This study investigated the impact of assimilating satellite data into atmospheric reanalyses on trends in ocean surface winds and waves. Two experiments were performed using a numerical wave model forced by near-surface winds: one derived from the Japanese 55-year Reanalysis (JRA-55; experiment A) and the other derived from JRA-55 using assimilated conventional observations only (JRA-55C; experiment B). The results showed that the satellite data assimilation reduced upward trends of the annual mean of wave energy flux (WEF) in the midlatitude North Pacific and southern ocean (30°–60°S), south of Australia, from 1959 to 2012. It was also found that the assimilation of scatterometer winds reduced the near-surface wind speed in the midlatitude North Pacific after the mid-1990s, which resulted in the reduced trend in WEF from 1959 to 2012. By contrast, assimilation of the satellite radiances for 1973–94 increased near-surface wind speed in the southern ocean, south of Australia, whereas the assimilation of the scatterometer winds after the mid-1990s reduced wind speed. The latter led to the reduced trend in WEF south of Australia from 1959 to 2012.
Greater understanding of ocean surface wind and wave climate is crucial in the design of harbors, ship routing, offshore industries, and offshore wind and wave energy farms. Therefore, long-term trends in ocean surface winds and waves have undergone intense study based on in situ data, satellite measurements, atmosphere and wave reanalysis data, and numerical wave hindcasts (Carter and Draper 1988; Graham and Diaz 2001; Wang and Swail 2001; Caires and Swail 2004; Gulev and Grigorieva 2004; Hemer et al. 2009; Gemmrich et al. 2011; Izaguirre et al. 2011; Young et al. 2011; Sasaki 2012, 2014; Kumar et al. 2013; Wu et al. 2014).
Homogeneity in marine meteorological data is essential to assess the marine wind and wave climate accurately. However, meteorological observation records suffer from temporal inhomogeneity due to changes in observational instruments or methods. Ramage (1984) found a spurious trend in the marine wind speed derived from long-term compilations of ship observations from the Comprehensive Ocean–Atmosphere Data Set (COADS; Slutz et al. 1985). Peterson and Hasse (1987) found that an upward trend of wind speed at the western entrance to the English Channel during the period 1859–1980 may result from the transition from the Beaufort wind estimates to anemometer observations. Cardone et al. (1990) showed that the strengthening of surface wind speed from the 1950s to 1980s along the South China Sea, North Pacific, and North Atlantic shipping routes is a consequence of the increasing use of anemometers in place of Beaufort wind estimates. Thomas et al. (2005, 2008) also showed that marine wind speeds from ship reports are inhomogeneous over time due to changing measurement methods and measurement heights. Tokinaga and Xie (2011) corrected ship-based measurements of marine wind speed derived from the International COADS (ICOADS; Woodruff et al. 2011) using ocean surface wave data. Inhomogeneity in the near-surface wind speed is also found in state-of-the-art atmospheric reanalysis data (Cox and Swail 2001; Sterl 2004; Krueger et al. 2013; Ferguson and Villarini 2014). For example, Krueger et al. (2013) revealed substantial temporal inhomogeneity in the quality of the Twentieth Century Reanalysis (20CR; Compo et al. 2011) product caused by low quality of observations used in reanalysis before 1950. Regarding the inhomogeneity in ocean surface wave data, the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40; Uppala et al. 2005) exhibits inhomogeneity in significant wave height due to the assimilation of different altimeter wave height data [European Remote Sensing Satellite-1 and -2 (ERS-1 and ERS-2, respectively)] and faulty ERS-1 fast delivery product (Bauer and Staabs 1998; Caires and Sterl 2005). Gulev et al. (2003) showed that voluntary observed ships underestimate significant wave height wherever the wave heights are large (particularly in the Southern Hemisphere), compared with altimeter wave data, because ships tend to avoid stormy conditions. Therefore, inhomogeneity potentially exists in marine meteorological data, which could lead to erroneous interpretation of a historical climate state.
