Effectiveness of Using Multisatellite Wind Speed Estimates to Construct Hourly Wind Speed Datasets with Diurnal Variations
1. Introduction
One of the keys to understanding air–sea interactions over the global ocean is sea surface wind speed (SSW), because it is crucial in determining the momentum, heat, and freshwater fluxes between the atmosphere and the ocean. These fluxes are driven by surface wind stress, which affects ocean circulation and wave generation and drives the latent heat fluxes that warm or cool the boundary layer, and by evaporation, which increases sea surface salinity and moistens the atmosphere. Given the complexity of these relationships, SSW datasets with high spatiotemporal resolution are required to better understand and predict these processes. Such datasets could also be applied to describe changes related to long-term fluctuations, such as El Niño–Southern Oscillation (ENSO) and climate variability trends.
Several recent modeling studies have suggested that high-resolution sea surface temperature (SST) datasets that reproduce diurnal variations are needed to understand climate variability over the global ocean. For example, Masson et al. (2012) demonstrated that SST without instantaneous diurnal variability could reduce ENSO amplitude by 15%. Clayson and Bogdanoff (2013) suggested that fluxes without diurnal variation induce errors of 25% or more in the seasonal signal of the modeled SST over most of the tropical oceans. These results indicate that the diurnal variation of SSW is also crucial in determining the surface energy balance, because diurnal variations of SST and heat flux are largely dependent on SSW (Donlon et al. 2007). Nevertheless, given its transient nature, the extent to which the diurnal variation of SSW influences climate variability and air–sea interaction remains unclear. To evaluate the importance of this variation, observational datasets with high spatiotemporal resolution and diurnal variation are required. Satellite-derived observations could adequately fulfill these requirements; however, currently, there are no such satellite-derived SSW datasets available.
Most instruments used by oceanographers and meteorologists to measure SSW and SST are aboard sun-synchronous polar-orbiting satellites, typically at altitudes of 700–800 km, that observe a swath of ocean 1500–1800 km wide (Atlas et al. 2011). It usually takes about 100 min for such satellites to complete one circuit around Earth, allowing just over 14 orbits daily. These satellites cross the equator at the same local time each day [this is often referred to as the local equatorial crossing time (LECT)], ascending once (traveling from south to north) and descending once. The characteristics of sun-synchronous polar-orbiting satellites restrict satellite observation duration, such that these satellites can observe geophysical parameters at the same location only once or twice per day. Therefore, we cannot investigate the importance of the diurnal variation of those parameters using single-satellite observations. In addition, if daily mean wind speeds were computed using a single sun-synchronous polar-orbit satellite, accuracy would be expected to be poor because of sampling errors, particularly for a region with high temporal variability (Tomita and Kubota 2011).
A simple method for avoiding problems caused by the limited observation time of polar-orbiting satellites is to merge the data derived from multiple satellite sensors. High-frequency sampling using multiple sensors is extremely useful for constructing datasets with high temporal resolution. Therefore, the first objective of this study was to verify the effectiveness of multisatellite observations for constructing an hourly SSW product that could capture diurnal variation. The second objective was to assess the level of accuracy of such a dataset in relation to the reduction of the sampling error because of the use of multiple satellite sensors.
2. Data and method
a. Sun-synchronous polar-orbiting satellites
High-accuracy hourly and daily SSW datasets were constructed by combining four types of Level 3 wind speed products: AMSR-2, operated by the Japan Aerospace Exploration Agency; WindSat, provided by Remote Sensing Systems (RSS); and ASCAT and OceanSat Scatterometer (OSCAT), furnished by the Royal Netherlands Meteorological Institute (KNMI). Table 1 presents detailed descriptions of these wind speed products. The different satellite observation times (Fig. 1) are very useful for constructing an hourly dataset with diurnal variations, and they are effective for removing sampling errors when daily mean values are computed. Sun-synchronous polar-orbiting satellites with different LECTs in each year (e.g., DMSP SSM/I) and nonsun-synchronous polar-orbiting satellites (e.g., the Tropical Rainfall Measuring Mission Microwave Imager) were not used because it remains uncertain whether their data would contribute to achieving the objectives of the present study. In fact, the inclusion of duplicated observations might produce a detrimental effect on the accuracy of the daily mean SSW values. The products contain ascending and descending average SSWs on a 0.25° × 0.25° spatial grid, although the OSCAT data were interpolated linearly to this grid spacing from a 0.5° × 0.5° spatial resolution. These datasets were merged using a spatiotemporal interpolation method to construct an hourly SSW dataset with diurnal variations (see section 2b).
