The RASM (Regional Arctic System Model) is a coupled high-resolution ice-ocean -ecosystem model that covers the north hemisphere north of 30°N with a horizontal resolution of 1/12° (~9 km) (Jin et al., 2018). The model is configured using the Los Alamos National Laboratory (LANL) Parallel Ocean Program version 2 (POP2) and Sea Ice (CICE) models, a mainstream sea ice model in the world (Wang et al., 2020). The initial conditions of temperature and salinity are from PHC (Steele et al., 2001), while nitrate from the gridded World Ocean Atlas (WOA2013) on the National Oceanic and Atmospheric Administration (NOAA) website (https://www.nodc.noaa.gov/OC5/woa13/woa13data.html).The coupled model is driven JRA-55 (Harada et al., 2016) reanalysis atmospheric forcing. The model includes 45 vertical layers (5m per layer for the top 20 m, with the layer thickness gradually increases to 300 m at the deepest ocean bottom). The model ocean biogeochemical (BGC) component is a medium-complexity Nutrients-Phytoplankton-Zooplankton-Detritus (NPZD) model (Jin et al., 2016; Jin et al., 2018). The model includes 26 state variables of phytoplankton, nutrient, zooplankton, and other carbon and nutrient pools. There are three phytoplankton types: diatoms, small phytoplankton (flagellates) and diazotrophs, with explicit carbon, iron, and Chl-a pools for each type, an explicit silicon pool for diatoms and an implicit calcium carbonate pool for small phytoplankton. Considering the purpose of this research, we focus on total net primary production of the three phytoplankton groups and NO₃.

In this study, the NPP and wind correlations based on satellite data (Crawford et al., 2020) were investigated using the model output during the summer (June-September) of 1990-2014. We also further examined the mechanisms how high wind and upwelling-favorable wind drive nutrients vertically and influence NPP in different seas. Out of the seven Arctic regional seas ( Figure 1 ), only four showed wind and NPP correlations in Crawford et al. (2020): (1) Barents Sea (Br), (2) Southern Chukchi Sea (SC), (3) Beaufort and Canada Basin (BC), and (4) Northern Chukchi Sea (NC). The analysis of this study will focus on those four regions.

The model NPP is vertically integrated in the upper 110m and we use the sum of NPP from the three phytoplankton species. Monthly and seasonal sum of NPP (g C/m²/month) are calculated for each region. Nitrate (NO₃) has been proved to be a major limiting nutrient of NPP in many Arctic waters (Tremblay et al., 2008; Zhuang et al., 2020). Monthly and seasonal average of nitrate, sea-surface temperature (SST) and sea ice concentration (SIC) for each region are also extracted from model. Ice-free days ratio (IFR) is calculated with a criterion of SIC<10%, same as Crawford et al. (2020). Note that the NPP in Crawford et al. (2020) are only available in ice-free areas, and the model data include NPP in both ice-free and ice-covered areas.

JRA-55 reanalysis data has a ∼0.5625^° latitude/longitude spatial resolution and 3-hr temporal resolution. Near-surface winds are for the period June-September 1990-2014 are used to calculated (a) high-wind frequency (HWF) corresponding to the percentage of time when wind speed exceeds 10 m/s and (b) westerly wind frequency (WWF) corresponding to the percentage of time when zonal wind speed exceeds 0 m/s.

