Access the full text.
Sign up today, get DeepDyve free for 14 days.
W. Rath (2013)
Reduction of near-inertial energy through the dependence of wind stress on the ocean-surface velocity, 118
J. M. Lilly (2021)
jlab: A data analysis package for Matlab, v. 1.7.0
S. Elipot (2016)
A global surface drifter data set at hourly resolution, 121
J. M. Lilly (2017)
Element analysis: A wavelet-based method for analysing time-localized events in noisy time series, 473
R. E. Thomson (1998)
Satellite-tracked drifter measurement of inertial and semidiurnal currents in the northeast Pacific, 103
D. Slepian (1978)
Prolate spheroidal wave functions, Fourier analysis, and uncertainty—V: The discrete case, 57
I. A. Renfrew (2020)
An evaluation of surface meteorology and fluxes over the Iceland and Greenland Seas in ERA5 reanalysis: The impact of sea ice distribution, 147
P. Niiler (2001)
The world ocean surface circulation
R. Lumpkin (2016)
Full access fulfilling observing system implementation requirements with the global drifter array, 33
S. Elipot (2009b)
Estimates of wind energy input to the Ekman layer in the Southern Ocean from surface drifter data, 114
M. M. Flexas (2019)
Global estimates of the energy transfer from the wind to the ocean, with emphasis on near-inertial oscillations, 124
W. G. Large (1981)
Open ocean momentum flux measurements in moderate to strong winds, 11
N. Maximenko (2013)
Ocean surface circulation
R. T. Pollard (1970)
Comparison between observed and simulated wind-generated inertial oscillations, 17
J. E. Stopa (2014)
Intercomparison of wind and wave data from the ECMWF reanalysis interim and the NCEP climate forecast system reanalysis, 75
V. Hormann (2014)
Response of upper ocean currents to Typhoon Fanapi, 41
L. St. Laurent (2006)
Estimates of power consumed by mixing in the ocean interior, 19
M. H. Alford (2001)
Internal swell generation: The spatial distribution of energy flux from the wind to mixed layer near-inertial motions, 31
M. H. Alford (2020)
Revisiting near-inertial wind work: Slab models, relative stress and mixed-layer deepening, 50
M. H. Alford (2003a)
Improved global maps 54-year history of wind-work on ocean inertial motions, 30
E. A. D’Asaro (1985)
The energy flux from the wind to near-inertial motions in the surface mixed layer, 15
S. Landwehr (2020)
Using global reanalysis data to quantify and correct airflow distortion bias in shipborne wind speed measurements, 13
R. W. Jones (2016)
Evaluation of four global reanalysis products using in situ observations in the Amundsen Sea embayment, Antarctica, 121
G. J. Goni (2017)
Autonomous and Lagrangian ocean observations for Atlantic tropical cyclone studies and forecasts, 30
S. R. Jayne (2001)
Parameterizing tidal dissipation over rough topography, 28
K. M. Schmidt (2017)
Evaluation of satellite and reanalysis wind products with in situ wave glider wind observations in the Southern Ocean, 34
S. T. Gille (2005)
Statistical characterization of zonal and meridional ocean wind stress, 22
Y. Liu (2019)
Wind power on oceanic near-inertial oscillations in the global ocean estimated from surface drifters, 46
A. J. Plueddemann (2006)
Observations and models of the energy flux from the wind to mixed-layer inertial currents, 53
S. Elipot (2006)
Spectral characterization of Ekman velocities in the Southern Ocean based on surface drifter trajectories
R. Lumpkin (2007)
Measuring surface currents with surface velocity program drifters: The instrument, its data, and some recent results
M. H. Alford (2003b)
Redistribution of energy available for ocean mixing by long-range propagation of internal waves, 423
L. Centurioni (2017)
A global ocean observing system for measuring sea level atmospheric pressure: Effects and impacts on numerical weather prediction, 98
D. V. Hansen (1996)
Quality control and interpolations of WOCE-TOGA drifter data, 13
B. F. Kilbourne (2015)
Quantifying high-frequency wind energy flux into near-inertial motions in the southeast Pacific, 45
H. L. Simmons (2012)
Simulating the long-range swell of internal waves generated by ocean storms, 25
S. R. Smith (1999)
Establishing more truth in true winds, 16
W. G. Large (1995)
Accounting for surface wave distortion of the marine wind profile in low-level Ocean Storms wind measurements, 25
J. M. Lilly (2021)
A unifying perspective on transfer function solutions to the unsteady Ekman problem, 6
W. Munk (1998)
Abyssal recipes II: Energetics of tidal and wind mixing, 45
P. M. Poulain (2015)
Direct measurements of World Ocean tidal currents with surface drifters, 120
J. F. Price (1986)
Diurnal cycling: Observations and models of the upper ocean response to diurnal heating, cooling, and wind mixing, 91
S. Elipot (2009a)
Ekman layers in the Southern Ocean: Spectral models and observations, vertical viscosity and boundary layer depth, 5
P. Niiler (1995)
Measurements of the water-following capability of holey-sock and TRISTAR drifters, 42
M. Belmonte Rivas (2019)
Characterizing ERA-Interim and ERA5 surface wind biases using ASCAT, 15
OCTOBER 2022 KL E N Z E T A L . 2417 Estimates of Near-Inertial Wind Power Input Using Novel In Situ Wind Measurements from Minimet Surface Drifters in the Iceland Basin a b c d e THILO KLENZ, HARPER L. SIMMONS, LUCA CENTURIONI, JONATHAN M. LILLY, JEFFREY J. EARLY, AND VERENA HORMANN College of Fisheries and Ocean Sciences, University of Alaska Fairbanks, Fairbanks, Alaska Applied Physics Laboratory, University of Washington, Seattle, Washington Scripps Institution of Oceanography, La Jolla, California Planetary Science Institute, Tucson, Arizona NorthWest Research Associates, Redmond, Washington (Manuscript received 2 December 2021, in final form 8 June 2022) ABSTRACT: The Minimet is a Lagrangian surface drifter measuring near-surface winds in situ. Ten Minimets were de- ployed in the Iceland Basin over the course of two field seasons in 2018 and 2019. We compared Minimet wind measure- ments to coincident ship winds from the R/V Armstrong meteorology package and to hourly ERA5 reanalysis winds and found that the Minimets accurately captured wind variability across a variety of time scales. Comparisons between the ship, Minimets, and ERA5 winds point to significant discrepancies between the in situ wind measurements and ERA5, with the most reasonable explanation being related to spatial offsets of small-scale storm structures in the reanalysis model. After a general assessment of the Minimet performance, we compare estimates of wind power input in the near-inertial band using the Minimet winds and their measured drift to those using ERA5 winds and the Minimet drift. Minimet-derived near-inertial wind power estimates exceed those from Minimet drift combined with ERA5 winds by about 42%. The results highlight the importance of accurately capturing small-scale, high-frequency wind events and suggest that in situ Minimet measurements are beneficial for accurately quantifying near-inertial wind work on the ocean. SIGNIFICANCE STATEMENT: In this study we introduce a novel, freely drifting wind measurement platform, the Minimet. After an initial validation of Minimet sea surface wind measurements against independent wind measure- ments from a nearby research vessel, we investigate their utility in context of the near-inertial work done by the wind on the ocean, which is important for the ocean’s energy budget. We find Minimet near-inertial wind work estimates ex- ceed those estimated using winds from a state-of-the-art wind product by 42%. Our results indicate that capturing storm events happening on time scales less than 12 h is crucial for accurately quantifying near-inertial wind work on the ocean, making wind measurements from platforms such as the Minimet invaluable for these analyses. KEYWORDS: Ocean; North Atlantic Ocean; Atmosphere-ocean interaction; In situ atmospheric observations; In situ oceanic observations; Wind profilers 1. Introduction The near-inertial band has been identified as important for transferring energy from inertial oscillations of the surface It has long been recognized that wind stress acting on mixed layer, through the shear zone at the base of the mixed the ocean surface accounts for a significant portion of the layer, and into vertical propagating near-inertial waves in the estimated 2 TW (Munk and Wunsch 1998; St. Laurent and stratified ocean interior (e.g., Plueddemann and Farrar 2006). Simmons 2006) needed to sustain abyssal mixing in the Multiple approaches have been taken to estimate how much deep ocean. Previous studies have highlighted the impor- energy the wind transfers to inertial oscillations of the surface tance of the wind for internal wave generation (D’Asaro mixed layer. D’Asaro (1985), in the first study of its kind, 1985; Simmons and Alford 2012), a process that provides a forced a slab model (Pollard and Millard 1970) with wind mechanism for mixing through shear instabilities and wave stress from moored meteorological buoys located in the North breaking, often far away from the waves’ generation region Pacific and North Atlantic to calculate the energy transfer (Alford 2003b). Theaccurateestimation ofthis vertically from the wind to mixed layer currents rotating at the local in- propagating energy fraction is crucial for correctly repre- ertial frequency. He showed that the bulk of the energy input senting abyssal mixing in general circulation and climate was provided by midlatitude winter storms on scales of 100 km models. and was as such highly temporally intermittent. Building on this study, Alford (2001) produced maps of near-inertial wind power input between 6508 latitude, using 6-hourly Denotes content that is immediately available upon publica- NCEP–NCAR gridded reanalysis wind stress to force a tion as open access. slab model similar to D’Asaro (1985). He found significant regional and seasonal variability, with the wind supplying Corresponding author: Thilo Klenz, tklenz2@alaska.edu an average of about 0.29 TW to inertial oscillations of the DOI: 10.1175/JPO-D-21-0283.1 Ó 2022 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses). 