Description of wind speed products used in this study.
Local equatorial crossing time (LECT) of each satellite.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Local equatorial crossing time (LECT) of each satellite.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Local equatorial crossing time (LECT) of each satellite.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
b. Method
First, by rounding the minute data of the observation time after changing the decimal number (i.e., 1903 became 1900 UTC), we constructed four types of hourly wind speed product for each hour using four different Level 3 wind speed products individually (Fig. 2). Thus, the datasets constructed in this study were not “hourly means” but were hourly datasets for each hour. This was performed to avoid the difficulty in computing hourly mean values using satellite-derived SSWs due to the different characteristics of the satellite observations (see section 1; Table 1; Fig. 1). Second, these products were simply merged onto each grid for each hour (Fig. 3a). Although there are few grids shown in Figs. 2 and 3a, if these represented multiple data from different satellites in the same grid, we simply averaged them. Figure 3a shows that there were substantial areas in this hourly dataset without values. Therefore, appropriate spatial and temporal interpolations were performed to address data coverage issues. Third, a linear interpolation method with a 6-h time scale was used for the temporal interpolation (Fig. 3b). For example, in this linear interpolation, to estimate the data at 1200 UTC in each grid, SSW data observed between 0600 and 1800 UTC were used. This time scale was determined based on each LECT in Fig. 1 and Table 1. Although the maximum difference of LECT between two satellites was 4 h, 30 min (AMSR-2 and WindSat), we used a longer time scale because it was not always possible for four satellites to observe the SSW in each grid eight times per day. The task of spatial interpolation was more difficult because data were missing over large areas, which is a common challenge when using such data to explain phenomena such as global climate variability, wind-driven current systems, and air–sea interactions. We addressed the problem using the same optimum interpolation method (OIM) used in Kako and Kubota (2006) and Kako et al. (2011) to produce a more useful dataset, in which areas without data were removed. The results after applying the OIM once and twice to the data illustrated in Fig. 3b are shown in Fig. 3c and 3d, respectively. Figure 3d demonstrates that all missing values were interpolated completely by the OIM. This was the final product used to accomplish the study’s objectives. A detailed description of the results of the OIM is presented in section 3a.
Horizontal distribution of wind speed observed at 1900 UTC 23 Apr 2013: (a) AMSR-2, (b) ASCAT, (c) OSCAT, and (d) WindSat.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Horizontal distribution of wind speed observed at 1900 UTC 23 Apr 2013: (a) AMSR-2, (b) ASCAT, (c) OSCAT, and (d) WindSat.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Horizontal distribution of wind speed observed at 1900 UTC 23 Apr 2013: (a) AMSR-2, (b) ASCAT, (c) OSCAT, and (d) WindSat.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Horizontal distribution of wind speed observed at 1900 UTC 23 Apr 2013: (a) multisatellite, (b) after applying a linear interpolation to (a), and (c) after applying OIM once to (b), and (d) after applying OIM twice to (b). Multisatellite indicates a product resulting from the simple merging of four types of products. The OIM was applied only to the regions outlined by red lines.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Horizontal distribution of wind speed observed at 1900 UTC 23 Apr 2013: (a) multisatellite, (b) after applying a linear interpolation to (a), and (c) after applying OIM once to (b), and (d) after applying OIM twice to (b). Multisatellite indicates a product resulting from the simple merging of four types of products. The OIM was applied only to the regions outlined by red lines.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Horizontal distribution of wind speed observed at 1900 UTC 23 Apr 2013: (a) multisatellite, (b) after applying a linear interpolation to (a), and (c) after applying OIM once to (b), and (d) after applying OIM twice to (b). Multisatellite indicates a product resulting from the simple merging of four types of products. The OIM was applied only to the regions outlined by red lines.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
A detailed description of the OIM has been published by Kako et al. (2011); therefore, we offer only a brief explanation here.
