3.1 Climatological biases and historical variations of air temperature and precipitation
Figure 2 shows biases of surface air temperature and precipitation in RCDSJRA-55 to the AMeDAS observation. Surface air temperature has cold and warm biases in January and August, respectively, especially in eastern and northern Japan (Fig. 2b, c). The annual mean temperature shows small biases (Fig. 2a) since the cold and warm biases in cold and warm seasons, respectively, are almost canceled out (Fig. 3a, b). Similar temperature biases are shown by DSJRA-55 (Kayaba et al. 2016).
The annual total precipitation shows small biases at most stations, with the exception of inland stations in central Japan (Fig. 2d). Total January precipitation, on the other hand, shows large positive biases over the inland areas along the Pacific Ocean and negative biases over the coastal areas along the Japan Sea (Fig. 2e). Note that the amount of winter precipitation along the Pacific Ocean is much smaller than that along the Japan Sea (figure not shown). The fact that the mountain ranges in the model are lower than the actual ones causes the biases of winter precipitation along the coasts of the Japan Sea and the Pacific Ocean, respectively (Ishizaki et al. 2012; Kawase et al. 2015). Moreover, the horizontal grid spacing of 5 km is too coarse to resolve the cumulus convections over the Japan Sea, which results in less precipitation over the coastal and more precipitation over mountainous areas along the Japan Sea (Kawase et al. 2019; 2020). In contrast, precipitation in August shows small biases over all of Japan. Only the inland stations in the central parts of Japan (Fig. 2f) show large positive biases, which corresponds to the prominent positive biases in annual total precipitation (Fig. 2d). As with the surface air temperature, the positive biases in cold seasons and negative biases in warm seasons are canceled out (Fig. 3c, 3d). Similar precipitation biases in January and August are shown by DSJRA-55 (Kayaba et al. 2016). The skills of RCDSJRA-55 are comparable to the previous dynamical downscaling datasets for climatology.
RCDSJRA-55 well reproduces the interannual variations of surface air temperature from 1959 to 2020 observed by the JMA in situ stations (Fig. 4a). Correlation coefficients are 0.981 and 0.992 in annual and 5-year running mean temperatures, respectively. A recent rapid warming since about 1980 is well reproduced by RCDSJRA-55. Linear trends of annual mean temperature are 1.25 ºC and 0.92 ºC per 60 years for the observations and the RCDSJRA-55, respectively, with a 99% confidence level. As with the annual mean temperature, the annual and 5-year running mean total precipitation show high correlations between RCDSJRA-55 and the observations (0.877 and 0.822, respectively). Annual total precipitation shows increasing trends in RCDSJRA-55 and the observation, although both are statistically insignificant.
Figure 4c shows the annual frequency of daily precipitation exceeding 100 mm/day (hereafter, referred to as nPR100). The nPR100 is underestimated by RCDSJRA-55 (Fig. 4c). The correlation coefficient (R) between RCDSJRA-55 and the observation is 0.638 for annual values and 0.649 for 5-year-running mean values. Both observation and simulation show increasing trends (0.34 and 0.23 days/60 years) with 95% and 90% confidence levels, respectively. The annual maximum daily precipitation (hereafter, referred to as Rx1d) also shows a similar interannual variation between the observation and RCDSJRA-55 (Fig. 4d). The correlation coefficients between observation and RCDSJRA-55 are 0.597 for the annual value and 0.726 for 5-year-running mean. Since Rx1d is an index of extreme events, the correlations are lower than those of the other indices in Fig. 4. Also, it is interesting to note that the variation of Rx1d is similar to that of nPR100 (Fig. 4c). The observed Rx1d shows an increasing trend (10.57 mm/60 year) with a 90% confidence level. The simulated Rx1d also shows an increasing trend, while it is not statistically significant.
Rx1d and nPR100 were remarkably underestimated by RCDSJRA-55 in 2004. In 2004, ten typhoons landed in Japan, which is the largest number of typhoons landing since 1951. Since we conducted one-year continuous simulations, typhoons sometimes move a little away from the observed tracks, which causes the underestimation of heavy rainfall induced by typhoons. In addition, two tremendously heavy rainfall disasters occurred in July 2004—the Niigata-Fukushima Heavy Rainfall of July 2004, and the Fukui Heavy Rainfall of July 2004 (Kato and Aranami 2005). These heavy rainfall events are caused by the quasi-stationary band-shaped precipitation systems, called “senjo-kousuitai” (Hirockawa et al. 2020; Kato 2020). It has been pointed out that a horizontal resolution of less than 2 km is required to resolve the senjo-kousuitai by the numerical model (Kato 2020), which also contributes to the underestimation of precipitation extremes in 2004.
