NUMERICAL STUDY OF EFFECTS OF WARM OCEAN EDDIES AND OCEANIC BARRIER LAYERS ON TROPICAL CYCLONE INTENSITY IN NORTHWEST PACIFIC

It is well recognized that evaporation from the sea surface, primarily within a tropical cyclone (TC) core, provides heat energy required to maintain and intensify the storm. The sea surface temperature (SST) typically decreases within the storm core due to the mixing and upwelling processes in the upper ocean thereby limiting the storm intensity. This negative feedback to the TC intensity depends on the oceanic thermal conditions and salinity stratification ahead of the storm. Upper oceanic heat content (OHC) has become widely accepted as a measure of the ocean energy available to the TCs. Observational and modeling studies note that some TCs rapidly intensify while passing over warm core eddies (WCEs) because of their high OHC. TC intensification is also significantly affected by salinity-induced barrier layers (BLs) formed when a low-salinity is situated near the surface in the upper tropical oceans. When storms pass over the regions with BL, the increased stratification and stability within the layer reduce storm-induced vertical mixing and SST cooling. This causes an increase in enthalpy flux from the ocean to the atmosphere and, consequently, leads to TC intensification. In this study, we applied the Hurricane Weather Research and Forecast (HWRF) v.4.0 system coupled to the Message Passing Interface Princeton Ocean Model (MPIPOM). We conducted the idealized experiments in which the WCE is embedded into the U.S. Navy's Generalized Digital Environmental Model (GDEM) climatology with a specified size using a feature-based initialization procedure. Idealized vertical ocean profiles from Hlywiak and Nolan (2019) are selected to conduct the sensitivity of TC intensity to BL thickness. The goal of this study is to quantify the impact of WCEs and BLs in the upper ocean on TC’s self-induced cooling and subsequent feedback on TC intensity in three TCs in 2018, Jebi, Trami, and Kong-Rey


