Surface Meteorology and Air–Sea Fluxes at the WHOTS Ocean Reference Station: Variability at Periods up to One Year

Abstract

An eighteen-year record of in situ surface meteorology and computed bulk air–sea fluxes of heat, freshwater, and momentum from an ocean site windward of the Hawaiian Islands is presented. Observations were logged every minute. The one-minute, one-hour, and one-day time series statistics are presented. The daily-averaged time series provide an overview of this trade wind site, with mean wind of 6.8 m s⁻¹ toward the west–southwest, mean ocean heat gain of 23.2 W m⁻², and freshwater loss of 1.2 m yr⁻¹. Energetic variability was found at the higher sampling rates, evidenced by spectral peaks in solar insolation and sea-level pressure and by striking transient signals including short-lived insolation values higher than clear-sky values, short periods with air warmer than the sea surface, and by series of downdrafts of dry air. At longer periods, the presence of moist air accompanying low winds and sunny skies enhanced ocean heating. Winter events with dry air and wind, resulting in large latent and net heat loss, led to ocean cooling. Signals of two hurricanes, Darby and Douglas, were recorded. Normalized by their duration, short-lived events have the potential to make significant contributions to the heat, freshwater, and mechanical energy exchanges.

1. Introduction

The oceanic trade wind regions cover nearly 50% of the ocean surface. The trade winds originate in the mid-latitude high-pressure regions and flow south-westward toward the equator, where there is ascension in the intertropical convergence zones. During their passage across the ocean, there are opportunities for exchanges of heat, freshwater, and momentum between the atmosphere and the ocean. There has long been interest in improving understanding of this coupling between the ocean and atmosphere in the trade wind regions and the role these regions play in weather and climate.

On shorter time scales, the daily cycle of solar insolation and the modulation of incoming radiation by clouds in the trade wind regions have been of interest. The collaborative ASTEX (Atlantic Stratocumulus Transition Experiment) in 1992, for example, sought to better understand cloud dynamics and marine boundary layer processes in the eastern tropical Atlantic [1]. The VOCALS-ReX field study [2] similarly investigated cloud dynamics, boundary layer processes, and the modulation of surface radiation by stratocumulus clouds in the trade wind region of the eastern South Pacific in 2008.

Interest in air–sea coupling in the trade wind regions extends from processes operating on daily to weather time scales to variability at interannual and decadal time scales. Studies have looked at links between trade wind regions and climate variability as well as the realism of models in representing surface meteorology and air–sea exchanges in these regions. Li et al. [3] used wind data from atmospheric reanalyses in the period from 1900 to 2010 to identify strengthening trade winds in the western equatorial Pacific and weakening trade winds in the eastern equatorial Pacific and to investigate the links between these trends and Pacific climate modes. Yang et al. [4] used monthly wind data from the ERA5 reanalysis to examine Pacific trade wind variability between 1950 and 2020 and found strengthening winds in the 1990s, accompanied by decreases in sea-level pressure and sea surface temperature (SST). Simpson et al. [5] compared surface wind stress in the trade wind regions estimated using global climate models to that from reanalyses, which assimilate observations, and found that the climate models yield stronger values than the observed zonal-mean surface winds. Weller et al. [6] compared surface meteorology and air–sea fluxes from three reanalyses to nearly 20-year-long, withheld, and observed surface meteorology and air–sea flux time series at three trade wind sites; they found that the mean stress magnitudes of reanalyses were higher than the observed magnitudes and that the reanalysis multi-year mean net air–sea heat fluxes were between 5 and 30 W m⁻² lower than the observed values.

Motivated by the desire to better understand the surface meteorology and air–sea exchanges of heat, freshwater, and momentum, long-term deployments of well-instrumented surface moorings have been carried out at three trade wind sites, two in the Pacific and one in the Atlantic [6]. The sites are called Ocean Reference Stations. The intent was to build long, climate-quality, withheld time series and to document the accuracy, statistics, and characteristics of these time series. Because the observations are withheld, these time series serve as benchmarks for comparison to models and to remote sensing and hybrid products.

In this paper, we report analyses of the time series from the Ocean Reference Station (ORS) at the Pacific site north of the island of Oahu in the Hawaiian Island chain. Our plan is to write a series of papers to first document the observed surface meteorology and air–sea flux time series at periods up to and including one year; second, to document the observed interannual variability; third, to investigate the role of sub-annual processes and events in contributing to the observed interannual variability; and finally, to present a more detailed comparison between the observations and atmospheric models across all time scales than presented earlier in Weller et al. [6].

The record has a basic sampling rate of once per minute, which began in 2004, and now spans over 20 years. Time series of surface meteorology and computed air–sea fluxes with one-minute, one-hour, and one-day sample rates are produced from the observations. This paper describes and examines the variability at periods ranging from one minute to one year. The duration and high temporal resolution contribute to the uniqueness of the time series. Some long records of surface meteorology from land sites exist and have been presented and discussed (e.g., [7,8,9]). In the past, time series of ocean surface meteorology were collected from ships at the Ocean Weather Stations [10]. Fissel et al. [11], for example, reported on 10 years of Ocean Weather Station P data. Since the end of the Ocean Weather Ship occupations, observations and reporting of surface meteorology and collection of long oceanic time series with collocated surface meteorology and air–sea fluxes have required sequential deployments of surface moorings. The U.S. National Data Buoy Center (NDBC), for example, maintains surface buoys that provide surface meteorological and surface wave observations that are telemetered and assimilated into operational models [12]. The Upper Ocean Processes Group (UOPG) at the Woods Hole Oceanographic Institution (WHOI) and other research groups have developed well-instrumented surface moorings to observe air–sea fluxes of heat, freshwater, and momentum, as well as surface meteorology, and have worked to maintain select observing sites. The time series discussed here is from one such site of sustained surface mooring deployments. This site, north of the island of Oahu, is easy to reach for maintenance cruises sailing out of Pearl Harbor; yet, being positioned 120 km north of the island in a region of prevailing northeast trade winds, it is not in the island’s leeward wake [13] and provides information about the broader local trade wind region.

Reported here is the analysis of data from deployments 1 to 17 (13 August 2004–25 July 2022), spanning 18 years, with a focus on characterizing the observed variability across the span of one minute to one year. Section 2 discusses the observational methods and the quality of the observations. Section 3 provides an overview of the 18-year record using daily-averaged time series to make plots and histograms. To explore variability not captured in daily averages, Section 4 compares the statistics of the one-minute and one-hour time series with those of the one-day time series. Large differences in several variables’ maxima and minima at different averaging periods motivated further discussion of sub-diurnal period variability, including frequency spectra of the one-minute time series and energetic but transient signals, including those responsible for one-minute minima and maxima. In Section 5, the 18-year one-hour time series are used to characterize the mean daily cycle of the surface meteorological and air–sea flux variables. Section 6 examines variability at periods of days to months, with the goal of characterizing the meteorological variability and air–sea exchanges during different regimes. Five events are described: ocean heating under low winds, ocean heat loss in fall, a second period of ocean heat loss in winter, a period of summer heating including Hurricane Darby, and Hurricane Douglas. To document strong seasonal cycles in many variables, the mean annual cycles are presented in Section 7. Section 8 provides a summary of the findings, comparing the cumulative heat, freshwater, and mechanical energy transfers for the events described herein.

2. Observational Methods and Air–Sea Flux Computation

The Upper Ocean Processes Group at WHOI has worked to maintain sustained observations from three ORSs [6], including the site north of Oahu. The goals are to recover a complete, ongoing, accurate time series of surface meteorology to support multidisciplinary studies, to contribute to the collection of oceanic surface meteorology and air–sea fluxes as part of the international OceanSITES component (https://ocean-sites.org, accessed on 24 February 2026) of the Global Ocean Observing System, and to provide an independent basis for assessing model-based and remote-sensing-derived ocean surface variables. To obtain an ongoing record, annual deployments of a fresh surface mooring are made, overlapping the previously deployed mooring. To maximize the likelihood of collecting a complete set of meteorological observations, redundant sensors are deployed. Supporting work has been carried out to quantify the uncertainties in surface meteorological observations and in the derived bulk formula air–sea fluxes (e.g., [14]). The high-quality time series are available through OceanSITES (https://dods.ndbc.noaa.gov/oceansites/, accessed on 24 February 2026) and from WHOI UOPG (https://uop.whoi.edu, accessed on 24 February 2026). The telemetered surface meteorology is withheld from the Global Telecommunications System (GTS) and thus from assimilation by operational models. These observations can thus serve as benchmarks or references for assessing models. We have previously noted [6] that comparisons of WHOTS observations and air–sea fluxes with those from the NCEP2 [15], ERA5 [16], and MERRA2 [17] show significant differences in long-term means and low-pass filtered time series.