2418 J OUR N A L O F P HY SI C A L O C E A N OGR A P HY VOLUME 52 mixed layer, most of it being supplied in the western part sufficient in situ wind measurements in combination with of the basins during winter. Alford (2003a) later modified mixed layer current measurements. Schmidt et al. (2017) com- the slab model approach slightly and extended the Alford pared several gridded satellite scatterometer and reanalysis (2001) calculations poleward, estimating the global near- wind products to in situ winds measured from an autonomous inertial wind power input to be around 0.47 TW. The ability measurement platform in the Southern Ocean. Their results of the slab model to accurately simulate mixed layer inertial emphasized the need for high resolution in situ wind measure- currents as a response to strong wind forcing depends upon ments to validate gridded wind products. Liu et al. (2019) the suitability of the choices of prescribed mixed layer depth compared near-inertial wind power estimates calculated from and a linear damping parameter (Alford 2020; Plueddemann surface drifters to that using in situ wind measurements from and Farrar 2006). Plueddemann and Farrar (2006) showed moored buoys located primarily in the North Atlantic and that the slab model systematically overestimates near-inertial Pacific Oceans and found power estimates from 6-hourly wind power input compared with observations, and hence ques- reanalysis winds interpolated onto drifter positions were tioned its utility for wind power calculations. consistently smaller than those from moored buoy estimates by More recently, Alford (2020) updated the Alford (2003a) up to a factor of 2. They attributed this discrepancy to reduced estimate again to 0.27 TW using hourly reanalysis winds and near-inertial variance in the wind product. Similarly, Elipot and the Price et al. (1986) mixed layer model (henceforth referred Gille (2009b) concluded that their near-inertial wind power cal- to as the PWP model). Using the PWP model resulted in bet- culations in the Southern Ocean are likely to be underestimates ter agreement with observations compared to the slab model, due to reduced near-inertial variance in both the 6-hourly due to an additional damping term acting on short time scales. drifter product and 6-hourly reanalysis winds (see also Gille Recently Alford (2020), using hourly, 0.68 resolution reanaly- 2005). sis winds, showed that the Pollard and Millard (1970) slab Since the transition to the Iridium satellite system in 2016, model overestimated near-inertial wind power input by 23% the modern GDP drifters have been configured to report their globally compared to PWP. observations, including their GPS location, at regular hourly From the above estimates, the contribution of near-inertial intervals (Centurioni 2018). The combination of hourly drifter wind power input is likely of the same order of magnitude as data with a new generation of hourly reanalysis wind products theestimated 1TW provided bythe tides (Jayne and St. promises significant improvements of the near-inertial wind Laurent 2001) and its importance in the global energy bud- power input estimate, particularly at latitudes above 508,where get is clear. However, the large range of estimates from a the 6-hourly resolution of some gridded wind products is insuffi- variety of methods is evidence of significant uncertainty. cient to resolve near-inertial wind variability due to the Coriolis A desirable option is to use mixed layer velocities obtained frequency approaching the Nyquist frequency of the wind prod- from satellite-tracked Lagrangian surface drifters [see Niiler uct (Alford 2001; Gille 2005). The hourly version of the GDP (2001) and Maximenko et al. (2013) for a review]. A large dataset has been shown to have between 25% and 50% more global array of approximately 1250 surface drifters is main- velocity variance than the previous, 6-hourly kriged version tained as part of NOAA’s Global Drifter Program (GDP; (Elipot et al. 2016). Flexas et al. (2019) showed that forcing Centurioni et al. 2017). These drifters are drogued to follow their model with hourly instead of 6-hourly reanalysis winds the currents at 15-m depth and hence represent a good ap- resulted in an inertial peak in the surface current spectrum proximation of ocean mixed layer velocities. This fact together that was 2–3 times higher. While the hourly model reanalysis with their extensive coverage (Lumpkin et al. 2016), recently winds are definitely expected to result in improved near-inertial improved hourly reporting (Centurioni 2018), and the recent wind power input estimates, to our knowledge, an intercompari- development of an enhanced resolution hourly interpolated son between hourly model reanalysis winds and hourly in situ product (Elipot et al. 2016), makes the GDP drifters conve- winds within the context of the near-inertial wind power prob- nient platforms for estimating near-inertial wind power input lem has yet to be made. It is unclear whether the use of hourly from observed mixed layer velocities directly, obviating the reanalysis winds will solve problems associated with underesti- need for simulated velocities from a mixed layer model. Sur- mated near-inertial variance, so validation against in situ meas- face drifters have previously been utilized in combination with urements remains vital. This is the point of the present study. gridded wind products to estimate wind power input on re- In this study we utilize the Minimet drifter, a satellite-tracked, gional and global scales. Elipot and Gille (2009b) used a subset freely drifting measurement platform. Building on the Surface of the standard, 6-hourly kriged GDP dataset (Hansen and Velocity Program (SVP; Niiler et al. 1995; Niiler 2001; Centurioni Poulain 1996; Lumpkin and Pazos 2007) in combination with 2018) drifter design, Minimet drifters (Centurioni 2018)were 6-hourly ERA-40 reanalysis wind stress to estimate wind configured to measure in situ wind speed and direction at vari- power input in the Southern Ocean. On a global scale, Liu able sampling rates in combination with position data. This et al. (2019) used the hourly GDP dataset of Elipot et al. presents us with a unique opportunity to estimate near-inertial (2016) and a 6-hourly assimilated wind product. They estimated wind power input from concurrent in situ wind and current the global near-inertial wind power input to be between 0.3 and measurements in the Lagrangian frame and compare it to that 0.6 TW, in general agreement with studies mentioned above. calculated using hourly ERA5 reanalysis winds interpolated The accuracy of gridded wind products in combination with along the Minimet drifter trajectories. To our knowledge, this surface drifters for near-inertial wind power calculations is, is the first estimate of near-inertial wind power input using however, rarely assessed directly, particularly due to a lack of Lagrangian drifting buoys to provide both in situ current OCTOBER 2022 KL E N Z E T A L . 2419 FIG. 1. Trajectories of Minimet drifters deployed during spring field campaigns in 2018 (red) and 2019 (blue). The inset shows the location of the study region and the time mean vector winds (arrows) and wind stress (colors) from ERA5 over the study period. Initial positions are shown by the colored dots and by the yellow star in the inset. and wind measurements. The data and methods used for deployed in front of tropical cyclones and is optimized to mea- analysis are described in section 2, followed by our results sure wind speed and direction in the most challenging condi- and a discussion thereof in sections 3 and 4, respectively. tions (Goni et al. 2017). Similar drifter configurations with an identical surface buoy design were successfully deployed in front of tropical cyclones and measured wind speeds in excess 2. Data and methods of 40 m s (Hormann et al. 2014). An internal algorithm is a. Minimet surface drifters used to filter out unrealistic or unrepresentative winds over a 7-min ensemble to ensure meaningful wind estimates. The A total of 10 Minimet surface drifters were deployed from wind velocity samples are discarded when the tilt of the sensor, the R/V Armstrong during the Near-Inertial Shear and obtained from the motion package, exceeds the manufacturer Kinetic Energy in the North Atlantic experiment (NISKINe). specifications, or when the anemometer is submerged. The Five Minimets were deployed during each of the two cruises wind samples are then preconditioned using a filter that re- in May 2018 and June 2019 (Fig. 1). The Minimet (Centurioni moves anomalous wind velocity values due, for example, to 2018)is based on the SVP (https://gdp.ucsd.edu/ldl/svp/) wave crest shielding effects and disturbances from seawater drifter configuration, consisting of a surface float drogued with spray, and wind speed is represented by the median value over a holey sock at 15-m depth and equipped with a temperature the ensemble. This preprocessing aims to reduce bias due to sensor to measure