We used daily mean wind vector data in the Reanalysis-1 dataset (NRA1) (Kalnay et al. 1996) of the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) for the first-guess value. Spatially, the NRA1 wind data were linearly interpolated to provide values 0.25° × 0.25° grid cell.
c. In situ measurements (moored buoy data)
To validate our SSW (OIM) dataset, we compared it with observations from 32 moored buoys, the locations of which are shown in Fig. 4. The comparison used 25 buoys operated by the National Data Buoy Center (NDBC) (NDBC 2009), three buoys of the Prediction and Research Moored Array in the Tropical Atlantic (PIRATA) (Bourlès et al. 2008), two buoys operated by the Research Moored Array for African–Asian–Australian Monsoon Analysis and Prediction (RAMA) (McPhaden et al. 2009), and two buoys of the Tropical Atmosphere Ocean (TAO) array (McPhaden et al. 1998). Hereafter, the PIRATA, RAMA, and TAO buoys are referred to as Autonomous Temperature Line Acquisition System (ATLAS) buoys for convenience. In addition to these buoys, we also used two Ocean Reference Station (ORS) buoys: Stratus and Northwest Tropical Atlantic Station (NTAS). The ORS project is conducted by the Woods Hole Oceanographic Institution to provide critical sustained observations of the key oceanic region of the trade winds (Fig. 4). ORS data are not entered into the Global Telecommunication System; therefore, they can be used as independent data for the validation of models by users such as the European Centre for Medium-Range Weather Forecasts and NCEP. Note that we used all buoy observations available during the period 8 March–31 May 2013 for the comparison.
Locations of NDBC (triangles), TAO (diamonds), PIRATA (circles), RAMA (stars), and Stratus and NTAS (squares) buoys used in the study.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Locations of NDBC (triangles), TAO (diamonds), PIRATA (circles), RAMA (stars), and Stratus and NTAS (squares) buoys used in the study.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Locations of NDBC (triangles), TAO (diamonds), PIRATA (circles), RAMA (stars), and Stratus and NTAS (squares) buoys used in the study.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Wind speed measurements by the aforementioned buoys at various heights above the sea surface are converted to neutral stability wind speeds at 10-m height using the Coupled Ocean–Atmosphere Response Experiment (COARE) 3.0 bulk algorithm of Fairall et al. (2003). These buoy data have different sampling periods; therefore, it was necessary to adjust the observations to a common averaging time. The sampling periods of the ATLAS, NDBC, and ORS instruments are 2 min (http://www.ndbc.noaa.gov/rsa.shtml), 8 min (http://www.pmel.noaa.gov/tao/proj_over/sampling.html), and 60 min (http://uop.whoi.edu/UOP_realtime_files.html), respectively. However, these high-temporal-resolution data are not necessarily available from their respective websites. For example, the highest temporal resolution of data available from ATLAS is 10 min and that from NDBC and ORS is 60 min. Thus, the same averaging periods cannot be defined for the different buoy observation products. As mentioned in section 2b, we constructed hourly datasets for each hour. Therefore, the data for each hour derived from the above-mentioned buoy observations were used for the comparison to validate the accuracy of the satellite observations; that is, we assumed that the hourly NDBC and ORS data were concurrent with our hourly datasets.
d. Reanalysis data
The dataset constructed in this study was also compared with two reanalysis products to evaluate the diurnal variation of each. The first product was the NCEP Climate Forecast System, version 2 (CFSv2), developed by the NCEP Environmental Modeling Center (Saha et al. 2014). This is a fully coupled model representing the interaction between the atmosphere, oceans, land, and sea ice that became operational at NCEP in March 2011. The CFSv2 product with 0.2° horizontal resolution was used herein. The CFSv2 data are available from the Computational and Information Systems Laboratory of the Research Data Archive (http://rda.ucar.edu). The second product used in this study was Modern-Era Retrospective Analysis for Research and Applications (MERRA), available from the National Aeronautics and Space Administration Global Modeling and Assimilation Office (https://gmao.gsfc.nasa.gov/reanalysis/MERRA/). MERRA output of SSW at 1-h intervals is available at the full spatial resolution (½° latitude × ⅔° longitude). As many buoy observations have been assimilated into these reanalysis products, we used the ORS buoy data for the comparison of the OIM dataset and reanalysis products because the ORS data are not assimilated within the latter products.
