1. Introduction
The Tropical Atmosphere Ocean (TAO) array is a system of moored ocean buoys conceived and implemented by the NOAA/Pacific Marine Environmental Laboratory (PMEL) between 1985 and 1994 to monitor the variability in the tropical Pacific Ocean and to better investigate the dynamics of El Niño–Southern Oscillation (ENSO) events. The management and quality control (QC) of the array were transferred to the NOAA/National Weather Service/National Data Buoy Center (NDBC) in the period 2005–07 (McPhaden et al. 1998, 2023).
The system currently includes 55 buoys regularly spaced in the tropical Pacific between 165°E–95°W and 8°S–8°N, but the exact number of buoys has varied in time from 20 to 70, with some experiencing long interruptions of data. The sensors on board the buoys observe several variables that characterize the atmosphere/ocean system, such as air and sea temperature, ocean heat content, salinity, humidity, and ocean surface wind speed at anemometer heights of 3–5 m. These variables are used for climate variability studies, the derivation of ocean surface fluxes, data assimilation into numerical weather prediction (NWP) models, and as ground truth for the verification of other observing systems. Additional information on the TAO buoy array is provided in appendix A.
In situ ocean surface wind measurements from buoys are very valuable as they are routinely used to verify the calibration and monitor the stability of decades-long time series for several satellite wind datasets. Additionally, in a global statistical sense, the calibration of satellite ocean surface winds is tied to buoy winds (see the section on satellite calibration/validation and appendix B for more details).
During routine analyses using satellite ocean surface wind data from microwave (MW) radiometers and scatterometers processed at Remote Sensing System (RSS), we detected a sudden significant bias between the satellite winds and those from the TAO array, starting in mid-2020 until the present. The comparison between these two types of observations described here was made by matching a satellite-based ocean observation within 25 km of a buoy location and within a short time interval of 30 min, where both the satellite-based wind observations and buoy anemometer measurements are converted to a common reference height of 10 m (see appendix B on satellite matchups with buoys). We refer to these matched satellite–buoy observations as “collocations.”
Over the past two decades, these satellite wind datasets have shown remarkable consistency with the global buoy winds, particularly with the TAO buoys, with differences typically within 0.2 m s−1 at the monthly global scale and less than 0.1 m s−1 for the annual averages. However, the collocated TAO buoys compared to several cross-calibrated satellite wind datasets started showing a significant difference of 0.5–0.8 m s−1 in mid-2020, as illustrated in Fig. 1.
(top) Smoothed (2-week) time series of the wind speed difference between all buoys in the TAO array and collocated rain-free wind speeds from the satellite MW radiometers WindSat, GMI, and AMSR2, and scatterometers ASCAT-A, ASCAT-B, and ASCAT-C, for the duration of each mission during the period 2003–23. (bottom) Time series of the number of satellite/buoy matchups within 2-week periods.
Citation: Bulletin of the American Meteorological Society 106, 2; 10.1175/BAMS-D-24-0072.1
After extensive investigation of both the satellite data and the code developed for the satellite monitoring analyses (discussed in the next section), we determined that the drifts did not arise from anomalies in the satellites’ calibration, code errors, or changes in the conversion of buoy winds to a reference height of 10 m. We suspect the drift to be related to calibration changes in the buoys in the TAO array, as the separation between satellite and buoy winds coincides with the period when many (44) of these buoys were serviced and a new type of anemometer was installed, primarily between March and September 2020. Following this finding in 2022, we promptly informed scientists at the NOAA/National Weather Service (NWS)/NDBC, who are currently investigating the changes. This is discussed in the final section of this manuscript.
The stability of the integrated satellite–in situ system is vital for international ocean observation programs such as the Global Ocean Observing System (GOOS, https://goosocean.org/; Moltmann et al. 2019) and the Tropical Pacific Observing System (TPOS, https://tropicalpacific.org/; Smith et al. 2019) which both aim at providing coordinated and sustained ocean observing networks that include moorings and satellites.
Therefore, the analyses presented here aim to inform the research community about the changes detected in the TAO buoy winds to avoid serious consequences such as misinterpreting them as climate signals or propagating them into either derived datasets or the calibration/verification of other observing systems (e.g., satellites).
An in-depth description of the changes observed in the TAO buoy winds is provided in the next sections, including an analysis of the wind speed dependency of the change. We discuss the potential impact of these changes on a wide community of buoy data users. Finally, we present the conclusions and suggest a course of action to minimize the impact of such spurious changes in buoy calibration to the user community and to help resolve the inconsistencies among different datasets.
2. Observed calibration change in TAO buoy winds
a. Calibration shift in wind speed and components.
After first observing the apparent drift of the TAO buoy winds versus satellite wind retrievals, we investigated the change in detail to identify the possible cause. Figure 2 illustrates a recent time series (2018–23) of the TAO buoy winds and collocations with satellite winds from the GMI radiometer on board NASA’s Global Precipitation Measurement (GPM) mission, considered to be the most accurate and stable space-based radiometer due to its on-orbit calibration maneuvers (Wentz and Draper 2016; Wentz et al. 2024). The time series in Fig. 2 shows that satellite and buoy winds started diverging in the period between March and September 2020. According to deployment information given in NDBC’s TAO buoy data files, this is the period when many (44/55) of the TAO buoys were serviced and their anemometers were upgraded. Among these, three buoys had limited observations, and they were not used in our analysis. The remaining 11 buoys in the TAO array were serviced in the following 2 years. The uncertainties in GMI and TAO wind observations were estimated over the preservice period shown in Fig. 2 (i.e., 1 January 2018 to each buoy’s service date) with a triple collocation method (Stoffelen 1998) that uses NCEP reanalysis surface winds. The resulting uncertainties are 0.39, 0.55, and 0.77 m s−1, respectively, for GMI, TAO, and NCEP. Note that the uncertainty over a sustained period (months to years) is much smaller as it is averaged over thousands of observations.
Averaged time series for the buoy wind speed collocated with GMI rain-free wind speeds within 30 min and 25 km, for the period 2018–23. Here, the buoy winds are converted to a reference height of 10 m, and daily time series for individual buoys are averaged over all the buoys in the TAO array. The black line refers to the daily variability of the buoys, while the colored lines refer to a 30-day running mean, for easier visualization of the difference between GMI (dark orange) and buoy (blue) wind speed time series. In mid-2020, the GMI and buoy average wind speed time series started to diverge, showing a bias of about 0.5–0.8 m s−1. Similar biases were observed even when comparing satellite winds to individual TAO buoys and for different satellites.