Recently, the Japan Meteorological Agency produced state-of-the-art atmospheric reanalyses, the Japanese 55-year Reanalysis (JRA-55; Kobayashi et al. 2015) and also JRA-55C (Kobayashi et al. 2014), which is based on the same atmospheric model and assimilation scheme as JRA-55 but assimilates in situ surface observations only. Comparing JRA-55 and JRA-55C enables us to clarify the impact of satellite data assimilation (SDA) on atmospheric reanalysis. In this study, we used the JRA-55 and JRA-55C to investigate the impact of inhomogeneity originating from SDA on the wind and wave climates.
2. Model, experimental design, and datasets
Significant wave height has been used as a descriptor of the state of ocean surface waves. However, significant wave height is not the sole variable used to characterize the wave climate. Other sea-state parameters are required to analyze the effects of climate change or wave energy resources, such as the wave period and wave direction. Wave energy is a better descriptor of storm strength than is significant wave height alone (Bromirski et al. 2013). Therefore, this study focused on the wave energy flux (WEF), which is the transport of energy by ocean surface waves. The WEF per unit crest length (kW m−1) is computed from the wave spectrum, as follows:
where Cg is the group velocity of waves, E(f, θ)is the spectrum of waves as a function of the frequency f and direction θ, ρw is the density of water, and g is the acceleration due to gravity. The WEF is computed using WAVEWATCH III (WW3; Tolman 2014), a third-generation wave model developed at the U.S. National Oceanic and Atmospheric Administration (NOAA)/National Centers for Environmental Prediction (NCEP). The model explicitly accounts for wind input, wave–wave interaction, dissipation due to whitecapping, wave–bottom interaction, and other shallow water processes, and solves the spectral action density balance equation for wavenumber–direction spectra. In this study, the input and whitecap dissipation source term developed by Ardhuin et al. (2010) is used. For the nonlinear wave–wave interaction, the discrete interaction approximation method (Hasselmann and Hasselmann 1985) is used, while the bottom friction is parameterized using the empirical JONSWAP model (Hasselmann et al. 1973).
The WW3 is configured at a grid spacing of 1.25° × 1.25° over almost the entire globe (78°S–78°N, 0°–360° in longitude). The frequency and direction of the wave spectrum is divided into 35 and 36 bins, respectively. The directions begin at 0° latitude and span 360° longitude clockwise in 10° increments. The frequencies begin with the lowest frequency at 0.0412 Hz and are logarithmically spaced with an increment factor of 1.1. The accuracy of the WW3 was validated against in situ wave observations obtained from the U.S. National Data Buoy Center (NDBC) (Sasaki 2014). In addition, Stopa et al. (2016) showed that all physical parameterizations for the wind input and whitecap dissipation in the WW3 performed well in terms of significant wave height. Therefore, the WW3 is capable of accurately simulating ocean surface waves.
To examine the impact of SDA on the wave climate, two numerical experiments were performed using WW3. In the first experiment (experiment A, hereinafter Exp.A), the wave model is forced by near-surface winds derived from the JRA-55 during 1958–2012 at 6-h intervals. JRA-55 is the second Japanese global atmospheric reanalysis covering the period from 1958 onward and is produced by applying four-dimensional variational data assimilation (4D-Var). The observations used in JRA-55 are those used in ERA-40 (Uppala et al. 2005) and those archived by the Japan Meteorological Agency. Specifically, JRA-55 assimilates conventional observations since 1958, such as surface observations from fixed land and sea stations and buoy observations; satellite observations since 1973, such as satellite radiances (e.g., infrared sounders, microwave sounders, and microwave imagers); atmospheric motion vectors (AMV) since 1979; and scatterometer ocean surface wind measurements [e.g., sea winds from QuikSCAT, ASCAT, and Active Microwave Instrument (AMI)] since 1997. In addition, radio occultation from the Global Navigation Satellite System has been used since 2006. Regarding the changes in the number of observations used for data assimilation in JRA-55, the global monthly mean counts of AMV increased gradually from 103 to more than 104 day−1 during the period 1979–2012. Scatterometer winds have been used for approximately 104 observations per day from 1997 onward. JRA-55 applies the variational bias correction (Dee and Uppala 2009) to satellite radiances. Its good performance in handling satellite radiances is reflected by the good agreement between the temporal variability of the stratospheric temperature in JRA-55 and that in microwave sounder measurements processed by Remote Sensing Systems. Regarding quality control of scatterometer ocean surface winds, poor-quality observed data are identified and excluded before the assimilation (e.g., because the data in heavy rain areas have less accuracy due to scatter noises by rain drops, the data flagged as rain are rejected). Refer to Kobayashi et al. (2015) and Ohhashi (2004) for further information regarding the observational data sources, quality control, and data selection for JRA-55.