3. Results
a. Comparison of hourly satellite-derived wind fields with buoy observations
To evaluate whether the four single SSW datasets are appropriate for the computation of the OIM, we compared them with observations obtained by 32 moored buoys between 18 March and 31 May 2013, during which all wind speed datasets were available. Figure 5 shows a scatterplot of the satellite-observed wind speed against those from all the buoys. The satellite-derived wind speeds are consistent with the buoy observations without systematic biases except for the winds observed by AMSR-2. In addition, high correlation coefficients (all significant at the 95% confidence level) indicate that the temporal wind variations derived from the buoys are consistent with those of the satellite-derived winds. Figure 6 shows the results of comparisons in the region of each buoy observation. It demonstrates that the accuracy of winds observed by AMSR-2 is extremely poor in buoy regions within 200 km of the nearest coast (see green color in Fig. 6). For example, the bias exceeds 1.0 m s−1 (Fig. 6a) and the root-mean-square (RMS) difference is almost twice that of other single-satellite products. Possible causes of the low accuracy of the AMSR-2 close to the coast remain unsolved, although the contamination of microwaves related to the land structure might be a contributing factor. As the OIM assumes that errors are unbiased and uncorrelated, if observational data have large biases, then the OIM leads to large biases in the analysis. Thus, winds observed by AMSR-2 in this region are inappropriate for the OIM computation and these data were deselected in the OIM operation. Although AMSR-2 wind speeds observed in the open ocean have slightly larger bias compared with other datasets (Fig. 6a), we used these data for the OIM computation after bias correction using linear regression equations. The regression coefficients were determined using all the buoy observations except for those buoys located within 200 km of the nearest coast. Figure 5e shows the results of the bias corrections and demonstrates that bias is removed from the original AMSR-2 dataset.
Scatterplots of wind speed observed by satellite vs data recorded by various buoys shown by various marks and colors. Data sources are given in upper portions of each panel. The line y = x is added to each panel.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Scatterplots of wind speed observed by satellite vs data recorded by various buoys shown by various marks and colors. Data sources are given in upper portions of each panel. The line y = x is added to each panel.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Scatterplots of wind speed observed by satellite vs data recorded by various buoys shown by various marks and colors. Data sources are given in upper portions of each panel. The line y = x is added to each panel.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Statistics for comparison of hourly wind speeds derived from satellite and buoy observations: (a) bias, (b) Taylor diagram representing the relationship between RMS and correlation coefficients, and (c) the number of data used for comparison. NDBC1 (NDBC2) denotes the region where buoys are located within (more than) 200 km from the nearest coast.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Statistics for comparison of hourly wind speeds derived from satellite and buoy observations: (a) bias, (b) Taylor diagram representing the relationship between RMS and correlation coefficients, and (c) the number of data used for comparison. NDBC1 (NDBC2) denotes the region where buoys are located within (more than) 200 km from the nearest coast.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Statistics for comparison of hourly wind speeds derived from satellite and buoy observations: (a) bias, (b) Taylor diagram representing the relationship between RMS and correlation coefficients, and (c) the number of data used for comparison. NDBC1 (NDBC2) denotes the region where buoys are located within (more than) 200 km from the nearest coast.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
As mentioned in section 2b, to construct an hourly wind speed dataset with diurnal variation in the absence of missing values, we first applied linear interpolation temporally with a 6-h time scale to the dataset shown in Fig. 3a, which was created by simply merging four types of product (see a more detailed description in section 2b). Figure 3b shows the result of the temporal interpolation, which illustrates there are several grids without data. To remove these missing values completely, we applied the OIM to the data shown in Fig. 3b (Fig. 3c). To reduce computation time, we applied the OIM only to regions with buoy observations (regions outlined by red lines in Fig. 3b; see also Fig. 4). Figure 3c shows that values were interpolated mostly over specific regions and that values for a few percent of the ocean remain unknown. The OIM was again applied to remove these data voids. In the second OIM, values were estimated only where data were missing. The results are shown in Fig. 3d, in which there are no missing values. This represents the final product that was used for the analyses.