2.1. The WHOTS ORS Site and Surface Mooring

The data collection started in August 2004 using a well-instrumented surface mooring maintained at a site 120 km north of Oahu, Hawaii, through cooperation between the WHOI UOP group and colleagues at the University of Hawaii at Manoa. The observations continued with sequential deployments of the WHOTS (Woods Hole Oceanographic Institution Hawaii Ocean Timeseries Station) surface mooring at Station Aloha, north of Oahu. The site is part of the long-running Station ALOHA (~4700 m water depth), where ongoing oceanographic studies began in 1988 (https://hahana.soest.hawaii.edu/stationaloha/, accessed 24 February 2026). The surface mooring also supports ongoing atmospheric CO₂ time series observations by NOAA PMEL [18], which complement the atmospheric CO₂ observations made on Mauna Loa since the 1950s (https://gml.noaa.gov/obop/mlo/, accessed 24 February 2026).

The WHOTS ORS is maintained in coordination with the ongoing Hawaii Ocean Timeseries (HOT) program. Two sites, 12–14 km apart, within the ALOHA site (Figure 1), are used on an alternating basis. A new mooring is deployed before the previous mooring is recovered to facilitate QC (quality control) and quantify consistency. Cruises to the site to recover and deploy the moorings are planned annually, and all but one have been conducted as planned. The WHOTS 16 mooring was deployed for 692 days when there was no cruise in 2020. There were sufficient battery and data storage capacity, so that surface meteorological data were not lost from WHOTS 16, and field comparisons and post-deployment calibrations confirmed that the data quality had not degraded. Appendix A provides a table listing the deployment and recovery times and anchor locations for each deployment.

The 3 m diameter buoy hull supports the mooring line and carries the meteorological instrumentation. Figure 2 shows the buoy tower with meteorological instrumentation. Two ASIMET (Air–Sea Interaction METeorological) systems [19] are deployed, which, with additional stand-alone ASIMET modules, have achieved 99.5% data completeness for the one-minute records of meteorological variables, except rainfall (the self-siphoning catchment rain gauges are susceptible to biofouling from birds). These time series include downward shortwave radiation (DSWR), downward longwave radiation (DLWR), air temperature (T_a), relative (RH) and computed specific humidity (SH), accumulated precipitation (P) and rain rate (Prate), barometric pressure (SLP), and wind speed (WSPD) and direction (WDIR). Sensor heights/depths are included in

Table 1. Sensors used at WHOTS in the ASIMET systems with typical heights. Actual heights above and, for SST, below the sea surface are measured and recorded for each deployment and then used when computing the bulk fluxes. The listed typical heights are close to the actual heights and indicate relative placements, such as the elevation of the radiometers above other sensors. Manufacturers: R.M. Young, Traverse City, MI, USA; Gill Instruments, Lymington, Hampshire, UK; Rotronic, Hauppauge, NY, USA; Eppley, Newport, RI, USA Heise, Stratford, CT, USA; Sea-Bird, Bellevue, WA, USA.

Observable	Sensor Make and Model	Typical Height Above Sea Surface	Notes
Wind (WSPD)	RM Young 5103	3.3 m	Propeller-vane anemometer, stock propeller bearing upgraded
Wind (WDIR)	Gill Instruments WindObserver II Ultrasonic Anemometer	3.3 m	Sonic anemometer used at times to mitigate data loss due to birds
Air temperature/humidity (Ta/RH)	Rotronic MP-101A	2.95 m	Porous Teflon filter and multiplate radiation shield
Incoming shortwave radiation (DSWR)	Eppley Precision Spectral Pyranometer	3.43 m	Case adapted to ASIMET module tubing
Incoming longwave radiation (DLWR)	Eppley Precision Infrared Radiometer	3.43 m	Case adapted to ASIMET module tubing
Barometric pressure (SLP)	Heise DXD	3.0 m	With parallel plate pressure port
Precipitation (P)	RM Young 50202	3.12 m	Self-siphoning rain gauge
Sea surface temperature and salinity (SST, SSS)	Sea-Bird 37 MicroCAT	−0.75 to −0.85 m	Mounted on buoy bridle

. There is no active ventilation of sensors; however, the temperature and humidity sensors are installed within a multiplate solar radiation shield, and the DLWR pyrgeometer is modified to provide thermopile voltage along with body and dome temperature for use with a black-body lab calibration [20], reducing error from solar heating. The post-deployment, calibrated, one-minute records are used to generate hourly, daily, monthly, and annual time series.

2.2. The Quality of ORS Surface Meteorological Observations

Recovery of the deployed mooring is generally preceded by deployment of a new surface mooring with freshly calibrated sensors at the alternate site. On most mooring turn-around cruises, high-quality meteorological sensors have been deployed on the ship to cross-compare buoy and ship-based measurements. These additional shipboard meteorological sensors were deployed by colleagues from the NOAA Physical Sciences Laboratory, Boulder, CO (e.g., [21]). The UOP group also routinely mounted freshly calibrated, stand-alone ASIMET modules on the vessel. For the cross-comparison between the two buoys and the ship, meteorological data spanning one to several days are collected with the ship near the moorings and, if possible, oriented to the wind. At the new buoy, comparisons with shipboard sensors provide an immediate check on the meteorological measurements. If the agreement between the ship and buoy meteorological sensors is poor, ASIMET modules on the buoy are replaced. At the old buoy, the shipboard observations provide an in situ end-of-deployment assessment of the sensors against the chance that they are damaged on recovery and not available for post-calibration. After the in situ intercomparison, the old mooring is recovered. Recovered ASIMET modules are photographed, and data spikes are induced (for example, by putting covers on the shortwave radiometers and placing SST sensors in ice water) to check the instruments’ clocks and refine the time bases. Once recovered, meteorological instruments are shipped back to WHOI for post-calibration.

One-minute time series are quality-controlled after each deployment, and observations from successive deployments are added to construct a merged, long record. In the merging process, differences between simultaneous observations from the two buoys are resolved by comparing the overlapping records with shipboard observations. When new information is available about calibration or processing procedures, the deployment-by-deployment data are reprocessed, and updated merged time series and accompanying metadata are created.

There are no gaps in the surface mooring occupancy of the WHOTS ORS. Sensor issues were restricted to wind and rainfall. Information on wind direction was lost during Hurricane Darby in July 2016 and remained unavailable until the next deployment in 2017. Shortly thereafter, wind speed observations stopped due to damage by sea birds, and replacement anemometers were installed. For the 2016–2017 deployment, the missing hourly wind speed data were filled using ERA5 wind speed by fitting overlapping WHOTS and ERA5 winds. The fit for wind speed was ERA5 WSPD = 0.97 × WHOTS WSPD − 0.5 m s⁻¹; WHOTS hourly east winds were filled using ERA5 WNDE = 1.064 × WHOTS WNDE + 0.3392 m s⁻¹; and WHOTS north winds were filled with ERA5 WNDN = 0.9782 × WHOTS WNDN − 0.2003 m s⁻¹. The later short gap was filled in the same way. Of the time periods and events discussed in this paper, only the discussion of Hurricane Darby and a period of ocean heating in July 2016 makes use of time series from the gap filled with adjusted ERA5 winds. The hourly gap-filling winds were interpolated to one-minute sampling when creating full-length one-minute wind and flux time series. The other sensor issue was related to the rain gauges, which are self-siphoning cylinders that can be clogged by guano. The two ASIMET rain gauges were compared, and for recent deployments, a third piezoelectric rain sensor was deployed to aid in QC. Overall, with these wind sensor issues and excepting rainfall, a 99.5% complete one-minute record has been running from 2004 to the present.

One additional post-recovery processing step was taken in this analysis. Buoy–ship and buoy–buoy comparisons, confirmed with laboratory calibrations, found that the amplifiers used in the first-generation DLWR ASIMET modules had small shifts in gain and offset when the module was powered off and then on again. These early DLWR modules were used in the first three deployments; later deployments used DLWR modules with new, stable amplifiers. The DLWR recorded in the first three deployments was corrected to bring its mean in line with the mean of the fourth deployment, and the amplitude of the swing between clear-sky and cloudy-sky DLWR values was scaled to match that of the fourth deployment.

Over the years, work to assess the accuracy of the ASIMET sensors has been conducted. Colbo and Weller [14] analyzed the differences between the redundant, coincident one-minute samples recorded on ORS buoys and found that differences occur from sensor aging, environmental impacts, and clock drift in the time bases of different sensors. They used probability distributions of the difference time series to obtain 50% and 95% confidence limits for each sensor type, as well as the standard deviation of each of the uncorrected one-minute time series, to quantify measurement accuracy in the one-minute time series. With averaging over several minutes, the accuracies improve greatly.

