The snow‐shadow effect of Sado Island on Niigata City and the coastal plain

Japan's Hokuriku region, near the Sea of Japan, typically experiences heavy snowfall; however, Niigata City, the largest city on the Sea of Japan side, experiences lower levels of snowfall than neighbouring cities. This study investigates the snow‐shadow effect of Sado Island on snowfall in Niigata City, located 45 km away leeward. Statistical analysis of long‐term radar data for 10 winters showed that snow‐shadow effects in the Niigata plain occurred in 151 (80%) of the 188 cases, during which a strong approaching wind reached the island. The location of this snow‐shadow effect was always downwind of Sado Island and depended on the wind direction. Numerical experiments using the Weather Research and Forecasting model predicted that snowfall over the Niigata Plain would be lighter with the island than without it. Additionally, the snow‐shadow effect occurs in areas more than 150 km downwind. The experiments showed that Sado Island reduces heat fluxes from the sea surface by weakening leeward winds. At the same time, the horizontal wind convergence downwind is weakened. Meanwhile, the orographic snowfall over Sado Island reduces the amount of water vapour, cloud water and cloud ice over the leeward sea. Therefore, Sado Island prevents cloud lines from redeveloping over the leeward sea and can further reduce snowfall over the leeward plain, including in Niigata City.


| INTRODUCTION
The Hokuriku region, situated on the Sea of Japan side of the country, often experiences heavy snowfall despite its location at approximately 38 N latitude (e.g., Steenburgh & Nakai, 2020;Yoshino, 1977).This is due to the air mass transformation effect caused by the cold north-westerly monsoon passing over the warmer Sea of Japan, leading to the development of cumulonimbus clouds and resulting in snowfall (e.g., Kato & Asai, 1983;Manabe, 1957;Nakamura & Asai, 1985;Ninomiya, 1964).The snowfall mechanism in this region is similar to that observed along the Great Lakes coast of North America (e.g., Andersson & Gustafsson, 1994;Niziol et al., 1995;Passarelli Jr & Braham Jr, 1981).
Although the Hokuriku region typically experiences heavy snowfall, Niigata City, located on the Niigata Plain and the largest city on the Sea of Japan side, experiences less snowfall than other cities (Figure 1; e.g., Matsumoto, 1967;Veals et al., 2019).One possible explanation for Niigata City experiencing less snowfall might be due to its nature being the largest plain on the Sea of Japan side and far from the mountains.Additionally, the city is outside the area where the Japan-Sea Polar-airmass Convergence Zone reaches a high frequency (e.g., Shimizu et al., 2017;Shinoda et al., 2021).It remains unclear whether additional factors affect snowfall levels in this area.
Figure 2 shows a weather chart, winds at the 850 hPa level, and precipitation distribution on a typical day with little snow over most of the Niigata Plain.Matsumoto (1967) and Yagi and Uchiyama (1983) investigated cloud movement and suggested that the reduced snowfall pattern leeward of Sado Island was formed by north-westerly winds bypassing the island.However, Veals et al. (2019) found that the snowfall amount downwind of Sado Island is less than its surroundings regardless of approaching wind direction and speed.The results of the days with strong winds, when air flows over the island, suggest that the factors proposed by Matsumoto (1967) and Yagi and Uchiyama (1983) may not be the primary reasons.It is still unclear how Sado Island reduces snowfall in cities 45 km away from the island.
Sado Island is situated approximately 45 km northwest of Niigata City and has an area of approximately 855 km2 (Figures 1 and S1).The summit of the largest mountain on the island is 1172 m above sea level.When the northwest monsoon is strong (e.g., wind speed >9 m/s and dividing streamline height is lower than the ground level), the approaching air is expected to flow over Sado Island.If wind speeds are very strong (or, more strictly, if the mountain Froude number is very large), snow can flow over the mountain range and reach its downwind side as spill-over (e.g., Sinclair et al., 1997).In contrast, winds below the dividing streamline height can travel around mountains and reconverge downwind, where clouds form again.However, due to the relatively short distance between Sado Island and the Niigata Plain, which is only 45 km, it is uncertain whether clouds can redevelop and produce snowfall in the Niigata Plain.It is also unclear if Sado Island has a snow-shadow effect that could reduce snowfall in the region.
Rain-or snow-shadow effects are typically observed in large mountain ranges on continents and the mountains on relatively large islands.During the northwest winter monsoon in Japan, the windward side of the backbone mountain range (Sea of Japan side) is often heavily snow-covered, whereas the leeward side (Pacific Ocean side) typically has clear skies.This is a typical snow-shadow effect.Rainshadow effects have been observed in Wales and the Peak Districts of the United Kingdom (e.g., Sawyer, 1956;Stockham et al., 2018), Taiwan (e.g., Yeh & Chen, 1998), Luzon Island, Philippines (e.g., Akasaka et al., 2007;Chang et al., 2005;Pullen et al., 2015), the Southern Alps of New Zealand (e.g., Chater & Sturman, 1998;Sinclair et al., 1997) and Sri Lanka (e.g., Puvaneswaran & Smithson, 1991).
Rain-shadow effects are typically observed downwind of islands.In Hawaii, which has mountains reaching heights of up to 4 km, precipitation amount and frequency decrease by approximately 40 km downwind of the island when trade winds are dominant (e.g., Kidd & McGregor, 2007;Nullet & McGranaghan, 1988).Although rain-shadow effects have been studied in various locations worldwide, it has not been extensively examined on small mountains without high elevations, such as Sado Island in Japan.
It remains unclear whether a small mountain on a small island can block strong winds and reduce snowfall on a leeward plain located 45 km away.To address this knowledge gap, we investigated the snow-shadow effect of Sado Island on snowfall in Niigata City.Our study aims to improve the understanding of mountain meteorology in terms of the snow-shadow effect on small islands rather than the large rain-shadow effect on medium-to-largesized mountains investigated previously.