Citation: Bulletin of the American Meteorological Society 106, 2; 10.1175/BAMS-D-24-0072.1
While the separation between buoy and satellite winds appears more like a drift rather than an immediate jump in the time series averaging all TAO buoys, this is actually a by-product of the cumulative effect of the changes as they were introduced to individual buoys at different times. Figure 3 displays the sudden changes for three sample buoys, highlighting the day they were serviced. Table C1 (in appendix C) lists the buoy IDs, location, and dates they were first serviced since the beginning of 2020. As of 2024, this jump is present in all the TAO buoy winds. This is illustrated in Fig. 4 which presents the shift of the buoy winds versus a reference satellite (GMI) for each individual buoy of the TAO array for two periods, pre- and postservice. The change displays a slight regional pattern. Also shown (in addition to being listed in Table C1) is the percentage change at each location compared to the preservice period. Note that when expressed as a percentage, which accounts for differences in the mean wind at each location, the changes have a much more uniform regional distribution, suggesting a wind speed dependency, which is explored in the next section. The exception is buoy 52003 at 5°N, 165°E (light blue circle), which showed a large negative bias after being serviced in 2020 that appeared to be unrelated to the bias described in this paper. The buoy stopped transmitting several months after its 2020 servicing. When the buoy was serviced again in August 2022, it began retransmitting data and showed a positive bias consistent with that of the other TAO buoys.
Time series of buoy minus GMI collocated winds for three sample TAO buoys (WMO IDs 32318, 51302, and 51019), for the period 2018–23, with their respective service days highlighted in yellow. The time series are shown as 2-week averages for better visualization.
Citation: Bulletin of the American Meteorological Society 106, 2; 10.1175/BAMS-D-24-0072.1
Map of the mean bias between each TAO buoy and collocated GMI wind speeds, but separated into two periods: (top) preservice and (middle) postservice, and (bottom) the corresponding percentage change at each location relative to the preservice mean wind speed. The preservice values are calculated from 1 Jan 2018 to the date each buoy was serviced while the postservice values were calculated from the buoy service date through the end of 2023. Note that every buoy had a separate service day (see Table C1, appendix C).
Citation: Bulletin of the American Meteorological Society 106, 2; 10.1175/BAMS-D-24-0072.1
b. Wind speed dependency of the calibration change.
An unaccounted change of 0.5–0.8 m s−1 is significant for buoy wind data users looking at climate variability signals or using them as ground truth for validating other datasets. If the change is not a constant shift (but, rather, proportional to wind speed), the consequences are more severe, as the change becomes larger at higher winds. To quantify a potential wind speed dependency, we analyzed a two-dimensional histogram of the collocations between all the TAO buoys and the satellite wind speeds from January 2018 until December 2023 and partitioned the collocations into two groups, either before each buoy’s service date or after it. These are illustrated in Fig. 5 for the TAO versus GMI wind speed collocations. To quantify the wind speed dependency, we calculated a regression line of the collocated data for the two periods. The coefficients and their uncertainties are listed in Table 1. Also listed in the table are the average bias and mean GMI winds for the two periods. Based on these statistics, the correspondent average percentage change for the TAO buoys compared to GMI in the post- versus preservice period is 9.3%. The change in the slope of the regression indicates that the change is speed dependent, but there is also a speed-independent offset. Note that while the standard deviation associated with a single observation is on the order of 0.7 m s−1, the uncertainty associated with the mean of 25 000 observations is much smaller, less than 0.01 m s−1. A change of about 0.6 m s−1 in the mean is therefore indicative of a systematic postservice bias.
(top) Two-dimensional histograms for TAO buoy winds vs collocated GMI winds, partitioned into two periods, (left) before and (right) after they were serviced, during 2018–23, as defined in Fig. 4. The color bars on the right indicate the number of observations per wind speed bin of width = 0.2 m s−1. The black diagonal lines represent the 1-to-1 (unbiased) line. The green dashed lines in all panels depict the linear regression calculated for each set of data (coefficients listed in Table 1). (bottom) For the same cases as the top panels, (left) before and (right) after service was performed, the mean wind speed bias is indicated by the blue line. The wind speed histograms (number of matchups) are depicted in yellow. These are displayed as a function of the average (buoy + GMI) wind speed, to avoid skewing the statistics over the tails of the distribution.
Citation: Bulletin of the American Meteorological Society 106, 2; 10.1175/BAMS-D-24-0072.1
Statistics and regression coefficients of TAO buoy vs satellite collocations, separated into two periods, preservice and postservice, for comparisons with GMI wind speeds and ASCAT-B wind components U, V, and wind speed W. The percentage change is calculated as the difference between the post- minus preservice bias divided by the preservice satellite mean.
There seems to be no significant directional dependence on these changes, as they appear in both the zonal and meridional wind components. Similar to Fig. 5, in Fig. 6, we illustrate the two-dimensional histograms of the TAO versus the wind components U and V from the ASCAT-B scatterometer, for the two periods, pre- and postservice, along with their respective regression lines, with the coefficients listed in Table 1. The separation of regression lines pre- and postservice is significant, as well as the percent change in both components, which is estimated to be about 9% and 13%, respectively, for the U and V components. This difference is not large enough to suggest a directional dependency of the changes.
Two-dimensional histograms for TAO buoy vs collocated ASCAT-B wind components (top) U and (bottom) V, partitioned into two periods, (left) before and (right) after they were serviced, similar to Fig. 5. The color bars on the right indicate the number of observations per wind speed bin of width = 0.2 m s−1. The black diagonal lines represent the 1-to-1 (unbiased) line. The green dashed lines represent the regression lines for each period (coefficients in Table 1).
Citation: Bulletin of the American Meteorological Society 106, 2; 10.1175/BAMS-D-24-0072.1
3. Potential impact of buoy wind changes to observing systems and research community
A systematic 9% change in buoy wind speeds is very significant, and it has the potential to affect a large community of researchers. Here, we briefly describe some of the most impactful ramifications.
a. Air–sea surface fluxes.
