In the present study, the winter precipitation in the vicinity of Japan is simulated by the Weather Research Forecasting model by using two sets of sea surface temperature (SST) data with different spatial resolutions. On comparing the simulated mean precipitations, we found that SST resolution has a significant influence on the simulated precipitation along the northwestern coast of Japan; in this region, the coarse-resolution SST data have a systematic cold bias. In the simulation using high-resolution SST data, the moisture supply to the atmosphere increases over the relatively warm coastal SST. The increase in the moisture supply leads to an increase in the moisture convergence near the mountain ranges in Japan on the Japan Sea side, leading to an increase in precipitation amount. The result suggests that coastal SST must be carefully used for dynamic downscaling of the climate simulation, in particular, in Japan, which is surrounded by boundary currents. We also found that a small-scale SST anomaly in the Kuroshio-Oyashio Extension (KOE) region near Japan enhances the interannual variance of local precipitation in the regions downwind of the SST anomaly. The associated anomalous ascent extends to the midtroposphere and is accompanied by an increase in cloud ice, suggesting that the interannual SST variation over the KOE region may affect the free atmosphere. Moisture budget analysis indicates the influence of moisture advection by mean wind on the spatial phase difference between the SST and precipitation anomalies.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The Kuroshio is the western boundary current of the North Pacific. Owing to this current, a large amount of heat is released into the atmosphere over the Kuroshio Extension (KE) region. Therefore the Kuroshio is important for maintaining the climate systems on the Earth. It has also been argued that sea surface temperature (SST) variation over the Kuroshio–Oyashio Extension (KOE) region may play an active role in driving climate variations in the North Pacific on a decadal time scale [Latif and Barnett, 1994; Nakamura et al., 1997]. Furthermore, it is documented that the maximum frequency of explosive cyclogenesis is observed in the vicinity of the warm water of the KE [Chen et al., 1992; Yoshiike and Kawamura, 2009]. This suggests that the SST in the KE region and the associated latent heat released into the atmosphere are important even for synoptic-scale extratropical cyclone development [Xu and Zhou, 1999].
 By using high-resolution satellite data, Nonaka and Xie  found that surface wind speeds increase (decrease) in association with warm (cold) SST anomalies of wavelengths less than 1000 km in the Kuroshio and KE regions. A similar positive correlation between the surface wind speed and the SST at oceanic fronts and eddies in other regions and the influence of the small-scale SST on the overlying marine boundary layer have been investigated in several studies [Chelton et al., 2004; Xie, 2004; Small et al., 2008]. Recent studies based on observations [Liu et al., 2007; Minobe et al., 2008; Kobashi et al., 2008; Tokinaga et al., 2009] and numerical models [Nakamura et al., 2008; Nonaka et al., 2009] further reveal that small-scale SST structures in oceanic frontal zones can affect the atmosphere above the boundary layer. However, it is not certain whether the small-scale SST structures over the KOE region affect local precipitation on an interannual time scale, because the precipitation in the KOE region is also controlled by the East Asian winter monsoon system [Yoshiike and Kawamura, 2009]. In the present study, we examine the influence of small-scale SST structures in the KOE region on precipitation by comparing the results of the simulations using coarse-resolution and high-resolution SST data, together with a regional atmospheric model.
 The present study also aims to examine the potential uncertainty in the changes in the winter climate in Japan observed in the simulations performed with a regional climate model (RCM) when the SST derived from climate models with a coarse resolution is used. With increasing public concern about climate changes, a regional assessment of climate change impacts has become an urgent task in various countries. Regional climate characteristics are influenced by the global climate change as well as by the local terrain and land use. Hence global climate models with a coarse resolution cannot be directly used to obtain high-resolution information required for the assessment in such areas as water supply and agriculture. Therefore climate change scenarios must be downscaled to the regional or local scale by statistical downscaling or high-resolution dynamical modeling using RCMs. Several previous studies have demonstrated that surface forcing caused by highly heterogeneous land surface conditions and steep orography of the land can help to improve small-scale climate features simulated by models [Antic et al., 2006]. However, very less attention has been paid to the role of the fine structures resolved in the high-resolution SST data in the climate simulations performed with RCMs. This is because the previous studies on RCMs have mainly focused on the climate changes in the continents.