3.2 Hourly precipitation extremes
The NHRCM with 5 km grid spacing can reproduce intense hourly precipitation that cannot be calculated by NHRCMs with 15 km and 20 km grid spacings (Sasaki et al. 2011; Yamada et al. 2018). The annual frequency of heavy hourly precipitation over 50 mm/h in RCDSJRA-55 is comparable to that observed by AMeDAS stations (Fig. 5). The variations of annual and 5-year running mean frequencies are similar to those observed (R = 0.584 and 0.715, respectively). Note that AMeDAS observation started in 1976. The linear trends show significant increasing trends with 95% and 99% confidence levels for observations and RCDSJRA-55, respectively.
Annual frequencies of 50 mm/h and their historical changes depend on the region in Japan. Figure 6 shows a time series of the backward 20-year running mean of hourly precipitation frequency over 50 mm/h simulated in RCDSJRA-55 in each region (Fig. 1c). The Pacific Ocean side of western Japan shows the largest frequency (Fig. 6a). The ocean and the Pacific Ocean side of eastern Japan follow the Pacific Ocean side of western Japan. The frequencies along the Japan Sea in eastern and northern Japan are smaller than those in other regions. Most regions show increasing trends, while the Japan Sea side in western Japan has no trend. The frequency anomaly from 1960 to 1979 has approximately doubled along the Japan Sea in northern Japan and has increased by approximately 1.5 times along the Pacific Ocean in northern Japan and the Japan Sea in eastern Japan (Fig. 6b, Table 2). Note that frequencies in northern Japan and along the Japan Sea in eastern Japan are much lower than those in other regions (Fig. 6a).
Table 2
Frequency of hourly precipitation over 50 mm and the ratio of 2001–2020 to 1961–1980. Asterisk indicates a significant difference with a 95% confidence level.
| 1961–1980 | 2001–2020 | 2001–2020/1961–1980 |
NJ–JS | 0.014 | 0.030 | 2.1* |
NJ–PO | 0.19 | 0.32 | 1.7* |
EJ–JS | 0.11 | 0.19 | 1.7 |
EJ–PO | 0.99 | 1.2 | 1.2 |
WJ–JS | 0.55 | 0.62 | 1.1 |
WJ–PO | 1.9 | 2.4 | 1.3* |
Sea | 1.5 | 1.8 | 1.2* |
3.3 Historical changes in snow depth and snowfall
As compared to the DSJRA-55, the biggest advantage of the RCDSJRA-55 is in evaluating historical snow cover with a 5 km grid spacing for all of Japan. First, we compare the reproducibility in annual maximum snow depth (hereafter, referred to as SDmax) in RCDSJRA-55 with in situ observations by the JMA. Figure 7 shows the time series of SDmax anomalies relative to 1991–2020 on the Japan Sea side, which has large amounts of snowfall. Note that most stations observing snow depth over the long term are located at low elevations (Fig. 1b). As with the previous figures, the grid points in RCDSJRA-55 near the snow observational stations are extracted and averaged in each region.
The observed SDmax shows decreasing trends in all regions. Eastern and western Japan show large decreasing trends (-15.0 and − 19.0%/10 years, respectively). The decreasing trend in northern Japan (-4.6%/10 years) is smaller than those in other regions. All decreasing trends are statistically significant with a 95% or 99% confidence level. RCDSJRA-55 also shows decreasing trends in eastern and western Japan (-10.0 and − 9.2%/10 years, respectively) with a 95% confidence level, although the decreasing trends are smaller than the observed ones. On the other hand, the decreasing trend is quite small and statistically insignificant in northern Japan. The difference between the SDmax before 1990 and after 1991 is relatively smaller in RCDSJRA-55 than that in the observation, resulting in the smaller decreasing trends in RCDSJRA-55.
