Satellite signature of the instantaneous wind response to mesoscale oceanic thermal structures

The thermal air–sea interaction mechanism that modulates the atmospheric mixing due to sea‐surface temperature (SST) variability is studied with long‐term consistent satellite records. Statistical analyses of daily and instantaneous wind and SST data are performed over the major western boundary currents (WBCs). This wind–SST coupling, which is mediated by atmospheric mixing, is found to be very relevant on daily, and even shorter, time scales. Co‐located and simultaneous SST and surface wind fields (from Advanced Very High Resolution Radiometer and Advanced Scatterometer data) reveal that the atmosphere responds instantaneously to the presence of SST structures with a larger coupling coefficient with respect to daily and monthly time‐averaged fields. The coupling strength varies seasonally over WBCs in the Northern Hemisphere, with wintertime coupling being the lowest. Reanalysis data show that this behaviour is related to the seasonality of the air–sea temperature difference over the region of interest. Over the Northern Hemisphere WBCs, dry and cold continental air masses drive very unstable conditions, associated with very weak thermal air–sea coupling.


INTRODUCTION
Mesoscale sea-surface temperature (SST) structures are known to modulate the atmospheric boundary layer (ABL) wind through two main mechanisms: downward momentum mixing (DMM Hayes et al., 1989;Wallace et al., 1989) and pressure adjustment (PA Lindzen and Nigam, 1987).
The DMM-mediated response is controlled by a reduced atmospheric stability over warm SSTs, which produces vertical mixing of horizontal momentum, resulting in a net surface wind acceleration.When DMM dominates, surface wind convergence is generated in correspondence to SST gradients (when the wind blows from warm to cold SST).The PA-mediated response is driven by a low surface pressure generated above a positive SST anomaly due to thermal air expansion.Thus, surface wind convergence is produced at a local SST maximum.Both mechanisms have been shown to be important on different time scales in different regions of the world's oceans (see Small et al. (2008) for a review).
Most works focus on the atmospheric response on long time scales (monthly to multiannual), showing that long-lasting SST structures modulate the surface wind field and its derivatives (Chelton et al., 2004), as well as surface turbulent heat fluxes (Small et al., 2019), cloud cover, and rainfall (e.g.Minobe et al., 2008;Nkwinkwa Njouodo et al., 2018).Seasonal variability of the atmospheric response over the Gulf Stream has been highlighted, with significant differences in the thickness of the layer characterized by wind convergence, the dominant height of cloud cover, the type of precipitation, and the heat-flux vertical profiles (Minobe et al., 2010).There is evidence that, even on shorter time scales (weekly and daily), the lower atmospheric dynamics (in terms of wind, clouds, and rainfall) is affected by SST patterns, such as mesoscale eddies in the Southern Ocean (Frenger et al., 2013) and mesoscale structures in the Mediterranean Sea (Meroni et al., 2020;Desbiolles et al., 2021) and warm pool of the western tropical Pacific (Li and Carbone, 2012).Seo (2017) and Gentemann et al. (2020) also highlight the positive correlation between local wind speed and SST on a daily scale over the global ocean, indicative of the action of DMM.
Using satellite data of opportunity, Gaube et al. (2019) show that the surface wind accelerates instantaneously over a submesoscale warm SST structure in the Gulf Stream region.The atmospheric response at the submesoscale is very complex and is characterized by turbulent heat fluxes that are significantly larger than those predicted by bulk approaches (Shao et al., 2019).Young and Sikora (2003), by exploiting remote-sensing data, find that the fast ABL response to SST gradients over the Gulf Stream can lead to the formation of wide stratocumulus bands, with strong implications for the open-ocean weather and, possibly, alteration of the radiative balance over large areas.Also, in situ observations from the Elucidating the Role of Clouds-Circulation Coupling in Climate (EUREC 4 A) field campaign (Stevens et al., 2021) show that the presence of a mesoscale cold SST patch can alter the lower atmospheric vertical mixing, with implications for moisture export out of the ABL and thus low-level cloud formation (Acquistapace et al., 2022).All these works highlight that the atmospheric response can happen on very fast time scales (daily, hourly, and even instantaneous).