 In winter, rainfall/snowfall often occurs over the central mountain ranges that are close to the Japan Sea, when cold and dry air masses outbreak from the Asian continent. While the air masses are crossing the Japan Sea, they absorb heat and moisture from the underlying relatively warm water in the Japan Sea. Subsequently, the modified air masses are forced to climb up the mountains of the Japan islands, causing precipitation. Therefore it is recognized that the SST of the Japan Sea substantially affects the winter climate in Japan [Manabe, 1957]. Chen et al.  and Yamamoto and Hirose  showed that the use of high-resolution SSTs based on satellite and assimilation data also using a high-resolution ocean general circulation model could help to improve the simulation of surface winds and cyclones over the Japan Sea. Thus the use of SST derived from global climate models with a coarse resolution may cause serious uncertainties in future regional climate change projections from RCMs. However, very few studies have investigated the manner in which high-resolution SST influences the RCM-simulated winter precipitation in Japan, which is surrounded by oceans. In the present study, we examine the potential impact of SST resolution on the RCM-simulated winter precipitation in Japan. The rest of this paper is organized as follows. Section 2 describes the model and the data. The results are presented in section 3 and discussed in section 4. Section 5 includes the summary of this study.
2. Model and Data Used
 The Advanced Research Weather Research and Forecasting (AR-WRF) model (version 3) (W. C. Skamarock et al., A description of the Advanced Research WRF version 2, 2005, http://wrf-model.org/wrfadmin/docs/arw v2pdf), which considers compressible nonhydrostatic equations, is used in the present study. The domain of the numerical simulation comprised the area around Japan divided into 121 × 121 grids with a spatial resolution of 20 km (Figure 1). Atmospheric radiative heating was calculated using the rapid radiative transfer model (RRTM) longwave parameterization [Mlawer et al., 1997] and the Goddard shortwave parameterization [Chou and Suarez, 1994]. The radiation physics was called only once every 30 min to reduce the computational time. The planetary boundary layer scheme, with countergradient terms to represent fluxes due to nonlocal gradients, was used; this scheme was developed at Yonsei University [Hong et al., 2006]. The land surface model (LSM) used in this study is called Noah. The modified version of the Kain-Fritsch scheme [Kain and Fritsch, 1990] was used to represent effects of convection on subgrid scales. The scheme proposed by Morrison et al.  was selected as the microcloud physics option.
 In the present study, the results of two experiments carried out with SSTs of different spatial resolutions are compared to examine the impact of small-scale SST structures on the atmosphere around Japan. In one experiment, the daily optimal interpolation SST (OISST) data [Reynolds et al., 2007] with a spatial grid resolution of 0.25° was used. In the other experiment, the reanalysis SST (RASST) data derived from the Japanese 25-Year Reanalysis Project (JRA25) [Onogi et al., 2007] for the period 1999–2004 and the Japan Meteorological Agency Climate Data Assimilation System (JCDAS) for the period 2005–2008 with a spatial resolution of 1.125° longitude by 1.125° latitude was used. The OISST data were obtained from the National Climate Data Center (ftp://eclipse.ncdc.noaa.gov/pub/OI-daily). For the period from 1985 to May 2002, the data include only the Pathfinder Advanced Very High Resolution Radiometer (AVHRR); after June 2002, the Advanced Microwave Scanning Radiometer (AMSR) was also included in the OISST data.
 The initial and boundary conditions of the model were derived from the JRA25 and JCDAS data for all experiments. Nine winters for the period from 1999–2000 to 2007–2008 were simulated, and a segmented integration of the models was conducted every year from 1 November to 1 March to reduce the computational burden. Note that winter is defined as the three months from December to the following February (DJF) in the present study. Hence the value of a parameter in winter is estimated as the 3 month average for the corresponding DJF period. Hereafter, the results obtained from the simulations using the RASST and the OISST data are referred to as the RASST-Run and the OISST-Run, respectively. The results of the experiments in the present study are not based on ensemble integrations, and the spectral nudging techniques [Von Storch et al., 2000] that help to reduce climate drift on a synoptic scale are not used. However, the differences in the results of the two experiments in terms of the temporal variation of winds, potential temperature, and humidity averaged over the model domain are negligibly small when compared to the variations in these variables under the corresponding experimental configurations (not shown). Thus the internal errors in both these experiments are assumed to be similar. Note also that the internal generated errors in RCMs over East Asia are relatively small during winter because stronger westerlies sweep away the internally generated variability within models [Giorgi and Bi, 2000]. In all simulations, the lateral boundary of the model represents the influence of the external large-scale atmospheric variability associated with the East Asian winter monsoon, which is strongly influenced by the Siberian high as well as the Aleutian low. Therefore we expect to isolate the atmospheric response to the local SST on an interannual time scale by considering the difference between the two simulations.
