importance of identifying the correct type of sea breeze in numerical weather/wind energy forecasting. 25 https://doi.org/10.5194/wes-2021-109 Preprint. Discussion started: 22 October 2021 c © Author(s) 2021. CC BY 4.0 License.


Introduction
Offshore wind began to be harnessed as a renewable energy source in the early 1990's, and has been experiencing a rapid growth ever since due to its high wind resource potential and virtually unlimited installation areas over the ocean (Boyle. 2004;Esteban et al., 2011;Costoya et al. 2020). Europe has a long history of development and investment in offshore wind energy and is expected to have a capacity of 74 GW offshore wind power installed by 2030 (van Hoof et al. 2017). However, 30 despite having a vast offshore wind resource that can reach up to 2000 GW of potential production, the United States currently only has a single 30-MW commercial wind farm installed (Schwartz et al. 2010, Costoya et al. 2020Musial et al., 2016;Draxl et al. 2015;Deepwater Wind. 2016). According to the U.S. Department of Energy (U.S. DOE), more than 25 offshore wind projects with a total generating capacity of 25 GW are being planned, mostly concentrated along the U.S. East Coast. As many offshore wind farms will be constructed in the foreseeable future, it is essential to identify the 35 meteorological research needs that are critical for the sustainability and growth of U.S. offshore wind energy.
Both Archer et al. 2014 and the Offshore Wind Workshop from DOE (DOE. 2019) addressed this issue and pointed out an essential need to accurately capture the dynamic coastal processes from both observational and modeling perspectives as they represent a significant source of the uncertainty in the offshore wind resource assessment. One of such costal processes 40 is the sea breeze, which is defined as a local circulation induced by a thermal contrast between the land and sea (Simpson, 1994;Miller et al. 2003). The basic structural, dynamic and physical properties of the sea breeze are well documented (Abbs and Physick. 1992;Miller et al. 2003;Crosman and Horel. 2010). Previous studies have mostly investigated the impacts of sea breeze on air quality, heat waves and even flooding events (Yerramilli et al., 2009;Bianco et al., 2006;Golding et al., 2005, Lombardo et al., 2016. Recently, more attention has been given to classifying the different types of sea breezes 45 (Miller et al. 2003;Steele et al. 2013; based on the orientation of the prevailing wind (PW) with respect to the coastline.
During offshore PW conditions, sea breezes can be classified into three categories: pure, corkscrew and backdoor (Miller et al. 2003). A pure sea breeze occurs when there is no alongshore wind component, whereas a corkscrew (backdoor) sea 50 breeze occurs when the PW has an alongshore component with land to the left (right). There have only been a few studies focusing on onshore PW as atmospheric conditions normally suppress the development of sea-breeze circulations during such cases (Hu and Xue. 2016;Arritt 1993;Finkele et al. 1995). With a large human population living in major cities in proximity to a coastline, as well as the growing development of offshore wind energy, detecting and forecasting sea breeze events is of high importance. Past studies have mostly investigated sea breeze detection methods from observational 55 perspectives (Azorin-Molina et al., 2011;Prtenjak and Grisogono et al. 2007;Laird et al. 2001;Gustavsson et al. 1995;Ryznar et al. 1981). To our best knowledge, Steele et al. 2015 is the only study that proposed a sea breeze detection algorithm from a modeling perspective. Using 11-year model simulations, they presented an identification method that https://doi.org/10.5194/wes-2021-109 Preprint. Discussion started: 22 October 2021 c Author(s) 2021. CC BY 4.0 License. distinguished sea breeze types along five different coastlines around the southern North Sea of England. Given the upcoming investment for large offshore wind projects along the U.S. East Coast, it is important to develop a numerical detection 60 method for sea breezes over this region as sea breezes constitute an important coastal wind climate and will play an important role in resource assessment during the pre-construction phase.
The objective of this study is to present a new two-step identification method to detect sea breezes using numerical weather simulations. The method is tested along the U.S. Northeastern seaboard, centered on the New York metropolitan area, using 65 a year-long Weather Research and Forecasting (WRF) model simulation. Section 2 describes the simulation setup as well as the sea breeze detection method. The results from the detected sea breeze events using the proposed method are analyzed in detail in Section 3. Section 4 examines the uncertainty associated with the method and discusses the offshore wind potential with respect to each type of sea breezes. followed by the conclusion in Section 5.