Table 2. Summary of the WHOTS accuracies for one-minute observations, daily averages, and annual means (after Colbo and Weller [14] and Weller [23]). For context, the mean values for the merged WHOTS 1 to 17 records are shown. Except for wind velocity and rain rate, all values represent means from 1 January 2005 to 31 December 2021. For those variables, calendar years that included gaps were omitted. Wind direction is based on the recorded mean WNDE and WNDN.

Sensor	WHOTS Mean	One-Minute	Daily	Annual
Downward longwave (W m−2) (DLWR)	388.8	7.5	4	4
Downward shortwave (W m−2) (DSWR)	238.1	20	6	5
Relative humidity (%RH) (RH)	75.6	1 3 (low winds)	1 3	1
Air temperature (°C) (Ta)	24.26	0.2 (more in low wind)	0.1	0.1
Barometric pressure (hPa) (SLP)	1017.0	0.3	0.2	0.2
SST (°C)	25.15	0.1	0.1	0.004
Wind speed (m s−1) (WSPD)	6.77	1.5% or 0.1 (more in low wind)	1%, 0.1 (max of these)	1%, 0.1 (max of these)
Wind direction (°) (WDIR)	264.0	6 (more in low wind)	5	5
Rainfall (% undercatchment) (Prate, mm h−1)	0.06	10%	10%	10%

summarizes the uncertainty assessments for the WHOTS surface meteorological records that have been stated and used to evaluate comparisons with models [6]. To provide context for the uncertainty estimates, Table 2 also shows the 18-year means. Further discussion of ASIMET sensor accuracies in the field can be found in Bigorre et al. [22], Weller [23], and Weller et al. [6]. Recently, Schlundt et al. [24] conducted a detailed intercomparison of wind observations from four UOP group surface moorings with scatterometer winds. They reported a root-mean-square difference of 0.56 to 0.76 m s⁻¹ between buoy and scatterometer winds. Part of the work included modeling flow distortion errors on the buoys. In processing the ORS wind record, small differences in wind direction (~2–5°) were noted from the redundant anemometers on the buoy (Figure 2). Modeling of flow distortion around the buoy hull and tower by Schlundt et al. [24] showed that, depending on the angle of the wind relative to the buoy, the anemometers were affected by flow distortion, with error up to 5% of the wind speed in the most affected sensor. The results from Schlundt et al. [24] allow the identification of the anemometer with the least error, but this has not yet been incorporated into the QC processing. In their analysis of the ORS in the North Atlantic, Bigorre and Plueddemann [25] included flow distortion error (4%) and buoy tilt error (4%) to estimate the WSPD error of 8% or 0.4 m s⁻¹. Other sensor errors reported by Bigorre and Plueddemann [25] were close to those in Table 2.

2.3. Air–Sea Flux Computation and Quality Assessment

The quality-controlled surface meteorological time series are used with the COARE (Coupled Ocean Atmosphere Response Experiment) 3.0 bulk formulae (Fairall et al. [26,27]) to compute heat, freshwater, and momentum fluxes. During the computation, the COARE 3.0 algorithm used measured sensor heights/depths, simulated the ocean temperature structure of the warm layer and cool skin, and provided air temperature and humidity at 2 m height and wind at 10 m height. Colbo and Weller’s [14] error propagation equations for the COARE 3.0 bulk formulae (Fairall et al. [26,27]) are used to estimate the uncertainties in the computed fluxes. The estimates of in-the-field sensor uncertainties, combining all error sources, are used in the error propagation equations for the bulk fluxes, and thus the resulting flux uncertainties include the known sources of error. The flux terms comprise net air–sea heat flux (Q_N), net longwave radiation (Q_l), net shortwave radiation (Q_s), latent heat flux (Q_H), sensible heat flux (Q_B), rain heat flux (Q_R), wind stress magnitude (|τ|), east wind stress component (τ_Ε), north wind stress component (τ_N), and wind stress direction (τ_dir). Q_R is computed assuming that rainwater hitting the sea surface is at the dew point temperature.

Table 3. Uncertainties in fluxes at one-minute sampling, averaged over one day, and averaged over one year from an ORS buoy based on Colbo and Weller [14] and Weller [23]. The accuracy of the freshwater flux (E-Prate) was calculated by computing evaporation (E) from latent heat flux and observed rainfall (Prate) at the WHOTS ORS. Uncertainty in Q_R stems from uncertainty in the rain rate. WHOTS 1–17 long-term mean values provide a context for the uncertainties. Except for wind stress and freshwater flux rate, all values represent means from 1 January 2005 to 31 December 2021. The mean stress direction is based on the mean stress components.

Flux	WHOTS Mean	One-Minute	Daily	Annual
Net longwave (QL, W m−2)	−57.0	7.5	2	2
Net shortwave (QS, W m−2)	225.0	10	3	3
Latent (QH, W m−2)	−137.5	5	4	4
Sensible (QB, W m−2)	−7.3	1.5	1.5	1.5
Rain heat flux (QR, W m−2)	−0.2	10%	10%	10%
Net heat flux (QN, W m−2)	23.2	15	8	8
Wind stress magnitude (|τ|, N m−2)	0.0938	0.007	0.007	0.007
East wind stress (τE, N m−2)	−0.0762	0.007	0.007	0.007
North wind stress (τN, N m−2)	−0.0115	0.007	0.007	0.007
Wind stress direction (τdir, °)	261.4	6	5	5
E-P (mm h−1)	−0.14	10%	10%	10%

summarizes the accuracy of the bulk formulae fluxes reported in Weller et al. [6]. Further discussion, including consideration of the use of the COARE algorithm, is available in Bigorre et al. [22] and Weller et al. [6]. When Bigorre and Plueddemann [25] included the additional error in wind speed, their error estimates of daily Q_B, Q_H, and Q_N were 2.5, 12, and 15.5 W m⁻², respectively.

3. An Overview of Surface Meteorology and Air–Sea Fluxes at WHOTS

3.1. Surface Meteorology Time Series—An Overview Using Daily Averages

Daily-averaged time series for WHOTS 1 to 17 provide an overview of the surface meteorological record (Figure 3). Annual cycles are evident in SST, T_a, SH, and DSWR. The annual cycles in DLWR are smaller in amplitude (~10 W m⁻²), with the minima lagging those of DSWR by about 3 months. In winter, the variability of T_a, SLP, SH, Prate, DSWR, and WSPD is higher, but the mean values of SLP, DSWR, and SH are lowest. On average, sea surface temperature is warmer than air temperature, with evident upward spikes, whereas the air temperature time series has downward spikes. The vector wind time series (Figure 3) shows that the variability in WDIR is greater during the winter, and departures from southwesterward flow are associated with synoptic weather events.

The wind rose (Figure 4) shows the dominance of ENE trade winds in the region of the WHOTS ORS. Figure 4 also presents the histograms of the daily surface meteorology averages. While specific humidity (SH), SLP, and DLWR are the most normally distributed, though with negative skewness, SST, and DSWR are bimodal. The bimodal SST distribution stems from its annual sinusoidal variability, with one peak centered near the long-term mean late winter SST minimum of 23.8 °C and the other near the long-term mean late summer maximum SST of 26.8 °C. Although T_a generally tracks SST, its histogram lacks the pronounced winter peak seen as for SST, and time series plots show cool T_a events during winter. The absence of a winter peak is due to the negative skew associated with variable cold air advection during winter.

DSWR also exhibits an annual cycle, but its distribution is not symmetric around the mean value. Being just inside the tropics (±23.46°), the WHOTS ORS experiences the sun passing overhead twice each year, once in late May and again in mid-July. Winter clouds can only reduce the DWSR relative to summer, fair-weather trade wind conditions, resulting in a negative skew in the daily-averaged values. The histogram peaks for DSWR are close to the long-term mean winter and the summer values of 158 W m⁻² and 305 W m⁻², respectively.

The distribution of daily rain rates is dominated by low rain rate events, at <4 mm h⁻¹, though several high rain rate events, up to 13 mm h⁻¹, extend the histogram. Rain is episodic, and the time series of rain rate (Figure 3) was plotted on a logarithmic scale to capture the range of rain rate events. The daily maximum rain rate was associated with Hurricane Darby in July 2016. SLP is negatively skewed by cyclonic systems (winter storms, Kona lows, and tropical cyclones) occasionally passing near WHOTS.