Buoy measurements are often used for verification of the air–sea flux products (Praveen Kumar et al. 2012; Jiang et al. 2005; Cronin et al. 2006; Wen et al. 2019; Chiodi et al. 2019; Chiodi and Harrison 2020). The uncertainties or errors in ocean surface wind observations can impact estimates of all types of air–sea fluxes, including carbon fluxes. Some air–sea flux datasets extend over several decades, and their long-term stability is very important. They include the OAFlux V3 (Yu and Jin 2014), IFREMER (Bentamy et al. 2013), HOAPS (Andersson et al. 2011), Japanese Ocean Flux Data Sets with Use of Remote Sensing Observations (J-OFURO) (Kubota et al. 2002), and SEAFLUX (Curry et al. 2004).
Therefore, the air–sea flux datasets derived or calibrated using the buoy data would be significantly affected by a systematic 9% increase in buoy wind speeds, whether directly when derived from the TAO buoys or indirectly when derived from satellite, numerical model, or assimilated datasets which have routinely been calibrated to be consistent with buoy wind observations.
b. Satellite calibration/validation.
Satellite sensors are able to retrieve ocean surface winds with indirect measurements using the relationship between ocean surface wind and either the scattering of a MW radar signal (scatterometers) or the MW thermal emission (radiometers) from a wind-roughened surface (Martin 2014). The MW retrieval algorithms are fully or partially empirical, meaning that the relationship between the observed quantities (backscatter and brightness temperature) is statistically tied to a large multiyear global reference set of anemometer winds from all the buoy arrays: These are often used for algorithm training (wind retrieval calibration) or for final verification after the retrievals are performed (wind retrieval validation). More details are provided in appendix B. In addition to the calibration of individual satellite sensors, ocean moored buoys play a very important role in verifying the stability of the satellite sensors used for wind climate data records (Wentz et al. 2017).
Therefore, any discontinuity or change in the calibration of a buoy array indirectly impacts all satellite wind datasets, which would need to be realigned to the buoys.
c. Tropical climate variability analysis.
The TAO array has been specifically designed for monitoring and studying the dynamics of tropical variability at a range of time scales varying from short weather events (tropical cyclones, wind bursts), to intraseasonal (Madden and Julian oscillation, tropical instability waves), to interannual (ENSO) and decadal scales.
A detailed review of the role of TAO buoys within the tropical observing system is presented in McPhaden et al. (2023), while Chen et al. (2018) focus specifically on the TAO observations for ENSO studies. There are a multitude of studies on tropical variability utilizing the TAO buoys for analyses or for verification of the variability inferred from satellite/model data, at all temporal scales. Some examples include the following: Chelton et al. (2001), Deng et al. (2009), Moum et al. (2013), Amaya et al. (2015), Chiodi and Harrison (2017a,b), Hasan et al. (2022), and Hackert et al. (2023).
A change of 0.5–0.8 m s−1 in the observed ocean surface wind speed would be significant for most of the studies at time scales longer than a few days. For example, as illustrated in Fig. 7, yearly regional wind anomalies in the tropical Pacific are typically within 0.8 m s−1, except in El Niño/La Niña years where they can exceed 2 m s−1 in the central or eastern equatorial Pacific. Wind trend signals at the global level are less than 0.1 m s−1 decade−1, but due to El Niño signals, trends in the tropical Pacific can reach 0.2 m s−1 decade−1, still much smaller than the recent systematic change experienced by the TAO buoy winds.
Global maps of (top) the average wind anomaly in 2021, during La Niña, (middle) the standard deviation of the annual mean wind anomaly in the tropical Pacific, and (bottom) trend for 1988–2023 from RSS satellite radiometers. These analyses were performed in the context of the AMS State of the Climate reports (Dunn et al. 2024) and use 1991–2020 as a baseline period for the anomalies.
Citation: Bulletin of the American Meteorological Society 106, 2; 10.1175/BAMS-D-24-0072.1
d. Numerical weather prediction models.
The TAO buoy array is equipped with a data transmission system that allows its observed meteorological parameters to be transmitted in real time via satellite and be promptly ingested into operational analysis and NWP models, such as the widely used ECMWF (De Rosnay et al. 2022) or NCEP GFS models (Kleist 2023). While the vast majority of assimilated observations that constrain the model fields are from satellites, and each assimilated dataset is assigned a different weight and bias correction, the assimilation of “inconsistently calibrated” buoy data for the periods before and after service was performed could have some minor regional impact and introduce small errors in the models.
e. Merged wind analysis datasets.
In addition to numerical models, data assimilation can also be used to combine observations and numerical model data to represent a geophysical field without the spatial/temporal gaps that would result from exclusively observational datasets. In observationally based wind datasets that directly assimilate buoys, the impact of a change would likely be seen mostly in the derived fields, like wind curl and divergence, close to the buoy locations (McGregor et al. 2017). In other satellite-based merged winds that do not directly use the buoys as input, but rather use them for verification [i.e., the latest version V3.1 of the Cross-Calibrated Multi-Platform (CCMP), wind vector analysis; Mears et al. 2022], the change in the consistency between satellites and buoys has a similar impact to what has been discussed for the satellite wind retrievals.
4. Discussion and conclusions
A large temporal instability in the TAO buoy wind time series has been detected using multiple satellite wind datasets to verify and identify its source. The change, introduced at each buoy’s first servicing since 2020, appears as a drift in the collective TAO array, as the buoys were individually serviced at different times. As of 2024, all the 55 TAO buoy anemometers under NOAA’s management have been serviced and display this change. From the analyses presented here, a sudden and systematic change is estimated to be about +9% relative to collocated satellite winds.
As mentioned in the introductory section, in 2022 we promptly informed NDBC scientists about our findings and engaged with them over the following 2 years. They confirmed the bias estimate and pointed to a change in the wind tunnel laboratory used for testing and resulting changes in the calibration of the buoys’ new anemometers as the possible source (Steve DiNapoli, email communications, June 2022–February 2024). The sudden and systematic step change affects all TAO buoys serviced in the period between mid-2020 and early 2023. At this time, it is not clear whether this change is meant to be permanent, and a more in-depth investigation is underway at NDBC.
It is important to note that the calibration of the other tropical buoy arrays Research Moored Array for African–Asian–Australian Monsoon Analysis and Prediction (RAMA) (McPhaden et al. 2009) and Prediction and Research Moored Array in the Tropical Atlantic (PIRATA) (Bourlès et al. 2019) is managed by NOAA PMEL, and it is independent from the TAO arrays, which are managed by NDBC. RAMA and PIRATA winds do not seem to have experienced any significant changes in the past few years. Despite the higher temporal variability due to the smaller and declining number of buoys (Sprintall et al. 2024), these two arrays have been stable in the last decade compared to satellite winds, as illustrated in Fig. 8, which compares buoy winds to those retrieved by GMI.