sea surface temperature. Minimet drifter positions are tracked with a GPS transponder with 2–10-m ac- the Minimet sampling in adverse conditions. Such a bias would curacy, and data are transmitted using a two-way Iridium sat- be expected to lead to Minimet wind speed being lower at ellite modem. In addition to the standard SVP configuration, high wind speeds relative to reanalysis winds (cf. Large et al. Minimets are equipped with a barometer and a 2D Gill Sonic 1995; Renfrew et al. 2020; Schmidt et al. 2017). anemometer, measuring wind speed up to 60 m s and wind To assess how well the Minimets measure winds, Minimet direction with 2% and 62.58 accuracy, respectively, at a nomi- wind speed measurements were compared against winds mea- nal height of 0.5 m above the sea surface. A 9-DOF (degrees sured by the meteorology package on board the R/V Armstrong of freedom) motion package is used to compute the attitude of when the Minimets were within 25 km of the vessel during the the buoy with respect to the east-north-up (ENU) frame of respective cruise periods. True winds were calculated from raw reference. The 9-DOF chip’s sensors are sampled simulta- R/V Armstrong ship winds following Smith et al. (1999) and neously at high frequency and fed into a sensor fusion algo- the median was calculated over the same 7-min interval as rithm that outputs the 3D attitude of the buoy. The the Minimets. Ship winds were adjusted from the ship-mast anemometer is sampled at the highest frequency allowed by sensor height of 17.9 m to the standard 10-m height using an the device and the horizontal wind is converted to the ENU assumed law of the wall profile (Large and Pond 1981). The frame of reference using the buoy orientation obtained from Minimet winds were similarly adjusted to 10 m by compar- the motion package. The Minimet was designed to be ing the 10-m ship winds to the Minimet wind measurements 2420 J OUR N A L O F P HY SI C A L O C E A N OGR A P HY VOLUME 52 utilizing an iterative process that sought to minimize the Here u is a vector of the drifter velocity at 15-m depth, root-mean-square deviation (RMSD) between the two meas- which we assume to be an adequate approximation of the sur- urements by varying the effective height of the Minimet wind face currents, and u~ is the 10-m wind vector from ERA5. era sensor. The resulting Minimet effective sensor height of 0.30 m Flexas et al. (2019) highlighted the importance of including was subsequently used for adjusting the Minimet winds to 10-m the effect of ocean currents in calculating the wind stress, and height assuming a law of the wall (Large and Pond 1981). Alford (2020) and Rath et al. (2013) reported a reduction of Drifter velocities were obtained by central differencing the wind power input into the near-inertial band of 13% and drifter positions in time. Outliers in the inferred velocity re- 20%, respectively, when accounting for this effect. Energy cords were identified, and removed, in postprocessing by first input by the wind can be calculated as the time integral of P applying a velocity standard deviation (s) criteria to the over some time interval. Minimet velocity time series. Velocities exceeding 4s were Trajectories were divided into 20-day segments with 50% eliminated from the data. A second analysis step involved ap- overlap. The segment length was chosen to account for the plying a 24-h running-mean filter to the drifter position time se- change of Coriolis parameter along a given trajectory while ries and eliminating positions that deviated from the running retaining adequate frequency resolution. Near-inertial wind mean by more than 0.058 in latitude or in longitude. The thresh- power input P was then calculated for each 20-day segment olds for these criteria were chosen subjectively. This procedure by multiplying wind stress t calculated using either Eq. (1) or resulted in outliers being removed without damping high- Eq. (2) along the drifter trajectory by the near-inertial drifter frequency signals in the time series through excessive filtration. velocity u : Minimet wind measurements for these outliers were discarded. P t · u : (3) After these processing steps, a total of 37 500 drifter hours of drifter velocities and in situ sea level winds from Minimet The velocity u was calculated using a bandpass filter. The drifters deployed during the NISKINe cruise periods were I specific choice of filter will be described in detail later on. available for analysis. b. Reanalysis winds 3. Results We made use of the European Centre for Medium-Range As mentioned previously, a total of 10 Minimet drifters Weather Forecasts (ECMWF) ERA5 reanalysis (ECMWF were deployed from the R/V Armstrong during NISKINe 2019). Hourly ERA5 10-m winds at 0.258 horizontal resolution Pilot and Process cruises in spring of 2018 and 2019 (Fig. 1). All were downloaded for the respective periods on 24 February 2020. Minimets recorded their positions, as well as in situ sea surface As per the ERA5 documentation, timestamps can be viewed wind speed and direction, for several months, 156 days on aver- as instantaneous, with winds available at the top of the hour, age. The Minimets were deployed in an active mesoscale eddy matching the temporal resolution of the Minimets. To calcu- field and hence experienced rapid dispersal over their lifetimes, late near-inertial wind power input and for comparison to in drifting in a mainly northeasterly direction from their initial situ measurements, ERA5 10-m winds were linearly interpo- deployment locations. Individual Minimet drifters occasionally lated onto drifter GPS positions at every time step. It is impor- 8 8 became trapped in eddies, particularly around 58 Nand 22 W, tant to note that while sea level pressure measurements from as indicated by their looping trajectories. The study region in the Minimets were assimilated into the reanalysis, the wind which the Minimets were deployed was characterized by strong measurements were not, therefore ensuring that the ERA5 wind stress (Fig. 1, inset), with winds mainly from a southwest- winds are independent from our in situ measurements. erly to westerly direction. As such, Minimet drifters would rou- c. Near-inertial wind power calculation tinely undergo inertial oscillations forced by passing storms, as can be seen upon zooming into Fig. 1 (not shown). Near-inertial wind power input was estimated by first calcu- To evaluate the utility of Minimet wind measurements, in lating wind stress t along the drifter trajectories. Since the situ wind speed measured by Minimet drifters and by the me- Minimets essentially measure wind speed relative to the ocean teorological package located on the bow of the R/V Armstrong velocity at 15-m depth, we can calculate the wind stress as are compared during the respective cruise periods. We are t c r u~ u~ , (1) comparing measurements during times when both platforms mm d air mm mm were within 25 km of each other, resulting in a total of 1137 ~ ~ ~ ~ with u ≡ u · u . Here, u is the horizontal wind individual wind speed data pairs recorded at 15-min inter- mm mm mm mm vector from Minimet wind measurements adjusted to 10-m vals. For quality-control purposes, ship winds were omitted height, r is a reference air density assumed to be a constant following Smith et al. (1999) during times when the ship was air at 1.25 kg m , and c is a drag coefficient obtained following rapidly changing course, accelerating or decelerating, or when Large and Pond (1981). To compare ERA5 winds to in situ thetruewind direction was within 6158 of the stern of the ship. winds from the Minimets, we have to subtract the drifter ve- Changing this true wind direction criterion to 6308 reduced the locities from the reanalysis winds and calculate ERA5 wind number of data points but did not significantly alter the results. stress as follows: The two measurements agreed well with each other (Fig. 2a) and the majority of measurements lay on or close to the 1:1 line t c r u~ 2 u(u~ 2 u): (2) era d air era era (Fig. 2b), with wind speeds up to 17 m s captured by both. OCTOBER 2022 KL E N Z E T A L . 