Evidence of the fast atmospheric response to the presence of SST gradients is also presented by Parfitt et al. (2016), who show that the sharp SST gradient of the Gulf Stream enhances the strength and frequency of atmospheric fronts through differential heating that reinforces the temperature difference of the air masses of the atmospheric front itself.To interpret the surface convergence long-term mean field better, Parfitt and Seo (2018) introduce a new framework, where the prominent role of wintertime atmospheric fronts in shaping the long-term mean appears.The fact that the frontal near-surface wind convergence is found to be related to the SST patterns suggest that there is a link through either DMM or PA, as also found in satellite data over the Mediterranean (Meroni et al., 2020) and in reanalysis data over the global ocean (Desbiolles et al., 2023, under review).These studies indicate that the atmospheric response to SST is rapid, but they only analyze the effects of mesoscale structures, characterized by a relatively long lifetime.An investigation of the atmospheric response to shorter lived surface-temperature signals has recently been carried out by Strobach et al. (2022), who analyzed high-resolution coupled simulations over the Gulf Stream.In that work, they underlined the existence of a nearly instantaneous atmospheric response to the meso-and submesoscale SST forcing through DMM, characterized by very intense turbulent heat bursts during cold-air outbreaks.
Air-sea connections are typically measured by computing the coupling coefficients (Small et al., 2008, and references therein).They are defined as the least-squares linear slope computed on the binned scatter plots of atmospheric and oceanic variables, appropriately chosen, as documented clearly in the literature.In particular, downwind SST gradient and wind divergence are considered to measure the relevance of DMM (e.g.Chelton et al., 2004) and SST Laplacian and wind divergence that of PA (Takatama and Schneider, 2016).
The primary goal of this work is to compute the coupling coefficients for DMM to explore whether simultaneous and co-located wind and high-resolution SST data reveal the existence of a fast atmospheric response to short-lived SST structures.To this aim, both instantaneous and daily SST products are used and the differences in the expression of the air-sea coupling emerging in the wind field are highlighted.Secondly, (Desbiolles et al., 2023, under review), by exploiting reanalysis data, highlight that background wind and air-sea temperature difference modulate the wind-SST coupling.For this reason, a link between the seasonality of the coupling coefficients and these environmental variables is sought.Data and methods are introduced in Section 2. Section 3.1 describes the results in terms of the comparison between instantaneous and daily coupling.Section 3.2 shows the seasonal variability of the wind-SST coupling and its connection with the environmental variables introduced above.Conclusions are drawn in Section 4.

DATA AND METHODS
SST data come from the European Space Agency Climate Change Initiative (ESA CCI) 1 data set, which covers the time frame from 1981-2016 and is based on data coming from the Along-Track Scanning Radiometer (ATSR) and Advanced Very High Resolution Radiometer (AVHRR) instruments carried on different satellites throughout the time span considered (Merchant et al., 2019).In particular, we use the Meteorological Operational-A (MetOp-A) AVHRR L3U (Level 3 Uncollated) v2.1 product (Embury et al., 2019) and the L4 (Level 4) v2.1 analysis product (Good et al., 2019), which are available from the Centre for Environmental Data Analysis (CEDA) archive. 2 Both L3U and L4 data are given on a regular longitude-latitude grid with a 0.05 • spacing.The L4 data are gap-free daily maps, obtained with a variational approach to combine a forecast field and all the available satellite observations (Merchant et al., 2019).L3U SST products are instantaneous data retrieved on the swath of the satellite (the instrument performs six measurement cycles per second).For this reason, they have gaps where there are clouds and sea ice.They are provided with a quality flag that describes the reliability of the observed value.The interest here is to maximize the spatial coverage, with a focus on the SST spatial gradients and not on the SST absolute values.In this case, Merchant et al. (2019) suggest keeping data with a quality flag larger than three, which is what is done here.It is important to note that, despite L3U and L4 SSTs having the same nominal resolution, the L4 data set has fewer fine-scale structures, as its data gaps are filled with relatively smooth fields.
Wind observations are taken from the L2 (Level 2) coastal Advanced Scatterometer (ASCAT) products, in the CDR version from the MetOp-A satellite (Verhoef et al., 2017) in the time frame between the beginning of January 2007 and the end of March 2014.They are available from the NASA JPL PODAAC platform 3 and are given on their native irregular grid with nominal 12.5-km grid spacing.They are instantaneous measurements and contain a quality flag that enables us to remove the unreliable pixels.The reasons for which pixels are flagged include very low and very high wind speed values, the presence of rainfall, and the failure of some quality checks, to list a few.The ASCAT instrument exploits microwave radar pulses from six antennas to retrieve the surface wind field.It performs a full measurement cycle in 0.21 s, which is interpreted to be the time scale over which the wind field 1 https://climate.esa.int/en/ 2 Accessible from https://dap.ceda.ac.uk/neodc/esacci/sst/data/CDR_ v2/ 3 https://podaac.jpl.nasa.gov/ in a single pixel is considered to be coherent.The wind components are linearly interpolated on the regular SST grid, so that first derivatives in the wind field are conserved.We justify the interpolation from a coarse to a fine grid with the fact that we want to preserve the full distribution of the SST gradient forcing field.The choice of bilinear interpolation method implies that the wind divergence values are repeated over contiguous pixels.This does not impact the results, as we perform statistical tests on the binned scatter plots.The analyses are performed in the time window covered by the ASCAT MetOp-A CDR, namely between January 2007 and March 2014.