 To compare the results of the simulations with the observations, we used the QuikSCAT wind data obtained from the Remote Sensor Systems (RSS) (http://www.remss.com) and the Tropical Rainfall Measuring Mission (TRMM) 3b42 precipitation obtained from the Goddard Earth Science Data and Information Services Center (http://disc.sci.gsfc.nasa.gov). The spatial grid resolution of both the data sets is 0.25°. To validate the simulated wind, we used 10 min averaged surface wind speed data recorded at the Kuroshio Extension Observatory (KEO) buoy (32.4°N, 144.6°E). This buoy was deployed by the National Oceanic and Atmospheric Administration (NOAA) Pacific Marine Environmental Laboratory (PMEL), and the data can be downloaded from http://www.pmel.noaa.gov/keo/data.html. Merged sea surface height (SSH) derived from the Archiving Validating and Interpretation of Satellite Oceanographic (AVISO) data (ftp.cls.fr), which includes the data obtained from two satellites (Jason-2 Envisat, Jason-1 Envisat, or TOPEX POSEIDON ERS), is also used to examine the influence of the ocean on the interannual SST variability.
3.1. Difference in SST Fields
Figure 2 shows the mean SST field and the standard deviation of the RASST and the OISST data along with the difference between the mean SST fields of the OISST and RASST data for nine winters from 1999–2000 to 2007–2008. With respect to the mean SST field, the difference between the OISST and RASST is most notable along the northwestern coasts of the Japanese islands; here the mean SST of the OISST is warmer than that of the RASST. This difference indicates that the penetration of warm water due to the first branch of the Tsushima Warm Current along the western coasts of the Japanese islands [Kawabe, 1982a, 1982b] can be inferred from the OISST but not from the RASST. The difference in the mean SST between the OISST and RASST data for the eastern coast of the Korean Peninsula is attributed to the third branch of the Tsushima Warm Current, which is known as the East Korean Warm Current. Another significant difference is observed around the subpolar front (SPF), which represents the thermal boundary between the cold water to the north and the warm water supplied by the Tsushima Warm Current [Matsumura and Xie, 1998; Isoda, 2003].
 The difference between the mean SSTs of the OISST and RASST data sets is associated with the difference in SST gradients of both these data sets. The difference in SST gradients between the OISST and RASST is observed along the path of the Kuroshio Current to the south of Japan and in the KE region (Figure 3). The mean SST gradients of these two data sets differ considerably around the SPF (not shown). Thus the sharp SST gradients at oceanic fronts are better represented by the OISST than by the RASST. The above-mentioned differences essentially result from the difference between the spatial resolutions of the two data sets, and the spatial pattern of the low-pass-filtered OISST is similar to that of the RASST (not shown).
 There is a remarkable difference between the two data sets in terms of the SST variance on an interannual time scale for the KOE region. The magnitude of the standard deviation of the OISST is considerably larger than that of the RASST; this difference is noted particularly near Japan, in the region that lies along 37°N and is north of the KE front that is indicated by the 18°C isotherm (Figures 2a and 2b). However, no significant difference is found between the mean SST fields of the two data sets in this region (Figure 2c). Note that the trend in the time series of the merged OISST used in the present study that is averaged over the area represented by a box in Figure 2c is similar to those of the OISST derived from only AVHRR and the TRMM Microwave Image (TMI) SST, although a small difference is observed between the merged OISST and the TMI SST in 2001–2002 (not shown). Thus the larger standard deviation of the OISST observed near Japan in the region along 37°N is considered to be independent of the satellite measurements.
Figure 4 shows the SST anomalies and SSH in individual winters. The warm SST anomalies in the KOE region near Japan and along 37°N in 1999–2000, 2000–2001, 2001–2002, and 2006–2007 are apparently related to the meridionally elongated meanders of the KE or the warm-core rings pinched off the meanders. The warm-core rings generally move westward and accumulate in the KOE region near Japan, resulting in local SST warming along the eastern coast of Japan. Thus the large amplitude of the interannual variation of SST in the KOE region near Japan may be attributed partially to oceanic mesoscale eddies. The path of the upstream KE is stable without warm-core rings in 2002–2003, 2003–2004, and 2005–2006 when cool SST anomalies are observed to the north of the KE front and sharp meridional SST gradients are observed at the KE front (Figure 3). The first branch of the Oyashio Current flowing southward along the coast of Japan may contribute to the interannual SST cooling near Japan because the intensified Oyashio Current enhances cold advection, which is accompanied by the southward migration of the subarctic front; this results in the cool SST in the KOE region [Nonaka et al., 2006].
 A large interannual variance is observed in the OISST along the SPF over the Japan Sea, but corresponding signals are not observed in the RASST. The spatial and temporal variability of the SPF on an interannual time scale may be attributed to both oceanic mesoscale eddies and atmospheric wind forcing [Park et al., 2007]. The above-mentioned differences between the standard deviations of the SST data sets are mainly significant in the region where ocean dynamics plays an important role in maintaining SSTs.