(top) Smoothed (3-week) time series of the wind speed difference between all buoys in the TAO (blue), PIRATA (yellow), and RAMA (green) arrays and collocated rain-free winds from the GMI radiometer since the beginning of the satellite mission, during the period 2014–23. (bottom) Time series of the number of satellite/buoy matchups within 3-week periods. Similar results (not shown) were obtained using the other satellites displayed in Fig. 1.
Citation: Bulletin of the American Meteorological Society 106, 2; 10.1175/BAMS-D-24-0072.1
If the currently detected change in TAO winds becomes permanent, it poses a dilemma: whether to realign the calibration of all other global buoy arrays to the new TAO calibration standard, and whether/how to recalibrate all the satellite wind retrievals and all the ocean surface flux datasets that depend on the ocean surface wind speeds to be consistent with this new standard. This would require a monumental effort to update datasets classified as climate data records (CDRs) that span multiple decades and for which a 9% “instrumental discontinuity” is unsustainable.
It is important to note that it is not uncommon for remote observing systems (especially satellite sensors) to experience calibration instabilities, drifts, and spurious signals. The typical course of action, after an initial investigation to confirm the changes and study the potential causes, is to address these changes in public statements, notifying the users whether a correction will be provided, and later document, in detail, any updated data release. Satellite data processing centers value any notification about possible errors/biases detected by the users and perform continuous monitoring of the satellite data quality. For example, when biases emerged in some RSS satellite data, they were documented on the public web page, or in peer-reviewed publications, and corrected data were released after a brief time [e.g., for a calibration jump detected in the ASCAT-A scatterometers winds (Ricciardulli and Manaster 2021) or a spurious drift in the AMSR2 geophysical variables (see https://remss.com/missions/amsr/)]. Similarly, in the early 2000s, while PMEL was responsible for operating and performing quality control of TAO array data, a bias in wind direction due to a problem with the anemometers was unexpectedly discovered, affecting about a decade’s worth of data. The error was documented, corrected where possible, and widely advertised (see https://www.pmel.noaa.gov/gtmba/sensor-specifications, and Freitag et al. 2001). With this article, we intend to reach a wide community of buoy data users to inform them about the change in TAO buoy winds and to stimulate a discussion on a possible course of action to maintain consistency among numerous datasets, data assimilation, and forecast models that use the buoys as a reference standard to constrain the model fields, or for verification of the stability of satellite retrievals.
The stability of the integrated in situ and satellite observing system is vital for the coordination of international programs such as GOOS and TPOS. We suggest that any relevant change in the buoys’ calibration for a specific array needs to be coordinated at the international level with other observing systems, whether they are other buoy arrays, satellite sensors, or newer in situ systems such as the Saildrones (Gentemann et al. 2020; Foltz et al. 2022; Zhang et al. 2023).
This is a time-sensitive problem: New satellite sensors for wind measurements have been recently launched [e.g., the scatterometer OceanSat scatterometer (OSCAT) on Ocean Satellite (OceanSat-3), December 2022; the radiometer Microwave Imager (MWI) on Weather Satellite Follow-On Microwave (WSF-M), April 2024] or are planned in the near future (e.g., the JAXA radiometer AMSR-3 in 2025 and the scatterometer SCA and radiometer MWI on MetOp-SG B in 2026). At this time, for continuity with previous satellite missions, we advise withholding observations from the TAO buoy winds for satellite verification and only use the other tropical (PIRATA, RAMA) and extratropical buoy arrays (e.g., other buoys maintained by NDBC or the Canadian Meteorological Service) until the mismatch with the other buoy arrays is resolved and a buoy calibration standard is finalized.
Although we have no insight into how exactly the calibration of the buoy anemometers was performed, the change in the buoy-reported wind speeds after servicing suggests an inconsistency with the calibration process. One way forward is for NDBC and the buoy anemometer manufacturer to repeat the laboratory testing in a common controlled environment, likely in a wind tunnel, with both the old and new sensors, in order to establish a consistent set of calibration parameters to link the two together. Additional factors that could play a role, such as potential differences in flow distortion, or in the response time of the old versus new instruments, should be investigated.
Over the long term, it would be wise to consider introducing an additional source of in situ calibration ground truth for surface ocean observations, such as from Saildrones. The Saildrones carry sonic anemometers that are calibrated in wind tunnels, independently from the buoy anemometers. Some of the Saildrone observations have already been cross-calibrated or verified in proximity of the buoy locations in the tropical Atlantic and many missions have been successfully deployed in other ocean basins; they are planned to continue and possibly expand in the future. According to a recent article (available at https://www.saildrone.com/news/ndbc-director-moored-uncrewed-systems-resolve-ocean-observation-gaps), NDBC Director Dr. William Burnett highlighted an urgent need for “a system of systems to resolve ocean observation gaps,” such as in the event of buoy “outages.” One example is a recent Saildrone long-term mission funded by NOAA Office of Marine and Aviation Operations (OMAO), to replace a buoy off the coast of Half Moon Bay, California, beginning in September 2023 and still ongoing in 2024 (see https://www.saildrone.com/news/weather-buoy-swapped-ocean-drone-protect-seafloor). Using Saildrones in conjunction with the buoys, for shorter cross-calibration periods, would add confidence in the accuracy of both observational methods and provide robust, continuous monitoring and more stability to the integrated in situ and satellite observing system.
It is also important to note the benefit of having multiple cross-calibrated satellite wind datasets in order to detect changes in consistency with the buoy observations, such as the one described here for the TAO array, and to rule out any specific satellite as being responsible for observed discrepancies. This reinforces the value of worldwide efforts from several groups in creating stable and accurate satellite climate data records that are able to detect changes as little as 0.1 m s−1 (or lower) when averaged in space and time (Wentz et al. 2017; Ricciardulli and Manaster 2021). These types of accurate satellite observations can also be used to set up a publicly available near-real-time monitoring of the stability of each buoy, accessible to users worldwide, or to provide corrections if needed.
Regarding the use of the TAO buoy winds for climate variability analyses, it is very important to be aware of the “spurious” systematic change of about 0.5–0.8 m s−1 in the buoy wind data record starting from mid-2020 until now. The magnitude of this change is very significant, considering that it is substantially larger than the trend and average yearly anomaly signals in the eastern and central tropical Pacific, where the array is located. Whether to discard or adjust the buoys’ wind data postchange is a possibility currently in the hands of the users. Ideally, the agencies distributing the buoy data should provide a public statement and guidance to the users regarding this change. A future of sustained and publicly accessible monitoring of the buoy stability is also recommended.