Experiment design
The WRF model version 4.12 (Skamarock et al. 2008;Powers et al. 2017) is used to conduct the model simulation in this study. Table 1 summarizes the main physical options used. Both the boundary and initial conditions are initialized with the ERA5 dataset ( (Hersbach et al., 2020). The initial sea surface temperature field is updated by the Operational Sea Surface 75 Temperature and Sea Ice Analysis product (Stark et al., 2007). Static fields, such as topography height, land use category, soil and vegetation, use the default options from the WRF Prepossessing System.  (Iacono et al., 2008) Longwave Rapid Radiative Transfer Model for GCMs (Iacono et al., 2008) Microphysics Eta Grid-scale Cloud and Precipitation (Tao et al. 2003) Boundary Layer Mellor-Yamada-Nakanishi-Niino (MYNN; Niino, 2006, 2009) Land Surface Noah Land Surface Model (Chen and Duhia., 2001) Surface Layer MYNN Cumulus Kain-Fritsch (Kain and Fritsch., 1990;Kain., 2004) The model has a two-way interactively nested domains with horizontal grid spacings of 6 km and 2 km respectively (Fig.1a).

80
The domains cover the U.S. Northeastern seaboard and near-coastal Atlantic Ocean. There are 61 vertical levels and the lowest 9 levels are within 200 meters (m) above the ground. The 1-year simulations span from September 2019 to September 2020 and are conducted using a monthly reinitialization method with a 2-day spin-up and 1-day extension for each monthly simulation. These three days are discarded, and the rest are kept for the analysis.

Sea breeze identification method
There are various sea breeze identification methods based on different approaches, such as using satellite remote sensing 90 (Damato et al..2003), a sea-breeze index (Biggs and Graves 1962;Lyons 1972) and conditional filtering (Azorin-Molina et al., 2011;Steele et al. 2015). Even though each approach differs in terms of sophistications, all methods ultimately determine the likelihood of sea breezes based on key meteorological principles (e.g. land-sea temperature contrast) and preceding/coexisting physical conditions associated with sea breezes (e.g. large-scale wind field or changes in wind speed/wind directions). Since the accuracy of the detection algorithm strongly depends on the features of the study area (e.g., geometry 95 of the coastline), data and testing criteria, the filtering approach is selected as the preferred method for this study as it allows for the systematic determinations of days when conditions favor sea breeze formation over a wide range of scenarios.
To automatically select sea breezes event from the model simulation, key meteorological variables that are sensitive to sea breeze existence must be evaluated. Therefore, hourly model output of 10-m wind speed (WS10), wind direction (WD10), 2-100 m air temperature (T2) as well as sea level pressure (SLP) are analyzed. The search for sea breeze events is temporally constrained to the hours between 08 to 20 local time (LT), as land breezes are more likely to occur at night. Figure 2 shows the flow diagram of the two-step approach to identify sea breeze events from the model simulation. The first step is applied at a regional scale to identify days with flow regimes that have potential for pure, backdoor and corkscrew sea 105 breeze development. Figure 1b shows the topography map for domain 2 in which the New York metropolitan area is highlighted as our targeted region; that domain is further separated into four quadrants. Quadrant 1 (Q1) is mostly over land while Q4 is over the ocean. Both Q2 and Q3 are over the coastal region. The four quadrants are designed in such a way that For each day at 08 LT, the mean SLP, WS10, and WD10 are calculated in all four quadrants. Days with cyclonic conditions over the targeted region are rejected, as sea breeze identification would be difficult due to rapid changes in wind direction in 130 these cases. Then, the mean SLP and WD10 for each individual quadrants, as well as those for the entire targeted region, are used to determine which wind regime is dominant for that particular day. Table 2 shows the total number of days identified for each wind regime over the course of the simulation. The NW-WR occurs the most often and N-WR is the least common.
The other three wind regimes have almost an equal amount of identified days. Figure 4 shows the spatial pattern of averaged SLP, WS10 and WD10 for all five different wind regimes at 08 LT. Evidently, each wind regime is characterized with a 135 unique SLP condition in term of spatial pattern and magnitude. All the identified days have the potential for sea-breeze development and are further examined in the next step. The unidentified days are rejected as they tend to have a strong onshore wind component and synoptic setups that in general suppress sea-breeze development (Hu and Xue. 2016). Westerly 51 Northerly 30 Backdoor 47 Corkscrew 51   145 The second step applies the conditional filters to detect sea breeze existence for the days identified from the first step. The following three empirical criteria, which are based on the most important meteorological characteristics associated with sea breeze development, are used. If all the criteria are satisfied, then it is assumed that a sea breeze forms on that particular day.
1) The daytime maximum land-sea temperature difference is greater than 1.5 K 2) The mean mid-day (8am to 3pm) offshore WS10 is smaller than 6.0 m/s.  The first filter examines the thermal contrast between land and sea, as this is the most important physical condition needed 160 for sea breeze formation. The land-sea temperature difference is determined by the mean T2 different between Q1 and Q4. In principle, a positive land-sea temperature difference ( ΔT > 0℃) is required for the sea breeze formation. Previous studies have selected different thresholds for ΔT (Azorin- Molina et al., 2011;Steele et al. 2013Steele et al. , 2015. In this study, the threshold for ΔT is calculated from March 2020, which generally marks the beginning of the warm season. The second filter aims to eliminate days when the magnitude of offshore PW during the daytime is too strong, as studies (Miller et al. 2003, Steele et 165 al. 2013) have shown that this condition would prevent a sea breeze from reaching inland. Therefore, the mean WS10 from Q1 is calculated and applied in this filter. The critical value of 6 m/s is selected following the work from Steele et al. (2013).
The third filter examines the changes in WD10 over the entire region as a shift in wind direction from offshore to onshore is a key characteristic accompanying the sea breeze formation. Due to the difference in wind regimes, the threshold for the change in WD10 varies by sea breeze type to ensure onshore wind flow.