Introduction
Tropical cyclones (TCs), known as hurricanes or typhoons, form over warm water in tropical oceans, especially in late summer, and cause devastating disasters such as heavy rains, floods, and storm surges resulting in immense damages in social and economic aspects. It has been well known that heat energy from the upper ocean is a source to generate and intensify the TCs (Emanuel et al., 1986;Cione and Uhlhorn, 2003). In the development of TC, the surface wind stress leads to an increase of evaporation from the sea surface, and it provides the latent heat energy to develop the TC. The surface water is, however, cooled by the imposed surface wind stress as the storm continues to intensify, which results in a decrease in the heat energy from the ocean surface and weakens the storm's intensity Bender and Ginis, 2000). Sea surface cooling is generated during the evaporation, but the main cooling processes are caused by the wind stress via shear-induced turbulent mixing of the upper ocean, upwelling and entrainment from the deep ocean into the oceanic mixed layer (Price 1981;Shen and Ginis, 2003;D'Asaro et al., 2007). It is evident that the magnitude of the sea surface cooling near the storm core contributes to the TC intensity. However, negative feedback from sea surface cooling to TC intensity is not merely exerted by storm forcing, but the combination with ocean thermal conditions (Price 1981;Schade and Emanuel, 1999;Bender and Ginis, 2000;Chan et al., 2001;Lin et al., 2005;Wang et al., 2018). Generally, the sea surface cooling is inhibited by a thicker subsurface warm water layer during the passage of the storm as the cold water from deep cannot be entrained into the surface layer. This results in a less sea surface temperature (SST) cooling that can potentially strengthen the storm intensity if all else being equal. Thus, understanding the upper ocean thermal structure plays a crucial role in examining storm intensification because TCs interact not only with the surface waters but with the entire upper ocean water column.
Warm core eddies (WCEs), which have a characteristic of higher temperatures than the surrounding waters with downwelling motion in the eddy regime, are the common mesoscale features in the ocean. WCE has been regarded as an insulator between TCs and the deep ocean water in that the thicker mixed layer in the WCE can limit the wind-induced mixing of the upper ocean from below (Lin et al., 2005). WCE can reduce the negative feedback and help to transport sufficient heat energy into the storm to intensify. Thus, understanding the interaction of TCs with WCE is critical for improving the understanding and prediction of the TC intensity change. Recent studies have identified that TC intensifications occur when passing over the WCEs (Hong et al., 2000;Shay et al., 2000;Emanuel et al., 2004;Lin et al., 2005;Wu et al., 2007;McTaggart-Cowan et al., 2007;Vianna et al., 2010;Yablonsky and Ginis, 2013;Jaimes et al., 2016). Lin et al. (2005) discussed the importance of WCE in Typhoon Maemi, which rapidly intensified from category 3 to category 5. Maemi intensified while passing over two WCEs in the northwest Pacific and became one of the most powerful Typhoons to strike South Korea since record keeping began in the country in 1904.
Northwest Pacific is well known as one of the most active TC basins in the world (Emanuel., 2005;Peduzzi et al., 2012;Lin et al., 2013) and the region where eminent WCEs exist, particularly in the eddy-rich zones (Qiu 1999;Roemmich and Gilson, 2001;Hwang et al., 2004;Lin et al., 2005). In this study, we focus on the interaction between WCEs and TCs in the southern eddy zone (18°-25°N, 122°-160°E) where Maemi intensified over WCEs ( Fig. 1.1). Previous studies conducted analyses of the WCEs in this region (Liu et al., 2012;Yang et al., 2013;Ma et al., 2017). According to Liu et al. (2012), the average size (radius) of anticyclonic eddies is 120-140 km at the latitude of 20°N in the southern eddy zone ( Fig. 1.2).
There have been many studies investigating the impact of WCE on the TC, however, most of the previous modeling studies are based on a single storm case (Hong et al., 2000;Shay et al., 2000;Lin et al., 2005;Wu et al., 2007;Wang et al., 2018) or using simple one-dimensional ocean models (Chan et al., 2001;Lin et al., 2005;Wu et al., 2007). For the TC prediction model to capture the effect of wind-induced sea surface cooling, it must be fully coupled to a three-dimensional ocean model to create an accurate SST field (Yablonsky and Ginis, 2009). In recent years, numerical atmosphere-ocean coupled models have been used to explore the impact of WCE on TC intensity. Yablonsky and Ginis (2013) suggested that the circulation of WCE can affect the TC intensity, and the WCE located to the right of the storm could even cause a less favorable condition for TC intensification due to the advection of cold water into the TC inner core. Ma et al. (2017) suggested that the effect of ocean eddies is related to the strength of eddy and TC intensity, and the effect is less pronounced when the eddy is located at one side of storm tracks than at the TC center. Anandh et al.
(2020) showed that eddies play an important role in the intensification and dissipation of TCs in the Bay of Bengal using an atmosphere-ocean coupled numerical model, consisting of the Weather Research and Forecast (WRF) and Regional Ocean Modeling Systems (ROMS). Sun et al. (2020) investigated the response of TC intensity change to the spatial distribution of WCE using WRF and the three-dimensional Price-Weller-Pinkel (3DPWP) ocean circulation model. The results revealed that TC is strongly intensified (weakened) with a WCE located in the inner (outer) TC eyewall area. All of these modeling studies have been conducted for idealized TCs and simplified ocean conditions. The impact of WCEs on storm-induced ocean response and TC intensity in a realistic environmental setting remains unexplored by numerical models.
Negative feedback from the ocean, which reduces TC intensification depends on the ocean thermal parameters as well as the upper ocean stratification. There are regions where a low-salinity water is located near the surface, and these low-salinity layers can induce barrier layers (BLs) when the isothermal layer is deeper than the mixed layer (Lukas and Lindstrom, 1991;Sprintall and Tomczak, 1992). BL can increase stratification in the upper ocean, and thereby reduce the surface cooling negative feedback on TC intensity. When a TC passes over a region with the BL, the increased stability within the upper ocean layer can reduce TC-induced vertical mixing. This causes a decrease in SST cooling and an increase in enthalpy flux from the ocean to the atmosphere and consequently leads to TC intensification. Previous studies have examined how the SST response to the TC-induced wind is affected by the BL (Wang et al., 2011;Balaguru et al., 2012;Vissa et al., 2013;Reul et al., 2014;Hernandez et al., 2016;Yan et al., 2017;Rudzin et al., 2018;Hlywiak and Nolan, 2019). Balaguru et al. (2012) discussed the impact of BL on the reduction of SST cooling and TC intensification on a global scale. One of the well-known regions where the BLs form is the Amazon-Orinoco river plume region. Previous studies found that the TC-induced vertical mixing and SST cooling significantly are inhibited over the plume area due to the presence of strong vertical stratification (Grodsky et al., 2012;Reul et al., 2014).
However, the recent studies of Newinger and Toumi (2015) and Hernandez et al. (2016) found that there was little difference in TC-induced cooling between the plume and open ocean experiments using a regional ocean model. Yan et al. (2017) showed that the BL can weaken the storm intensity when the surface wind stress is too weak to break through the mixed layer. Hlywiak and Nolan (2019)