3.2. Air–Sea Flux Time Series—An Overview Using Daily Averages

Daily-averaged time series provide an overview of the air–sea flux record (Figure 5). Gaps in the hourly wind time series were filled with adjusted hourly ERA5 winds, as described earlier, to compute the flux time series. Strong annual cycles are evident in Q_N and Q_S, with ocean heating (Q_N > 0) occurring from spring to fall and ocean cooling in the winter. Smaller annual cycles are seen in Q_H and Q_l. Air–sea fluxes in winter–spring and fall show greater variability and larger amplitude signals. Cloud cover events caused downward spikes in Q_S. Together with strong Q_H events, they contributed to strong, short-lived ocean cooling events seen in daily Q_N but not as evident in the 10-day smoothed Q_N. The 18-year accumulation of rain approached 10 m, while evaporation reached nearly 30 m.

Figure 6 shows the histograms of daily-averaged heat fluxes and wind stress. Q_N and Q_H are strongly negatively skewed, with asymmetric distribution due to a few strong ocean heat loss events. The asymmetry in Q_B also reflects the occurrence of a small number of the strongest sensible heat loss events seen. Wind stress magnitude is strongly positively skewed, with the asymmetry arising from a small number of strong wind events compared to the mean. The histogram of Q_S resembles that of DSWR, as Q_S is computed from DSWR and albedo. The histogram of Q_L is similar to DLWR and close to normal.

4. Sub-Daily Variability in One-Minute Observations

The WHOTS ORS observations were logged at one-minute intervals. The initial focus of analysis was on the one-minute time series for the following reasons: First, during the quality control of observations, outliers stemming from instrumental errors were identified in the one-minute time series. Second, because such rapid sampling is not common, we focused on extreme values in the one-minute data to explore high-frequency variability. Third, the analysis provides other researchers with upper and lower bounds for quality control and establishes observed ranges for comparison with models.

4.1. Statistics of One-Minute Surface Meteorology and Fluxes

The minima and maxima from the one-minute time series were compared with those from hourly and daily-averaged time series (

Table 4. Means, minima, and maxima for full calendar years from 2005 through 2021. Minima and maxima are shown for 1 min, 1 h, and 1 day time series. Height-adjusted air temperature and specific humidity (to 2 m), wind (to 10 m), and ocean skin temperature are also given, along with delta T (skin temperature from the COARE algorithm minus the 2 m T_a). Wind and ocean current directions are toward.

	Mean	Min			Max
		1 Min	1 h	1-Day	1 Min	1 h	1 Day
obs Ta (°C)	24.26	15.89	16.95	18.62	29.54	29.17	27.96
2 m Ta (°C)	24.31	16.14	17.17	18.79	29.65	29.21	27.99
obs SST (°C)	25.15	21.85	21.86	21.97	31.28	30.09	26.68
Tskin (°C)	24.93	21.68	21.72	21.82	30.97	30.16	28.53
delta T (°C)	0.62	−5.65	−2.74	−0.60	7.04	5.88	4.26
SLP (hPa)	1017.0	994.9	995.6	1001.0	1026.8	1026.5	1025.0
obs RH (%RH)	75.6	37.6	42.6	51.6	99.3	98.0	94.2
obs SH (g kg−1)	14.26	5.89	6.78	7.63	20.81	20.77	20.18
2 m SH (g kg−1)	14.41	6.18	6.95	7.84	23.21	20.80	20.23
DSWR (W m−2)	238.1	−2.3	0.0	25.4	1469.5	1134.5	358.8
DLWR (W m−2)	388.8	302.5	312.9	332.7	476.1	453.6	441.1
Prate (mm h−1)	0.061	0.000	0.000	0.000	208.651	70.412	13.333
WSPD (m s−1)	6.77	0.00	0.01	0.05	22.56	19.55	12.61
Weast (m s−1)	−5.42	−22.33	−18.74	−12.30	17.21	15.22	10.40
Wnorth (m s−1)	−0.63	−17.13	−15.35	−10.09	16.51	14.52	9.58
10 m WSPD (m s−1)	7.41	0.03	0.03	0.07	26.08	22.31	14.10
Curspd (m s−1)	0.17	0.00	0.00	0.00	0.99	0.91	0.63
Cureast (m s−1)	−0.03	−0.75	−0.70	−0.56	0.88	0.84	0.61
Curnorth (m s−1)	0.02	−0.87	−0.85	−0.60	0.99	0.91	0.60
QN (W m−2)	23.4	−873.1	−723.8	−413.0	1246.4	900.5	215.9
QH (W m−2)	−137.5	−662.0	−539.7	−414.0	−0.2	−3.2	−14.4
QB (W m−2)	−7.3	−192.0	−137.8	−93.2	95.6	32.4	9.3
QS (W m−2)	225.0	−2.2	0.0	24.0	1388.7	1072.1	339.1
QL (W m−2)	−57.0	−129.7	−119.7	−102.0	−3.6	−5.4	−12.4
QR (W m−2)	−0.20	−938.0	−154.3	−22.6	0.0	0.0	0.0
|τ| (N m−2)	0.094	0.000	0.000	0.000	1.918	1.307	0.490
τE (N m−2)	−0.076	−1.899	−1.259	−0.486	1.033	0.661	0.329
τN (N m−2)	−0.011	−1.326	−1.023	−0.275	0.796	0.556	0.266
salinity (PSS-78)	35.06	31.68	32.74	34.18	35.77	35.64	35.61

). To provide a reference for comparison with models and other products, the table includes T_a and SH adjusted to 2 m above the sea surface, WSPD at 10 m above the sea surface, and skin temperature; these height adjustments were made based on the boundary layer structures embedded in the COARE 3.0 bulk formulae. Additionally, statistics for the ocean surface current (using measurements from the shallowest current meter record as a proxy) and salinity are included to provide insight into the use of wind relative to the surface current and adjustments to the saturation vapor pressure over salt water [28] when using the COARE bulk formulae.

Some meteorological variables SLP, WSPD, RH, SH, DSWR, DLWR) are positive definite, and negative values are not expected. However, in the one-minute DSWR time series, thermal gradients across the sensor can occasionally produce negative values [29]. With WHOTS being near the Tropic of Cancer and considering the impact of the eccentricity of the Earth’s orbit around the sun (±45 W m²), the noontime summer solstice maximum at the WHOTS ORS is ~1360.4 − 45 = 1315 W m². With a typical minimum value of clear-sky atmospheric attenuation of 0.8 [30], this corresponds to a surface irradiance maximum of 1052 W m⁻². The observed one-minute maximum DSWR of 1469.5 W m⁻² is significantly larger than this. DSWR values above the clear-sky reference curve have been reported by Dutton et al. [31] and are associated with reflection and additional diffuse radiation associated with clouds [32,33]. Comparison of WHOTS and the incoming shortwave radiation seen at the Mauna Loa Baseline Surface Radiation Network (BSRN) site shows similar short-lived, high-amplitude upward spikes of DSWR on Mauna Loa.

T_a exhibits a record minimum ~6 °C lower than SST, with stronger negative excursions and greater variability at one-minute than SST. Downward spikes in T_a and generally cooler values have been observed during cool air downdrafts [34,35]. In addition to the cool spikes, transient warm spikes in T_a were observed. Daytime excursions of T_a during low winds can occur due to solar heating of the sensor inside the unventilated multiplate sun shield [36]; no correction was applied to T_a. An investigation of warm spikes in T_a found that, when winds were not low, short-lived periods occurred when T_a had warmed significantly to be above SST. During these transient warm events, the SST minus air temperature difference was large, and the one-minute minimum delta T (skin temperature minus 2 m T_a) reached −5.65 °C, as shown in Table 4. Positive Q_B occurred during the warm air events. Strong, transient events contributing to the maxima and minima are discussed further in Section 4.2.

4.2. Frequency Spectra of One-Minute Time Series

The magnitude differences across the one-minute, one-hour, and one-day minima and maxima reported in Table 4 indicate signals that are not resolved in the daily-averaged data and not apparent in the overview plots of daily time series, and whose amplitudes are not reflected in the daily time series histograms. To investigate repetitive signals on the sub-diurnal scale, the frequency spectra of the one-minute time series from several different full calendar years of the record were computed. All these spectra are red, though some exhibit significant spectral peaks (Figure 7). The DSWR spectra show a strong 24 h solar peak and a series of higher-frequency peaks at harmonics (2, 3, 4, 5, and 6 times 1/24 cph), reflecting the generally half sine-wave nature of DSWR [37]. The SLP spectra have a strong peak for the solar diurnal tide at 0.0833 cph [38,39] and smaller, significant peaks at 0.0417 cph, 0.0126 cph, 0.0167 cph, and 0.0208 cph. The first two peaks correspond to the solar semi-diurnal (S2) and lunar diurnal (M2) atmospheric tides [39,40]. The higher frequency peaks are harmonics of the solar tidal, as reported by He et al. [40]. The small peak at 24 h in the wind spectrum reflects the wind variability associated with the atmospheric tide [41]. SST shows a 24 h peak associated with diurnal heating and smaller peaks at 12 h and 8 h periods noted in the spectrum of DSWR. Wind, T_a, and humidity spectra show small peaks at 24 h. The spectra for SST, RH, and SH are red up to the highest frequency, whereas the slope of the spectrum of SLP is less red above about 1 cph. The spectra for DSWR, DLWR, T_a, and rain rate become redder above ~1 to 5 cph. The spectra of the one-minute flux observations are red like those for surface meteorology. Q_N, Q_s, and Q_l exhibit a slope change around 2 to 5 cph, similarly to T_a, Prate, DSWR, and DLWR. The spectral peaks in DSWR are reflected in Q_s and Q_N, and a small but significant peak at the 1-day period appears in Q_B.