| Data used in this study
Ground-based radar data from the Japan Meteorological Agency (JMA) were used to understand the spatial patterns of snowfall in the study area.The observational sampling interval and spatial resolution were 10 min and 1 km, respectively.Cloud image data measured by the Himawari satellite, operated by the JMA, were also used.The 12-hourly sonde data observed at the JMA Wajima Aerological Observatory and the 6-hourly surface weather charts generated by the JMA were used to understand the synoptic weather conditions and prevailing winds around Niigata City.
The 6-hourly National Center for Environmental Prediction Final (NCEP-FNL) data and daily NCEP-real-time global sea surface temperature data were used to create the initial and boundary conditions for the numerical simulations using the Weather Research and Forecasting (WRF) model.

| Statistical analysis
We investigated how snowfall distribution around Niigata City is affected by the island.As dates and times for the analysis, we selected snowfall events that occurred during 10 winters (December-February 2005-2014) and met the following three conditions: (i) the typical winter-type pressure pattern of Japan, with an anticyclone to the west of Japan and a cyclone to the east.(ii) A wind direction at the 850-hPa level observed using the GPS-Sonde from the Wajima Aerological Observatory between west and northwest, with a wind speed of 9 m/s or higher; where 9 m/s is the average wind speed when the height of the dividing streamline is at ground level.Because wind data at the 850-hPa level were available only every 12 h, data at times other than the observation time were assumed to be the same as those at the nearest observation time.(iii) Cloud lines and precipitation bands were present on the windward side of Sado Island in the same direction as the wind at 850 hPa for >3 h.This was visually examined using images created from geostationary satellites and radar data.Specifically, a qualified event met the above criteria for at least one of the 3-hourly time bins: 0-2, [3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20] In the Sea of Japan, the wind direction and cloud bands may be parallel (L-mode) or orthogonal (T-mode), and their selection conditions are complex and involve vertical wind shear, atmospheric stability and buoyancy (e.g., Asai, 1972;Eito et al., 2010).Therefore, we visually confirmed the events and did not use an automated approach, using quantitative criteria for atmospheric conditions.