3.4 Altitudinal dependency of historical snow cover changes
Snowfall and snow depth are strongly influenced by air temperature, especially at approximately 0 ºC, meaning that elevation is an important factor when considering historical changes in snowfall and snow depth. Climatology of SDmax exceeds 150 cm over the mountainous areas in eastern and northern Japan (Fig. 8a). Table 3 summarizes climatology of SDmax at each elevation range in each region. Above 1,000 m, SDmax exceeds 300 cm in northern Japan and exceeds 250 cm in eastern Japan. Both decreasing and increasing trends are simulated by RCDSJRA-55 (Fig. 8b), while most trends are statistically insignificant in each grid except for western Japan. We evaluated the annual variations of regional mean SDmax from low elevations (0–250 m) to high elevations (1000–1500m or over 1500 m). SDmax shows similar variations at all elevations on the Japan Sea side of northern Japan, and no significant trends were detected (Fig. 9a). On the other hand, on the Japan Sea side of eastern Japan, SDmax shows a decreasing trend more clearly at lower elevations, as with observations (Fig. 7), and no trend or increasing trends at higher elevations (Fig. 9c). In western Japan, all elevations show decreasing trends in SDmax, with a 95% confidence level. As stated in previous studies (Suzuki 2006; Ishii and Suzuki 2011; Kawase et al. 2012), SDmax can be largely decreased by historical warming at the lower elevations in eastern and western Japan where the surface air temperature is around 0 ºC even in winter.
Table 3
SDmax and Sx1d climatology and linear trends. One, two, and three asterisks mean a statistical significance with 90%, 95%, and 99% confidence levels, respectively.
Elevation(m) | SDmax climatology [cm] (1991–2020) | Sx1d mean climatology [cm] (1991–2020) |
NJ–JP | NJ–PO | EJ–JS | WJ–JS | NJ–JP | NJ–PO | EJ–JS | WJ–JS |
0–250 | 89.5 | 59.7 | 49.6 | 13.1 | 22.9 | 25.0 | 26.1 | 9.3 |
250–500 | 142.6 | 93.8 | 103.7 | 31.6 | 28.6 | 31.1 | 36.9 | 19.4 |
500–1000 | 230.5 | 157.7 | 161.1 | 54.7 | 36.3 | 37.2 | 40.6 | 25.4 |
1000–1500 | 326.6 | 320.9 | 275.4 | 41.8 | 43.3 | 44.4 | 45.6 | 26.9 |
1500– | | 667.2 | 443.0 | | | 63.4 | 50.6 | |
| SDmax trend [%/10 years] | S1d trend [%/10 years] |
0–250 | -0.92 | -1.02 | -8.70*** | -11.25*** | 0.11 | -0.32 | -2.24 | -6.97*** |
250–500 | -0.34 | -0.46 | -3.77 | -7.48** | 0.42 | 0.12 | -0.18 | -3.43** |
500–1000 | -0.36 | -0.20 | -0.98 | -5.90** | 0.52 | 0.26 | 0.79 | -2.42* |
1000–1500 | -0.37 | 1.00 | 0.04 | -7.62** | 1.40 | 0.96 | 1.46* | 0.12 |
1500– | | 2.40** | 1.10 | | | 2.23* | 1.39 | |
Generally, the total snowfall on the Pacific Ocean side is much less than that on the Japan Sea side. However, in northern Japan, the SDmax on the Pacific Ocean side is comparable to that on the Japan Sea side (Table 3). It is noteworthy that the trends and variations in SDmax strongly depend on the elevation on the Pacific Ocean sides of northern Japan (Fig. 9b), in contrast to that on the Japan Sea side (Fig. 9a). Over the Pacific Ocean side of northern Japan, extratropical cyclones passing around northern Japan cause heavy snowfall (Inatsu et al. 2021). In this synoptic condition, warm and moist air masses transport from the warm Pacific Ocean to northern Japan. On the other hand, heavy snowfall occurs on the Japan Sea side of northern Japan when cold air breaks out. Therefore, the air temperature on the Pacific Ocean side is relatively higher than that on the Japan Sea side when heavy snowfall occurs. This results in the difference in elevation dependency of SDmax between the Japan Sea and the Pacific Ocean sides in northern Japan.