Acknowledgments.
This analysis was funded by the NASA Ocean Vector Wind Science Team (OVWST) Contract 80NSSC23K0984. We are extremely grateful to the BAMS editor Michael McPhaden (NOAA/PMEL) and three anonymous reviewers for their in-depth comments, additional information on the TAO array, and suggestions for improvement of this manuscript. We are also very grateful to Steve DiNapoli (NOAA/NDBC) for written communications over the period 2022–24 regarding the possible source of the observed buoy wind changes and to Cheyenne Steinberger (NOAA, Global Ocean Monitoring and Observing Program, and TPOS Program Manager) for additional information and help reaching out to NDBC scientists in 2024. We would also like to thank Tony Lee (NASA JPL), Carl Mears (RSS), Steve Swadley (NRL), Mark Bourassa (FSU), Ad Stoffelen (KNMI), Mohamed Dahoui (ECMWF), and several members of the OVWST for their insights and discussions on the detected instability and its impact to users.
Data availability statement.
All radiometer and scatterometer wind datasets used in this study are publicly available at https://www.remss.com/. The TAO buoy data were originally collected in netCDF4 format and were made freely available by NOAA/NDBC via their OceanSITES Global Data Assembly Center at https://dods.ndbc.noaa.gov/thredds/catalog/oceansites/catalog.html.
APPENDIX A The TAO Buoy Array
The TAO array is a series of 55 moored buoys located in the equatorial Pacific (8°S–8°N; 165°–265°E). Initially conceived by NOAA PMEL (NOAA Research) as a project to help monitor climate variability related to ENSO, the implementation by PMEL began in 1985 in collaboration with international partners during the Tropical Ocean and Global Atmosphere (TOGA) program and continued until its completion in 1994 (McPhaden et al. 1998, 2023). The moored buoys in the TAO array help provide researchers with insight into both short- and long-term climate and ENSO-related phenomena thanks to their ability to measure variables describing the air–sea interaction, such as wind speed and direction, sea surface temperature (SST), relative humidity, and air temperature.
The buoys are maintained and periodically updated through service trips from ships belonging to a variety of stewards. Since the TAO array’s inception, the TAO buoys have undergone many changes as new sensor and mooring technologies improve and as previous buoy moorings deprecate. One major change was the management shift from PMEL to NDBC NWS over the period 2005–07, during which NDBC became responsible for operations and quality control of the TAO buoy array.
Some recent changes have occurred during the NDBC “TAO Refresh campaign” (Bernard et al. 2008) during the mid-2000s to early 2010s, which significantly upgraded buoy mooring technology (McPhaden et al. 2023): NDBC’s moorings replaced the PMEL ATLAS moorings, but they were designed to be functionally equivalent and compatible with the previous ATLAS moorings to ensure historical continuity in the data. An additional campaign, “TAO Recap,” started in 2023 (and therefore, it is unrelated to the wind change described in this manuscript) and is planned to be completed by 2027 (see https://www.climate.gov/news-features/climate-tech/landmark-buoys-across-tropical-pacific-ocean-get-makeover/). The latter came to fruition, in part, due to the decommissioning of the Ka’Imimoana NOAA ship in 2012, which had serviced the TAO buoys since 1996 (McPhaden et al. 2023). As a result, a significant dip in available TAO buoy data occurred as buoy moorings/payloads suffered without regular maintenance trips. This decrease in data underscored the need for consistent in situ observations of the tropical Pacific and led to the TPOS 2020 Project (Cravatte et al. 2016; Kessler et al. 2019, 2021). From 2014 to 2020, this project evaluated the state of the observing systems in the equatorial Pacific and aimed to identify observational gaps and shortcomings in that region, providing some guidelines for redesigning the array configuration in a way that would benefit from the diverse remote and in situ observing capabilities that had become available since the array began in 1985. The TAO Recap is a direct result of TPOS and will involve repositioning some of the buoys and upgrading instruments to improve or add observations of the ocean mixed layer, barometric pressure, precipitation, solar radiation, and surface or deeper currents.
Ideally, these new or upgraded sensors and moorings will help to advance the TAO array and further its goal of providing consistent and accurate information related to climate variability and air–sea interaction in terms of sampling, measurement accuracy, and equipment durability, which reduces the need for frequent service trips to the buoys.
APPENDIX B Satellite Matchups with Buoys
Accurate satellite observations of ocean surface winds have been continuously monitored at the global scale since 1987 with radiometers, which are sensitive to the wind-induced emissivity of the ocean surface, and with scatterometers, which are space-based radars that measure the backscatter from the wind-roughened ocean surface. Typically, the radiative transfer models (RTMs) or backscatter empirical models needed for the satellite wind retrievals are developed using many years of ground truth observations, including from the global moored buoy arrays, to train the models in a global statistical sense. Once the models are developed, the satellite wind retrievals are not tuned to match individual buoys; rather, the buoys are only used for statistical verification of the stability of satellite data. One critical phase during which tuning might occur is the first few months after the launch of new satellite sensors, mostly as a result of sensor calibration issues. In any case, it is important to see the satellite–in situ surface winds observations as an integrated calibration system for optimal accuracy and consistency among all platforms.
Satellite datasets are available from different sensors, which are carefully “cross-calibrated” during overlapping periods in order to achieve high accuracy. Some long-standing radiometer missions include the U.S. Department of Defense SSM/I and SSMIS series, NASA TMI and GMI, JAXA AMSR-E and AMSR-2, and the U.S. Navy WindSat, while long-lasting scatterometer missions include NASA QuikSCAT and the EUMETSAT ASCAT series (A, B, C). Some examples of how ocean buoys are used for calibration/validation of common satellite wind datasets are provided in Stoffelen (1998), Freilich and Dunbar (1999), Mears et al. (2001), Ebuchi et al. (2002), Kelly et al. (2005), Verspeek et al. (2010), Stiles and Dunbar (2010), Ricciardulli and Wentz (2015), Xing et al. (2016), Yang and Zhang (2019), Ricciardulli and Manaster (2021), Ye et al. (2021), and Wentz et al. (2024).