3.2. Mean Fields
Figures 5a and 5b show the 9 year average winter surface wind fields of the models in which the RASST and the OISST data are used, respectively. In addition to these, the surface wind fields of the JRA25/JCDAS and the QuikSCAT are shown for comparison. Mountains higher than 1000 m are present along the Russian coast, with an orographic gap at Vladivostok (Figure 1). This land topography affects the surface winds over the Japan Sea in winter [Nagata et al., 1986; Nagata, 1991]; this leads to further cooling of the water to the north of the SPF [Kawamura and Wu, 1998]. Intense winds are observed off Vladivostok, while wind shadows are observed on the southern side of the Hamgyong mountain ranges that extend for more than 200 km in the northern part of the Korean Peninsula. This remarkable contrast is observed in the wind speeds simulated by the regional models. Furthermore, the gap winds along the southern coast of Japan are captured in both simulations. Thus the RCMs can simulate the orographically induced local-scale surface wind fields over coastal oceans, as has been revealed by the comparing of the wind fields of the JRA25/JCDAS with relatively coarse topography. The comparison of the simulation with the surface wind speed observed at the KEO buoy revealed that the RCMs reasonably simulate the temporal variation of the surface wind speed (Figures 6a and 6b and Table 1). However, the surface winds in both simulations are weaker than the QuikSCAT winds (Figure 5). This discrepancy may result from low frequency of the intense winds simulated by the models. The low frequency may be due to the insufficient resolution of these models. The discrepancy also results from the fact that low wind speeds are not considered in the QuikSCAT data used in the present study (Figure 6c). Further, the relatively strong model wind speed off the mountain ranges of the Sikhote-Alin mountain ranges may be due to the bias in the large-scale fields used as the lateral boundary conditions.
Table 1. Mean Bias, Root-Mean Square Difference, Root-Mean Square Difference Minus Mean Bias, and the Correlation Coefficients of Wind Speed Obtained From Satellite Data and the Two Simulations With the NOAA PMEL KEO Buoy Dataa
Unit is m/s. Daily averaged data derived from 10 min average KEO buoy wind and model winds sampled every 1 h are used to estimate the statistical values. RMSD, root-mean square difference; RMSD (-Bias), root-mean square difference minus mean bias; Corr. Coeff., correlation coefficient.
December 2006 to February 2007
December 2007 to February 2008
Figure 7 shows the mean precipitation field and the standard deviation for the RASST-Run, OISST-Run, and TRMM. The corresponding fields obtained from the simulation (hereafter referred to as the MEAN-OISST-Run), in which the 9 year average winter OISST was used as the lower boundary condition, are also shown; the initial and lateral boundary conditions in this simulation are the same as those for the OISST-Run. All the simulations capture the climatology feature, i.e., less precipitation on the Pacific Ocean side of Japan and more snowfall or precipitation on the Japan Sea side. However, the mean precipitation has been overestimated in all simulations over the entire model domain; this can be observed by comparing the simulated mean precipitations with the TRMM. However, the reason for the overestimation in the models is unclear. Furthermore, all the simulations have the precipitation bias over the Japan Sea to the east of the Korean Peninsula. This bias may be due to the convergence related to the excess weakening of the wind speed to the south of the Hamgyong mountain ranges, as observed in the models (Figure 5). The local maxima of the precipitation along the western coast of Japan are related to the local topography and are common biases in all simulations. However, the spatial patterns of the mean precipitation fields in all the simulations are similar.
3.3. Difference in Mean Fields
 To detect the impact of the spatial resolution of the SST data on the mean fields of the simulated precipitation in Japan, the difference between the mean fields in the OISST-Run and the RASST-Run is shown in Figure 8. The mean precipitation amounts in the OISST-Run and the RASST-Run differ significantly along the northwestern coasts of the Japanese islands, particularly in the region north of 35°N (Figure 8a). The precipitation amounts in the OISST-Run exceed those in the RASST-Run by approximately 10%. Along the northwestern coastal ocean, the latent heat flux to the atmosphere is greater and the surface wind speed is stronger in the OISST-Run than in the RASST-Run (Figures 8b and 8c). The increase in the latent heat flux and the strength of the northwesterlies is favorable for the enhancement of moisture flux convergence, which would result in increased precipitation in Hokuriku (Figure 1).