4.3. Strong Transient Signals in One-Minute Surface Meteorology and Air–Sea Fluxes

If the shorter-period minima and maxima resulted from large but infrequently repeated transient signals, the frequency spectra would not have shown significantly higher energy levels associated with such events. Further analysis focused on understanding and characterizing the events contributing to the one-minute and one-hour maxima and minima in Table 4. T_a, for example, exhibits nearly a 3 °C difference between the one-minute and one-day minima and a 2 °C spread across the maxima. The humidity minima exhibit a large spread, as do the minima and maxima of DSWR, Weast, Wnorth, Q_N, Q_B, t_E, and t_N. The Q_S maxima range from 1388.7 W m⁻² (one-minute) to 339.1 W m⁻² (one-day), while the PRATE maxima range from 208.7 mm hr⁻¹ (one-minute) to 13.3 mm. hr⁻¹ (one-day). These large differences in statistics across different averaging periods indicate the presence of strong transient signals resolved by one-minute sampling that would not be apparent in sampled records with longer averaging periods. These strong transient signals are described in this section.

Figure 8 shows observations from midday 6 May 2020 into 7 May (UTC). The one-minute DSWR observations showed transient upward spikes reaching above the expected clear-sky DSWR, resembling those reported by Dutton et al. [31]. The one-minute maxima in DSWR (1469.5 W m⁻²), Q_s (1388.7 W m⁻²), and Q_N (1246.4 W m⁻²) occurred just before 2300 UTC on 6 May 2020. The clear-sky DSWR was estimated following Iqbal [42], adjusting to match the amplitude of observed DSWR on cloud-free days. For the period in Figure 8, WSPD was between 4 and 7.5 m s⁻¹, but even with potential wind-driven mixing, the short-lived high values in DSWR and Q_N were accompanied by increases in SST and higher values of the rate of change in SST. Heat gain, the time integral of the heat flux, during this “High sun” event was 9.5 × 10⁶ J m⁻², five times that on an average day. The times when DSWR exceeded the estimated clear-sky downwelling shortwave contributed 3.2% of the accumulated DSWR on 6–7 May 2020. Heat gain was accompanied by evaporation of −0.065 m m⁻² and moderate mechanical energy transfer by wind stress. Using the tuning of the clear-sky downwelling shortwave for 2020, the full one-minute DSWR time series was searched for data points exceeding the estimated clear-sky shortwave by more than 20 W m⁻² (50 W m⁻²). For two years chosen at random, 2005 and 2011, 6.5% (2.7%) and 10% (4.3%) of the DSWR values exceeded the clear-sky estimates by more than 20 W m⁻² (50 W m⁻²), respectively. The most common occurrence of these high values was in the fall and winter.

At times, wind speeds dropped close to 0. Table 4 shows a minimum of 0.0 m s⁻¹ in one-minute WSPD, and 0.05 m s⁻¹ and 0.01 m s⁻¹ in one-hour and one-day minima, respectively. The histogram of one-day WSPD (Figure 4) shows a small number of low wind days. On cloud-free days (Figure 9), transient daytime periods with low WSPD and Q_H and strong insolation produced upward spikes in SST. The record SST maximum of 31.28 °C was observed very early on 7 September 2017, just after two short periods (~4 h, then less than 1 h) when the anemometer registered no wind. The period covered in Figure 9 is referred to as “Low wind, clear”. Heat gain during the 0.2-day-long period of low wind was strong, at 1.2 × 10⁷ J m⁻², which was more than 2.5 times that on a full, average day, while exhibiting one-fifth of an average day’s evaporation and very low mechanical energy transfer. For contrast, Figure 9 also shows the preceding day, when stronger wind and more latent heat loss resulted in lower amplitude midday warming in SST, despite cloud-free conditions.

Both downward and upward spikes marked the one-minute T_a record. Figure 10 shows several cool-air downdrafts in February 2019. The cool air, with a record minimum of 15.89 °C in T_a just after 1500 UTC on 10 February 2019, combined with upward spikes in WSPD, led to short but strong increases in Q_B, including the record minimum of −192.0 W m⁻² at the same time. Such cool air temperatures did not persist, as shown by the absence of a temperature less than 18 °C in the histogram of daily T_a and by the higher one-hour and one-day minima in Table 4. These transient cool air events are interpreted as being due to cool-air downdrafts [34,35]; the minimum T_a was accompanied by rain in the early morning before dawn, suggestive of convective activity. The period in Figure 10 is labeled as “Downdrafts”.

February 2019 also saw a period of dry air (midday on the 10th to midday on the 11th), in conjunction with WSPDs above 11 m s⁻¹ that spiked to above 16 m s⁻¹, which led to a Q_H record minimum of −662.0 W m⁻² on the 11th (Figure 11, “Dry air event”). SH showed a minimum of approximately 6.25 g kg⁻¹, which was close to the record minimum one-minute SH of 5.89 g kg⁻¹, and was below 9 g kg⁻¹ during the 24 h period. Latent heat loss occurred as winds persisted at and above 10 m s⁻¹ during the dry air. Net longwave radiation loss was elevated, with a record minimum one-minute Q_l of −129.7 W m⁻² observed on the 11th. The combination of large latent heat loss and Q_l losses produced the record minimum Q_N of −873.1 W m⁻². There was a net heat loss of −3.6 × 10⁷ J m⁻², strong evaporative loss of −0.339 m m⁻², and mechanical energy transfer four times that of an average day. The daily SH histogram (Figure 4) and increase in SH minima with averaging period (Table 4) indicate that the driest events generally did not persist for many hours.

Figure 12 illustrates a transient warm event (“TWE”) in T_a on 16 January 2011, showing surface meteorological and air–sea flux time series. As DSWR fell in the late afternoon, WSPD increased above 10 m s⁻¹, toward the SSE. After 5:00 UTC, SLP, SH, and WSPD started to decrease. Just before 6:00 UTC, T_a rose by ~7 °C, and SH fell by ~4 g kg⁻¹; the warm, dry air persisted for ~40 min as the wind shifted to the north and briefly northwest. T_a exceeded SST during and immediately after the event, and the record maximum one-minute Q_B of 95.6 W m⁻² was reached during this warm air event. Ocean heating from positive Q_B was offset by increased latent heat loss associated with the dry air of the event. Transient warm air events have been observed over land at nighttime and are called nocturnal warming events (NWEs) [43,44,45]. To assess the timing and frequency of transient warm events at WHOTS, the one-minute T_a was compared to the 36 h smoothed T_a; 68 events were identified, where the one-minute T_a exceeded the smoothed T_a by 1.5 °C or more for 20 min or longer. TWEs occurred every year from January to April; six years (2004, 2005, 2008, 2009, 2014, and 2021) also exhibited TWEs in October through December. There were no TWEs from May to August. Although the signature of these TWEs was evident in the one-minute statistics (Table 4), their uncommon occurrence, short duration, and offsetting signals in heat fluxes (Q_B vs. Q_H) meant they did not have a significant impact on calculated ocean heating.

The histogram of rain rate (Figure 4) shows a small number of the highest-amplitude events, and the rain rate maxima (Table 4) drop quickly with the averaging period. The strongest rain events are short-lived. The record maximum of 208.7 mm h⁻¹ was associated with Hurricane Lane on 29 August 2018. Another high rain rate event occurred on 1–2 December 2013, as shown in Figure 13 (“Rain event”). The short-lived atmospheric high pressure was accompanied by dense cloud cover and an hour and a half of heavy rain (maximum of 172.9 mm h⁻¹), which contributed to cooling because the falling rain was cooler than the SST. Figure 13 shows a drop in both SST and salinity during the rainfall. The COARE algorithm computes Q_r, the rain heat flux, assuming that the rain temperature is at the dew point. In December 2013, PRATE approached 173 mm h⁻¹, with the accompanying ocean heat loss, Q_r, close to −804 W m⁻². The rainfall resulted in rapid decreases in surface salinity and SST. More typically, Q_r is a minor contribution to net heat flux; Q_r is reported here for these heavy rain events to show its range. Heat loss associated with Q_N was −1.2 × 10⁷ J m⁻², with additional loss of −5.5 × 10⁵ J m⁻² from Q_r. Despite the rain, evaporation dominated at −0.058 m m⁻². Another heavy rain event yielded a one-minute maximum Q_r of −938.0 W m⁻², whereas the one-day maximum was only −22.6 W m⁻² and the record-length mean Q_r was −0.20 W m⁻², or 0.8% of the record-length mean Q_N.