In the past decade, these satellite sensors have reached a level of accuracy and consistency that is required to be considered “climate data records” and are able to identify climate signals on the order of 0.1 m s−1, when averaged over large regional and temporal scales (Wentz et al. 2017). This level of accuracy is also beneficial to global observing systems, as any inconsistency within a group of multiple simultaneous observations can help identify and isolate satellite drifts or spurious changes in other observational systems, including in situ.
To compare satellite wind observations with in situ point measurements such as those from moored buoys, it is important to take into account the differences in spatial/temporal scales and in elevation above the surface for the two types of measurements. Here, we describe how the satellite matchups (or “collocations”) with buoys were achieved for our analysis.
The satellite wind retrievals are derived from emissivity or backscatter measurements measured over a satellite footprint, which has a spatial resolution on the order of 25 km, and they are reported as 10-m equivalent neutral winds, i.e., the wind speed at a height of 10 m in a neutrally stable atmosphere (Liu and Tang 1996; Mears et al. 2001).
The buoys measure wind speed and direction with anemometers mounted at heights of 3–5 m above the ocean surface. The wind observations are distributed as 10-min and hourly averages. For comparison with satellite winds, we use hourly averaged buoy winds, as they are more representative of what the sensors measure within a satellite footprint of 25 km. For example, an average wind speed of 7 m s−1 covers about 25 km in an hour. This is explained in detail in Schlundt et al. (2020), which also addresses flow distortion effects for open ocean buoys. While these effects can reach 5% for individual cases, when averaged over many observations, wind directions, and buoys, the flow distortion contributions mostly cancel out and cannot explain the magnitude of the TAO bias described in this manuscript.
The distributed buoy hourly wind speeds undergo additional quality checks at RSS for any missing, repeated, or out-of-bound data. The wind data at the anemometer height are then adjusted to the standard 10-m height using a logarithmic vertical profile (Mears et al. 2001).
For every satellite sensor, and every orbit, we search all the observations from each buoy of the TAO array that fall within a 25-km grid box and within 30 min of a satellite pass. We call these matchups “satellite collocations with buoys.” The collocations used in this study are performed separately for each sensor and for the duration of each satellite mission from the early 2000s to the end of 2023. The collocations are separated by buoy WMO ID and satellite sensor and are stored as time series. Note that for each day, some buoys fall in the gaps of the satellite orbital swaths (i.e., the satellite does not pass over the buoy location). In those cases, while the buoy recorded wind observations, there are no valid collocations because the satellite did not have an orbital pass over that specific buoy at that time.
APPENDIX C TAO Buoy Locations, Recent Service Dates, and Observed Wind Change
Table C1 shows the TAO buoys considered in our analysis, including statistics before and after servicing.
List of all TAO buoy nominal locations, first service dates since the beginning of 2020, average preservice buoy 10-m wind speed, wind speed biases when compared to the GMI radiometer matchups within the period January 2018 until December 2023, separated in two periods: before and after each buoy was serviced. The last column lists the percent change at each buoy location compared to the average collocated GMI wind speed at the same location. Note that while the nominal longitude and latitudes are listed here for simplicity, in the analyses the actual buoy location was used, as it can vary over time typically within 1°, but often much less. The following buoys are not listed here and they were not used in the analysis because of very limited data postservice: 43001, 51310, and 52006.
References
Amaya, D. J., S.-P. Xie, A. J. Miller, and M. J. McPhaden, 2015: Seasonality of tropical Pacific decadal trends associated with the 21st century global warming hiatus. J. Geophys. Res. Oceans, 120, 6782–6798, https://doi.org/10.1002/2015JC010906.
Andersson, A., C. Klepp, K. Fennig, S. Bakan, H. Grassl, and J. Schulz, 2011: Evaluation of HOAPS-3 ocean surface freshwater flux components. J. Appl. Meteor. Climatol., 50, 379–398, https://doi.org/10.1175/2010JAMC2341.1.
Bentamy, A., S. A. Grodsky, K. Katsaros, A. M. Mestas-Nuñez, B. Blanke, and F. Desbiolles, 2013: Improvement in air–sea flux estimates derived from satellite observations. Int. J. Remote Sens., 34, 5243–5261, https://doi.org/10.1080/01431161.2013.787502.
Bernard, L., K. Kern, J. Zhou, and C.-C. Teng, 2008: Refreshed data system for tropical atmosphere ocean (TAO) buoy array. OCEANS 2008 – MTS/IEEE Kobe Techno-Ocean, Kobe, Japan, Institute of Electrical and Electronics Engineers, 1–7, https://doi.org/10.1109/OCEANSKOBE.2008.4531028.
Bourlès, B., and Coauthors, 2019: PIRATA: A sustained observing system for tropical Atlantic climate research and forecasting. Earth Space Sci., 6, 577–616, https://doi.org/10.1029/2018EA000428.
Chelton, D. B., and Coauthors, 2001: Observations of coupling between surface wind stress and sea surface temperature in the eastern tropical Pacific. J. Climate, 14, 1479–1498, https://doi.org/10.1175/1520-0442(2001)014<1479:OOCBSW>2.0.CO;2.
Chen, D., N. Smith, and W. Kessler, 2018: The evolving ENSO observing system. Natl. Sci. Rev., 5, 805–807, https://doi.org/10.1093/nsr/nwy137.
Chiodi, A. M., and D. E. Harrison, 2017a: Observed El Niño SSTA development and the effects of easterly and westerly wind events in 2014/15. J. Climate, 30, 1505–1519, https://doi.org/10.1175/JCLI-D-16-0385.1.
Chiodi, A. M., and D. E. Harrison, 2017b: Simulating ENSO SSTAs from TAO/TRITON winds: The impacts of 20 years of buoy observations in the Pacific waveguide and comparison with reanalysis products. J. Climate, 30, 1041–1059, https://doi.org/10.1175/JCLI-D-15-0865.1.
Chiodi, A. M., and D. E. Harrison, 2020: Tropical Pacific surface wind energy spectra and coherence: Basinwide observations and their observing system implications. J. Climate, 33, 7141–7154, https://doi.org/10.1175/JCLI-D-19-0836.1.
Chiodi, A. M., J. P. Dunne, and D. E. Harrison, 2019: Estimating air-sea carbon flux uncertainty over the tropical Pacific: Importance of winds and wind analysis uncertainty. Global Biogeochem. Cycles, 33, 370–390, https://doi.org/10.1029/2018GB006047.