Ocean Model Initialization
Prior to the coupled model integration of the HWRF system, MPIPOM-TC is initialized with a realistic, three-dimensional temperature and salinity field from the Generalized Digital Environmental Model (GDEM) monthly climatology (GDEMv3; Carnes 2009), which has a 0.5° horizontal grid spacing and 78 vertical z levels and subsequently integrated to generate realistic ocean currents (Teague et al., 1990). The GDEM climatology is modified by interpolating it in time to the MPIPOM-TC initialization date (using 2 months of GDEM) and onto the MPIPOM-TC grid, assimilating a land/sea mask and bathymetry data (Falkovich et al., 2005;Yablonsky and Ginis, 2008). The ocean temperature field is generated after assimilating with the real-time daily SST data (with 1° grid spacing) that is used in the operational NCEP Global Forecast System (GFS) global analysis (Reynolds and Smith, 1994;Yablonsky and Ginis, (2008, section 2)). Three-dimensional ocean initial temperature and salinity fields are then interpolated from GDEMv3 levels onto MPIPOM-TC vertical sigma levels. During the ocean spin up of 48 h in phase 1, the SST is held constant, and adjusted currents are generated. During phase 2, a cold wake at the sea surface is produced prior to the start of the coupled model forecast. Phase 2 is skipped in this study and the output after spin-up is used for the initial ocean component of the HWRF model.

Atmospheric Model Initialization
The location of the HWRF atmospheric component parent and inner domains is based on the observed TC's current and center position based on the NHC storm message. Once the environment fields in the parent domain are derived from interpolating the GFS analysis fields, the vortex replacement cycle and HWRF Data Assimilation System (HDAS) are used to create the initial nest fields (Domain size for parent nest is 77°x77°, inner nest 1 is 20°x20°, and 11°x11° for inner nest 2). The vortex-scale fields are generated by inserting a vortex corrected using TC vitals data (Trahan and Sparling, 2012). The analyses are interpolated onto the HWRF outer domain and two inner domains to initialize the forecast. (Tallapragada et al., 2014).

Coupled Model Run
After the ocean and atmosphere initializations are completed, the coupled HWRF is launched. During the atmosphere-ocean coupling, the momentum fluxes and total heat at the air-sea interface are passed from the atmosphere to the ocean, and the SST is passed from the ocean to the atmosphere as an independent interface between the HWRF ocean and atmospheric component (Fig 2.3). It highlights that the primary purpose of coupling a three-dimensional ocean model to HWRF is to create an accurate SST field for input into the atmospheric model. The total simulation time is 126 h and the output is provided every 6 hours.      SSHA data on 21 September 2017 is used to determine the spatial structure and magnitude of WCE (Fig. 3.1a).