 To understand the reason for the difference between the mean precipitation amounts along the northwestern coastal region of Japan in the OISST-Run and RASST-Run, we examine the moisture budget. The moisture budget equation can be written as follows:
where pt is pressure at the model top level; ps is the model surface pressure; g is the acceleration due to gravity; q is the moisture; V is the horizontal wind vector; E is the evaporation; and P is the precipitation. Equation (1) states that the time rate of change of the total precipitable water in the column (W) is equal to the difference between evaporation and the sum of precipitation and the vertically integrated moisture flux divergence (the first term on the right-hand side). In our analysis, all the terms except P in equation (1) are computed from the instantaneous model output sampled every 6 h. Note that exact balance cannot be achieved because of a small horizontal diffusion and the error due to the sampling interval. This residual value is diagnosed from all the terms in equation (1).
 The land and ocean areas near Hokuriku for calculating the mean moisture balance are shown in Figure 1. The precipitation over Hokuriku in both the simulations shows similar interannual variability associated with the interannual variability of the East Asian winter monsoon; however, the interannual variability of the difference between the precipitation in the two simulations differs from the interannual variability of the East Asian winter monsoon and is similar to the variability of the difference between the evaporation in the two simulations over the ocean along the northwestern coast of Japan (not shown). Further, the interannual variability of the difference between the precipitation over Hokuriku in the two simulations is smaller than the magnitude of the difference in the 9 year average winter precipitation. Thus the difference between the two simulations in terms of the 9 year average winter moisture balance over the land and ocean near Hokuriku is shown in Figure 9. It is noted that since the difference in the time rate of change of the total precipitable water between the two simulations is considerably smaller than the other terms, it is not shown. The greater precipitation over the land in the OISST-Run is mainly due to an increase in the moisture flux convergence (Figure 9a). The contribution from the evaporation over the land is negligible. The increase in the moisture flux convergence over the land is mainly related to an increase in the moisture flux divergence over the ocean, which is balanced by an increase in the evaporation (Figure 9b). The enhanced evaporation is explained mostly by the difference in sea surface specific humidity between the OISST and RASST data sets, and hence by the SST difference. Thus it is suggested that the difference in the simulated mean precipitation along the northwestern coast of Japan is caused by the SST difference along the coast between the two data sets.
 Further, less precipitation is observed near the SPF in the OISST-Run (Figure 8a), accompanied by the higher sea level pressure (SLP) (Figure 8d). The enhanced 850 hPa poleward eddy heat flux, which is a measure of baroclinicity, is also observed in the region to the south of the SPF (Figure 8h). Note that the eddy heat flux is computed by using the eddy components of meridional wind and temperature, which are obtained by applying a high-pass filter with a cutoff period of 8 days to the 6-hourly outputs. To determine whether the differences between the two simulations in terms of the SLP and baroclinicity around the SPF influence the difference in the precipitation along the northwestern coast along the Japan Sea, the results of other simulations were used for comparison. In one simulation, the 9 year average winter SSTs of the OISST were used. In another simulation, only the SST around the SPF was replaced by the 9 year average winter RASST to eliminate the frontal structure of the SPF; however, the initial and lateral boundary conditions were the same as those in the OISST-Run. The difference between the precipitation amounts in the two simulations is insignificant over the land along the northwestern coasts of the Japanese islands, and the differences in the SLP and baroclinicity around the SPF are almost same as that between the OISST-Run and the RASST-Run (not shown). This suggests that the difference between the coastal SSTs in the OISST and RASST data sets mainly affects the difference between the precipitation amounts in these simulations along the northwestern coasts of the Japanese islands.
 The mean precipitation amounts in the OISST-Run and RASST-Run also differ in the vicinity of the Kuroshio Current, in the region to the south of Kyushu and Shikoku. This difference accompanies the difference in surface wind convergence (Figure 8e) which closely resembles the pattern of the Laplacian of the SLP and SST differences (Figures 8f and 8g). These features suggest that pressure adjustment to the SST gradients may cause the difference in the surface wind convergence and precipitation [Minobe et al., 2008]. These relationships between the SST gradients and the atmospheric fields are more clearly observed, even over the KE region in 2003–2004 (Figure 10) when there is a notable difference in the SST gradients in this region (Figure 3), than the difference in the mean fields. An increase in 850 hPa poleward eddy heat flux is observed around the KE front (Figure 10f). However, the corresponding difference in the mean field over the KE region is statistically insignificant (Figure 8h) presumably because of the relatively small difference between the 9 year average winter meridional SST gradients at the KE front in the OISST and the RASST (Figure 3j). Note that the present results of the simulations are not based on ensemble integrations and hence the difference in the mean eddy heat flux may be underestimated.