Cravatte, S., and Coauthors, 2016: First report of TPOS 2020. 202 pp., https://tropicalpacific.org/tpos2020-project-archive/reports/.
Cronin, M. F., C. W. Fairall, and M. J. McPhaden, 2006: An assessment of buoy-derived and numerical weather prediction surface heat fluxes in the tropical Pacific. J. Geophys. Res., 111, C06038, https://doi.org/10.1029/2005JC003324.
Curry, J. A., and Coauthors, 2004: SEAFLUX. Bull. Amer. Meteor. Soc., 85, 409–424, https://doi.org/10.1175/BAMS-85-3-409.
Deng, Z., L. Xie, B. Liu, K. Wu, D. Zhao, and T. Yu, 2009: Coupling winds to ocean surface currents over the global ocean. Ocean Modell., 29, 261–268, https://doi.org/10.1016/j.ocemod.2009.05.003.
De Rosnay, P., and Coauthors, 2022: Coupled data assimilation at ECMWF: Current status, challenges and future developments. Quart. J. Roy. Meteor. Soc., 148, 2672–2702, https://doi.org/10.1002/qj.4330.
Dunn, R. J. H., and Coauthors, 2024: Global climate. Bull. Amer. Meteor. Soc., 105 (8), S12–S155, https://doi.org/10.1175/BAMS-D-24-0116.1.
Ebuchi, N., H. C. Graber, and M. J. Caruso, 2002: Evaluation of wind vectors observed by QuikSCAT/SeaWinds using ocean buoy data. J. Atmos. Oceanic Technol., 19, 2049–2062, https://doi.org/10.1175/1520-0426(2002)019<2049:EOWVOB>2.0.CO;2.
Fairall, C. W., E. F. Bradley, J. E. Hare, A. A. Grachev, and J. B. Edson, 2003: Bulk parameterization of air–sea fluxes: Updates and verification for the COARE algorithm. J. Climate, 16, 571–591, https://doi.org/10.1175/1520-0442(2003)016<0571:BPOASF>2.0.CO;2.
Foltz, G. R., C. Zhang, C. Meinig, J. A. Zhang, and D. Zhang, 2022: An unprecedented view inside a hurricane. Eos, 103, https://doi.org/10.1029/2022EO220228.
Freilich, M. H., and R. S. Dunbar, 1999: The accuracy of the NSCAT 1 vector winds: Comparisons with National Data Buoy Center buoys. J. Geophys. Res., 104, 11 231–11 246, https://doi.org/10.1029/1998JC900091.
Freitag, H. P., M. O’Haleck, G. C. Thomas, and M. J. McPhaden, 2001: Calibration procedures and instrumental accuracies for ATLAS wind measurements. NOAA Tech. Memo. OAR PMEL-119, NOAA/Pacific Marine Environmental Laboratory, Seattle, WA, 20 pp., https://repository.library.noaa.gov/view/noaa/11011.
Gentemann, C. L., and Coauthors, 2020: Saildrone: Adaptively sampling the marine environment. Bull. Amer. Meteor. Soc., 101, E744–E762, https://doi.org/10.1175/BAMS-D-19-0015.1.
Hackert, E., S. Akella, L. Ren, K. Nakada, J. A. Carton, and A. Molod, 2023: Impact of the TAO/TRITON array on reanalyses and predictions of the 2015 El Niño. J. Geophys. Res. Oceans, 128, e2023JC020039, https://doi.org/10.1029/2023JC020039.
Hasan, N. A., Y. Chikamoto, and M. J. McPhaden, 2022: The influence of tropical basin interactions on the 2020–2022 double-dip La Niña. Front. Climate, 4, 1001174, https://doi.org/10.3389/fclim.2022.1001174.
Jiang, C., M. F. Cronin, K. A. Kelly, and L. Thompson, 2005: Evaluation of a hybrid satellite- and NWP-based turbulent heat flux product using Tropical Atmosphere-Ocean (TAO) buoys. J. Geophys. Res., 110, C09007, https://doi.org/10.1029/2004JC002824.
Kelly, K. A., S. Dickinson, and G. C. Johnson, 2005: Comparisons of scatterometer and TAO winds reveal time-varying surface currents for the tropical Pacific Ocean. J. Atmos. Oceanic Technol., 22, 735–745, https://doi.org/10.1175/JTECH1738.1.
Kessler, W. S., and Coauthors, 2019: Second report of TPOS 2020. 266 pp., https://tropicalpacific.org/wp-content/uploads/2021/02/TPOS-2020-Second-Report_2019May30.pdf.
Kessler, W. S., and Coauthors, 2021: Final report of TPOS 2020. 87 pp., https://tropicalpacific.org/wp-content/uploads/2021/08/TPOS2020-Final-Report-2021.08.23.pdf.
Kleist, D., and Coauthors, 2023: Current state of data assimilation capabilities at NCEP’s Environmental Modeling Center. 63 pp., https://doi.org/10.25923/PJS0-4J42.
Kubota, M., N. Iwasaka, S. Kizu, M. Konda, and K. Kutsuwada, 2002: Japanese Ocean Flux Data Sets with Use of Remote Sensing Observations (J-OFURO). J. Oceanogr., 58, 213–225, https://doi.org/10.1023/A:1015845321836.
Large, W. G., and S. Pond, 1981: Open ocean momentum flux measurements in moderate to strong winds. J. Phys. Oceanogr., 11, 324–336, https://doi.org/10.1175/1520-0485(1981)011<0324:OOMFMI>2.0.CO;2.
Large, W. G., and S. Pond, 1982: Sensible and latent heat flux measurements over the ocean. J. Phys. Oceanogr., 12, 464–482, https://doi.org/10.1175/1520-0485(1982)012<0464:SALHFM>2.0.CO;2.
Liu, W. T., and W. Tang, 1996: Equivalent neutral wind. JPL Publication 96-17, 22 pp., https://ntrs.nasa.gov/api/citations/19970010322/downloads/19970010322.pdf.
Martin, S., 2014: An Introduction to Ocean Remote Sensing. 2nd ed. Cambridge University Press, 521 pp.
McGregor, S., A. Sen Gupta, D. Dommenget, T. Lee, M. J. McPhaden, and W. S. Kessler, 2017: Factors influencing the skill of synthesized satellite wind products in the tropical Pacific. J. Geophys. Res. Oceans, 122, 1072–1089, https://doi.org/10.1002/2016JC012340.