The recorded maximum in WSPD (22.56 m s ⁻¹) occurred during Hurricane Douglas in July 2020. Douglas, the strongest hurricane in the eastern Pacific since Hurricane Lane in 2018, moved west–northwestward, passing north of the Hawaiian Islands on 26–28 July 2020. Douglas passed within 60 nm of Oahu with 80-knot wind speeds late on the 26th local time [46]. Figure 14 illustrates the several-hour-long peak period of Douglas’ impact and the accompanying high-amplitude signals (labeled “Douglas 1 event”). The high winds were accompanied by a period of rain and low SLP. Dry air persisted through the beginning of 28 July, as wind speeds rose, cloud cover reduced Q_s, and Q_N was close to −500 W m⁻² for several hours. The strong heat loss stemmed from the combination of reduced insolation and larger latent heat flux. Evaporation exceeded the hurricane rain, producing a loss of −0.166 m m⁻²; heat loss was −1.6 × 10⁷ J m⁻² from Q_N and −1.3 × 10⁵ from Q_r.

5. The Mean Daily Cycle

Peaks at the 24 h period were seen in the spectra of several meteorological observables, and it is of interest to characterize the mean daily cycle, both to explore variability associated with the daily cycle and to provide researchers with an observation-based exposition. Observed mean daily cycles can be compared, for example, with those obtained from models or hybrid air–sea flux products. Of particular interest here is how the daily cycles of surface meteorology and air–sea fluxes contribute to forcing the ocean.

5.1. Surface Meteorology

The mean daily cycles for each surface meteorological variable were computed from hourly-averaged time series for the full years from 2005 to 2021, inclusive, yielding the average for each hour of the day, as shown in Figure 15. For wind speed and wind components, the years 2016 and 2017 were incomplete and thus excluded. The plots start at 1000 UTC to place the local solar noon at the midpoint in the plots. Over the seasons and years, local solar noon at WHOTS ORS varied from roughly 12:15 to 12:40 Hawaii Standard Time or 22:15 to 22:40 UTC.

SST and T_a exhibit minima in the morning and maxima in the afternoon. The difference between the maximum and minimum is 0.58 °C for T_a and 0.22 °C for SST, which is small compared to the mean diurnal range over land, as surface heating in the ocean is distributed over depth [47]. Relative humidity tracks T_a, and SH reaches a minimum near dawn and a maximum near sunset, as the air warms and moistens during the day, then cools at night. The diurnal range of RH is small, less than 3% RH, compared to that (~30% RH) reported by Betts [47] over land. Rain is more prevalent in the late afternoon and at night when the air is more humid. The solar semi-diurnal tidal signal is evident in SLP, along with a corresponding signal in wind speed, showing more southward flow mid-afternoon.

Mean daily DSWR peaks during local noon at approximately 800 W m⁻², with a daily mean of 238.1 W m⁻². Although the mean of the DLWR daily cycle is large, 388.8 W m⁻², the variation during the day is very small, about 3 W m⁻², with a peak close to local noon. The pyrgeometer used at WHOTS provides body and dome temperatures as well as thermopile voltage in order to to reduce errors from differential heating of the sensor. Tian et al. [48] found a larger-amplitude diurnal variation, ~20 W m⁻², based on observations on land, with diurnal changes in emissivity and lower atmospheric heat storage contributing to this signal, and a maximum in the local afternoon. Burleyson et al. [49] found a decrease in the DLWR daily cycle in the southeast Pacific Ocean of about 50 W m⁻² in the late afternoon, after the DSWR daily peak, and associated that signal with the daily cycle of cloud cover. The coincidence of the DLWR peak at WHOTS ORS with the daily DSWR peak suggests a signal associated with DSWR at midday, but the mean daily DLWR also shows a small increase near local midnight in the absence of solar heating. DLWR at WHOTS, with a daily mean of 388.8 W m⁻², is a strong signal, close to the nighttime value (~380 W m⁻²) in the southeast Pacific [49].

5.2. Air–Sea Fluxes

The mean daily cycles in the heat fluxes and wind stress were computed from hourly-averaged time series in the same fashion as for the surface meteorology, using full calendar years from 2005 to 2021, inclusive, except for stress, which was computed excluding 2016 and 2017. The net heat flux (Figure 16) reached a daily maximum close to 550 W m⁻² at local noon, matching the timing of the maxima in Q_s. The nighttime oceanic heat loss is computed as the total of Q_H, Q_l, and Q_B. Q_H and is close to −200 W m⁻² Q_H is the dominant heat loss, being 2.4 times larger than Q_l and almost 20 times larger than Q_B. Q_H remained between about −136 and −139 W m⁻² through the night and day, with the largest latent heat loss coinciding with the lowest SH in the morning. Net longwave loss (Q_l) was between −56 and −60 W m⁻², with the pattern of change over the day reflecting the signal in DLWR. Q_B was small, at −6 to −10 W m⁻², showing the largest loss when T_a was coolest. The heat loss due to rain falling at the dewpoint temperature, Q_r, was small in the mean daily cycle, close to −0.3 W m⁻² through the day. Strong Q_s shifted the mean daily Q_N cycle from ocean loss to ocean gain from around 1700 UTC to around 0300 UTC (~7 am to ~5 pm local). The daily mean cycle in wind stress was dominated by atmospheric tides, peaking a couple of hours before local midnight. For the mean daily cycle shown in Figure 16, the ocean gains 1.9 × 10⁶ J m⁻² of heat, loses 0.003 m m⁻² of freshwater, and gains 8.1 × 10³ N s m⁻² in mechanical energy per day.

6. Energetic Events—Time Scales of Days to Months

Sustained variability in the “weather band” of roughly 3 to 7 days was not apparent, consistent with the dominant fair-weather Tradewind regime. Hourly time series did not show significantly higher band-averaged spectral energy levels in the ~3- to 7-day band nor at periods of several days to months. However, wavelet scalograms of some 1 min time series show episodic energetic events for time scales longer than 1 day. Below, we describe the surface meteorology and air–sea fluxes for a selection of different such events, including how they impact the accumulation or loss of heat and freshwater in the ocean. Five events are described below: (1) a period of ocean heating under low winds (6–21 October 2009), (2) a period of ocean heat loss (9 November–21 December 2013), (3) a second period of ocean heat loss (9 January–28 February 2010), (4) Hurricane Darby (11–26 July 2016), and (5) Hurricane Douglas (20–30 July 2020).

6.1. Ocean Heating During a Low Wind Period (Low Wind, Moist Air)

We refer to 6–21 October 2009, as “Low wind, moist air”. Notable during this period is the drop in wind speed and stress for two several-day periods, accompanied by increased SST with a diurnal peak near midday (Figure 17). From 8 to 11 October, the air was moist, and the wind speed was low. The period 16–17 October again exhibited low wind, with the air regaining moisture after a dry period late on the 13th. SST and T_a showed midday warming during both events. Latent heat flux was modulated (Figure 17). During 8–11 October, nighttime heat losses were low, allowing the ocean to accumulate heat while wind-driven mixing in the upper ocean was low. In the second low-wind period (16–17 October), latent heat flux was larger due to drier air, contributing to greater nighttime heat loss, though the ocean still accumulated heat over the period. Between these two low wind periods, stronger winds and drier air caused larger latent heat loss and greater net heat loss at night, resulting in a plateau in ocean heat gain. Episodic, several-day-long events marked by low wind and clear skies, like the Low wind, moist air event, occurred at the WHOTS ORS and contributed to greater heating when the air was moister and latent heat loss was lower.