| Configurations of numerical experiments
The WRF model version 4.2.2 (Skamarock et al., 2019) was used for numerical simulations.The four nesting domains of the WRF model are shown in Figure 1a.The horizontal grid spacings for the domains D01, D02, D03 and D04 were 27, 9, 3 and 1 km, respectively.There were 61 vertical layers.The other configurations are listed in Table 1.
Numerical simulations were conducted for 12 typical precipitation band events listed in Table 2.A numerical experiment with real terrain is referred to as CTRL.Sensitivity experiments were conducted to evaluate the effect of Sado Island on the precipitation around Niigata City (Table 3).In Case No_SD, simulations were performed  (Dudhia, 1989) Long-wave radiation RRTM (Mlawer et al., 1997) Land surface Noah-LSM (Chen & Dudhia, 2001) Boundary layer turbulence YSU (Hong et al., 2006) Cloud microphysics WSM6 (Hong et al., 2004) Cumulus parameterization Kain-Fritch (Kain, 2004) only for D01 and D02 by removing Sado Island.In Case NE_SD, Sado Island was moved 150 km northeast of its actual location while maintaining the distance between the island and the mainland.
In this case, Sado Island is located upwind of the plain, where the Sakata local meteorological observatory is situated.In Case NW_SD, Sado Island was shifted 200 km upwind (northwest) from its actual position, away from Niigata City.These numerical experiments employed the same settings as the CTRL experiment, with the only difference being the location of Sado Island.

| Statistical and composite analyses
A total of 564 h of precipitation bands met all three conditions described in Section 2.1.In 453 (i.e., 151 Â 3-h  samples) (80%) of these 564 h (i.e., 188 Â 3-h samples), the precipitation in Niigata City was lower than that in the surrounding areas.Figure 3 shows a precipitation composite for each event, classified according to wind direction.In all figures, precipitation was low downwind of Sado Island, regardless of the wind direction.2.
Specifically, the snow-shadow effect was observed in different areas for westerly winds compared to other wind directions, such as north-westerly and westnorth-westerly winds.These results suggest that a snowshadow effect occurs on Sado Island.The values for all figures were averaged for the 12 typical events listed in Table 2.