2421 FIG. 2. (a) Coincident Minimet (blue) and ERA5 (red) wind speed vs R/V Armstrong wind speed. Minimet winds within 25 km of R/V Armstrong were considered. (b) Number of observations for Minimet vs Armstrong winds in 1 m s bins. (c) As in (b), but for ERA5 vs Armstrong winds. (d) Time series of hourly wind speed from a single Minimet (blue circles), R/V Armstrong (green diamonds) and ERA5 reanalysis interpolated onto the vessel’s position (red squares) during the 2018 Pilot cruise. The distance between the respective Minimet drifter and R/V Armstrong is shown in black, and the black dotted line depicts 25-km distance. Opacity indicates when the Minimet is within 25 km of R/V Armstrong. (e) As in (d), but during the 2019 Process cruise. ERA5 winds interpolated onto the position of R/V Armstrong components, the square of the bias (systematic differences) revealed that ERA5 wind speed generally tended to be and variance (spread around the mean). The term bias is used lower than that from R/V Armstrong (Figs. 2a,c)for the in a strict statistical sense, with no assumption about which same times considered in the Minimet and R/V Armstrong measurement platform represents the true wind speed. These comparison. intercomparisons during the cruise periods aid in evaluating Time series of adjusted 10-m wind speed indicate that wind the observed differences between ERA5 and Minimet wind speeds from all three sources generally agreed well during times speed during the entire observation period. when Minimets and the research vessel were within 25 km of Wind speed differences between Minimets and R/V each other (Figs. 2d,e), while, unsurprisingly, agreement be- Armstrong when both were within 25 km of each other were 21 21 tween Minimets and R/V Armstrong winds generally wors- generally small, 1 m s at wind speeds of 17 m s (Fig. 3a). ened with increasing distance. Temporal variability as well Differences between ERA5 and R/V Armstrong during those as the magnitude of wind events was generally well captured times were larger, around 2–4m s in the same wind speed by all. However, during individual high wind events, ERA5 range (cf. Fig. 2). Wind speed differences generally increased reanalysis wind speed was significantly lower compared to with wind speed. Over the whole observation period, ERA5 that measured from R/V Armstrong and the Minimets, as and Minimets wind stress differences increased linearly to about 21 21 was evident during wind events on 27 May 2018, 28 May 2018 1m s forwindspeeds upto20 m s , indicating that ERA5 (Fig. 2d), and 13 June 2019 (Fig. 2e). During these wind wind speed was generally lower than that from Minimets. For events, ERA5 wind speed differed from R/V Armstrong by wind speed in excess of 20 m s , ERA5 wind speed differed 21 21 up to 5 m s while wind speed differences between Minimets from Minimet wind speed by up to 3 m s . Note that wind and R/V Armstrong were smaller. speeds in excess of 20 m s were only observed for 0.5% of the To quantify the wind speed differences between Minimets, Minimet data (not shown) and were hence very rare. However, ERA5, and R/V Armstrong, we compute the mean square de- comparable wind speed differences between ERA5 and R/V viation (MSD) between wind speeds sorted into wind speed Armstrong were observed during the cruise periods for wind 21 21 bins of 1 m s . The MSD is further separated into two speeds of 10–20 m s . 2422 J OUR N A L O F P HY SI C A L O C E A N OGR A P HY VOLUME 52 FIG. 3. (a) Wind speed difference between Minimet and R/V Armstrong wind speed measurements during the cruise periods when Minimets were within 25 km of R/V Armstrong (red), between ERA5 and R/V Armstrong wind speed values for times when Minimets were within 25 km from R/V Armstrong (green), between all ERA5 and R/V Armstrong wind speed values during the whole cruise periods (blue), and between all ERA5 and Minimet wind speed values during the whole observation period (yellow). (b) Wind speed standard deviation between wind speed values according to the legend in (a). Error bars in all panels depict the standard error, and wind speed differences and stan- dard deviations are plotted against the respective reference wind speed measurement, either from R/V Armstrong (red, green, blue) or the Minimets (yellow). The standard deviation (calculated as the square root of the center of the cyclone and in a region of less pronounced local variance) between Minimet and R/V Armstrong wind speed gradients, recorded wind stress equal to or larger than the re- was 0.97 m s when both platforms were within 25 km of analysis model. This suggests the possibility that a mismatch each other (Fig. 3b). In comparison, for ERA5 wind speed of the spatial location of maximum winds between the obser- values interpolated onto Minimet positions and coincident vations and the reanalysis model could account for the differ- with Minimet wind measurements within 25 km of R/V Arm- ences between the reanalysis winds and those observed by the strong the standard deviation between Minimet wind speed Minimets. and ERA5 wind speed was 1.22 m s (cf. Fig. 2). Over the To further examine this hypothesis, we can look at wind entire observation period, the standard deviation between speed differences between ERA5 and Minimets in the vicinity Minimet and ERA5 wind speed was 1.26 m s . Both the ob- of wind speed gradients in the ERA5 reanalysis fields. We served differences and standard deviations between Minimets find that differences between the two are larger in the vicinity and R/V Armstrong wind speed suggest that the two measure- of larger local horizontal gradients in the ERA5 wind field ments agreed well, while agreement between ERA5 and R/V (Fig. 5). Further, spatial gradients in the ERA5 fields map Armstrong was poorer, leading us to conclude that the Minimet onto the time domain, and, because these energetic storms winds are reliable. are translating rapidly, a simple gradient analysis addresses We want to further investigate the differences between both spatial and temporal discrepancies. This leads us to con- ERA5 and Minimets at high wind speeds by highlighting an clude that the observed differences between individual events energetic wind event in August 2018 that was sampled by four in the Minimet and ERA5 data could at least in part be ex- of five Minimets (Fig. 4). Peak wind speed for this wind event plained by a difference in spatial structure of individual wind as measured by the Minimets on 17 August 2018 was around events between observations and the reanalysis model, or a 20 m s . The Minimets here are numbered 1–5 for the sake potential lack of small-scale structure in the reanalysis model of simplicity. During this wind event, the center of the cyclone fields. moved eastward just north of the Minimets, with maximum Having established how Minimet and ERA5 winds differ, wind stress passing directly over four of the five the Minimets we want to investigate the impact of these differences on the (Fig. 4a). Minimets 1–3 encountered maximum wind stress at near-inertial wind power input calculations. First of all, near- 1700 UTC 17 August 2018, while Minimet 5 was furthest to inertial wind power input into the ocean can be represented the west and sampled the same wind event slightly earlier. as a time series by calculating the dot product in Eq. (3) or in The largest difference between wind stress estimated from frequency space, through the real part of the complex cross spectrum between the wind stress and the surface velocities Minimet winds and ERA5 winds was found for Minimets 1 and 3. However, these Minimets, being closest to the center (Elipot and Gille 2009a; Flexas et al. 2019), as will be further of the cyclone, were sampling in regions characterized by investigated later. Hence, it makes sense to first look at the large local gradients in the reanalysis wind field. Additionally, frequency-domain representation of the drifter velocities and Minimets 2 and 5, which were sampling furthest from the wind stress from both ERA5 and Minimets. OCTOBER 2022 KL E N Z E T A L . 2423 FIG. 4. (a) Wind stress (colors) and wind vectors (arrows) from ERA5 at 1700 UTC 17 Aug 2018. The locations of five Minimets are shown. (b) Time series of wind stress from Minimet winds (blue) and ERA5 winds (red) for each of the five Minimets shown in (a). The vertical black dashed line corresponds to the time of the peak of the event shown in (a). We estimated rotary spectra for 20-day segments of Minimet and frequencies were normalized by the mean inertial fre- and ERA5 winds, as well as Minimet drifter velocities, by first quency f along the respective 20-day trajectory segments multiplying each 20-day time series by a Slepian taper (Slepian (Fig. 6). Since both the Minimet measurements and ERA5 1978), calculating the Fourier transform, and taking the squared reanalysis had hourly resolution, the Nyquist frequency was modulus of the result. The time-bandwidth product p was cho- v = 12 cpd, far from the average inertial frequency over Ny sen to be p = 2. A total of 138 twenty-day, half-overlapping the entire record of 1.7 cpd. segments were then averaged to produce the spectral estimates Both wind stress spectra were red, but Minimet wind stress showed a significantly shallower spectral slope at high fre- quencies compared to ERA5 (Fig. 6a). Around the inertial frequency the two spectra start to diverge, with spectral en- ergy at the local inertial frequency higher in the Minimet data compared to ERA5 winds by a factor of 1.8 (Fig. 6a, inset). Spectral energy at superinertial frequencies up to the Nyquist frequency was significantly larger in the Minimet dataset com- pared to ERA5. There is no reason to believe that this dif- ference at superinertial frequencies is due to noise in the Minimetwind data. Rather, itis indicativeofthe Minimets capturing high-frequency variability related to wind events happening on time scales smaller than the inertial period that are seemingly not captured in the ERA5 model. Minimet drifter velocities are used to estimate near-inertial wind power input in combination with both Minimet winds and ERA5 reanalysis winds. Their rotary spectra (Fig. 6b) showed a distinct peak around the average inertial frequency f on the clockwise side of the rotary spectrum, as expected for these Northern Hemisphere drifters, while a peak at the semidiurnal tidal frequency dominated the counterclockwise side. A superinertial shoulder on the clockwise side of the FIG. 5. Wind speed difference (blue circles) and standard devia- velocity rotary spectrum coincides with the semidiurnal tidal tion (red diamonds) between Minimet and ERA5 vs the local frequency. This shoulder is an order of magnitude larger than ERA5 wind speed gradient magnitude in the vicinity of the Mini- mets. Error bars depict the standard error. the semidiurnal peak on the counterclockwise side and is 2424 J OUR N A L O F P HY SI C A L O C E A N OGR A P HY VOLUME 52 FIG. 