The second experiment (experiment B, hereinafter Exp.B) is the same as Exp.A, but uses near-surface winds derived from JRA-55C to force the wave model. JRA-55C is a subproduct of JRA-55, assimilating only conventional observations from 1958 onward. See Kobayashi et al. (2014). The atmospheric model and data assimilation scheme used for JRA-55C are exactly the same as those used for JRA-55. Therefore, comparing Exp.A and Exp.B enables one to assess the impact of SDA on long-term trends in marine winds and waves. Satellite data may exhibit inhomogeneity attributed to changes in observational platforms, inaccuracy of the retrieval algorithms, and the calibration techniques. In this study, we investigated the combined effect of inhomogeneity in satellite data rather than the individual effects.
In this study, the region poleward of 30°–60°N (30°–60°S) is defined as the northern (southern) ocean, and the region between 30°S and 30°N as the tropical ocean. Linear trends were statistically tested by the Mann–Kendall test (Kendall 1975). The differences in mean values were statically tested by the Mann–Whitney test (Mann et al. 1947).
a. Validation of JRA-55C against similar products
ERA-20C (Poli et al. 2013) and 20CR, version 2 (20CRv2; Compo et al. 2011), are used to examine the validity of JRA-55C. ERA-20C is the atmospheric reanalysis of the twentieth century from 1900 to 2010 produced by the ECMWF. ERA-20C employs a coupled atmosphere–land surface–ocean wave model assimilating observations of surface pressure and surface marine winds only. The wave model used for ERA-20C is a derivative of the Wave Model (WAM) from Komen et al. (1994), with dependence of the surface roughness on the wave state (Janssen 1991). In comparison, 20CRv2 is a twentieth-century atmospheric reanalysis spanning 1871–2012 developed by the NCEP. The model assimilates only surface pressure reports and uses observed monthly sea surface temperature and sea ice distributions as boundary conditions. ERA-20C, 20CRv2, and JRA-55C integrate conventional observations only and are thus free from inhomogeneity resulting from the impact of SDA.
First, the trends of the annual mean near-surface wind speed from 1959 to 2012 derived from JRA-55C, ERA-20C, and 20CRv2 are compared. The JRA-55C trend of near-surface wind speed is similar to that of ERA-20C, specifically: upward trends in the midlatitude North Pacific and North Atlantic exceeding 0.1 m s−1 decade−1 and in the Indian Ocean and southern ocean exceeding 0.2 m s−1 decade−1, and a downward trend in the central-eastern equatorial Pacific (Figs. 1b,c). Meanwhile, 20CRv2 shows no major trend in near-surface wind speed in the midlatitude North Pacific and North Atlantic for the period 1959–2012 (Figs. 1d,e). Additionally, the upward trend of wind speed in the southern ocean from 20CRv2 is much greater than that seen in JRA-55C or ERA-20C (Figs. 1d,e). The discrepancy between 20CRv2 and the other two reanalyses may be due to the weaker model constraints for the former reanalysis compared with the latter two, as surface wind observations are not assimilated into 20CRv2.
To further compare the near-surface wind of JRA-55C with that of the ERA-20C, the temporal variation of the annual mean of wind speed averaged over the tropical ocean, northern ocean, and southern ocean is investigated. The annual mean wind speed in the three basins after the mid-1970s is greater than the climatological annual mean during 1959–72 in JRA-55C and ERA-20C (Fig. 2a; hereinafter we use the term “climatological annual mean” for the climatological annual mean for the period 1959–72 during which satellite data were not available). In detail, the average wind speed of the tropical ocean has been gradually increasing since the mid-1970s (black lines in Fig. 2a). The average wind speed of the southern ocean has also been increasing since the mid-1970s, although there is a bias in the wind speed between JRA-55C and ERA-20C (Fig. 2a). The average wind speed in the northern ocean has been strengthening since the mid-1970s, although there appears to be stagnation after the mid-1990s (Fig. 2a). Thus, JRA-55C long-term trends in near-surface winds appear to be qualitatively similar to those in ERA-20C.