3.4. Interannual Variability Over the KOE
 The notable difference in standard deviation of the precipitation field between the OISST-Run and the RASST-Run is observed in the region to the north of the KE front, around 145°E (see Figures 7a and 7b). The precipitation in the OISST-Run shows a large interannual variation over the region to the north of the KE front and to the east of the region with the largest SST variance (Figure 2); a relatively small variance is observed in the RASST-Run. A small precipitation variance is observed in the MEAN-OISST-Run (Figure 7c), as in the case of the RASST-Run. Figure 11 shows the time series of winter precipitation amount in the OISST-Run, RASST-Run, MEAN-OISST-Run, and TRMM, averaged over the region denoted by a box area in Figure 7; in this region, the standard deviation of precipitation in the OISST-Run is relatively large. The time series of the KOE-SST index that is defined here as the SST anomalies averaged over the KOE region near Japan (Figure 2b) is also shown.
 The precipitation amounts in the OISST-Run and RASST-Run tend to increase (decrease) when the SST over the KOE region near Japan is warmer (cooler) than normal, in general agreement with TRMM. The variability in the precipitation amounts in the OISST-Run is overestimated when compared to the TRMM, while that in the RASST-Run is slightly underestimated. The correlation coefficient of precipitation amount in TRMM with the OISST-Run is 0.71, which is lower than that with the RASST-Run (0.78), at 95% confidence level. On the other hand, the correlation coefficient of precipitation amount in the MEAN-OISST-Run with TRMM is 0.19, and the variability in the precipitation amount is considerably less than that in TRMM. Although the variation of precipitation amounts in the simulations somewhat differs from that in the TRMM, the difference among the simulations in terms of the amplitude of interannual precipitation over the KOE region seems to be linked with the difference among the SST anomalies in the KOE region in the simulations (Figure 11b). Thus it is implied that the SST anomalies in the KOE region near Japan may enhance the interannual variability of the local precipitation in the region to the east of the SST anomalies.
 To clarify the relationship between the SST and precipitation anomalies in the KOE region on an interannual time scale, we examine the regression maps of the difference in the atmospheric fields between the OISST-Run and the RASST-Run against the difference between the KOE-SST indices in the OISST and the RASST. The precipitation anomaly is observed to the east of the SST anomaly in the KOE region (Figures 12a and 12b). Furthermore, the precipitation anomaly coincides with the locations of the SLP anomaly and the convergence of surface wind anomaly in the KOE region (Figures 12d and 12e). The wind speed anomaly is roughly in phase with the warm SST anomaly (Figures 12a and 12c).
 The difference in atmosphere response between the two simulations is also found above the boundary layer. Figure 13 shows the vertical section of the difference in the atmospheric fields between the OISST-Run and the RASST-Run regressed on the difference between the KOE-SST indices in the simulations. An increase in convergence around 145°E associated with enhanced wind speed below 900 hPa is observed in the region of the SST anomaly (Figure 13a); this is accompanied by an increase in the equivalent potential temperature below 800 hPa (Figure 13b) and the cloud-liquid water mixing ratio around 800 hPa (Figure 13c) to the east of the region of the SST anomaly. Interestingly, the anomalous convergence is further linked to the anomalous ascent extending toward midtroposphere around 500 hPa, accompanied by an increase in the cloud ice (Figure 13d). Figure 13 suggests that the SST in the KOE region could affect the atmosphere not only within the boundary layer but also in the free atmosphere on an interannual time scale.
 To further understand the relationship between the SST and the precipitation anomalies in the KOE region on an interannual time scale, we examined the moisture budget in the KOE region. Here the time average difference between the moisture fluxes in the OISST-Run and the RASST-Run is decomposed as follows:
Here the overbar indicates the time average of the corresponding quantity, and the prime indicates deviation from this time average. In the present study, the 3 month average for the December–January–February (DJF) period is simply used as the period of a time average. Δ represents the difference between the values in the OISST-Run and the RASST-Run. Equation (2) is further decomposed as follows:
Here the subscript R stands for the RASST-Run. The first, second, and third terms on the right-hand side are the contributions from the difference between the mean wind, mean moisture, and both mean wind and moisture, respectively, in the OISST-Run and the RASST-Run. The last term represents the contribution of the transient components.
Figure 14 shows the differences between the two simulations in terms of the contribution of each term to the precipitation anomalies in the KOE region to the north of the KE front, regressed on the difference between the KOE-SST indices in the OISST and RASST. The difference between the interannual precipitation amounts in the two simulations is mainly due to the difference between the simulations in terms of evaporation, which in turn depends on the difference between the corresponding SSTs. This is because the difference in evaporation is explained mostly by that in sea surface specific humidity (not shown). However, the spatial phase difference between the SST and precipitation anomalies in the two simulations cannot be explained by the difference in evaporation. This spatial phase lag is caused by the difference in moisture flux convergence between the two simulations. A large contribution by the moisture flux convergence to the precipitation difference between the two simulations is observed to west as well as east of the location of the maxima of the precipitation anomalies. The anomalous divergence of moisture flux contributes to a decrease in the precipitation on the western side of the evaporation anomalies, while the anomalous moisture flux convergence on the eastern side results in an increase in precipitation.