On the seasonal time scale, the results ( Table 1 ) based on equation (3) show further enhanced correlation (both coefficient and significance) between NPP and HWF mainly in the Br (r=0.48, p<0.05) and SC (r=0.34, p<0.05), versus remote sensing-based results (Br: r=0.31, p<0.10; SC: r=0.32, p<0.10). On monthly time scales, the correlations are also enhanced. NPP and HWF are significantly correlated in two months (June: r=0.71, p<0.05, August: r=0.34, p<0.19) in the Br; and in all four months (June: r=0.40, p<0.05, July: r=0.33, p<0.05, August: r=0.55, p<0.05 and September: r=0.37, p<0.05) in the SC. The correlation between NPP and HWF (r=0.03, p<0.19) passed significant level in the NC, but the correlation coefficient is almost zero, thus it is not considered as an effective significant correlation in this study. The model results can reflect the phenomenon of high winds enhanced NPP in the Br and SC, similar to the observed results (Crawford et al., 2020), and also enable us to include the NO₃ in the regression as one of the possible mechanisms. Our discussion on the correlation between NPP and HWF will therefore focus on the Br and SC.

Meanwhile, the correlation between NPP and WWF on the seasonal time scale is significant in the BC for both model (r=-0.33, p<0.19) and remote sensing (r=-0.38, p<0.05), but not significant in the NC for both model and remote sensing ( Table 1 ). On monthly time scale, NPP and WWF are significantly correlated in all four months (June: r=-0.47, p<0.05, July: r=-0.23, p<0.19, August: r=-0.42, p<0.05 and September: r=-0.26, p<0.19) in the BC; in contrast, only in two months (July: r=-0.45, p<0.05, September: r=0.27, p<0.19) in the NC. Therefore, our discussion of the correlation between NPP and WWF will focus on the BC. The modeled correlation between NPP and WWF is similar as observations (Crawford et al., 2020) on seasonal scale, but stronger on monthly scales. This indicates that the model can capture NPP increase by the wind-driven upwelling mechanism in the BC, which will be analyzed in the latter section. The BC is strongly influenced by the Beaufort High, and the clockwise wind is the main driver of the upwelling along the shelf break. To facilitate the subsequent discussion, the WWF in equation (3) is changed to Easterly Wind Frequency (EWF), and the magnitude and significance of the NPP-EWF correlation remain the same as NPP-WWF, but the signs of correlation coefficients change from negative to positive. NPP in the Br, SC and BC are positively correlated with IFR and NO₃ seasonally and statistically significant for most months (p<0.05), indicating that increasing IFR and NO₃ in these three seas will lead to increasing NPP.

It is necessary to analyze the modeled monthly average climatology of all variables in equation (3) and validate those based on remote sensing by Crawford et al. (2020). Spatially, the modeled NPP is strongest in the Arctic coastal seas, especially in the Br and SC ( Figures 2A–D ), similar to the remote sensing results. Temporally, the NPP of the four seas is relatively higher in June, July and August than in September ( Figure 2E ), and the stronger NPP in June than remote sensing results is because model results include NPP under sea ice cover, while remote sensing do not have data under ice-covered areas. Despite the differences, the magnitude, spatial and temporal distribution of NPP still reflect the general patterns seen from remote sensing results in Crawford et al. (2020).

IFR generally increases from June to September, but the time of ice retreat and the number of ice-free days are different by regions ( Figure 3 ), which is partially responsible for regional monthly NPP differences in Figure 2 . In June, the number of ice-free days is highest in the Br ( Figure 3A ), with over 25 ice-free days in the south Br. In July, ice-free days increase significantly in the SC, followed by the BC and NC ( Figure 3B ). In August and September, ice-free days in most of the Br and SC exceed 25 days, in contrast to less than 10 days in large part of the BC and NC ( Figures 3C, D ). Compared to the remote sensing (Crawford et al., 2020), the modeled monthly ice-free days and IFR distribution are very similar in the Br, but are slightly different in the other seas in some months. For example, in the SC in June ( Figure 3A ) as well as the BC and NC in July-September ( Figures 3B–D ), the modeled areas with more than 25 ice-free days are smaller and the differences of IFR between model and remote sensing results are ~5%-12% ( Figure 3E ).