6.2. A Period of Heat Loss During November–December 2012 (Heat Loss, Strong)

Figure 18 shows the surface meteorology and air–sea fluxes during a 40-day period marked by 5- to 15-day variations in SLP, SH, WSPD, and DSWR. The period was accompanied by sustained, moderate ocean heat loss (9 November–19 December). In contrast to the ocean heating event, the air was often drier, with periods of cool air and specific humidity near 10 g kg⁻¹. The 26–29 November period of cool, dry air was accompanied by reduced DSWR. As a result, the lowest nighttime Q_N and strongly reduced midday ocean heating were seen on 27–29 November. During the same period, the strongest Q_H and Q_l values were observed. Drier air and stronger winds were associated with stronger Q_H; the cooler, drier air led to lower DLWR and more negative Q_B. The downward trend in ocean heat loss was interrupted from 23 to 25 November when WSPD fell, and nighttime heat loss decreased. An increase in WSPD and a return to larger nighttime heat loss facilitated the downward trend in heat loss. The net heat loss over the period was −2.6 × 10⁸ J m⁻², accompanied by freshwater loss of −0.034 m m⁻². Within this period, 1–7 December had strong variability. Wind speed and wind stress magnitude dropped close to 0 on 1 December, and Q_l became less variable, and from 4 to 7 December 2012, the air became very moist. Q_H initially fell under low wind, and although the wind eventually increased, the moistening air kept Q_H low. DSWR and Q_s decreased on 4 December, and the decrease was maintained from the 4th to the 6th. This led to a flattening of the slope of the cumulative heat loss curve (Figure 18).

6.3. A Period of Ocean Heat Loss from 9 January to 28 February 2010 (Heat Loss, Moderate)

A second period of prolonged heat loss, referred to as Heat loss, moderate (50 days long, 9 January–28 February 2010), was also examined (Figure 19). The energy in wavelet scalograms (not shown) during this interval was enhanced in the 5- to 15-day time scale. A series of 1- to 5-day pulses of stronger winds was accompanied by cooler, drier air, as seen at the WHOTS ORS every ~6 to 10 days. T_a was often cooler than SST by more than 1 °C and was at times 3 to 4 °C cooler. SH approached the 18-year 1 h minimum of 6.78 g kg⁻¹ by dropping to 7.29 g kg⁻¹ on 19 February. Occasional rain and cloudier conditions occurred between the dry events. This is a classical sequence of frontal passages associated with midlatitude storms passing eastward well north of the islands. During the stronger winds and at times between them, DSWR was less reduced by cloud cover and SH was lower, resulting in lower nighttime net heat loss. Though the 10-day duration was longer than Heat loss, strong, the net heat loss was half, at −1.3 × 10⁸ J m⁻², while the freshwater loss was greater, at −0.034 m m⁻². Compared to Heat loss, strong, the temporal evolution of heat loss showed a series of plateaus with slight heat gain separated by 3–6 days of increased heat loss. Thus, in this winter setting, it was the series of cold and drier air events accompanied by stronger winds that led to heat loss over 50 days; otherwise, heat loss was typically small or absent.

6.4. Summer Heating and Hurricane Darby from 10 to 28 July 2016 (Heat Gain + Darby)

Two hurricanes, Darby and Douglas, passed near the WHOTS ORS. Both occurred during the summer net warming of the ocean (see Section 8), and it was, therefore, of interest to document their impacts on surface meteorology and air–sea fluxes, including how they altered summertime ocean heat gain. After landfall on the island of Hawaii, the remnant tropical depression from Hurricane Darby passed between Oahu and Kauai from 0 to 6 UTC 25 July 2016 [50]. Darby’s passage near WHOTS on 25 July was accompanied by low SLP, rainfall, and short-lived stronger winds (Figure 20). A 15-day period was examined, referred to as “heat gain + Darby”. Rain accumulation reached 0.33 m on 25 July.

An earlier low SLP was observed on 19 July, along with cooler, drier air and steady winds from 10 to 19 July. Both lower SLP periods were accompanied by periods of relatively low T_a and, initially, during these periods, drier air. As a result, nighttime Q_N loss was lower in the period 22–24 July preceding Darby than during Darby. The reduction in ocean heat gain in mid-July stemmed from both the increased latent heat flux prior to Darby and the reduced Q_s and Q_N during Darby. Over the 19-day period, cumulative heat gain was 1.6 × 10⁸ J m⁻², freshwater gain was 0.27 m m⁻², and mechanical energy transfer was 1.8 × 10⁵ N s m⁻².

6.5. Hurricane Douglas from 20 to 30 July 2020 (Douglas 2)

The one-minute time series from 27 to 29 July 2020 (Figure 14) spanned the period when Hurricane Douglas approached the closest (~30 nm) to the WHOTS ORS, passing midway between WHOTS and Kahuku Point, Oahu (https://www.nhc.noaa.gov/data/tcr/EP082020_Douglas.pdf, accessed on 24 February 2026). To assess Douglas’ impact on the summer heating regime at WHOTS, the broader 20–30 July 2020 period, referred to as “Douglas 2”, was examined (Figure 21). Prior to the close approach of Douglas, strong summer insolation had supported ocean net heating from 20 to 27 July despite cooler, drier air, and Q_H and Q_l contributed to a nighttime Q_N of ~−250 W m⁻². At the WHOTS ORS, SLP dropped late in the day on 26 July and reached a low value close to 1005.0 hPa early on the 27th, accompanied by rain (accumulating ~0.015 m), and higher winds on 27 July 2020. Winds reached up to 19.3 m s⁻¹, and the large latent and sensible losses, together with reduced Q_s from cloud cover on the local afternoon of the 26th, drove Q_N to be below −350 W m⁻² for several hours, with a minimum of −438 W m⁻² early on 27 July. The nighttime heat loss on 26 July, combined with the reduced solar gain, started a period of ocean cooling that persisted through the 30th. There was a net heat loss of −8.9 × 10⁶ J m⁻² for the Douglas 2 period. The mechanical energy transfer of 1.6 × 10⁵ N s m⁻² was comparable to that analyzed for Darby. However, compared to Darby, the longer persistence of stronger winds and the extended period of heat loss had more impact on arresting heat gain by the ocean. Despite Douglas’ heavy rainfall, ongoing evaporation resulted in net ocean freshwater loss of −0.046 m m⁻².

7. The Mean Annual Cycles

The mean annual cycles at daily resolution for WHOTS 1 to 17 were computed from the daily-averaged time series. As with the mean daily cycles, documenting the observed mean annual cycles provides a metric for comparisons with models. Bigorre and Plueddemann [25], for example, documented the mean annual cycles at the ORS in the North Atlantic and used these mean annual cycles to gauge how well models perform. With only 17 samples of daily-averaged variables for each day of the year, there is scatter in each day’s record mean, discussed below.

7.1. The Mean Annual Cycles in Surface Meteorology

The mean annual cycles of surface meteorological variables are shown in Figure 22. The greater variability in winter, spring, and fall noted above is reflected in the annual cycles of wind, T_a, SLP, humidity, and rain rate. Wind direction during June–August showed the least variability and was toward ~263°; wind direction in earlier and later periods was more variable. Wind speed varied between 5 and 8 m s⁻¹, with lighter winds in May and again in October. T_a and SST showed distinct annual cycles, with SST-T_a also varying between 0.8 and 1.3 °C. The largest air–sea difference was in late winter when cold air was advected from more northern latitudes. The temperature maxima lagged the annual DSWR maximum by about 3 months. The summer mean DSWR was close to 300 W m⁻², roughly twice the daily mean DSWR in winter. DLWR showed a smaller annual cycle and was ~30 W m⁻² larger during the warmest temperatures in the late summer to fall. Humidity was higher when the temperature was higher. The record maximum one-day-averaged rain rate (24 h average of 13.3 mm h⁻¹) during Douglas yielded a one-day spike in the annual evolution of daily mean rainfall, while late spring through summer generally had less rain than other seasons. SLP peaked in spring and dropped throughout the summer to a low in the fall; a low SLP was also seen in January–February. The prominent high SLP in April was associated with eastward migrating high-pressure systems passing to the north of WHOTS, along with a seasonal maximum in trade winds.

7.2. The Mean Annual Cycle in Air–Sea Fluxes

The mean annual evolution of wind stress reflects the stronger early spring, mid-summer, and fall levels noted in the wind speed and the more westward flow in the summer (Figure 23). Net ocean heating typically occurs from late March to mid-October, with the annual cycle determined largely by that of Q_s. Q_H is close to −140 W m⁻² throughout summer and fall but is more variable and shows larger heat loss in winter and early spring. Q_B is also larger in winter through early spring. Q_l has an annual cycle resembling that of Q_B, though 5 to 10 times larger. Like Q_H, the radiative heat losses are larger in fall to winter and into early spring. Integrating Q_N (Figure 24) shows net ocean heat loss from the first of the year until mid-March, when Q_N becomes positive and after earlyMay cumulative heat stays positive for the rest of the year. In mid-October, the net heat changes sign, and the accumulated heat begins to decrease. The mean annual heat gain is 7.3 × 10⁸ J m⁻². Evaporation dominates precipitation throughout the year, and the mean freshwater loss is 1.2 m m⁻².