McPhaden, M. J., and Coauthors, 1998: The Tropical Ocean-Global Atmosphere observing system: A decade of progress. J. Geophys. Res., 103, 14 169–14 240, https://doi.org/10.1029/97JC02906.
McPhaden, M. J., and Coauthors, 2009: RAMA: The research moored array for African–Asian–Australian monsoon analysis and prediction. Bull. Amer. Meteor. Soc., 90, 459–480, https://doi.org/10.1175/2008BAMS2608.1.
McPhaden, M. J., K. J. Connell, G. R. Foltz, R. C. Perez, and K. Grissom, 2023: Tropical ocean observations for weather and climate: A decadal overview of the Global Tropical Moored Buoy Array. Oceanography, 36 (2–3), 32–43, https://doi.org/10.5670/oceanog.2023.211.
Mears, C., T. Lee, L. Ricciardulli, X. Wang, and F. Wentz, 2022: Improving the Accuracy of the Cross-Calibrated Multi-Platform (CCMP) ocean vector winds. Remote Sens., 14, 4230, https://doi.org/10.3390/rs14174230.
Mears, C. A., D. K. Smith, and F. J. Wentz, 2001: Comparison of Special Sensor Microwave Imager and buoy-measured wind speeds from 1987 to 1997. J. Geophys. Res., 106, 11 719–11 729, https://doi.org/10.1029/1999JC000097.
Moltmann, T., and Coauthors, 2019: A Global Ocean Observing System (GOOS), delivered through enhanced collaboration across regions, communities, and new technologies. Front. Mar. Sci., 6, 291, https://doi.org/10.3389/fmars.2019.00291.
Moum, J. N., A. Perlin, J. D. Nash, and M. J. McPhaden, 2013: Seasonal sea surface cooling in the equatorial Pacific cold tongue controlled by ocean mixing. Nature, 500, 64–67, https://doi.org/10.1038/nature12363.
Praveen Kumar, A., J. Vialard, M. Lengaigne, V. S. N. Murty, and M. J. McPhaden, 2012: TropFlux: Air-sea fluxes for the global tropical oceans—Description and evaluation. Climate Dyn., 38, 1521–1543, https://doi.org/10.1007/s00382-011-1115-0.
Ricciardulli, L., and F. J. Wentz, 2015: A scatterometer geophysical model function for climate-quality winds: QuikSCAT Ku-2011. J. Atmos. Oceanic Technol., 32, 1829–1846, https://doi.org/10.1175/JTECH-D-15-0008.1.
Ricciardulli, L., and A. Manaster, 2021: Intercalibration of ASCAT scatterometer winds from MetOp-A, -B, and -C, for a stable climate data record. Remote Sens., 13, 3678, https://doi.org/10.3390/rs13183678.
Schlundt, M., J. T. Farrar, S. P. Bigorre, A. J. Plueddemann, and R. A. Weller, 2020: Accuracy of wind observations from open-ocean buoys: Correction for flow distortion. J. Atmos. Oceanic Technol., 37, 687–703, https://doi.org/10.1175/JTECH-D-19-0132.1.
Smith, N., and Coauthors, 2019: Tropical Pacific observing system. Front. Mar. Sci., 6, 31, https://doi.org/10.3389/fmars.2019.00031.
Sprintall, J., and Coauthors, 2024: COVID impacts cause critical gaps in the Indian Ocean observing system. Bull. Amer. Meteor. Soc., 105, E725–E741, https://doi.org/10.1175/BAMS-D-22-0270.1.
Stiles, B. W., and R. S. Dunbar, 2010: A neural network technique for improving the accuracy of scatterometer winds in rainy conditions. IEEE Trans. Geosci. Remote Sens., 48, 3114–3122, https://doi.org/10.1109/TGRS.2010.2049362.
Stoffelen, A., 1998: Toward the true near-surface wind speed: Error modeling and calibration using triple collocation. J. Geophys. Res., 103, 7755–7766, https://doi.org/10.1029/97JC03180.
Verspeek, J., A. Stoffelen, M. Portabella, H. Bonekamp, C. Anderson, and J. F. Saldana, 2010: Validation and calibration of ASCAT using CMOD5.n. IEEE Trans. Geosci. Remote Sens., 48, 386–395, https://doi.org/10.1109/TGRS.2009.2027896.
Wen, C., A. Kumar, and Y. Xue, 2019: Uncertainties in reanalysis surface wind stress and their relationship with observing systems. Climate Dyn., 52, 3061–3078, https://doi.org/10.1007/s00382-018-4310-4.
Wentz, F. J., and D. Draper, 2016: On-orbit absolute calibration of the Global Precipitation Measurement microwave imager. J. Atmos. Oceanic Technol., 33, 1393–1412, https://doi.org/10.1175/JTECH-D-15-0212.1.
Wentz, F. J., and Coauthors, 2017: Evaluating and extending the ocean wind climate data record. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 10, 2165–2185, https://doi.org/10.1109/JSTARS.2016.2643641.
Wentz, K., T. Meissner, F. Wentz, and A. Manaster, 2024: Absolute intercalibration of spaceborne microwave radiometers. J. Atmos. Oceanic Technol., 41, 1121–1138, https://doi.org/10.1175/JTECH-D-24-0024.1.
Xing, J., J. Shi, Y. Lei, X.-Y. Huang, and Z. Liu, 2016: Evaluation of HY-2A scatterometer wind vectors using data from buoys, ERA-Interim and ASCAT during 2012–2014. Remote Sens., 8, 390, https://doi.org/10.3390/rs8050390.
Yang, J., and J. Zhang, 2019: Comparison of Oceansat-2 scatterometer wind data with Global Moored Buoys and ASCAT observation. Adv. Meteor., 2019, 1–9, https://doi.org/10.1155/2019/1651267.
Ye, H., and Coauthors, 2021: Evaluation of CFOSAT scatterometer wind data in global oceans. Remote Sens., 13, 1926, https://doi.org/10.3390/rs13101926.
Yu, L., and X. Jin, 2014: Insights on the OAFlux ocean surface vector wind analysis merged from scatterometers and passive microwave radiometers (1987 onward). J. Geophys. Res. Oceans, 119, 5244–5269, https://doi.org/10.1002/2013JC009648.
Zhang, C., and Coauthors, 2023: Hurricane observations by uncrewed systems. Bull. Amer. Meteor. Soc., 104, E1893–E1917, https://doi.org/10.1175/BAMS-D-21-0327.1.