We also compared 32 buoy observations with this newly constructed SSW (OIM) product. Figures 5f and 6 demonstrate that the data accuracy of our product is slightly poorer than that of the single-satellite products except for winds observed by AMSR-2. However, the OIM dataset has up to 43 times more data at buoy locations than the other four datasets (e.g., AMSR-2 has 1081 data points, whereas the new dataset has 46 704; Fig. 5). The benefit offered by this increase in data availability more than compensates the disadvantage of the slight decrease in data accuracy. This issue is discussed in section 3c.
b. Comparison of daily satellite-derived wind fields with buoy observations
The new hourly dataset is expected to be effective only for obtaining accurate daily mean values by the removal of sampling error. As mentioned above, sun-synchronous satellites are capable of observing SSWs at the same location only once or twice per day. Thus, if we use data only from a single satellite to estimate daily mean values, a large sampling error will result. To evaluate the issue of sampling error induced by the use of sun-synchronous satellites, we compared the daily mean wind speeds derived from buoy and satellite observations. We calculated daily average wind speeds based on hourly observations of buoys and the OIM product, whereas daily mean wind speeds originating from single-satellite products (i.e., AMSR-2, ASCAT, OSCAT, and WindSat) were just averaged values of ascending and descending observations.
Figure 7 shows scatter diagrams comparing the daily mean buoy observations with four satellite-derived daily mean wind speed products and our new dataset. Statistics associated with this comparison in each buoy location are illustrated in Fig. 8. These results demonstrate that temporal wind variations derived from single- and multisatellite observations are consistent with those from buoy observations, determined by correlation coefficients significant at the 95% confidence level. In addition, biases in those datasets are within 0.5 m s−1 except for winds observed by AMSR-2. The Taylor diagram in Fig. 8b, representing the relationship between RMS differences and correlation coefficients, indicates that the largest correlation coefficient and the smallest RMS differences are between the OIM dataset and buoy observations in each buoy location. This result should be emphasized. The RMS differences in the single-satellite daily products are large compared with those in the OIM dataset (Fig. 8b), even though the RMS differences in the single-satellite hourly products are smaller than in the OIM dataset. This suggests that a sampling error in sun-synchronous satellite data induces a large RMS error in daily mean values, but this does not result in a large bias value. These results also indicate that in computing daily mean values, the removal of sampling error is extremely important to the construction of a highly accurate dataset of those values.
As in Fig. 5, but for daily mean data.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
As in Fig. 5, but for daily mean data.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
As in Fig. 5, but for daily mean data.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
As in Fig. 6, but for daily mean data.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
As in Fig. 6, but for daily mean data.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
As in Fig. 6, but for daily mean data.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
c. Intercomparison of OIM dataset and reanalysis products using ORS buoy data
To evaluate the use of multisatellite observations and the effectiveness of the OIM in filling missing values, we further compared the OIM dataset and reanalysis SSW hourly products, because there are no satellite-derived hourly datasets to date. In this comparison, we did not use wind speed data derived from TAO, PIRATA, RAMA, and NDBC buoys. Instead, the ORS buoy data (Stratus and NTAS; Fig. 4) were used as independent data for the validation of SSW products, because they were not assimilated in the reanalysis products. Figure 9 shows scatter diagrams that compare the hourly buoy observations with two reanalysis products and the OIM hourly dataset. This reveals that the data accuracy of the latter dataset is considerably greater than that of the reanalysis datasets. A comparison of the daily mean products shows the same results as the hourly product comparison (Fig. 10).