The quantities considered to measure the dynamical effect of the DMM mechanism are introduced below, following Meroni et al. (2022).In particular, DMM and PA are considered to be acting mostly in the along-wind and across-wind directions, respectively.Firstly, a background wind field is defined by means of a Gaussian spatial filter with standard deviation equal to 50 km, so that the along-wind and across-wind directions can be defined.They are named {r, ŝ}, with r being the along-wind unit vector and ŝ being the across-wind unit vector, positive at 90 • counterclockwise with respect to r.Secondly, the wind-field anomaly is obtained by subtracting the background wind from the full field, and is then projected on the {r, ŝ} frame of reference.Thus, the wind field can be written as with { î, ĵ} denoting the standard eastward and northward Cartesian unit vectors, U the background wind speed, and u r , u s the wind anomaly components in the rotated frame of reference.Thirdly, the along-wind divergence is defined to be u r ∕r.Meroni et al. (2022) show that the action of DMM can be detected by computing the correlation or the coupling coefficient between the along-wind SST gradient SST∕r and the along-wind divergence u r ∕r.
With the L4 SST data, a single daily SST map is analyzed together with all the available L2 wind swaths (both ascending and descending) of the same day.In what follows, this is named daily coupling.With the L3U SST data, instead, the analyses are performed on co-located and simultaneous observations of SST and wind field (instantaneous coupling).This is possible because both AVHRR and ASCAT are carried by the same satellite, MetOp-A.Because of the coarser resolution of the wind data and the presence of small-scale noise in the SST products (mainly in the L3U), the SST fields are smoothed by applying a Gaussian filter with standard deviation  = 10 km (comparable with the grid spacing of the scatterometer data).ERA5 reanalysis data (Hersbach et al., 2020) are also used to compute the coupling coefficients of interest (using monthly averaged fields) and to evaluate the seasonality of the relevant environmental variables introduced previously (using daily averaged fields).
As previously introduced, the coupling coefficients are computed as the least-squares slope of the binned distributions of the relevant variables.With daily and instantaneous fields, such binned distributions show a significant degree of nonlinearity.For this reason, the least-squares slope, denoted by , is computed on the central 90% points of the distributions.Thus, only values between the 5% and the 95% percentiles of the forcing variable (the along-wind SST gradient for DMM) are considered in the computation of .

RESULTS
Previous works indicate that, in the presence of strong SST fronts, the signature of ocean thermal effects on the atmosphere is stronger (e.g.Sullivan et al., 2020).Thus, in this work we focus on the regions of the four major western boundary currents (O'Neill et al., 2012, WBCs), as they contain strong SST gradients associated with the fronts of intense currents, many mesoscale eddies, and submesoscale structures (Figure 1).

Instantaneous versus daily coupling
The annual DMM statistics are computed for the various data sets in the regions of interest (Figure 2).The use of daily and instantaneous data highlights that the DMM mechanism holds on very short time scales, as indicated by the monotonic relationship that emerges from the binned scatter plots of Figure 2. Compared with monthly mean data, which are often analyzed in the literature (e.g.Chelton and Xie, 2010), the range of the SST forcing field is extended by roughly a factor 3. In terms of coupling coefficients, it is found systematically that the shorter the time scale, the stronger the coupling.In particular, Figure 2 shows that the monthly coupling coefficient (ERA5) is lower than the daily one (L4), which is lower than the instantaneous one (L3U).For example, over the Gulf Stream and the Kuroshio, the coupling coefficient is 0.07 and 0.08 m⋅s −1 ⋅K −1 for ERA5 monthly mean data, reaches 0.18 and 0.16 m⋅s −1 ⋅K −1 at the daily time scale, and increases to 0.25 and 0.30 m⋅s −1 ⋅K −1 for L3U co-located instantaneous data (Figure 2).This shows that, from a statistical point of view, the instantaneous wind response is stronger than in longer time frames.