Comparison of the wave component is made in terms of significant wave height, since WEF is not available output field in ERA-20C. The strengthening of the near-surface wind speed since the mid-1970s caused a significant rise in wave height (Fig. 2b). ERA-20C and Exp.B show increased significant wave heights in the three basins after 1980, compared with 1960–79 (Fig. 2b). Note, however, that there are some differences in the trends of significant wave height between ERA-20C and Exp.B. For example, ERA-20C shows a rising trend in significant wave height in the tropical ocean from 1980 to 2010, while Exp.B displays no marked trend in the significant wave height during the same period (Fig. 2b, top). This may be attributed to the difference between the wave model used for ERA-20C (WAM-based wave–atmosphere coupled model) and WW3. The magnitude of a rising trend in significant wave height in the southern ocean differs between ERA-20C and Exp.B (Fig. 2b, bottom). This may be due to the difference in the trend of the near-surface wind speed (Fig. 2a, bottom).
Consequently, the temporal variability in the near-surface wind speed and significant wave height in Exp.B is generally consistent with that of ERA-20C. Hereafter, the study focuses on comparing trends in the marine wind and wave using JRA-55 and JRA-55C. Since the atmospheric model and assimilation scheme used for JRA-55C are exactly the same as those used for JRA-55, a comparison between these data enables us to quantify the impact of SDA on the marine wind and wave climate. The use of the same model and assimilation scheme is important for the assessment because of concern that use of a different model–assimilation scheme may cause further inhomogeneity attributed to the bias.
b. Influence of satellite data assimilation on trends in marine winds and waves
The spatial pattern of the trend in annual mean near-surface wind speed from JRA-55 and JRA-55C is similar from 1959 to 2012 (Figs. 1a,b). However, the magnitude of the trends is lower in JRA-55 than in JRA-55C in the midlatitude North Pacific, midlatitude North Atlantic, and south of Australia (Figs. 1a,b). The difference in wind speed trends between JRA-55 and JRA-55C is seen more clearly in the trend for WEF and significant wave height (Figs. 3a,b and 5a,b). Figure 3b shows a spatial map of the linear trend in the annual mean WEF for the period 1959–2012 in Exp.B. There are marked upward trends in the WEF in the midlatitude North Pacific and midlatitude North Atlantic (>3.0 W m−1 decade−1) and southern ocean (>8 W m−1 decade−1), and those upward trends in Exp.A are lower than those in Exp.B (Figs. 3a,b). The trend in WEF is reduced by 70% and 18% by SDA in the midlatitude North Pacific and southern ocean, south of Australia, respectively, from 1959 to 2012 (Table 1).
To further understand the changes in WEF induced by SDA, the temporal variation of the WEF of Exp.A is compared to that of Exp.B. There is no significant change in the average WEF in the tropical ocean resulting from the impact of SDA after 1973; it varies from −5% to 5% (Fig. 4b). The impact of SDA is also small in the northern ocean from 1973 to the mid-1990s. However, the difference is enhanced by more than −5% after the mid-1990s (Fig. 4b). The average WEF in the southern ocean increased by approximately 5% from 1973 to the mid-1990s resulting from the impact of SDA, whereas it is reduced by SDA after the early 2000s (Fig. 4b). These results indicate that different inhomogeneity of SDA exists before and after the mid-1990s.
The impact of SDA on WEF is further investigated by splitting the period into two parts, 1973–94 and 1995–2012, with emphasis on the midlatitude North Pacific and southern ocean, south of Australia, where there are large differences in the annual mean WEF trends between Exp.A and Exp.B (Figs. 3a,b). Figures 3c and 3d show spatial maps of the trend in annual mean of WEF for the period 1973–94 in Exp.A and Exp.B, respectively. An upward trend of WEF in the midlatitude North Pacific, caused by increasing storm activity in the midlatitude North Pacific (Graham and Diaz 2001), is reduced 5.6% by SDA (Table 1). An upward trend of the WEF in the southern ocean, south of Australia, is also reduced by 15.8% (Figs. 3c,d; Table 1). Thus, SDA from 1973 to 1994 (i.e., the assimilation of satellite radiances) is found to reduce the trend of the WEF in the northern and southern oceans during that period.