Simulated Tropical Cyclones in the Northwest Pacific
The upper ocean temperature profiles at the center of WCE and the background are shown in Fig. 3.1b. The temperature profile of the WCE center is from the most prominent WCE that super typhoon Maemi passed over and intensified (Lin et al., 2005). The background profile is from GDEM climatology at the location where the WCE is implemented. The WCE center profile in Lin et al. (2005) was provided by the U.S. Naval Research Laboratory's NPACNFS nowcast model, which is the ocean model with near real-time operational assimilation of satellite observations. The WCE is initialized by assigning a series of specified temperatures from the background temperature profile to the WCE center (Yablonsky and Ginis, 2008;Yablonsky and Ginis, 2013). The horizontal temperature field in WCE at 75 m depth and the zonal vertical cross section after the initialization are shown in Fig. 3.2. The radius of the idealized WCE is set to be about 200 km based on the observed SSHA shown in Fig.   3.1a. In this study, the size of eddy is defined to be the radius of the circle that has the same area as the region within the eddy edge, and the outmost closed SSHA contours are used to define the eddy edge. After the WCE is created, a temperature anomaly field is obtained by subtracting the background temperature. The anomaly field is horizontally interpolated onto the POM grid and added to the GDEM climatology in such a way that the simulated storm track crosses the center of WCE.

Control Experiments
Before comparing the results between CTRL and WCE experiments, it is important to first investigate the results of each CTRL experiment to examine the different characteristics of each storm, interacting with the different ocean conditions.
Here we compare the control experiments initialized at 0600 UTC 30 August for Jebi, 1200 UTC 23 September for Trami, and 1800 UTC 30 September for Kong-rey. In Kong-rey CTRL experiment, the storm rapidly intensified during the first 12 hours as seen in the increased maximum wind speed and decreased central pressure (Fig. 3.5). (1) where 26 is the depth of the 26°C isotherm, is the seawater density, * is a specific heat at constant pressure, is the ocean temperature, and is the change in depth (Leipper and Volgenau, 1972). We assume a constant density of 1025 kg/m 3 for our calculations. Considerably high OHC of around 100 kJ/cm 2 at the beginning of Kongrey track indicates more heat energy available for the storm and explains the initial rapid intensification (Fig. 3.6b). This is consistent with previous studies proposed that SST in advance of the TC does not account for the storm-induced SST cooling, and OHC ahead of the storm is a better measure of the available ocean energy for TC intensification Mainelli et al., 2008).
In the Jebi CTRL experiment, the pre-storm SST field is around 30°C along the track which is higher than that in Kong-rey (Fig. 3.8a). As for Kong-rey, the evolution of Jebi intensity (Fig. 3.7) can be interpreted by the distribution of the OHC field rather than SST. The gradual intensity increase at the early stage occurs as the storm passes over the high OHC region during the first two days (Fig. 3.8b). The maximum wind speed decreases from 1800 UTC 2 September until landfall at 1200 UTC 4 September due to the lower OHC along the track. In the Trami CTRL experiment, there is a significant intensity decrease in the middle of the simulation (Fig. 3.9). This is because the Trami propagated with a very slow translation speed during that time period. Figure   3.10 compares the distribution of SST and a zonal vertical cross section along 20.4°N at 1800 UTC 23 September and 1200 UTC 26 September. The slow TC propagation speed generates strong vertical mixing as well as upwelling. Vertical mixing occurs due to wind stress driven ocean currents and the resulting vertical current shear leading to entrainment of the colder water from thermocline into the ocean surface layer (Price 1981;Ginis 2002). Ocean surface currents are also diverged by the TC cyclonic wind stress above causing the upwelling of the colder water toward the surface and making thermocline to lift (Fig 3.10d). The TC induced upwelling increases the efficiency of the vertical mixing and cooling of the SST. This explains why TC Trami rapidly weakened.