If the SST data are not filtered, the instantaneous coupling is weaker than the daily one (not shown).The reason for this is twofold.Firstly, the coarse resolution of the current scatterometer data (Δx ∼ 12.5 km) does not allow us to detect wind structures at the scales of the SST variability (Δx ∼ 5 km).Secondly, the SST L3U data contain some small-scale noise.This calls for an effort to develop new missions to observe the co-located and simultaneous wind, SST, and current fields properly at higher spatial resolution.Examples of such missions are the European Space Agency (ESA) Earth Explorer X Harmony

Seasonal modulation
The DMM binned scatter plots can be computed for each season separately (defined in the standard meteorological way: December-January-February (DJF), March-April-May (MAM), June-July-August (JJA), and September-October-November (SON)).By considering co-located and simultaneous wind and SST data (L3U), it is found that, over the Northern Hemisphere (NH) WBCs, the DMM coupling coefficient displays a seasonal variability, whereas over the Southern Hemisphere (SH) WBCs it does not (Figure 3).In particular, the lowest coupling is found in DJF over the NH WBCs, corresponding to the winter season.The clearest example is the Kuroshio current system, in which the DJF slope is 0.23 m⋅s −1 ⋅K −1 , that is, roughly 30% smaller than the coupling coefficients of the other seasons, which are around 0.  are used to estimate the dependence of the DMM and PA coupling coefficients as a function of two relevant environmental variables.In particular, they consider background wind speed and air-sea temperature difference (defined as ΔT = T 2m − SST, with T 2m being the air temperature at 2 m).The air-sea temperature difference is chosen as a proxy for atmospheric stability, as it is directly related to the ABL dynamics through turbulent air-sea fluxes and the consequent excitation or inhibition of large eddies (Kettle, 2015).With their approach, Desbiolles et al. (2023) can derive a general dependence of the strength of DMM according to the large-scale atmospheric state.They find that the DMM coupling is enhanced in moderate to strong background wind conditions (between 10 and 20 m⋅s −1 ) and for near-neutral air column stability conditions (for ΔT between −1 and 0.5 K).For the present work, it is relevant to highlight that they find a decrease in the strength of DMM coupling in unstable conditions.According to their interpretation, such a decreased DMM strength is related to the strong vertical atmospheric mixing produced by the large negative ΔT.Three years (2007-2009 inclusive) of daily ERA5 data are used to compute the median, 10th, and 90th percentiles of the seasonal cycles of air-sea temperature difference, wind speed, SST, and T 2m (Figure 4).Summer is characterized by near-neutral conditions and relatively low winds, whereas winter is characterized by unstable conditions and moderate to large winds.It appears that the amplitude of the seasonal cycle of all variables considered is larger in the NH than in the SH.In particular, the amplitude of the median ΔT in the NH is at least twice as large as in the SH (between 3 and 4 K with respect to roughly 1.5 K, Figure 4a).For the wind speed, instead, the presence of strong and steady winds over the Southern Ocean makes the SH seasonal median quite flat and larger than in the NH (Figure 4b).Following the results of Desbiolles et al. (2023), the L3U coupling coefficient being maximum in summer in the NH is interpreted as due to the fact that ΔT is the controlling environmental variable.In fact, the lowest coupling coefficient is observed in winter, when the air column is in unstable conditions (Figure 4a).If the background wind modulation was the dominant one, the peak coupling coefficient would be observed in winter, when the winds are stronger (Figure 4b).The fact that in the SH the ΔT seasonal amplitude is weak and the coupling coefficients do not show a seasonal variability appears to confirm further the ΔT control on the DMM coupling.With a closer look at the seasonal cycle of SST and T 2m (Figure 4c,d), it is visible that the larger amplitude of the T 2m cycle is responsible for the large ΔT variability in the NH.In particular, it is because of the low T 2m winter values, associated with cold-air outbreaks that flow from the continent over the oceans, that the air column is so unstable in DJF (Young and Sikora, 2003;Strobach et al., 2022).

CONCLUSIONS
It is known that various mechanisms can control the atmospheric response to the presence of SST patterns (Small et al., 2008;Seo et al., 2023).In this work, we focused on the role of DMM, which has been shown to be relevant over strong SST gradients and short time scales (Skyllingstad et al., 2007;Sullivan et al., 2020).Multiannual analyses of surface wind and SST satellite data over the four major WBCs are performed.They reveal that a statistically significant response of near-surface wind to mesoscale SST structures can be detected, both in data that match within a 24-hr overpass and in simultaneous data.Firstly, the monotonic relationship between along-wind SST gradient and surface wind divergence is found to extend over a larger range (at least of a factor 3) of both forcing and response fields with respect to monthly fields, as recently pointed out by Strobach et al. (2022) with numerical simulation data.Secondly, the instantaneous coupling is found to be roughly 50% larger than the daily coupling, indicating that the atmosphere responds to SST through the mechanism of downward mixing of momentum over very short time scales.This has been pointed out mostly in idealized setups, such as those in Skyllingstad et al. (2007) and Sullivan et al. (2020), but, to the best of our knowledge, consistent evidence of it in satellite observations is still lacking.Further analyses should be performed to quantify the role of other mechanisms known to impact the wind-SST coupling on such short time scales (Small et al., 2008;Seo et al., 2023).