 The difference between the moisture flux divergences in the simulations estimated from the time average moisture and wind contributes to the difference between the time average moisture flux divergences. The contribution by the transient components shows the opposite tendency to the difference between the time average moisture flux convergences in the simulations on the eastern side of the precipitation anomalies and is negligible on the western side (see also Figure 12f). This implies that a sharp meridional SST gradient at the KE front can enhance the transient components because the meridional SST gradient is relatively weaker during winter when the KOE-SST index is positive (Figure 3). The anomalous mean moisture flux divergence in the western side results mainly from the differences between the mean moisture in the two simulations, while the contribution of the difference between the mean winds in the two simulations results in anomalous mean moisture flux convergences on the downwind side (Figure 14b). This suggests that the moisture advection by mean wind strongly influences the spatial phase lag between the SST and precipitation anomalies [Small et al., 2003].
 Various mechanisms have been proposed to explain the relationship between the surface wind changes within the atmospheric boundary layer and the SST [Small et al., 2008]. In the present case, the SLP anomalies are observed to the east of the SST anomaly. This spatial phase relationship is inconsistent with the mechanism proposed by Lindzen and Nigam , in which the thermodynamically induced sea level pressure gradients below the atmospheric boundary layer that are associated with SST variations induce surface wind changes. The spatial phase relationship between the surface wind speed and the SST anomalies in the KOE region to the west of 145°E is roughly in phase, which is consistent with the momentum mixing mechanism [Wallace et al., 1989]. In the present study, however, the atmospheric response to the SST is observed even above the boundary layer. The location of SLP anomaly coincides with that of the precipitation anomaly. The wind anomalies converge at the anomalous SLP minima. Thus the anomalous circulation on the downwind side of the SST anomaly in the KOE region near Japan partially appears to be a consequence of the influence of the pressure gradient [Small et al., 2003].
 In the present study, we examined the impact of small-scale SST structures on the atmosphere by comparing the results of the two simulations in which the two sets of SST data with different spatial resolutions were used. The simulation results suggest that small-scale SST anomalies in the KOE region near Japan may enhance the interannual variability of the local precipitation in the areas downwind of the SST anomalies. To confirm this, we computed the regression coefficients of surface wind speed, surface wind convergence, and precipitation anomalies in the observations based on satellite measurements with the KOE-SST index based on the OISST, as shown in Figure 15. As shown in Figure 12, the observed surface wind speed anomalies are roughly in phase with the SST anomalies in the KOE region near Japan (Figure 15a). The anomalous surface wind convergence and increased precipitation amounts are observed to the east of the region with the SST anomalies (Figures 15b and 15c). This relationship in observations partially supports the results of the present study (Figure 12). Furthermore, it is suggested that the SST anomaly in the KOE region near Japan is induced by ocean dynamics; this is inferred from the significant positive correlation between the SST and the wind speed anomalies in the KOE region near Japan. In fact, Kako and Kubota  have shown that a change in the mixed layer temperature in the KOE region is mainly induced by the lateral heat flux.
 However, the intensification of the East Asian winter monsoon is accompanied by the equatorward extension of the storm track axis in the western North Pacific; the axis tends to shift poleward as the East Asian winter monsoon weakens [Yoshiike and Kawamura, 2009]. The East Asian winter monsoon is relatively weak in the years when the KOE-SST index is positive, while it is often stronger in the years when the SST in the KOE region near Japan is cooler than the average (Figure 11b). Hence the spatial pattern of the observed precipitation anomalies in the KOE region regressed on the East Asian winter monsoon index is similar to that of the observed precipitation anomalies regressed on the KOE-SST index (Figures 15c and 15d). Therefore analysis of the magnitude of the contributions of the small-scale SST anomalies in the KOE region to the actual observed interannual variability of precipitation must be carried out by using long-term high-resolution observational data.
 In this study, we focused on the interannual variability of precipitation in the region north of the KE front; this interannual variability is mainly affected by the small-scale SST anomalies induced by the oceanic eddies. On the other hand, Tokinaga et al.  have shown that in winter, clouds tend to develop on the warm flank of the KE front and reach the midtroposphere. Furthermore, a sharp meridional SST gradient enhances the baroclinicity of eastward moving disturbances [Hoskins et al., 1985; Nakamura et al., 2008]. The results of the present study also suggest that a sharp SST gradient in the KE region has the potential to enhance the precipitation on the warm flank of the KE front (Figure 10a). Hence it is necessary to identify the relationship between the SST gradients at the KE front and the precipitation in the KOE region on an interannual time scale.