The modeled surface (0-20 m) averaged NO₃ is high in most of the Arctic shelf seas (over 5 mmol/m³) in June ( Figure 4A ) as they are still in pre-bloom stage except the south Br (<2 mmol/m³). NO₃ is relatively lower in the south Br in June ( Figure 4A ), because there are less sea ice cover to limit phytoplankton production ( Figure 3A ). Therefore, the input of nutrient-rich Pacific inflow water (Woodgate et al., 2012; Lin et al., 2019) are still evident in the SC in June, but the intrusion of nutrient-rich Atlantic water are less pronounced in the Br. NO₃ decrease in all four seas in July and August as nutrients are consumed by primary production ( Figures 4B, C, E ). In September, NO₃ starts to recover ( Figures 4D, E ), which may be associated with the deepening of the mixed layer due to sea surface cooling and higher HWF ( Figure 5A ). The magnitudes and spatial-temporal distributions of the modeled NO₃ are generally comparable to the gridded World Ocean Atlas 2018 (WOA2018, https://www.ncei.noaa.gov/archive/accession/NCEI-WOA18) in June-September. The modeled NO₃ in the BC in August and September (<0.5 mmol/m³) is close to the measurements (<1 mmol/m³) in August-September from 2008 to 2014 (Zhuang et al., 2018). The low NO₃ in the BC is due to the convergence of surface low nutrient water toward the center of the Beaufort Gyre (Carmack et al., 2008; Zhuang et al., 2018).

The HWF and EWF ( Figure 5 ) are from the JRA-55 reanalysis data in this study. The temporal distribution of JRA-derived HWF is similar to that from ERA-I in Crawford et al. (2020), but magnitudes are generally higher in the SC and Br from June to September ( Figure 5A ), partly because JRA is in higher resolution (3 hourly, 0.25°) than ERA-I (6 hourly, 0.75°). JRA-derived HWF ranges of 7% -12% (SC) and 8%-10% (Br) in June-August, while ERA-derived HWF are 4% -7.5% (SC) and 5% -8% (Br). HWF in September is much higher at 20% (SC) and 18% (Br) ( Figure 5A ), which are higher than the ERA-derived HWF of 10% (SC) and 11% (Br). JRA-derived EWF is close to 50% in July and August with no dominant east-west wind, while larger (over 58%) in June and September, indicating that upwelling-favorable wind dominant in the BC. However, the magnitudes of JRA-derived EWF are notably smaller than those from ERA-I in June (58% vs. 66%) ( Figure 5B ).

The HWF influence on NPP is through its redistribution of NO₃, but increasing NPP also means increased consumption of NO₃, therefore the NO₃ response to HWF is both physical and biological. Consider the HWF influence and NPP activity are depth-sensitive, we divided the shallow Br and SC into the surface (0-20 m) and subsurface (20-50 m) layers for analysis. The seasonal (June to September) averaged NO₃ and vertically-integrated NPP in five highest (Br: 1991, 1994, 2006, 2009, 2010; SC: 1990, 1996, 2001, 2011, 2012) and five lowest (Br: 1990, 1993, 1998, 2000, 2011; SC: 1992, 1995, 2002, 2004, 2007) HWF years were calculated, as well as the statistical significance (using t-test) of the NO₃ and NPP relative differences of high minus low HWF years. Since sea ice retreat is widely considered to be the main cause of the increase in NPP in the Arctic Ocean, it is necessary to compare the relative magnitude of the impact of HWF and sea ice on NPP. Here the NPP relative differences (high minus low IFR years) were also calculated for the five highest (Br: 1992, 2006, 2007, 2012, 2013; SC: 1990, 1996, 2003, 2007, 2009) and five lowest (Br: 1993, 1997, 1998, 1999, 2003; SC: 1994, 1998, 2000, 2001, 2008) IFR years.