6. (a) Wind stress rotary spectra for Minimet in situ winds (blue) and ERA5 reanalysis winds (red). Solid and dashed lines represent the clockwise and counterclockwise sides of the rotary spectrum, respectively. The inset high- lights the near-inertial regime between 0.7 # v/f # 1.3 on linear axes. (b) Clockwise (dark blue) and counterclock- wise (light blue) rotary spectral components for drifter velocities. The dashed blue line shows the effect of the band- pass applied to the clockwise side of the spectrum and the dashed black line shows the clockwise rotary spectral component of the residual velocities obtained by subtracting the bandpassed velocities. The vertical dashed red line marks the location of the M tidal frequency. indicative of an anticyclonic polarization of the semidiurnal The onset of inertial oscillations with peak amplitudes of tide. A small peak at v = 0.6f coincides with the diurnal tide. 0.28 and 0.2 m s (Figs. 7a,b) coincided with the wind events in To calculate near-inertial wind power input using Eq. (3), August 2018 and October 2019, respectively. Near-inertial wind drifter velocities were bandpassed using a generalized Morse power P during the event in 2018 showed a distinct peak corre- wavelet filter (Lilly 2017), applied only to the anticyclonic sponding to this short-lived wind event, while near-inertial wind side of the drifter velocity rotary spectrum. The parameters of power amplitude remained high for several days due to the pro- the filter were chosen subjectively as g =0.4 and b =12in longed high wind stress following the onset of the event in order to effectively eliminate the observed inertial peak 2019 (Figs. 7e,f). Additionally, peak Minimet wind stress (Fig. 6b, gray dotted line). The advantage of using such a during the 2019 wind event was delayed by 8 h compared to one-sided bandpass is that energy on the cyclonic side, ro- ERA5, as can be seen by zooming into Fig. 7d (not shown). tating opposite the inertial oscillations, is completely elimi- Peak Minimet wind stress coincided with maximum inertial nated and hence does not contaminate the estimates. The current amplitude and consequently lead to larger near- resulting bandpassed drifter velocities captured the spec- inertial wind power input compared to ERA5 (Fig. 7f). tral energy in the near-inertial band around the local iner- Time-averaged estimated near-inertial wind power input tial frequency f on the clockwise side of the rotary spectrum P from Minimet and ERA5 winds over the two periods con- 22 22 (Fig. 6b, blue dotted line), while excluding such energy from the sidered were P =0.4 mW m and P =0.9 mW m MM ERA residual velocities, that is, the original signal minus the band- between 10 and 25 August 2018, and P =1.2 mW m MM pass. It should be noted that our filter does not eliminate the and P =0.4 mW m between 3 and 18 October 2019. ERA semidiurnal tidal signal on the clockwise side of the rotary Minimet-derived energy input into the mixed layer esti- spectrum. mated along the Minimet trajectories over the same time We now want to highlight two wind events that are repre- period was about 2.3 times lower in August 2018, and sentative of when differences between ERA5 and Minimet- about 3.7 times higher in October 2019 (Figs. 7g,h)com- derived near-inertial wind power estimates arise. Example pared to that calculated from ERA5 winds. The impact of time series of bandpassed drifter velocities, Minimet and the two wind events presented here was a net input of en- ERA5 wind stress magnitude, and the associated estimated ergy into the ocean by the winds as captured by both the near-inertial wind power and energy input for two wind Minimet and ERA5 winds, but their magnitudes differ events captured by Minimets in August 2018 and October 2019 substantially. are shown (Fig. 7). In these example time series, Minimet The two example time series above (Fig. 7) highlight one drifters underwent inertial oscillations forced by wind events case in which ERA5 overestimates, and one case in which it around 18 August 2018 and 7 October 2019 (Figs. 7a,b). There underestimates, the time-averaged near-inertial wind power is indication of the semidiurnal tide being present in the band- input relative to that computed using the Minimet winds. passed drifter velocities and of possible leakage from the diur- Over the whole dataset, the time-averaged near-inertial nal tide (cf. Fig. 6b), particularly when the inertial signal is weak wind power input P estimated from Minimet winds was 22 22 and the tides might dominate. While both ERA5 and Minimets 0.26 6 0.06 mW m , 42% higher than the 0.18 6 0.05 mW m seem to capture the general temporal variability well, Minimet estimated from ERA5 winds. For both Minimet and ERA5 wind stress magnitude was lower compared to ERA5 wind stress winds the estimated effect of the wind acting on the ocean sur- in 2018 and higher in 2019 (Figs. 7c,d), and high-frequency face was, unsurprisingly, a net input of power into the ocean variability was more pronounced in the Minimet winds. over the observation period considered here. OCTOBER 2022 KL E N Z E T A L . 2425 0.3 0.3 a) b) 0.2 0.2 0.1 0.1 0 0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 1 1 c) d) Minimet ERA5 0.75 0.75 0.5 0.5 0.25 0.25 0 0 225 225 e) f) 150 150 75 75 0 0 -75 -75 -150 -150 2.5 2.5 g) h) 2.0 2.0 1.5 1.5 1.0 1.0 0.5 0.5 0 0 -0.5 -0.5 08/10 08/12 08/14 08/16 08/18 08/20 08/22 08/24 10/04 10/06 10/08 10/10 10/12 10/14 10/16 10/18 FIG. 7. Example time series for two wind events recorded in (left) August 2018 and (right) October 2019. Shown are (a),(b) bandpassed near-inertial currents from the Minimets, (c),(d) wind stress magnitude calculated from Minimet winds (blue) and ERA5 reanalysis winds (red), (e),(f) near-inertial wind power input, and (g),(h) energy input. To investigate how the wind-speed-dependent differences Minimet-derived estimates were generally larger than those in wind speed presented above affect the near-inertial wind from ERA5 winds (Fig. 8a). Significantly larger differences be- power input estimates, we averaged instantaneous near-inertial tween the two estimates were apparent at wind speeds larger wind power input differences between Minimets and ERA5 in than 20 m s (Fig. 8a). However, these were again associ- 21 21 1m s wind speed bins. At wind speeds below 20 m s these ated with rare events. The bulk of the discrepancy in time- differences were small and largely negative, indicating that averaged near-inertial wind power input from Minimets 60 60 a) b) 40 40 20 20 0 0 -20 -20 -40 -40 -60 -60 Diff ( - ) era mm STD -80 -80 0 5 10 15 20 25 -250 -200 -150 -100 -50 0 50 100 150 200 250 -1 -2 Minimet wind speed [m s ] [mW m ] mm FIG. 8. (a) Instantaneous near-inertial wind power input difference and standard deviation between ERA5 and Minimet- derived near-inertial wind power input estimates vs Minimet wind speed in 1 m s bins. (b) As in (a), but plotted against instantaneous Minimet-derived near-inertial wind power input estimates in 20 mW m bins. Error bars in both panels denote the standard error. -1 -2 -2 -2 -2 Difference, STD [mW m ] u [m s ] [mW m ] dt [kJ m ] | | [N m ] I 2426 J OUR N A L O F P HY SI C A L O C E A N OGR A P HY VOLUME 52 and ERA5 stems from the differences at smaller wind 1.6 speeds, with Minimet-derived time-averaged near-inertial 1.4 wind power input for wind speeds below 20 m s being larger than that estimated from ERA5 winds by 75%. At wind speeds 1.2 larger than 20 m s , time-averaged near-inertial wind -1 power estimated from Minimets was larger by only 10%. f Standard deviation between instantaneous near-inertial 0.8 wind power input estimates increased for wind speeds larger than 10 m s suggesting that, while differences averaged in 21 0.6 wind speed bins below 20 m s were small and generally indi- cating near-inertial wind power input estimated from 0.4 Minimets to be larger than that from ERA5, discrepancies grew more pronounced with increasing wind speed. 0.2 Near-inertial wind power input differences averaged in in- stantaneous Minimet-derived near-inertial wind power input 0 2 4 6 8 10 12 14 bins showed that particularly at the high positive and negative Duration of Event in Hours ends of the distribution, Minimet-derived estimates showed FIG. 9. Relative number of events lasting between 1 and 14 h con- considerably larger near-inertial wind power input magnitudes tributing to the discrepancy between Minimet and ERA5 near- compared to those estimated from ERA5 winds (Fig. 8b). inertial wind power input, defined as P varying by more than one mm ERA5 underestimated both energy input and energy loss, with standard deviation from the record mean and estimates of instanta- the former being more pronounced. Some physical intuition neous wind power input P from Minimet and ERA5 winds differing might help interpret this effect. Strong wind events happening by more than one root-mean-square deviation. The average Coriolis on time scales shorter than an inertial period could lead to large period over the record is shown by the black dashed line. near-inertial wind power input during the first half of an inertial period, when winds and currents are in phase. Differences in importance of accurately capturing energetic wind events oc- wind speed magnitude or the onset of these wind events be- curring on time scales smaller than the local inertial period to tween the Minimet observations and ERA5 data could explain accurately represent near-inertial wind power input, as noted the asymmetry in Fig. 