 The results of the present study also suggest that the amount of winter precipitation along the northwestern coasts of the Japanese islands is sensitive to the neighboring coastal SST. Recent advanced microwave remote sensors can be used for SST measurements even in the presence of clouds, and therefore global SST and surface wind field data can be continuously recorded. However, the accuracy of the microwave remote sensors close to a coast is still limited. Thermal infrared sensors, on the other hand, can be used to measure the SST near the coasts, although they are not well suited for obtaining temporal and spatial high-resolution SST data in cloudy and rainy weather. In the present study, although the OISST, which included data from the AVHRR and AMSR, was used as the lower boundary condition in the OISST-Run, the SST in the regions near the western coast of Japan were mostly derived from the AVHRR data; hence the temporal resolution of the data may be insufficient. Therefore for quantitative confirmation of the present results, it is necessary to carry out sensitivity experiments in which high-resolution SST derived from assimilation data obtained by using a high-resolution ocean general circulation model as well as satellite measurements [e.g., Yamamoto and Hirose, 2007] are used or experiments in which RCMs are coupled with ocean models [e.g., Sasaki et al., 2006].
 In the present study, we examined the influence of small-scale SST anomalies on winter precipitation near Japan by comparing the results of simulations in which coarse and high-resolution SST data are used along with a regional atmospheric model. Two SST data sets with horizontal resolutions of 1.125° and 0.25° were used as surface boundary conditions over oceans; the initial and lateral boundary conditions used in both simulations were the same.
 The difference in the SST resolution has a significant influence on the simulated mean precipitation along the northwestern coast of Japan. The difference in the mean precipitation between the two simulations is induced by the difference in the coastal SST along the Japan Sea sides of the Japanese islands, where the SST with a coarse resolution has a systematic cold bias. This is because the effect of the intrusion of warm water carried by a branch of the Tsushima Warm Current is not captured. In the simulation performed using high-resolution SST data, the moisture supply into the atmosphere increases in response to the relatively warm SST. This results in an increase in the precipitation amounts along the Japan Sea sides of the Japanese islands.
 The difference in the SST resolutions also influences the interannual variance of the local precipitation over the KOE region. The warm (cold) SST anomaly in the KOE region near Japan, caused by ocean processes, induces an increase (decrease) in the local precipitation to the east of the SST anomaly. This increase is accompanied by the enhancement of surface wind convergence (divergence). Furthermore, the anomalous ascent associated with the enhanced surface wind convergence in response to the warm SST anomalies extends to the midtroposphere up to around 500 hPa; this is accompanied by an increase in the cloud ice. This suggests that the SST in the KOE region can affect the atmosphere above the planetary boundary layer.
 The moisture budget indicates that the quantitative difference between the simulated interannual variability of precipitation amounts in the KOE region is mainly caused by the differences in the local evaporation through the differences in the SST between the two simulations. However, the spatial phase difference between the SST and precipitation anomalies cannot be explained by the difference in evaporation. The anomalous moisture flux divergence on the upwind side caused by the eastward advection of moisture by the mean westerly winds and the associated anomalous convergence on the downwind side of the SST anomalies seems to be responsible for this spatial phase lag between the SST and precipitation anomalies.
 The results of the present study suggest that winter precipitation amounts along the northwestern coasts of the Japanese islands are sensitive to the neighboring coastal SST. Despite an increase in computational resources, the horizontal resolutions of most global climate models, from which the SSTs to be used in RCMs for projecting future regional-scale climate scenarios are derived, are insufficient for resolving the spatial structure of SST along coastal areas. Therefore the use of SSTs derived from coarse global climate models for dynamical downscaling cause uncertainties in the results of climate simulations performed using RCMs. This is particularly true for coastal areas located downwind of the mean flow. The importance of the use of high-resolution SST data for climate simulations has been demonstrated by Lenderink et al. , who focused on the simulation of summer coastal precipitation in the Netherlands. The results of their study and the present study suggest that the coastal SST must be used very carefully for dynamical downscaling of the climate in coastal regions.
 I would like to thank R. Kawamura and T. Matsuura of the University of Toyama, M. Seki of RESTEC, I. Takayabu of MRI, K. Dairaku of NIED, and S.-I. Kako of the University of Ehime for their valuable comments. The comments and suggestions of the referees who perused the earlier version of the paper helped achieve a significant improvement of the manuscript. This research was supported partially by the Global Environment Research Project Fund (S-5-3) of the Ministry of the Environment, Japan, the Study on Long-term Projection of Typhoon Disasters of NIED and grants-in-aid (22310111) from the Japanese Ministry of Education, Science, Sports, and Culture.