170
If all the three criteria are fulfilled for the BD-WR and SC-WR cases, these are then automatically identified as backdoor sea breezes and corkscrew sea breezes with no further steps needed. Those falling within the other three wind regimes, however, still need one more filter to distinguish between the corkscrew sea breeze and the pure sea breeze. This filter examines whether WS10 near the coast has increased significantly (by 4.0 m.s -1 ) in the late afternoon (18 to 20 LT) compared with 175 that in the early morning (08 to 10 LT). The point of reference for this filter is at the center of the targeted region (Figure 1b), which is about 20 km away from the coastline. If this condition is satisfied, the identified event will be considered a corkscrew sea breeze. Otherwise, the event is classified as a pure sea breeze. The uncertainties associated with the selected thresholds are discussed in detail in Section 4. 180 3. Results

1 Temporal variability of the sea breeze events from the model simulation
Using the proposed identification method, a total number of 61 sea breeze events have been identified from the year-long WRF simulation. This includes 28 pure sea breeze events, 24 corkscrew sea breeze events and 9 backdoor sea breeze events.  Out of the three sea breeze types, the backdoor sea breeze is the least common of the three types, which aligns with the findings of Steele et al. (2015). In addition, they are quite evenly distributed from April to August. The most frequent sea 195 breeze types are the pure sea breeze and the corkscrew sea breeze. Note that the total numbers of these two types of sea breeze are relatively even. However, there is a significant difference between the two in terms of seasonal distribution. In general, the pure sea breezes mostly occur during the spring and fall season whereas the corkscrew sea breezes appear more frequently during the summer months. This suggests that the modeled sea breezes show a gradual change from pure sea breeze to corkscrew sea breeze during the warm season. This could be partially associated with the increase in the land-sea 200 thermal contrast. As the land-sea temperature difference becomes more strongly positive, there is greater potential for corkscrew sea breeze development over pure sea breeze development along the U.S. Northeast coast. Figure 6 shows the spatial pattern of the development of the averaged pure sea breeze events from 08 LT to 20 LT. At 08 LT, 205 the simulated WD10 is mostly northwesterly with weaker wind speed over land than that over the ocean. The calm zone, which is the key precursor for pure sea breeze, forms at around 10 LT between 0 to 20 km offshore, reaches its mature stage at around noon and then moves away from the coastline. In conjunction with the development of the calm zone, the near coastal wind direction shrifts from offshore (northwesterly) to onshore (southerly) and the associated WS10 firstly decreases and then increases, reaching a maximum wind speed about 7 to 8 m/s during the late afternoon. Even though the composite wind speed over the calm zone is between 2 and 4 m/s, it falls primarily between 0 and 1 m/s for each individual case. In addition, the location of the calm zone varies by cases, although most calm zones develop relatively close to the coastline.  Previous studies have indicated that the potential causes for a coastal jet are Coriolis acceleration acting on a shore-parallel flow (Hunt et al. 2004) and the presence of a significant topographic barrier (Moore and Renfrew. 2005). The magnitude of WS10 at the jet core can reach up to 11 m/s, whereas the onshore wind speed falls between 5 and 6 m/s. With regards to offshore wind energy, accurately forecasting the occurrence and timing of the corkscrew sea breeze would be of high value for the wind energy industry. This will be further discussed in the Section 4.  Figure 8 shows the development of averaged backdoor sea breeze events. During the morning hours, the WD10 is mostly northeasterly. In addition, the WS10 over the ocean is significantly faster than those during the pure and corkscrew sea 230 breezes due to less inland deacceleration. In contrast to the corkscrew sea breeze, the arrival of the backdoor sea breeze is marked by a gradual veering (rotation in the clockwise direction) of the wind direction. Additionally, a relatively weak calm zone begins to appear near the coast around noon time as the wind shifts onshore. These features continue to persist throughout the rest of the analysis period. Compared with the other two sea breeze types, the backdoor sea breeze is the least common sea breeze type over the study region and has the weakest coastal wind speed later in the day.