8b. previously by Plueddemann and Farrar (2006). To investigate the short-time-scale hypothesis further, we To investigate the frequency-domain representation of now look at time scales over which differences between near-inertial wind power input into the mixed layer, we ERA5 and Minimet winds contributed to the observed near- calculate complex-valued rotary cross spectra between inertial wind power input discrepancies. We analyzed those wind stress and drifter velocities. Lilly and Elipot [2021, events during which instantaneous near-inertial wind power their Eq. (12)] have shown that under the assumptions of input estimates in the Minimet record differed from the the Ekman problem, the rate of change of the vertically in- record mean by more than one standard deviation tegrated mixed layer kinetic energy due to the wind forc- ing is given by |P 2 P | . (P 2 P ) , (4) mm mm mm mm R{t(t)u (t)} ≡ P(t): (6) and also simultaneously during which the difference be- Here t(t) and u(t) are complex-valued time series of the form tween instantaneous Minimet and ERA5 near-inertial wind z(t)= x(t) 1 iy(t), with x(t) being the zonal and y(t) the merid- power input estimates was greater than one root-mean-square ional components of the drifter velocity or wind stress vector, deviation and (·) denotes the complex conjugate. Equation (6) is thus interpreted as the instantaneous wind power input. We can |P 2P | . (P 2P ) : (5) mm era mm era express the wind stress and drifter velocities as their spectral representations In both criteria N represents the record length. Testing the ivt entire record for (4) and (5) identified those energetic wind t(t) e dT(v) 1 t, 2p 2‘ events that contributed to significant differences in time- averaged near-inertial wind power input estimates from Minimet and ERA5 winds. Their duration is defined as the 1 ivt u(t) e dU(v) 1 y, number of consecutive hours during which both of the 2p 2‘ abovecriteriaweremet. [Lilly and Elipot 2021, their Eqs. (2) and (3)] and substitute The bulk of events that contributed most to the observed them into the definition of the cross covariance: discrepancies happened on time scales much shorter than the local inertial period (Fig. 9), with many individual events R (n)≡ E[t(t)u (t 2 n)]: lasting less than half an inertial period, further highlighting the tu 10 Relative number of events OCTOBER 2022 KL E N Z E T A L . 2427 Together with the definition of the cross spectrum 1.2 Minimet ERA5 S (v)d(v 2 y)dv dy ≡ E[dT(v)dU (y)], tu 2p 0.8 we readily obtain 0.4 ivn R (n) e S (v)dv, tu tu 2p 2‘ which states that the cross covariance and the cross spectrum are a Fourier transform pair (Emery and Thomson 2001). -0.4 Combining this with Eq. (6) we see that the time-averaged 0.5 0.7 1 1.3 1.5 /f wind power input is given by * FIG. 10. Coincident clockwise rotary spectrum between Minimet P≡ E[P(t)] E[R{t(t)u (t)}] R (0) tu drifter velocities and Minimet winds (blue) and between Minimet drifter velocities and ERA5 winds (red). The near-inertial fre- R{S (v)}dv: t u 2p quency band between 0.7f and 1.3f is indicated by the vertical 2‘ 0 0 thin black dotted lines and the vertical dashed red line marks the This implies that the total expected wind power input is the in- location of the M tidal frequency (cf. Fig. 6b). tegral of the real part of the cross spectrum, the cospectrum, be- tween the wind stress and the surface current, and contributions measurements, as suggested by Schmidt et al. (2017) as a due to fluctuations in the vicinity of the local inertial frequency method to verify in situ measurements from platforms such can be obtained by bandpassing either the wind stress or the as the Minimet. While ship winds are not without issues surface currents. (e.g., Landwehr et al. 2020; Smith et al. 1999), they provide We estimated complex-valued rotary cross spectra by first an additional independent wind measurement in the vicinity multiplying each 20-day segment of Minimet wind stress, ERA5 of the Minimets and aid in evaluating the wind speed differ- wind stress, and Minimet drifter velocities with a Slepian taper ences observed between ERA5 and Minimet winds outside (Slepian 1978). The time–bandwidth product p was again cho- of the cruise periods. Comparisons to R/V Armstrong winds sen to be p = 2. Shown here is only the clockwise side of the co- during the cruise periods indicated that Minimet winds dif- spectrum; the counterclockwise side showed little to no spectral fered slightly from ship winds, with both positive and negative energy in the near-inertial band, as is expected since inertial os- differences of up to 1 m s occurring over the observed wind cillations rotate clockwise in the Northern Hemisphere. speed range up to 17 m s . However, considering these meas- Cospectra averaged over all available segments from both urements were separated by up to 25 km, we would not expect Minimet and ERA5 winds showed a local maximum at or around them to agree perfectly. Overall, validation against indepen- the local Coriolis frequency, with maximum cospectral values at dently measured ship winds have shown that Minimets captured v/f = 1 being larger by a factor of 2.4 for those estimated from the general temporal variability and wind speed magnitude Minimet winds compared to those from ERA5 winds (Fig. 10). well. Since wind stress spectra differed by only a factor of 1.8 at v/f =1 Potential problems with platforms like the Minimet mea- (Fig. 6a), the larger discrepancy in the cospectra is indicative of suring winds close to the sea surface include a sheltering effect larger coherence between the Minimet drifter velocities and by wave crests and the periodical submersion of the instru- Minimet winds compared to ERA5 winds (not shown). Between ment leading to anomalous wind speed values. Based on com- 0.7f and 0.95f the ERA5 cospectrum was larger compared to 0 0 parisons to reanalysis winds, Schmidt et al. (2017) suggested the Minimet-derived estimate. At superinertial frequencies both that autonomous platforms like the Minimet measuring winds cospectra were negative, particularly around 1.11f , indicating in challenging conditions characterized by high wind and sea the wind essentially removing energy from the ocean at these states would likely underestimate true wind speed. Generally, frequencies. Negative cospectra at 1.11f for both the Minimet these differences would be expected to manifest as reanalysis and ERA5 winds coincide with the superinertial shoulder on the wind speed being larger than those measured in situ and af- clockwise side of the velocity rotary spectrum and a spectral fected by wave crest sheltering (Large et al. 1995; Renfrew peak on the counterclockwise side (Fig. 6b, red dashed line), et al. 2020; Schmidt et al. 2017). suggesting that the wind acting on the semidiurnal tide could act As mentioned previously, the internal preprocessing of the to remove energy from the ocean. For frequencies v .. f ,co- Minimet winds aims to eliminate anomalous wind values due spectral values were nearly equal to zero for both Minimet and to sheltering effects and instrument submersion. Intercompar- ERA5 estimates. isons between Minimet, R/V Armstrong, and ERA5 winds in- terpolated onto the respective drifter and vessel positions 4. Discussion revealed ERA5 winds were generally smaller than both R/V Novel in situ wind measurements from Lagrangian Minimet Armstrong and Minimet winds over a range of wind speeds, surface drifters were validated against coincident ship wind and wind speed differences were wind speed dependent. The -2 -1 PSD [mW m cpd ] 2428 J OUR N A L O F P HY SI C A L O C E A N OGR A P HY VOLUME 52 observed wind speed differences between Minimets and ERA5 by wind events happening on time scales far smaller than the were consistent during both the cruise periods and when the en- local inertial period that were not captured in the reanalysis tire observation period was considered. The wind speed differ- model and lead to large discrepancies between wind power es- ences observed between ERA5 and Minimet winds were hence timates from Minimet and ERA5 winds. Plueddemann and inconsistent with wind sheltering effects. Similar to the results Farrar (2006) showed that wind events with the largest contri- in this present study, wind-speed-dependent differences be- bution to time-averaged near-inertial wind power input occur tween reanalysis winds and buoy measurements were reported on time scales much shorter than the local inertial period. We by Jones et al. (2016) and Stopa and Cheung (2014),as well as found events lasting less than half the average inertial period between ship winds and ERA5 (Renfrew et al. 2020). in our study contributing significantly to the discrepancies be- The validation of Minimet winds against coincident ship tween ERA5 and Minimet-derived near-inertial wind power winds from R/V Armstrong has shown that the Minimets mea- input estimates. These events comprised about 5% of our da- sure winds reliably. Additionally, the fact that wind speed taset. As such, these results imply that high-resolution wind differences between ERA5 and Minimets increased in the products are necessary to capture small-scale wind events and presence of larger local gradients in the ERA5 reanalysis to adequately resolve near-inertial wind power input. Lower- model wind fields suggest the observed discrepancies be- resolution wind products may fail to accurately represent tween Minimets and ERA5 are likely due to potential limita- these short wind events (e.g., Kilbourne and Girton 