Scatterplots of wind speed derived from the OIM dataset and reanalysis products vs data recorded by ORS buoys. Data sources are given in upper portions of each panel. The line y = x is added to each panel.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Scatterplots of wind speed derived from the OIM dataset and reanalysis products vs data recorded by ORS buoys. Data sources are given in upper portions of each panel. The line y = x is added to each panel.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
Scatterplots of wind speed derived from the OIM dataset and reanalysis products vs data recorded by ORS buoys. Data sources are given in upper portions of each panel. The line y = x is added to each panel.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
As in Fig. 9, but for daily mean data.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
As in Fig. 9, but for daily mean data.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
As in Fig. 9, but for daily mean data.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
To assess whether the OIM dataset can resolve the diurnal variation of buoy-derived wind speed, we computed the power spectral density (PSD) of each observation at the ORS buoy locations. A comparison of the PSD and the period is illustrated for hourly wind speeds for the two ORS observation sites in Fig. 11. The PSD at period of more than 100 h is not illustrated because our focus is on the high-frequency variation of SSW. Figure 11 demonstrates that significant PSD peaks for the buoy observations can be found at around the periods of 12 and 24 h. The energy spectra for the OIM dataset and the reanalysis products also have significant peaks at 24 h. This indicates that the OIM and reanalysis products are capable of resolving the diurnal variations in actual wind speed—a result not reproduced using the other single-satellite products. However, although the PSD peaks for the reanalysis products are also significant at the period of 12 h, the OIM datasets have no peaks around that period. Nevertheless, the energy spectra for all wind products depart from the buoy wind spectrum for periods shorter than 12 h; that is, both the OIM dataset and the global reanalysis wind speed products have a disadvantage in that they do not contain realistic energy at such short periods. For the OIM dataset, it is reasonable that the OIM dataset does not contain realistic high-frequency energy, because we applied temporal linear interpolation with a 6-h time scale to the multisatellite observations (see section 2). These results suggest that the selection of appropriate temporal resolution for the dataset with diurnal variation might reduce the interpolation error shown in Fig. 5f. However, it is not possible to know the appropriate temporal resolution in advance.
PSD vs period for SSW at ORS buoy locations: (a) Stratus and (b) NTAS.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
PSD vs period for SSW at ORS buoy locations: (a) Stratus and (b) NTAS.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
PSD vs period for SSW at ORS buoy locations: (a) Stratus and (b) NTAS.
Citation: Journal of Atmospheric and Oceanic Technology 34, 3; 10.1175/JTECH-D-16-0179.1
4. Summary
We constructed an SSW dataset with hourly temporal resolution on a 0.25°× 0.25° grid by applying an OIM to four remote sensing products (AMSR-2, ASCAT, OSCAT, and WindSat). In the derived wind fields, data voids were completely filled using linear interpolation with a 6-h time scale and the OIM (in the absence of artificial patterns) depending on the satellite orbit. The accuracy of our newly constructed SSW (OIM) dataset was then examined by comparison with wind data recorded by NDBC, TAO, PIRATA, and RAMA buoys. The wind field derived from the OIM dataset was in reasonable agreement with that from the buoy observations. Although the RMS differences between the OIM dataset and buoy observations were slightly greater than with single-satellite products, the OIM dataset was found capable of resolving the diurnal wind speed variations. This result was not reproduced using the other single-satellite products. The validity of the OIM dataset was also confirmed by the comparison of two reanalysis wind products with ORS buoy observations. This comparison showed that both CFSv2 and MERRA have obvious systematic biases. In particular, MERRA has large biases in the tropics, for example, wind biases exceeded −0.74 m s−1 at the NTAS buoy (see Figs. 9f and 10f).
These results demonstrate that the application of multisatellite observations could be a very useful and effective approach for the construction of datasets with high temporal resolution. These findings also indicate that the wind dataset constructed herein would be more useful than single-satellite and reanalysis products in driving ocean general circulation models and in computing momentum, heat, and freshwater fluxes between the atmosphere and the ocean. As mentioned above, several recent modeling studies have suggested that high-resolution SST and SSW datasets with diurnal variation are needed to understand the seasonal variation over most of the tropical oceans. Nevertheless, the effect of the diurnal variation of SSW on climate variability and air–sea interaction in midlatitudes remains unclear. The effectiveness of using an hourly dataset with diurnal variations to investigate air–sea interactions, climate variability, and numerical simulations should be addressed in a future study. As AMSR-2 cannot observe wind directions, a reasonable estimate of wind direction using scatterometer (e.g., ASCAT and OSCAT) and radiometer (WindSat) data is another important topic for future research.
Our hourly dataset was shown to be effective for obtaining accurate daily mean values. A comparison of daily mean wind speeds derived from satellite observations with those from buoy observations demonstrated that we could obtain greater accuracy in daily mean SSW data using multisatellite observations compared with single-satellite observations. Therefore, the use of the multisatellite observations could be a very useful and effective approach to improve the accuracy of daily mean values. Using actual satellite observations, we also demonstrated that an important objective in constructing very accurate daily mean data is the reduction in sampling errors, as shown in Tomita and Kubota (2011) (although they did not use real satellite data).