The HWF relative differences ( Table 3 ) between high and low HWF years are 54.25% and 62.88% in the Br and SC, respectively, and are both statistically significant (p<0.05). The NPP relative increase due to high HWF ( Table 3 ) is very similar in the Br (6.72%) and SC (6.53%). The magnitude of increase in surface NO₃ due to high HWF (Br: 0.18, SC: 0.15 mmol/m³) is smaller than the magnitude of decrease in the subsurface layer (Br: -0.78, SC: -0.88 mmol/m³) ( Table 3 ). Assuming that all nutrients lost in the subsurface were mixed into the surface layer, then 0.60 (Br) and 0.73 mmol/m³ (SC) were consumed by primary production and other processes, which led to increases in NPP. For comparison, the IFR relative differences (Br: 35.67%, SC: 43.69%) are both statistically significant (p<0.05) and similar to those for HWF. The NPP relative increase caused by high IFR ( Table 3 ) is much greater in the SC (17.46%) than in the Br (9.98%), and both are greater than the NPP relative increase caused by high HWF in these two seas, respectively. This indicates that in the Br and SC, particularly the SC, IFR is currently still the dominant process leading to an increase in NPP. However, with further reduction in Arctic sea ice in the future, the relative influence of IFR will decrease, while the relative influence of HWF will continue to become greater.

Since the increased highly nutrient Atlantic inflow into the Br from 1993 to 2016 (Oziel et al., 2020) and the increased highly nutrient Pacific inflow into the SC from 1990 to 2015 (Woodgate, 2017), there will be an increasing trend in nutrients in both seas. To evaluate how much the effect of HWF on NO₃ is influenced by these long-term trends, we recalculated the differences using detrended data. The results show only a small change in NO₃ differences after detrending compared to differences before detrending, with no change in statistical significance, indicating that the long-term trends have less influence on the results of our analysis.

The statistical analysis above shows that high HWF leads to high surface and low subsurface NO₃ as well as high vertical-integrated NPP. The results of equation (3) indicate that NPP in the SC is significantly correlated with HWF. However, in high and low HWF years, the HWF relative difference is statistically significant, while the NPP relative difference is statistically insignificant. We choose a zonal average section of the SC as an example to analyze the spatial variation of NO₃ and NPP responses to high and low HWF ( Figures 6 , 7 ). In all years, the seasonally averaged distribution of HWF increased gradually from south to north ( Figure 8A ), while the seasonal NPP ( Figure 8B ) and NO₃ ( Figures 6A, B ) decrease gradually, but the differences in HWF, NPP and NO₃ between high and low HWF years are not remarkable in the south and higher in the north ( Figures 8 , 6C ). In high HWF years, the locations of the increases in surface NO₃ ( Figure 6C ) and NPP ( Figure 7C ) in the north coincide, suggesting that the higher HWF in the north led to increases in surface NO₃ and NPP.

Subsurface NO₃ (>16 mmol/m³) in all years is much higher than the surface layer (<8 mmol/m³) in most areas ( Figures 6A, B ). The vertical distribution and magnitudes of modeled NO₃ in SC are consistent with observations, such as Figures 3 - 5 in Lee et al. (2007) and Figures 4 - 5C and 6 - 10C in Lowry et al. (2015). In August, section across the Chukchi Sea along 170°W is generally characterized with high-temperature, low-salinity, nutrient-poor Alaskan Coastal Water (ACW) in surface layer and low-temperature, high-salinity, nutrient-rich Chukchi Summer Water (CSW) in the subsurface layer (Qi et al., 2022). The strong vertical stratification in summer causes a rapid attenuation of turbulent mixing (Randelhoff et al., 2017), which maintains the strong vertical gradient of NO₃, along with other factors, such as the high-nutrient pacific inflow, the supplementation of organic matter decomposition on the adjacent shelf (Tremblay et al., 2015) and biological consumption (Zhuang et al., 2021).The strong vertical gradient of NO₃ remains even in high HWF years ( Figure 6A ). The distribution of NPP in the south region is higher than that in the north region, and the surface layer is much higher than the subsurface layer ( Figures 7A, B ). The higher NPP in the south region is due to higher nutrients and more ice-free days ( Figures 3A-D ), that increase light availability (Song et al., 2021) and growth time of phytoplankton (Arrigo and Van Dijken, 2011). Lower subsurface NPP is due to lower light that limits phytoplankton growth (Tremblay and Gagnon, 2009).