Uncoupled Experiments
In  in the uncoupled and CTRL experiments are summarized in Table 1. As expected, the largest differences are found in TC Trami, which is the slowest-moving storm among the three simulated TCs (Fig. 2.5). The slow translation speed in Trami allows sufficient time to mix and cool the upper ocean beneath the storm, which is the largest within the storm core. This leads to the largest differences in maximum SST, heat flux, and intensity compared to the uncoupled, fixed SST, experiments.

Impact of Warm Core Eddy on TC intensity
Here we discuss the TC simulations in which the ocean model is initialized with  Table 4. MWPI as a function of WCE size is shown in Fig 3.24. Overall, when the storm is interacting with a larger WCE a larger MWPI is found in SST and HF in all three storms. Consequently, the largest MWPI is found in the WCE experiments with 300 km, especially 77% of maximum potential wind speed for Jebi, and 64% and 63% for Trami and Kong-rey, respectively (Fig. 3.24d).

Conclusion
The impact of WCE on TC intensity is investigated in three tropical cyclones in

Initialization and Experiment Design
There are regions where low-salinity water is located near the surface, and these low-salinity layers can induce barrier layers (BLs) when the isothermal layer depth (ILD) is deeper than the mixed layer depth (MLD).
The contribution of low-salinity water in the upper ocean to storm intensification was observed during the KIOST field experiment in 2019. TC Lingling rapidly intensified from a weak category 1 at UTC 00:00 4 September to category 4 over 24 hours while passing along the western side of WCE ( Fig. 4.1) (Kang et al., 2020).
Vertical temperature and salinity cross section at each CTD casting along the section E of WCE before and after Lingling passage are shown in Fig 4.1. The high temperature of 28-29.5˚C and relative lower salinity of 34.4-34.5 psu in the western side of WCE before Lingling passage suggest that the low salinity water is wrapping around the west of WCE (Figs 4.1b and 4.1c). It is suggested that the weak cooling of less than 0.5˚C occurred due to the low salinity water, which served as the barrier layer, led to the rapid intensification of Lingling (Figs 4.1d and 4.1e).
We conducted simulations to explore the effect of BL on three TCs in 2018, Jebi, Trami, and Kong-rey in the northwest Pacific. Each simulation is initialized using one temperature and one of four different salinity profiles indicating different BLT.  Table 5. In every simulation, the initial temperature is constant down to the ILD at 50 m depth.