Short-term heavy snowfalls bring risks of traffic hindrances, the isolation of villages through road closures, collapse of houses, and surface avalanches in mountainous areas. Compared with SDmax, the altitudinal dependency of annual maximum daily snowfall (hereafter, referred to as Sx1d) is smaller (Fig. 8c). Sx1d widely exceeds 20 cm in most areas of northern Japan and in some parts of eastern and western Japan, especially the areas facing the Japan Sea. Note that Sx1d over 20 cm is also found in the mountainous areas of western Japan facing Pacific Ocean. Heavy daily snowfall occurs over the mountainous areas in eastern Japan and the northern part of Japan (Hokkaido). Long term trends of Sx1d are similar to those of SDmax (Fig. 8d). Increasing trends are found more widely in eastern and western Japan. As with SDmax, most trends are statistically insignificant in each grid, except for the coastal areas along the Japan Sea in western Japan. Figure 10 shows the interannual variations and trends in annual maximum daily snowfall (Sx1d). Sx1d shows variations similar to those of SDmax among the elevations, as in the Japan Sea side of northern Japan. All elevations show increasing trends, especially at higher elevations above 1000 m (1.40%/10 years), although they are not statistically significant (Table 3). The Pacific Ocean side of northern Japan has large interannual variations of Sx1d, and the variations differ among the elevations (Fig. 10b). As with the Japan Sea side, all elevations except for 0–250m show increasing trends of Sx1d, and elevations higher than 1500m show a significant increasing trend with a 90% confidence level (Table 3).
Sx1d of eastern Japan has a clear elevation dependency, which is similar to those of SDmax (Fig. 10c). In contrast to SDmax, the decreasing trend at the lowest elevations is insignificant, and higher elevations show significant increasing trends with 90% and 95% confidence levels (Table 3). These results are consistent with the enhancement of daily snowfall extremes in the mountainous areas due to global warming (Kawase et al. 2016; Kawase et al. 2021). In western Japan, the decreasing trends of Sx1d are significant at elevations lower than 1000 m, but the decreasing trends are smaller than those of SDmax. At elevations higher than 1000 m, the Sx1d shows insignificant decreasing trends, although SDmax shows a significant decreasing trend.
3.4 Historical changes in snow-covered days
Finally, we investigate changes in the number of days with snow cover. Here, we define a snow-covered day as one in which the hourly snow depth exceeds 1 cm. Most areas in northern Japan and mountainous areas in eastern Japan are covered with snow for more than 100 days (Fig. 8c). Table 4 shows the climatology of snow-covered days from 1961 to 1990 and from 1991 to 2020 at four or five elevation ranges in each region. The climatological number of snow-covered days increases at higher latitudes and elevations. In contrast to SDmax and Sx1d, the number of snow-covered days has been reduced in most regions in Japan. (Fig. 8f and Table 4). In northern Japan, the decrease in snow-covered days in 1991–2020 is less than 5% relative to 1961–1990 at all elevations. In contrast, snow-covered days decreased by 18% and 32% at 0–250 m in eastern and western Japan, respectively. In western Japan, snow-covered days decreased by 16% even at elevations higher than 1000m. Our results indicate that rapid decreases in snow-covered days have already occurred in eastern and western Japan.
Table 4
Climatology of snow-covered days in each elevation. Round blanks show ratios of 1991–2020 relative to 1961–1990.
| NJ–JS | NJ–PO | EJ–JS | WJ–JS |
Elevation(m) | 1961–1990 | 1991–2020 | 1961–1990 | 1991–2020 | 1961–1990 | 1991–2020 | 1961–1990 | 1991–2020 |
0–250 | 139.4 | 133.3 (0.96) | 113.2 | 107.5 (0.95) | 71.7 | 58.6 (0.82) | 20.9 | 14.2 (0.68) |
250–500 | 158.9 | 154.5 (0.97) | 134.5 | 130.6 (0.97) | 108.4 | 98.3 (0.91) | 51.7 | 41.4 (0.80) |
500–1000 | 178.7 | 174.9 (0.98) | 164.9 | 161.7 (0.98) | 134.3 | 128.3 (0.96) | 76.8 | 66.0 (0.86) |
1000–1500 | 202.5 | 198.8 (0.98) | 220.2 | 218.5 (0.99) | 179.6 | 176.5 (0.98) | 102.2 | 85.8 (0.84) |
1500– | | | 350.2 | 349.7 (1.00) | 219.0 | 216.9 (0.99) | | |