On the other hand, there is a marked downward trend of WEF in the midlatitude North Pacific from 1995 to 2012 (Fig. 3f), attributed to strengthened trade winds and La Niña–like conditions in the tropical Pacific (Sasaki 2014). The downward trend in WEF from 1995 to 2012 is enhanced by SDA (Table 1; Fig. 3e). By contrast, upward trends of WEF exceeding 8 W m−1 decade−1 are reduced by SDA in the southern ocean, south of Africa and southwest of South America, from 1995 to 2012 (Figs. 3e,f). In particular, the trend of WEF south of Australia is considerably reduced. Thus, SDA after 1995 (i.e., the assimilation of the satellite radiances and scatterometer winds) is found to enhance (reduce) the trend of WEF in the northern (southern) ocean from 1995 to 2012.
These changes in the trend in WEF induced by SDA are explained by changes in significant wave height to a large extent (Fig. 5) and in wave period to a lesser extent (Fig. 6), since WEF is proportional to wave period and to the square of significant wave height.
Figure 7a shows a time series plot of the average WEF in the midlatitude North Pacific in Exp.A and Exp.B. As previously described, the impact of the assimilation of satellite radiances on the trend of WEF is small in the midlatitude North Pacific from 1973 to 1994, whereas the downward trend of the WEF is enhanced by SDA from 1995 to 2012 (Fig. 7a; Table 1). Note that a systematic bias in the WEF between Exp.A and Exp.B is enhanced gradually after the mid-1990s. As a result, the trend of WEF in the midlatitude North Pacific is reduced by about 70% from 1959 to 2012, which is mostly induced by the assimilation of the scatterometer winds after the mid-1990s. In the southern ocean, south of Australia, there is a bias in the WEF between Exp.A and Exp.B during 1973–94, while the trends during this period are not drastically changed by the SDA (Fig. 7b; Table 1). Conversely, SDA is found to induce a downward trend in WEF south of Australia from 1995 to 2012 (Fig. 7b). Thus, the trend in WEF in the southern ocean, south of Australia, decreases from 1959 to 2012, mainly due to the assimilation of the scatterometer winds after the mid-1990s.
This study investigated the impact of SDA on trends in near-surface wind speed and WEF from 1959 to 2012 by comparing two experiments using WW3, forced by near-surface winds derived from the JRA-55 (Exp.A) and JRA-55C (Exp.B); the impact of the presence or absence of SDA after 1973 was noted. The results showed that SDA impacted the trend of the annual mean WEF significantly. The upward trends in WEF in the northern and southern oceans during 1959–2012 were reduced by SDA. The reduced trend of WEF in the midlatitude North Pacific was mainly due to the assimilation of the scatterometer wind measurements into the JRA-55 after the mid-1990s, whereas the satellite radiance assimilation after 1973 had no major effect on the trend of the WEF in the midlatitude North Pacific. In contrast, an upward trend in the WEF was reduced in the southern ocean, south of Australia, from 1959 to 2012, mainly due to the assimilation of the scatterometer winds after the mid-1990s. These results demonstrate that SDA affected the trends of wind speed and WEF for two decades as well as for the last half century.
Our results suggest that atmospheric variables related to near-surface winds (e.g., the sensible, latent, and momentum fluxes) in JRA-55 need to be used with caution when long-term trends are investigated. Note also that JRA-55 and JRA-55C may have inhomogeneity due to the assimilation of conventional observation data as 20CR (Krueger et al. 2013).
This study focused on the impact of SDA on basin-scale wind and wave climate. The regional wave climate may also have changed, as coastal areas facing the open ocean are affected by waves propagating from offshore. This will be explored in future studies.
We thank three anonymous reviewers whose comments have led to a much improved manuscript. Wave hindcast experiments were performed on the SGI ICE X and UV1000 systems at the Japan Agency for Marine-Earth Science and Technology (JAMSTEC). This work was partially supported by JSPS KAKEN-HI Grants 25340131 and 16K00665.