Impact of Barrier Layer on TC intensity
Previous studies found that the SST response to a passing TC depends on the ocean thermal structure, and TC conditions such as its size, intensity, and translation speed (Price 1983;Shay et al., 1989;Yablonsky and Ginis, 2009). To investigate the impact of the BL on TC intensity, which is expected to be a favorable ocean condition for storm intensification, it is important to examine different TC conditions because the upper ocean response is strongly influenced by the TC intensity and translation speed. The total simulation time is 96 h for each experiment and the result is provided every 6 hours. Before discussing the impact of BLT, it is important to examine first the sea surface cooling trend due to the different characteristics of each storm. Overall, the magnitude of SST cooling is larger for the slower moving storm (Figs 4.4a,4.4c,and 4.4e). Unlike the gradual increase in SST cooling trends in TC Kong-rey and Jebi, rapid cooling occurs in TC Trami from the time at 0600 UTC 25 September as the translation speed slows down (Fig. 2.5). The peak of the cooling is reached at 1200 UTC 26 September after which the cooling is reduced due to do acceleration of the translation speed and storm weakening (Fig 4.4c).
In all experiments, TC-induced cooling has been suppressed by the presence of BL, especially in OBL20 and OBL25 cases.  (Fig 4.5). In Kong-rey, the maximum averaged wind speed is found further from the storm center compared to Jebi and Trami.
This implies that a larger area in the ocean is affected by the strong wind under Kong-rey, which can explain why the impact of BL in reducing SST cooling due to the BL occurs earlier in the Kong-rey case.
Spatial distribution of SST anomaly between OBL25 and OBL00 cases for three TCs is shown in Fig. 4 where Z is the storm translation speed and is the inflow angle (the angle between the wind vector and the azimuthal direction) and is Coriolis parameter (Ginis 2002). In the case of TC Jebi, the maximum upwelling occurs further from the center than in Kong-rey because Jebi's translation speed is higher. As a result, the BL induced SST cooling is found at a larger distance in Jebi compared to Kong-rey (Fig. 4.6b and Fig.   4.6f). After the 24 hours of the Trami simulation, SST cooling is reduced due the BL ( Fig. 4.6c), however, after 60 h SST field shows greater cooling near the storm center. This is because upwelling occurs closer to the center due to Trami's slow translations speed. In this case, the BL is completely eroded by mixing and upwelling, and the upper ocean is well mixed with the subsurface layer underneath the TC. The upper ocean mixing is seen in the vertical cross section in Fig. 4.7. At the beginning of the simulation, the cooling in the mixed layer is reduced because of the BL (Fig. 4.7e). Nevertheless, the increased cooling due to the BL is found after 60 h of simulation at the right of the storm center around 130°E and left around 128°E (Fig. 4.7f). This demonstrates that the presence of BL contributes to the reduced upper ocean mixing beneath the TC but also to increase the upwelling generated by the divergent ocean surface currents.  The maximum area-averaged values of SST and HF within 100 km of the storm center, as well as storm intensity (Pmin and Vmax) between the OBLx and OBL00 experiments are summarized in Table 6. In all the experiments, the presence of BL lead to reduction of SST cooling and increased heat fluxes which lead to an increase of TC intensity. Overall, the largest intensity differences are found in the OBL20 and OBL25 cases as expected due to larger BLT. However, no direct correlation between the SST and HF changes and TC intensity can be identified. In Jebi, increasing BLT leads to an increase in the maximum difference of SST and enthalpy flux. However, Trami and Kong-rey have the largest enthalpy flux difference due to BL in OBL20 cases, and the largest intensity differences are not consistently found in minimum pressure and maximum wind speed. Comparing the maximum difference values may not be the best indication of the impact of the BL on intensity in real TCs.

Conclusion
The influence of BL on TC development is investigated in three real tropical cyclones in the northwest Pacific. In our results, the presence of the BL has an effect on the upper ocean response by reducing the entrainment of colder water below into the sea surface and thus increasing the intensity of the storm. This finding agrees with idealized coupled model results in Hlywiak and Nolan (2019) and other previous studies. In the BL experiments, the maximum increased intensity is around 6 to 9 hPa compared to the CTRL experiments. However, we have inconsistent results in sensitivity tests to storm translation speed and oceanic barrier layer thickness. Hlywiak and Nolan (2019) found that the degree of the barrier layer favorable effect on TC intensification increases with increasing the BL thickness and increases for decreasing translation speed. We find maximum increased intensity in minimum pressure in the 51 OBL25 cases for Jebi (fast moving TC) and Trami (slow moving TC). Nevertheless, the impact on TC intensity was the largest in OBL20 for Kong-rey. Besides, the impact of BL on the intensity of Trami is not greater compared to that of other TCs. Since the three TCs used in this study are the real storms occurring in 2018 the results are more complicated than using the idealized background wind. Additionally, it is found that the BL may increase the magnitude of upwelling along the storm track depending on the TC translation speed. Our results show that the enhanced upwelling in slow moving TCs can increase SST cooling underneath the storm and thus compensate for the SST cooling decrease due to the BL. This effect needs to be examined further in future studies.