Large vertical differences in NO₃ in the SC make high winds more efficient in raising surface NO₃, which results in a significant increase in NPP from high HWF. Higher HWF increases NO₃ and NPP in the upper 0-10m in the south and 0-20m in the north, but reduces NO₃ and NPP in the subsurface layer ( Figures 6C , 7C ). The decrease in subsurface NO₃ decrease is 1.2-1.7 mmol/m³ (relative change of -7.5% to -10.6%), which is larger than the increase in surface NO₃ of 0.5-1.0 mmol/m³ (relative change of 3.1% to 6.2%) ( Figure 6C ).The decrease in subsurface NPP of about 0.1-0.3 g C/m³/month (relative change of -5% to -15%) is smaller than the increase in surface NPP of about 0.2-0.4 g C/m³/month (relative change of 10% to 20%) ( Figure 7C ). The relatively large increase in surface NPP reduces the relative increase in surface NO₃. The differences of NO₃ and NPP are statistically significant only in sparse areas ( Figures 6C , 7C ) in both surface and subsurface layers, probably because the NO₃ brought up by higher HWF and the corresponding variation in NPP are quickly altered by other factors in most areas.

Although the NPP differences between high and low HWF years are not statistically significant, the sections of NO₃ and NPP still reflect the mechanism by which HWF affects NPP through redistribution of NO₃.The mechanism of NPP response to higher HWF through vertical mixing of NO₃ from nutrient-rich subsurface to nutrient-poor surface layer, causing an increase in surface NO₃ and a decrease in subsurface NO₃. However, the amount of nutrient modification by other physical processes and biological growth can alter the extent and degree of high wind impacts, limiting statistically significant changes in surface and subsurface NO₃ and NPP in high HWF years only in small patches.

We used the seasonal (June-September) averaged NO₃ and NPP for the five highest (1998, 2007, 2008, 2010, 2011) and five lowest (1991, 1992, 1994, 1996, 2002) EWF years to analyze the redistribution of NO₃ by the EWF and the effect on NPP. To compare the relative effects of sea ice and EWF on NPP, we also calculated the NPP relative differences (high minus low) for the five highest (1998, 2008, 2010, 2011, 2012) and five lowest (1991, 1992, 1994, 1996, 2002) IFR years

The EWF relative difference (high minus low) between high and low EWF years is 51.52% and statistically significant (p<0.05) ( Table 4 ). The NPP relative increase due to high EWF in the BC (67.64%) ( Table 4 ) is much greater than that due to high HWF in the Br and SC ( Table 3 ). This is because the BC is much deeper than either of these seas ( Figure 1 ) and the upwelling caused by high EWF can carry deeper and higher NO₃ to the surface (0.09 mmol/m³) and subsurface (0.10 mmol/m³) ( Table 4 ). The magnitude of the NO₃ increase is smaller than in the Br and SC, since the increase in NO₃ is mainly at the shelf break where upwelling is strongest and is relatively low in the north Canada Basin. As a comparison, the IFR relative difference (158.37%) is statistically significant (p<0.05) and is almost three times the difference in EWF. However, the NPP relative increase caused by high IFR (64.76%) ( Table 4 ) is close to that caused by high EWF (67.64%), suggesting that the increase in NPP due to EWF in the BC is comparable to the increase caused by IFR.