Acknowledgments
This research was supported by the Global Change Observation Mission: 5th Research Announcement (GCOM-RA5-D03) of the Japan Aerospace Exploration Agency (JAXA) and the Institute of Oceanic Research and Development of Tokai University (2016-5). We are grateful to JAXA, the Royal Netherlands Meteorological Institute (KNMI), and Remote Sensing Systems (RSS) for providing the SSW products used in this study. We also thank the Woods Hole Oceanographic Institution (WHOI) and the Tropical Atmosphere Ocean (TAO) Project Office of NOAA/Pacific Marine Environmental Laboratory (PMEL) for providing the moored buoy data, and the National Centers for Environmental Prediction and the National Aeronautics and Space Administration (NASA) for providing the reanalysis products. AMSR-2 data were obtained from the JAXA GCOM-W website (http://suzaku.eorc.jaxa.jp/GCOM_W/w_amsr2/whats_amsr2.html). ASCAT and OSCAT data, furnished by KNMI, were obtained freely from MyOcean (
REFERENCES
Atlas, R., R. N. Hoffman, J. Ardizzone, S. M. Leidner, J. C. Jusem, D. K. Smith, and D. Gombos, 2011: A cross-calibrated multiplatform ocean surface wind velocity product for meteorological and oceanographic applications. Bull. Amer. Meteor. Soc., 92, 157–174, doi:10.1175/2010BAMS2946.1.
Bourlès, B. R., and Coauthors, 2008: The PIRATA program: History, accomplishments, and future directions. Bull. Amer. Meteor. Soc., 89, 1111–1125, doi:10.1175/2008BAMS2462.1.
Clayson, C. A., and A. S. Bogdanoff, 2013: The effect of diurnal sea surface temperature warming on climatological air–sea fluxes. J. Climate, 26, 2546–2556, doi:10.1175/JCLI-D-12-00062.1.
Daley, R., 1991: Atmospheric Data Analysis. Cambridge University Press, 457 pp.
Donlon, C. I., and Coauthors, 2007: The Global Ocean Data Assimilation Experiment High-Resolution Sea Surface Temperature Pilot Project. Bull. Amer. Meteor. Soc., 88, 1197–1213, doi:10.1175/BAMS-88-8-1197.
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 algorithms. J. Climate, 16, 571–591, doi:10.1175/1520-0442(2003)016<0571:BPOASF>2.0.CO;2.
Kako, S., and M. Kubota, 2006: Relationship between an El Nino event and the interannual variability of significant wave height in the North Pacific. Atmos.–Ocean, 44, 377–395, doi:10.3137/ao.440404.
Kako, S., A. Isobe, and M. Kubota, 2011: High resolution ASCAT wind vector dataset gridded by applying an optimum interpolation method to the global ocean. J. Geophys. Res., 116, D23107, doi:10.1029/2010JD015484.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471, doi:10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.
Kuragano, T., and A. Shibata, 1997: Sea surface dynamics height of the Pacific Ocean derived from TOPEX/POSEIDON altimeter data: Calculation method and accuracy. J. Oceanogr., 53, 583–599.
Masson, S., P. Terray, G. Madec, J.-J. Luo, T. Yamagata, and K. Takahashi, 2012: Impact of intra-daily SST variability on ENSO characteristics in a coupled model. Climate Dyn., 39, 681–707, doi:10.1007/s00382-011-1247-2.
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, doi: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, doi:10.1175/2008BAMS2608.1.
NDBC, 2009: Handbook of automated data quality control checks and procedures. NDBC Tech. Doc. 09-02, 78 pp. [Available online at http:// www.ndbc.noaa.gov/NDBCHandbookofAutomatedDataQualityControl2009.pdf.]
Saha, S., and Coauthors, 2014: The NCEP Climate Forecast System Version 2. J. Climate, 27, 2185–2208, doi:10.1175/JCLI-D-12-00823.1.
Tomita, H., and M. Kubota, 2011: Sampling error of daily mean surface wind speed and air specific humidity due to Sun-synchronous satellite sampling and its reduction by multi-satellite sampling. Int. J. Remote Sens., 32, 3389–3404, doi:10.1080/01431161003749428.