2015) and tions of the latter. While ERA5 reanalysis winds represent a hence underestimate wind work. Our results further empha- significant improvement overall compared to previous genera- size the utility of platforms like the Minimet for validating tion ERA reanalysis models (Belmonte Rivas and Stoffelen and possibly correcting regional and global estimates from re- 2019), our results indicate that Minimet winds better repre- analysis products in combination with surface drifter data. sented true wind speed magnitudes, particularly at superinertial Cospectra of Minimet wind stress and drifter velocities frequencies, than ERA5 during the observation period consid- showed significantly increased wind work in the near-inertial ered here. These results highlight the fact that verification of band compared to those estimated from Minimet drifter veloc- gridded wind products through in situ wind measurements, par- ities and ERA5 winds. Negative cospectral values estimated ticularly in poorly sampled regions, remains vital. from Minimet drifter velocities and both Minimet and ERA5 Despite its hourly resolution, near-inertial and particularly winds further suggested that the wind effectively acted to re- superinertial spectral energy on both sides of the ERA5 ro- move energy from the mixed layer at superinertial frequencies, tary wind spectrum was considerably smaller than that esti- reducing overall wind power input in the near-inertial band. mated from hourly Minimet winds. This is in agreement with The negative wind power at superinertial frequencies coincided results by Liu et al. (2019), who found near-inertial wind with a shoulder at the semidiurnal tidal frequency on the clock- power estimates derived from surface drifters in combination wise side of the drifter velocity rotary spectrum (cf. Fig. 6b), with a 6-hourly gridded wind product to be systematically suggesting that the wind acting on the semidiurnal tides acts to smaller than in situ buoy estimates. The authors concluded remove energy from the ocean. As mentioned previously, this these differences were due to reduced near-inertial variance peak was larger on the clockwise side compared to the counter- in the wind product and additionally found no significant im- clockwise side, pointing to an anticyclonic polarization of the provement in their near-inertial wind power estimates when semidiurnal tide (see also Thomson et al. 1998; Poulain and using an hourly reanalysis product. The results from the pre- Centurioni 2015). These results are further in agreement with sent study further suggest that, despite their improved hourly recent results by Flexas et al. (2019), who showed negative resolution, wind variability in the ERA5 reanalysis winds at superinertial wind power occurring at the semidiurnal tidal fre- near-inertial and superinertial frequencies is likely underesti- quencies across a range of wavenumbers at several geographi- mated in and poleward of the study region considered here. cal locations in a global ocean model. The exact mechanisms Near-inertial wind power input was estimated using Mini- underlying this negative wind power at superinertial frequen- met drifter velocities and coincident in situ sea surface winds cies exceed the scope of this paper but certainly warrant fur- measured by Minimet drifters. These were compared to esti- ther investigation. mates from hourly ERA5 reanalysis winds linearly interpo- The present study aimed to validate in situ wind measure- lated onto drifter positions. While both estimates showed a ments from the Minimet drifter against coincident ship meas- net input of energy into the mixed layer by the wind over the urements. Further, in the first study of its kind, it applied study period, the use of in situ winds measured by Minimet these novel wind measurements to the near-inertial wind drifters resulted in a 42% higher time-averaged near-inertial power problem. Comparisons to ship winds during the two wind power input over the observation period relative to that cruise periods indicate that there is a general utility for Mini- estimated using hourly ERA5 reanalysis winds. The approach mets to be deployed in regions other than the one highlighted taken in this study involved using in situ wind measurements here to reliably measure winds along their drifter track. The along the track of the Minimet surface drifter and represents results of this study do not, however, permit to make a gen- the most desirable option to estimate near-inertial wind power eral statement about discrepancies between ERA5 reanalysis input into the mixed layer, obviating the need for modeled winds and in situ wind measurements, and any potential limi- mixed layer velocities and gridded wind products. tations of the former might be regional. The importance of high-frequency wind variability for accu- Wind measurements from Minimet drifters can aid in fur- rate near-inertial wind power estimates was further evidenced ther investigating the skill of reanalysis model winds and OCTOBER 2022 KL E N Z E T A L . 2429 inertial motions. J. Phys. Oceanogr., 31, 2359–2368, https:// potential discrepancies with in situ winds, particularly in sparsely doi.org/10.1175/1520-0485(2001)031,2359:ISGTSD.2.0.CO;2. sampled or difficult to sample regions. Sustaining a Minimet ar- }}, 2003a: Improved global maps 54-year history of wind-work ray over at least one season at a basin scale would lead to an on ocean inertial motions. Geophys. Res. Lett., 30, 1424, interesting verification, or otherwise, of near-inertial wind power https://doi.org/10.1029/2002GL016614. input estimates from reanalysis products and highlight the need }}, 2003b: Redistribution of energy available for ocean mixing for maintaining arrays of Minimets at key locations. Similar to a by long-range propagation of internal waves. Nature, 423, correction method employed by Liu et al. (2019), who used the 159–163, https://doi.org/10.1038/nature01628. ratio of the variances of their wind product and buoy winds aver- }}, 2020: Revisiting near-inertial wind work: Slab models, rela- aged over the near-inertial band as a correction factor, a suitable tive stress and mixed-layer deepening. J. Phys. Oceanogr., 50, approach in our case would be to follow a method similar to 3141–3156, https://doi.org/10.1175/JPO-D-20-0105.1. that outlined in Elipot (2006) and Elipot and Gille (2009b).The Belmonte Rivas, M., and A. Stoffelen, 2019: Characterizing ERA- ratio of ERA5- and Minimet-estimated cospectra could provide Interim and ERA5 surface wind biases using ASCAT. Ocean Sci., 15, 831–852, https://doi.org/10.5194/os-15-831-2019. a frequency-dependent correction for the ERA5-estimated Centurioni, L., A. Horan ´ yi, C. Cardinali, E. Charpentier, and near-inertial wind power input and provide a measure of un- R. Lumpkin, 2017: A global ocean observing system for mea- certainty in the near-inertial range. This could be a way to suring sea level atmospheric pressure: Effects and impacts on utilize high-resolution in situ wind measurements from Minimet numerical weather prediction. Bull. Amer. Meteor. Soc., 98, drifters to estimate the degree to which reanalysis products under- 231–238, https://doi.org/10.1175/BAMS-D-15-00080.1. or overestimate near-inertial wind power input and apply correc- Centurioni, L. R., 2018: Drifter Technology and Impacts for Sea tions when necessary. This approach would require a lot more Surface Temperature, Sea-Level Pressure, and Ocean Circula- data, but regional deployments could aid in quantifying the error tion Studies. Springer International Publishing, 37–57, https:// associated with using gridded wind products and hence improve doi.org/10.1007/978-3-319-66493-4_3. our estimates of near-inertial wind power input globally, impor- D’Asaro, E. A., 1985: The energy flux from the wind to near-inertial tant for accurately depicting abyssal mixing in general circulation motions in the surface mixed layer. J. Phys. Oceanogr., 15, and climate models. 1043–1059, https://doi.org/10.1175/1520-0485(1985)015,1043: TEFFTW.2.0.CO;2. ECMWF, 2019: ERA5 reanalysis (0.25 degree latitude-longitude Acknowledgments. T. Klenz and H. L. Simmons were sup- grid). Research Data Archive at the National Center for Atmo- ported by NSF Grant 1658302 and ONR Grant N000141812386. spheric Research, Computational and Information Systems Lab- L. Centurioni and V. Hormann were supported by ONR Grant oratory, accessed 24 February 2020, https://doi.org/10.5065/ N000141812445. J. M. Lilly and J. J. Early were supported by BH6N-5N20. NSF Grant 1658564. Minimet drifters used in this study were Elipot, S., 2006: Spectral characterization of Ekman velocities in funded by ONR Grant N000141712517 and NOAA Grant the Southern Ocean based on surface drifter trajectories. NA150AR4320071 “The Global Drifter Program.” We thank Ph.D. thesis, University of California, San Diego, 126 pp., the captain and crew of the R/V Neil Armstrong.Drifter veloci- https://escholarship.org/uc/item/1ks7v6wm. ties and spectral estimates were calculated using co-author J. M. }}, and S. T. Gille, 2009a: Ekman layers in the Southern Ocean: Lilly’s MATLAB toolbox, jLab (Lilly 2021). The authors are Spectral models and observations, vertical viscosity and boundary layer depth. Ocean Sci., 5, 115–139, https://doi.org/ grateful to Mark Inall and an anonymous reviewer for their 10.5194/os-5-115-2009. comments and suggestions during the review process, which }},and }}, 2009b: Estimates of wind energy input to the led to an improved paper. Ekman layer in the Southern