The relative differences in EWF and NPP for both high and low EWF years are statistically significant, consistent with the statistical correlation between NPP and EWF in the BC from equation (3). We calculated the zonal average of seasonal NO₃ and NPP and their differences (high minus low) for the five highest and five lowest EWF years, in order to further analyze the spatial variability of the NO₃ and NPP responses to high and low EWF ( Figures 10 , 11 ). Both seasonal EWF and NPP show a gradual southward increase from the Canada Basin to the Beaufort Sea shelf break in all years, but the NPP near the shelf break in high EWF years is almost twice as high as in low EWF years ( Figure 9 ). The higher NPP in the south is so co-regulated multiple factors: (1) more ice-free days ( Figures 3A–D ), (2) more light, (3) more nutrients in all years ( Figures 10A, B ) and (4) more additional nutrients brought up by upwelling in high EWF years ( Figure 10C ). The EWF difference (high minus low) is smaller in the south (near the shelf-break) than in the north, suggesting that high EWF at the shelf break is a key factor in causing the increase in NPP. In the Canada Basin, while the EWF difference increases from south to north, the NPP difference decreases exponentially.

The strong vertical stratification caused by the accumulation of freshwater (Proshutinsky et al., 2009) prevents the upward transport of high NO₃ from subsurface Pacific summer waters (Tremblay et al., 2008; Timmermans and Toole, 2022). Modeled NO₃ is low (<2 mmol/m³) in most areas of the surface and subsurface layers, and higher (>8 mmol/m³) below the subsurface layer to 110 m. NO₃ in the south (<73.5°N) is higher (<12-16 mmol/m³) than in the north (<8-10 mmol/m³) ( Figures 10A, B ). The deeper vertical distribution of high NO₃ reduces the efficiency of high HWF in raising surface NO₃, resulting in a non-significant correlation between NPP and HWF in the BC.

The high EWF supports more upwelling effects at the shelf break to below 110 m depth ( Figures 10A, C ). Surface NO₃ differences near the shelf break are 1-2 mmol/m³ (relative change of 6.25% to 12.5%), less than the 2-4 mmol/m³ (relative change of 12.5% to 25%) of subsurface ( Figure 10C ). The smaller surface NO₃ differences are due to higher biological depletion, and thus the corresponding surface NPP differences are 0.10-0.45 g C/m³/month (relative change of 10% to 45%), larger than 0.05-0.20 g C/m³/month (relative change of 5% to 20%) of subsurface ( Figure 11C ). From the shelf break to the Canada Basin center, as the high NO₃ transported northwards by Ekman decreases with distance, the surface and subsurface NO₃ differences decrease exponentially to near zero in the center of the Basin ( Figure 10C ). The NPP differences corresponds to the NO₃ difference, which also decreases northwards in the surface and subsurface layers, with the surface NPP difference approaching zero by the center of the Basin and remaining positive in the subsurface layer ( Figure 11C ). In high EWF years, the surface NPP decreases faster than the subsurface, resulting in a greater south-north gradient than the subsurface ( Figure 11A ). High EWF tends to correspond to stronger Beaufort gyre strength and stronger down-welling in the Canada Basin center (Proshutinsky et al., 2009; Yang, 2009), with subsidence of higher production and nutrient-poor surface water leading to positive NPP differences in the subsurface and negative NO₃ differences in the deeper (>50 m) layers. NO₃ differences show a pathway of nutrients movement along the upwelling of the shelf break, northward Ekman transport in the surface layer and subsidence in the north basin region ( Figure 10C ). On this pathway, NPP differences are clearly causally related to NO₃ differences. NO₃ differences in the surface and subsurface layers are positive in most areas, except in the Canada Basin center where the NO₃ differences are close to zero, and the NPP differences in these areas are positive and statistically significant ( Figure 11C ).

The NPP difference is statistically significant in high and low EWF years. The cross-section of EWF, NO₃ and NPP reveal that the mechanism for the response of NPP to higher EWF is the uplift of lower high nutrients at the shelf break, northward Ekman transport in the surface layer and subsidence in the north basin region. Although the EWF differences are lower at the shelf break than in the north Canada Basin, the effects of EWF on NO₃ and NPP are strongest at the shelf break.