| Numerical experiments
averaged for the simulated results of the 12 typical events listed in Table 2.In the CTRL experiment, the WRF model reproduced a critical feature of the observations: the appearance of a minimum snowfall downwind of Sado Island.Interestingly, both the observations and simulations showed that the precipitation minima extended not only over the Niigata Plain where Niigata City is located but also over the mountain slopes behind the plain; in contrast, in the No_SD experiment (Figure 4c), there was no substantial difference between the snowfall in Niigata City and the surrounding areas in the plain, and the snowfall in the Niigata Plain was comparable to that in other plains.The impact of Sado Island on the precipitation patterns is shown in Figure 4d.As depicted in this figure, large precipitation anomalies were found only downwind of Sado Island.Although the model bias is found in the mountain areas, the results presented in Figure 4d can be considered acceptable because of the following reasons: precipitation at high elevations estimated from the radar is likely underestimated (e.g., Veals et al., 2019), and model bias can be cancelled out and reduced when comparing the experiments CTRL and No_SD.
To investigate the hypothesis that Niigata City experiences low snowfall levels because of its location on a relatively large plain near the Sea of Japan, a simulation was conducted in which Sado Island was moved upwind of a smaller plain, specifically to the northeast (Case NE_SD).The results are shown in Figure 5a area appeared downwind of the new Sado Island but not around Niigata City.As shown in Figure 5b, there was a substantial decrease in snowfall in the plain downwind of the new Sado Island compared to that seen for Case No_SD.These results show that snowfall decreases under the influence of Sado Island, regardless of whether the downwind area of Sado Island is a large or small plain.Furthermore, Figure 5b shows that Sado Island F I G U R E 7 Vertical cross-section along A-B line (shown in Figure S4) of winds, relative humidity and mixing ratio of precipitating and in cloud liquid and frozen particles over 0.001 g kg À1 .Here, precipitating and in-cloud liquid and frozen particles are defined by the sum of cloud water, cloud ice, rain, snow and graupel in the microphysics scheme.(a) Winds and relative humidity in Case CTRL, (b) winds and relative humidity in Case No_SD and (c) differences in winds and relative humidity between Cases CTRL and No_SD.(d) Mixing ratio of precipitating and in cloud liquid and frozen particles (over 0.001 g kg À1 ) in Case CTRL, (e) mixing ratio of precipitating and in cloud liquid and frozen particles (over 0.001 g kg À1 ) in Case No_SD and (f) difference in the mixing ratios between Cases CTRL and No_SD.The values for all figures were averaged for the 12 typical events listed in Table 2. Solid lines in (d) and (e) indicate the cloud water mixing ratio greater than 10 À3 g kg À1 .The solid and dashed lines in (f) indicate the difference in cloud water mixing ratio between CTRL and No_SD greater than 10 À3 g kg À1 and less than À10 À3 g kg À1 , respectively.
influences not only the plains but also the mountainous areas downwind.
To understand how far the influence of Sado Island extends, we moved Sado Island upwind, specifically to the northwest 200 km (Case NW_SD).The results show that a minimal snowfall area appeared downwind of Sado Island (Figure 5c).The difference between cases NW_SD and No_SD illustrated in Figure 5d indicates that Sado Island can reduce snowfall by more than 150 km downwind; thus, the potential impact on Sado Island is more than 125-fold farther than the elevation of Sado's largest mountain.This is probably because the change in winds around Sado Island caused a convergence line shift, which also shifted the snowfall area over the plains.
The results of precipitation simulations are often influenced by the selected physics schemes.Therefore, we validated the CTRL and No_SD results using physics scheme ensemble experiments.Ensemble experiments were conducted with 12 ensemble members created using four different cloud microphysics schemes, WSM6 (Hong et al., 2004), WDM6 (Sunny Lim & Hong, 2010), Thompson (Thompson et al., 2008) and Morris (Morrison et al., 2009) schemes and three different planetary boundary layer schemes, YSU (Hong et al., 2006), MYJ (Janjic, 1994) and ACM2 (Pleim, 2007a(Pleim, , 2007b)).The configurations in the ensemble experiments with and without Sado Island were the same as those in the CTRL and No_SD experiments described in Section 2.2, except for the selected physical schemes.The experiments were conducted for the 26 December 2008 event.The ensembleaveraged results from the 12 members shown in Figure S2 are similar to those of the CTRL and No_SD experiments shown in Figure 4.
The results of the precipitation simulations were influenced by both the initial/boundary conditions selected as well as the physics schemes.Therefore, we conducted CTRL and No_SD experiments using the initial/boundary conditions created from the ERA5 data instead of the NCEP-FNL data.The configurations and 12 target events in the experiments were the same as those in the CTRL and No_SD experiments described in Section 2.2.The results from the experiments shown in Figure S3 are almost identical to those in Figure 4.The results of the physics ensemble and initial/boundary condition ensemble experiments increased the robustness of the conclusions drawn in Section 3.2.
Finally, to examine the factors contributing to the snow-shadow effect, we examined how the winds and fields of water vapour and cloud water and ice content differ depending on the presence or absence of Sado Island.Figure 6 indicates that Sado Island weakens the winds over the leeward ocean, consequently reducing the heat flux from the sea surface to the atmosphere.The reduced heat fluxes were primarily caused by a reduction in the latent heat flux (Figure omitted).Additionally, Sado Island weakened horizontal wind convergence downwind.Another important aspect is that Sado Island produces orographic snowfall, reducing water vapour and cloud water and ice content over the leeward ocean (Figure 7).Thus, we believe that Sado Island prevents precipitation band from redeveloping over the leeward ocean and plains, reducing snowfall.