Coastal impact from the modeled sea breeze events
To estimate the impact of the three types of sea breezes on the coastal region, the 10-m surface divergence is calculated for 240 all the identified sea breeze events. In general, an inland convergence line would form as the sea breeze circulation reaches onshore. Such a convergence line is indicative of a sea breeze front which is the landward edge of the sea breeze system and is associated with sharp changes in temperature, moisture and wind. Figure 9 shows the spatial map of 10-m divergence from the composite analysis for the three types of sea breezes at 12, 13 and 14 LT. The selected three hours represent the development stage for each type of sea breeze. From the pure sea breeze composite, a significant sea breeze front appears 245 along the coast of New Jersey, Long Island and Connecticut. As for the corkscrew sea breeze, the sea breeze front is strongest along the coast of New Jersey but is much weaker elsewhere. Regarding the backdoor sea breeze, there is a strong sea breeze front at the center of Long Island and a weaker front along the coast of Connecticut. However, no sea breeze front forms at New Jersey. Furthermore, the inland propagation of the sea breeze front is the greatest for the backdoor sea breeze compared to the other two types. These results suggest that there is a significant difference in terms of location of sea breeze 250 front with respect to sea breeze type over this region, which further emphasizes the importance of identifying the correct sea breeze type in the numerical weather forecasting.

Discussion
The preceding evaluation examines the temporal and spatial variability of the three types of sea breezes. In addition, the possible impact for each type of sea breeze has been discussed. Overall, the results indicate that the proposed sea breeze 260 identification method can detect and characterize sea breeze events from the WRF simulation.
To evaluate the uncertainty of the identified sea breeze events to the chosen criteria, a similar analysis is conducted using first a conservative and then a non-conservative approach. In the conservative (non-conservative) assessment, the threshold for land-sea temperature contrast, midday offshore wind speed, the change in WD10 for each wind regime and the change in 265 WS10 near the coastline is increased (reduced) by 10% of the original value. Table 3 shows the number of detected sea breeze events using the original, conservative and non-conservative approach. Overall, there is a 10 % change in the number of identified backdoor sea breeze and pure sea breeze events, whereas a greater degree of sensitivity (~25 %) is found for the corkscrew sea breezes. Further analysis indicates that the increase in sensitivity in the case of the corkscrew sea breezes is more associated with the threshold selection for the coastal jet filter. This indicates that more offshore wind observations are 270 needed to determine a more robust threshold value. Figure 10 shows the monthly timeseries of sea breeze events using the conservative and non-conservative approach. In general, the distribution pattern of sea breezes using both criteria look very similar to that from Figure 5, suggesting consistency in the transition from pure sea breezes to corkscrew sea breezes during the warm season. The spatial composites (figures not shown) show that the key sea breeze features, such as the calm zone for pure sea breezes and coastal jet for corkscrew sea breezes, remain evident. Backdoor sea breeze 8 9 10