Ocean from surface drifter data. J. Geophys. Res., 114, C06003, https://doi.org/10. Data availability statement. European Centre for Medium- 1029/2008JC005170. Range Weather Forecasts ERA5 reanalysis wind data used }}, R. Lumpkin, R. C. Perez, J. M. Lilly, J. J. Early, and A. M. during this study are openly available from the Research Sykulski, 2016: A global surface drifter data set at hourly res- Data Archive at the National Center for Atmospheric Re- olution. J. Geophys. Res. Oceans, 121, 2937–2966, https://doi. search, Computational and Information Systems Laboratory org/10.1002/2016JC011716. at https://doi.org/10.5065/BH6N-5N20. All shipboard and Emery, W., and R. Thomson, 2001: Data Analysis Methods in drifter data collected as part of the U.S. Office of Naval Re- Physical Oceanography. Elsevier, 729 pp. search Near-Inertial Shear and Kinetic Energy in the North Flexas, M. M., A. F. Thompson, H. S. Torres, P. Klein, J. T. Farrar, H. Zhang, and D. Menemenlis, 2019: Global estimates Atlantic experiment (NISKINe) is still in the process of being of the energy transfer from the wind to the ocean, with em- organized and archived. All data will be fully released to the phasis on near-inertial oscillations. J. Geophys. Res. Oceans, public when the program formally concludes. NISKINe data 124, 5723–5746, https://doi.org/10.1029/2018JC014453. presented in this study are available from the corresponding Gille, S. T., 2005: Statistical characterization of zonal and meridional author upon reasonable request. ocean wind stress. J. Atmos. Oceanic Technol., 22, 1353–1372, https://doi.org/10.1175/JTECH1789.1. Goni, G. J., and Coauthors, 2017: Autonomous and Lagrangian REFERENCES ocean observations for Atlantic tropical cyclone studies and Alford, M. H., 2001: Internal swell generation: The spatial distri- forecasts. Oceanography, 30,92–103, https://doi.org/10.5670/ bution of energy flux from the wind to mixed layer near- oceanog.2017.227. 2430 J OUR N A L O F P HY SI C A L O C E A N OGR A P HY VOLUME 52 Hansen, D. V., and P.-M. Poulain, 1996: Quality control and Munk, W., and C. Wunsch, 1998: Abyssal recipes II: Energetics of interpolations of WOCE-TOGA drifter data. J. Atmos. tidal and wind mixing. Deep-Sea Res. I, 45, 1977–2010, https:// Oceanic Technol., 13, 900–909, https://doi.org/10.1175/ doi.org/10.1016/S0967-0637(98)00070-3. 1520-0426(1996)013,0900:QCAIOW.2.0.CO;2. Niiler, P., 2001: The world ocean surface circulation. Ocean Circu- Hormann, V., L. R. Centurioni, L. Rainville, C. M. Lee, and L. J. lation and Climate: Observing and Modelling the Global Braasch, 2014: Response of upper ocean currents to Typhoon Ocean, International Geophysics Series, Vol. 77, Academic Fanapi. Geophys. Res. Lett., 41,3995–4003, https://doi.org/10. Press, 193–204, https://doi.org/10.1016/S0074-6142(01)80119-4. 1002/2014GL060317. }}, A. S. Sybrandy, K. Bi, P. M. Poulain, and D. Bitterman, Jayne, S. R., and L. C. St. Laurent, 2001: Parameterizing tidal dissi- 1995: Measurements of the water-following capability of pation over rough topography. Geophys. Res. Lett., 28,811–814, holey-sock and TRISTAR drifters. Deep-Sea Res. I, 42, https://doi.org/10.1029/2000GL012044. 1951–1964, https://doi.org/10.1016/0967-0637(95)00076-3. Jones, R. W.,I.A.Renfrew,A.Orr, B. G. Webber, D. M. Holland, Plueddemann, A. J., and J. T. Farrar, 2006: Observations and and M. A. Lazzara, 2016: Evaluation of four global reanalysis models of the energy flux from the wind to mixed-layer iner- products using in situ observations in the Amundsen Sea em- tial currents. Deep-Sea Res. II, 53,5–30, https://doi.org/10. bayment, Antarctica. J. Geophys. Res. Atmos., 121, 6240–6257, 1016/j.dsr2.2005.10.017. https://doi.org/10.1002/2015JD024680. Pollard, R. T., and R. C. Millard, 1970: Comparison between ob- Kilbourne, B. F., and J. B. Girton, 2015: Quantifying high- served and simulated wind-generated inertial oscillations. frequency wind energy flux into near-inertial motions in the Deep-Sea Res. Oceanogr. Abstr., 17,813–821, https://doi.org/ southeast Pacific. J. Phys. Oceanogr., 45,369–386, https://doi. 10.1016/0011-7471(70)90043-4. org/10.1175/JPO-D-14-0076.1. Poulain, P. M., and L. Centurioni, 2015: Direct measurements Landwehr, S., I. Thurnherr, N. Cassar, M. Gysel-Beer, and of World Ocean tidal currents with surface drifters. J. J. Schmale, 2020: Using global reanalysis data to quantify and Geophys. Res. Oceans, 120,6986–7003, https://doi.org/10. correct airflow distortion bias in shipborne wind speed meas- 1002/2015JC010818. urements. Atmos. Meas. Tech., 13,3487–3506, https://doi.org/ Price, J. F., R. A. Weller, and R. Pinkel, 1986: Diurnal cycling: 10.5194/amt-13-3487-2020. Observations and models of the upper ocean response to di- Large, W. G., and S. Pond, 1981: Open ocean momentum flux urnal heating, cooling, and wind mixing. J. Geophys. Res., 91, measurements in moderate to strong winds. J. Phys. Oceanogr., 8411–8427, https://doi.org/10.1029/JC091iC07p08411. 11, 324–336, https://doi.org/10.1175/1520-0485(1981)011,0324: Rath, W., R. J. Greatbatch, and X. Zhai, 2013: Reduction of near- OOMFMI.2.0.CO;2. inertial energy through the dependence of wind stress on the }}, J. Morzel, and G. B. Crawford, 1995: Accounting for surface ocean-surface velocity. J. Geophys. Res. Oceans, 118,2761–2773, wave distortion of the marine wind profile in low-level Ocean https://doi.org/10.1002/jgrc.20198. Storms wind measurements. J. Phys. Oceanogr., 25, 2959–2971, Renfrew, I. A., and Coauthors, 2020: An evaluation of surface me- https://doi.org/10.1175/1520-0485(1995)025,2959:AFSWDO.2. teorology and fluxes over the Iceland and Greenland Seas in 0.CO;2. ERA5 reanalysis: The impact of sea ice distribution. Quart. J. Lilly, J. M., 2017: Element analysis: A wavelet-based method for Roy. Meteor. Soc., 147, 691–712, https://doi.org/10.1002/qj.3941. analysing time-localized events in noisy time series. Proc. Schmidt, K. M., S. Swart, C. Reason, and S. A. Nicholson, 2017: Roy. Soc., 473,1–28, https://doi.org/10.1098/rspa.2016.0776. Evaluation of satellite and reanalysis wind products with in }}, 2021: jlab: A data analysis package for Matlab, v. 1.7.0. situ wave glider wind observations in the Southern Ocean. J. http://www.jmlilly.net/software. Atmos. Oceanic Technol., 34, 2551–2568, https://doi.org/10. }}, and S. Elipot, 2021: A unifying perspective on transfer func- 1175/JTECH-D-17-0079.1. tion solutions to the unsteady Ekman problem. Fluids, 6, Simmons, H. L., and M. H. Alford, 2012: Simulating the long-range 2311–5521, https://doi.org/10.3390/fluids6020085. swell of internal waves generated by ocean storms. Oceanogra- Liu, Y., Z. Jing, and L. Wu, 2019: Wind power on oceanic near- phy, 25,30–41, https://doi.org/10.5670/oceanog.2012.39. inertial oscillations in the global ocean estimated from sur- Slepian, D., 1978: Prolate spheroidal wave functions, Fourier analysis, face drifters. Geophys. Res. Lett., 46, 2647–2653, https://doi. and uncertainty}V: Thediscretecase. Bell Syst. Tech. J., 57, org/10.1029/2018GL081712. 1371–1430, https://doi.org/10.1002/j.1538-7305.1978.tb02104.x. Lumpkin, R., and M. Pazos, 2007: Measuring surface currents Smith, S. R., M. A. Bourassa, and R. J. Sharp, 1999: Establishing with surface velocity program drifters: The instrument, its more truth in true winds. J. Atmos. Oceanic Technol., 16, data, and some recent results. Lagrangian Analysis and Pre- 939–952, https://doi.org/10.1175/1520-0426(1999)016,0939: diction in Coastal and Ocean Processes, A. Griffa et al., Eds., EMTITW.2.0.CO;2. Cambridge University Press, 39–67, https://doi.org/10.1017/ St. Laurent, L., and H. Simmons, 2006: Estimates of power consumed CBO9780511535901.003. by mixing in the ocean interior. J. Climate, 19, 4877–4890, https:// }}, L. Centurioni, and R. C. Perez, 2016: Full access fulfilling doi.org/10.1175/JCLI3887.1. observing system implementation requirements with the Stopa, J. E., and K. F. Cheung, 2014: Intercomparison of wind global drifter array. J. Atmos. Oceanic Technol., 33, 685–695, and wave data from the ECMWF reanalysis interim and the https://doi.org/10.1175/JTECH-D-15-0255.1. NCEP climate forecast system reanalysis. Ocean Modell., 75, Maximenko, N., R. Lumpkin, and L. Centurioni, 2013: Ocean 65–83, https://doi.org/10.1016/j.ocemod.2013.12.006. surface circulation. Ocean Circulation and Climate: A Thomson, R. E., P. H. Leblond, and A. B. Rabinovich, 1998: 21st Century Perspective, International Geophysics Series, Satellite-tracked drifter measurement of inertial and semidi- Vol. 103, Academic Press, 283–304, https://doi.org/10. urnal currents in the northeast Pacific. J. Geophys. Res., 103, 1016/B978-0-12-391851-2.00012-X. 1039–1052, https://doi.org/10.1029/97JC02374.
Journal of Physical Oceanography – American Meteorological Society
Published: Oct 26, 2022
You can share this free article with as many people as you like with the url below! We hope you enjoy this feature!
Read and print from thousands of top scholarly journals.
Already have an account? Log in
Bookmark this article. You can see your Bookmarks on your DeepDyve Library.
To save an article, log in first, or sign up for a DeepDyve account if you don’t already have one.
Copy and paste the desired citation format or use the link below to download a file formatted for EndNote
Access the full text.
Sign up today, get DeepDyve free for 14 days.
All DeepDyve websites use cookies to improve your online experience. They were placed on your computer when you launched this website. You can change your cookie settings through your browser.