A seasonality in the strength of the DMM coupling emerges over the two Northern Hemisphere WBCs (the Gulf Stream and the Kuroshio current), with minimum coupling observed during the winter months.Over the Agulhas and Malvinas currents, no seasonality is found.As demonstrated by the recent work of Desbiolles et al. (2023), a large atmospheric static instability (large negative ΔT, the air-sea temperature difference) strongly reduces the DMM coupling, because the overall stronger atmospheric mixing makes the lower atmosphere less sensitive to the surface SST gradients.Reanalysis data support the interpretation that the air-sea temperature difference is responsible for the seasonality of the coupling, because a strong ΔT seasonal cycle is observed in the NH only.In particular, the very cold air temperature winter values in the NH, associated with the presence of cold and dry continental air masses that reach the humid and warmer ocean, induce strongly unstable conditions, which lower the efficiency of the DMM coupling.Despite the weaker coupling observed in winter in the NH, the fact that the instantaneous coupling is strongest suggests that DMM might be involved directly in the dynamical response of an atmospheric front over a WBC.To evaluate this, further specific analyses should be performed.
The DMM coupling coefficients estimated in the present work exploiting robust and long-term intercalibrated satellite products can be used as a benchmark to evaluate global numerical model performance.In fact, it is known that there are some critical numerical parameterizations, such as those related to atmospheric mixing, that are responsible for a large fraction of the spread among climate models (e.g.Sherwood et al., 2014).The significance of the present work lies in the fact that we provided some metrics based on observational data, with their seasonal and geographical dependences, which can be compared directly with the corresponding quantities obtained from model simulations.This can shed light on how well various numerical models represent ABL mixing.Further improvements on these parameterizations require a knowledge of air-sea coupling at an even finer scale than is presently feasible, which is limited by the resolution of the wind observations (more than one order of magnitude coarser than SST data).Planned satellite missions, such as the ESA Earth Explorer X Harmony and the candidate Earth Explorer XI SEASTAR, will provide data that can be used directly for this purpose.For example, Harmony will provide simultaneous observations of the wind field and SST at kilometric scales, enabling us to estimate the strength of the air-sea coupling on ephemeral oceanic structures.

F
I G U R E 1 L4 SST on October 1, 2010, shown in the areas of interest in which the analyses are performed.
U R E 2 DMM binned scatter plots computed with all data between 2007 and 2014.L3U and L4 scatter plots are based on daily or instantaneous data, while ERA5 scatter plots are based on monthly averaged data.Transparent markers are used where the number of values in the bin is lower than 100.Thin lines in the lower part of the panels show the histogram of the forcing fields shown with a log scale (right side of the panels).The region over which the statistics are computed is indicated in the title of the panels.The coupling coefficients, computed as defined in the main text, are displayed in the legend.mission (ESA (2020)) and the candidate Earth Explorer XI SEASTAR mission.
3 m⋅s −1 ⋅K −1 .However, by looking at the tails of the binned scatter plots, the seasonality clearly also emerges over the Gulf Stream, suggesting that there is a seasonal modulation of the coupling in the NH and not in the SH.The seasonality of the DMM coupling coefficient is interpreted in the light of the recent findings of Desbiolles et al.(2023).In their work, global ERA5 daily fields U R E 3 DMM binned scatter plots computed between 2007 and 2014, distinguished by season and based on L3U data.The region over which the statistics are computed is displayed in the title of the panels.Transparent markers are used where the number of values in the bin is lower than 100.Vertical lines indicate the 90% central points.
Seasonal cycle of (a) air-sea temperature difference, (b) large-scale wind speed, (c) SST, and (d) air temperature at 2 m.They are shown as median values (thick lines) and 10th and 90th percentiles (thin dashed lines) for the WBC regions of interest, as indicated in the legend.Light and dark blues are for the NH and light and dark red are for the SH.The number of markers does not correspond to the time resolution of the data used to compute the seasonal cycle (daily).