| CONCLUSIONS
In this study, we examined whether Sado Island reduced precipitation on Niigata City, located 45 km downwind of the island (snow-shadow effect), by analysing long-term radar data and performing numerical experiments.The results of the radar data analysis for the past 10 winters showed that the snow-shadow effect occurred in 453 (151 3-h samples, 80%) of 564 h (188 3-h samples) when the monsoon reached Sado Island.The location of this snow-shadow effect depends on the wind direction and was always observed downwind of Sado Island.
Numerical experiments using the WRF model indicated that snowfall over the Niigata Plain would be less with Sado Island than without it.When the island was moved 200 km windward, the snow-shadow effect appeared more than 150 km downwind of the sea.
These results indicate that even small islands without high mountains, such as Sado Island, can reduce snowfall over leeward plains.
The numerical experiments also showed that (i) Sado Island reduces heat fluxes from the sea surface by weakening leeward winds, (ii) horizontal wind convergence is weakened in the leeward plain of Sado Island and (iii) Sado Island produces orographic snowfall and reduces the amount of water vapour and cloud water and ice over the leeward sea.These factors prevent cloud lines from redeveloping over the leeward ocean and create the snow-shadow effect on Sado Island.
However, there are several other reasons for low snowfall in Niigata City, as discussed in Section 1.Thus, the percentage of the total contribution from the snowshadow effect on Sado Island remains unclear.To understand this, long-term climate downscaling experiments for 10-20 years with and without islands are necessary.

F
I G U R E 1 (a) Domains of numerical simulations.Overview of the four nested domains.(b) Magnified view of Domain 4 (D04).Ak, Sa, Ni, Ta and To indicate Akita, Sakata, Niigata, Takada and Toyama local meteorological observatories of Japan Meteorological Agency (JMA), respectively.Wa indicates the JMA's Wajima Aerological Observatory.The values indicate the climatological normal of snowfall amount in winter (December to February).The climatology is defined by 1991-2020.

F
I G U R E 2 (a) Surface weather chart at 12 UTC on 26 December 2008.(b) Geopotential height (contours at 30-m intervals) and horizontal wind (wind barbs) at 850 hPa level.(c) 12-h accumulated precipitation from 00 UTC to 12 UTC with ground-based radar.The thick line indicates the coastline and contours indicate elevation at intervals of 500 m.The full and half barbs represent 10 and 5 kt/h, respectively.

NE_SD
Sado Island was moved to 150 km northeast of its actual location while maintaining the distance between the Island and the mainland NW_SD Sado Island was moved to 200 km upwind (northwest) from its actual position F I G U R E 3 Composites of observed 3-h accumulated precipitation amounts for (a) westerly winds (45 h), (b) westnorth-westerly winds (258 h) and (c) north-westerly winds events (150 h).Grey areas indicate areas out of radar observation.

F
I G U R E 4 (a) Observed 3-h accumulated precipitation.(b) Simulated 3-h accumulated precipitation for Case CTRL.(c) Simulated 3-h accumulated precipitation for Case No_SD.(d) Differences in 3-h accumulated precipitation between Cases CTRL and No_SD.Only regions with a 95% confidence level in the Welch's t-test are shown in Figure 4d.The values for all figures were averaged for the 12 typical events listed in Table

Figure
Figure 4a,b depict the spatial distributions of 3-h precipitation from observations and numerical experiments (Case CTRL), respectively.All values in Figure 4 are . A minimal snowfall F I G U R E 6 Distribution of wind speed at 10 m, heat fluxes from the surfaces and horizontal convergence of winds at 950 hPa.(a) Wind speed in Case CTRL, (b) wind speed in Case No_SD and (c) differences in wind speed between Cases CTRL and No_SD.(d) Surface heat fluxes in Case CTRL, (e) surface heat fluxes in Case No_SD and (f) differences in surface heat fluxes between Cases CTRL and No_SD.(g) Horizontal convergence of winds in Case CTRL, (h) horizontal convergence of winds in Case No_SD and (i) differences in horizontal convergence of winds between Cases CTRL and No_SD.The values for all figures were averaged for the 12 typical events listed in Table 2. Vector arrows in (a) and (b) indicates wind direction only.
Model configuration.
T A B L E 1 List of 12 typical events used for the numerical experiments.
T A B L E 2