285
The offshore wind energy potential associated with each type of sea breeze is first assessed by examining the spatial pattern of hub-height (100 meters above ground level) level wind in the sea breeze composites (Figure 11). At the hub-height level, the general features of the different types of sea breezes, such as the calm zone, coastal jet, veering (backing) of the wind, are still present. Furthermore, similar wind pattens are also apparent at the 200-m level (figure not shown). This suggests that the rotor-layer area of the offshore wind turbine are subject to potential impact from the sea breeze events. To estimate the 290 potential impact of sea breezes on offshore wind power, power output is calculated using the power coefficient associated with a 10 Megawatt (MW) offshore wind turbine is used. Figure 12a) shows the timeseries of hub-height wind speed associated with each type of sea breeze near the coast (at center point of the targeted region in Figure 1b). During a pure sea breeze event, the hub-height wind speed decreases during the morning hours due to the development of the calm zone and recovers back to approximately 7-8 m/s in the afternoon. In the case of the backdoor sea breeze, the strongest hub-height 295 wind speed (8-9 m/s) occurs in the morning and then starts to decrease through the late afternoon. In contrast, the hub-height wind speed associated with the corkscrew sea breeze is the weakest in the morning, after which it increases as the sea breeze develops, and finally reaches its maximum value (10-11 m/s) around the late afternoon. Figure 12b) shows the associated power output for each type of sea breeze. The most significant difference among the cases occurs in the afternoon, when the coastal jet associated with the corkscrew sea breezes forms, and the associated power output is almost 3 to 4 times as large as 300 that of the pure and backdoor sea breeze cases. Note that the backdoor sea breeze is associated with the highest power output during the morning hours. Selecting a different reference point, either closer or further away from the coast, does not change the overall pattern of the results. This highlights the importance of the predicting the correct type of sea breeze, especially the corkscrew sea breeze, for the wind energy application. 305 Figure 11: Spatial pattern of hub-height level wind speed and wind direction from the composite average for the pure sea breezes (a-c), corkscrew sea breezes (d-f) and backdoor sea breezes (g-i) at 12, 13 and 14 local time (LT). The colored contours indicate the magnitude of wind speed (m/s) and the vectors show the wind direction. Figure 12: Timeseries of (a) offshore hub-height wind speed (m/s) and (b) the associated power output from a 10 MW wind turbine with respect to sea breeze type between 08 LT and 20 LT.
Even though the focus of this paper is not to validate the simulated sea breezes with observations, some characteristics 315 associated with the modeled sea breezes do differ from observations documented in the past literatures. For instance, due to the nature of forced balance between local sea breeze circulations and synoptic scale forcing, the timing of landfall arrival for each type of sea breeze should vary among cases, with the corkscrew sea breeze being the earliest and the backdoor sea breeze being the latest (Adam, 1997, Miller et al. 2003. However, such feature is not evident in this study. The inland convergence line forms around noon time for all three cases. In addition, the sea breeze cycle starts, develops and manifests 320 around the same time regardless of the type (Figures 6-8

Conclusion
With the vastly planned offshore wind farm construction along the U.S. East coast, identifying and understanding key coastal process, such as sea breezes, has become a critical need for the sustainability and development of the U.S. offshore wind 330 energy. In this study, a new two-step identification method has been proposed to detect and characterize three types of sea breezes (pure, corkscrew and backdoor) from a year-long WRF simulation.
The results suggest that the proposed detection method is able to identify different types of sea breeze events in the model simulations. Key sea breeze features, such as the calm zone associated with pure sea breezes and the coastal jet associated 335 with corkscrew sea breezes, are evident in the sea breeze composite output. In addition, the simulated sea breeze events indicate a transition from pure sea breezes to corkscrew sea breezes during the warm season as the land-sea thermal contrast increases. Furthermore, the location and extension of sea breeze fronts is different for each type of sea breeze, suggesting that the coastal impact of sea breeze varies with sea breeze type. Using the power coefficient from a 10 MW offshore wind turbine, the power production associated with the corkscrew sea breeze is almost 3 to 4 times higher than the power 340 generated by other two types of sea breezes. This points out the importance of forecasting the correct type of sea breeze, especially the corkscrew sea breeze, in numerical weather/wind energy forecasting.
Even though this is not a validation study, the modeled sea breeze events do show discrepancy with the observations from the past literatures in certain aspects, such as the landfall timing. There are many possible reasons for this; one potential 345 source is the lack of ocean-atmosphere dynamic coupling in the current model setup. This motivates the development and usage of a fully coupled ocean-atmosphere model in sea breeze research, which will be the focus of a subsequent study.

Code/Data availability
OSTIA data are publicly available from the Physical Oceanography Distributed Active Archive Center. ERA-5 data are publicly available 350 from the ECMWF's MARS archive. The WRF simulations and code are available from the lead author upon request.

Author contributions
Geng Xiaprepared the manuscript with contributions from all co-authors; conceptualization, formal analysis, investigation,