8. Discussion and Summary

The WHOTS ORS provides a unique resource by successfully collecting rapidly sampled surface meteorology and providing derived bulk air–sea fluxes. An element of its uniqueness is that the observations are not telemetered to be available for assimilation into models. There are a number of surface moorings in the tropics (e.g., the Tropical Atmosphere Ocean array [51]), but their observations are assimilated consistent with their mission of supporting real-time modeling and prediction. In addition, sensor redundancy and work at sea and in the laboratory to assess meteorological and flux accuracies of the WHOI UOP ORS in the field have provided a near-complete record with documented uncertainty. WHOTS was deployed along with two other ORS [6] to support our effort to collect high-quality time series in the trade wind regions that can be used as independent observations to assess models as well as to investigate the temporal variability of surface meteorology and air–sea fluxes at these locations. This paper focuses on characterizing the temporal variability up to one-year at the WHOTS ORS, drawing on the data collected from 2004 to 2022.

Eighteen-year means indicate that the WHOTS ORS is in a trade wind regime, with mean winds close to 7 m s⁻¹ toward the west–southwest, occasional heavy rainfall averaging 0.06 mm h⁻¹, sustained evaporation averaging −0.08 mm h⁻¹, and a mean ocean heat flux of ~23 W m⁻². The heat gain results from a small excess of solar heating over combined latent, sensible, and net longwave loss. The availability of the 18-year record with basic sampling of once per minute reflects the success of the effort to collect long time series of surface meteorology and computed bulk air–sea fluxes at the WHOTS ORS. One remaining challenge is the protection of anemometers from damage from wind, waves, or birds, but sensor redundancy kept the wind loss down to the loss of wind speed after Hurricane Douglas and during one other short interval. Wind gaps were filled using ERA5 winds adjusted to fit WHOTS ORS-observed winds. Birds also contribute to clogging of self-siphoning rain gauges. The 18-year record captured the major rain events, but freshwater gain from less intense rain events may at times have been under-reported. Work is being carried out on quality control software to add corrections to wind speed and direction for flow distortion [24] and to build improved rainfall time series using data from the additional impact rain gauge now deployed, combined with data from both siphon rain gauges. Improved protection from birds is being added to the rain gauges.

A number of the results presented here provide metrics for assessing the realism of the surface meteorology and air–sea fluxes produced by models, by remote sensing, and by hybrids based on combinations of models and remote sensing. One can compare histograms, mean daily and annual cycles, spectra, and statistics from these other sources with those from WHOTS 1 to 17. It is important to note that observations from the gap filled with ERA5 winds were only used in the Hurricane Darby + heat gain event (Figure 20, Section 6.4). None of the extreme values identified here came from the gap-filled period. Thus, statistics and results such as the histograms, daily and annual cycles, and illustrative events and regimes can indeed serve as independent metrics to assess models. In this work, the availability of rapid sampling and statistics from time series with different sampling periods motivated part of the analysis.

Comparison of statistics from the 18-year one-minute, one-hour, and one-day sampled time series pointed to the presence of strong signals resolved by sub-diurnal sampling. Spectral peaks were, as expected, found associated with the diurnal cycle of insolation and solar heating, and semi-diurnal and diurnal atmospheric tides. However, some transient signals were striking and brought challenges to preliminary quality control efforts for the one-minute time series. First, a guideline sometimes is that DSWR does not typically exceed the clear-sky value of downwelling shortwave radiation estimated at the sea surface, and second, that air temperature is rarely much warmer than SST. Some recorded DSWR values did peak above the estimated clear-sky DSWR, including a record maximum of 1469.5 W m⁻². Based on comparison with land-based DSWR observations and other reports, these high DSWR values are accepted as valid. Including the high DSWR values added 3.2% to the time-integrated DSWR during the 6–7 May 2020 event in Section 4.3. Second, it is often found that SST is warmer than T_a and that Q_B has a negative sign associated with ocean heat loss. Yet, in the 18-year record, the difference between one-minute ocean skin temperature and 2 m T_a ranged from −5.65° C to 7.04° C. In initial quality control processing, values of air warmer than 2 °C above SST were flagged. Checking the redundant sensors during warm air events confirmed replication of the T_a observations. Searching the literature on terrestrial surface meteorology located descriptions of warm air events seen over land, and quality control processing was revised not to exclude warm T_a data points. The time series thus include transient warm events (TWEs), such as those presented in Figure 12a,b, and consider them to be valid events. Sixty-eight such events were found; as a result, further study of these TWEs is planned. Retaining these high DSWR values contributed to peak values in Q_N, and the TWEs led to unusual ocean heating from Q_B.

Refining the quality control process yielded one-minute time series that were used to characterize sub-diurnal events with signatures in air–sea heat, freshwater, and momentum transfer. On a cloudy day (Figure 7, “High sun”), incoming solar radiation showed transient spikes above clear-sky values, but a cloud-free “Low wind” (Figure 8) event was much more effective at transferring heat into the ocean. Events with dry air, such as “Downdrafts” (Figure 9) and “Dry air” (Figure 10), showed strong heat and freshwater losses and were accompanied by higher wind stress. Strong “Rain events” (Figure 12) were associated with ocean freshening and cooling, yet with ongoing evaporation, the net result over the period examined was freshwater loss. The two-day period of Hurricane Douglas’ closest passage (“Douglas 1”, Figure 13) illustrated a strong mechanical forcing event. The desire to better understand how the atmosphere and ocean couple in the trade wind regions had motivated the deployment of the ORS. The finding of the considerable strength of heat, freshwater, and mechanical energy transfers associated with these sub-diurnal events spawns questions about whether or not the lack of resolving these signals introduces error in means of air–sea fluxes sampled at slower sampling rates.

Additional variability on periods of several to tens of days, with episodic occurrences of elevated energy levels, was also investigated. Wavelet scalograms showed periods of energetic variability at these time scales, identifying a number of such periods that also contribute to air–sea coupling. Several of these events were investigated, including events characterized by ocean cooling, ocean heating, and close passage by hurricanes. A period of low wind (“Low Wind, moist air”, Figure 17) in October was marked by sustained ocean heat gain, but a period from 9 November to 19 December was marked by sustained heat loss in 2021 (Figure 18, “Heat loss, strong”). Figure 19 shows a period of heat loss later in winter (“Heat loss, moderate”, 9 January–28 February). In summer 2016 (Figure 20, “Heat gain + Darby”), despite the passage of Darby, heat gain persisted. The longer “Douglas 2” (Figure 21) period illustrated how the hurricane ended a seven-day period of sustained ocean heat gain. The passage of Hurricane Darby (Figure 20) provided a contrast, where moister air reduced latent heat loss, and consequently, summer heat gain was not greatly impacted. Overall, these events illustrate the role of near-surface humidity together with wind speed in controlling the magnitude of Q_H as the primary offset to heat gain from Q_s. The role of the presence of moist air and its resulting reduction in Q_H and increase in DLWR, with accompanying reduction in Q_B increase awareness of the need for accurate observations of RH, SH, and T_a and realistic representations of them in models.

Normalizing total gains by their respective durations provides (Figure 25) a way to gauge impacts. The potential impact of energetic, short-lived events is striking. In regard to heating the ocean, clear skies and low wind on a spring/summer/fall day—especially in the presence of moist air that reduces latent heat loss—provide strong ocean heating. Windy, cloudy conditions with associated dry, cool air lead to large ocean heat losses. Yet, annual and longer period mean air–sea fluxes and accumulations remain small. This raises the question of how the combination of transient strong events, well-defined daily cycles, longer periods of sustained but more modest fluxes, and well-defined annual cycles yields modest heat gain, persistent evaporation, and moderate wind stress over the 18-year record. From this, follow-on questions arise: how does the population of events across all sub-annual periods vary year to year? Is this variation the reason for the interannual variability in low-pass filtered SST, T_a, Q_N, |τ|, and other variables at the WHOTS ORS?

The goal of this paper is to describe the 18-year WHOTS ORS time series and to characterize events within it at periods ranging from one minute to one year. This paper provides a foundation for further analyses. Next steps are to document and characterize interannual variability observed at the WHOTS ORS and place these findings in the context of interannual variability and trends in the central, tropical North Pacific. After that, comparisons between WHOTS ORS surface meteorology and air–sea fluxes and models are planned. Questions to be addressed include the following: Do models resolve sub-diurnal and multi-day events seen at WHOTS ORS? Does the failure to reproduce these events lead to errors in model mean meteorology and fluxes? How well do models replicate SH and T_a, and do biases in humidity and temperature lead to errors in fluxes?