We present in this paper the results of the Russian Academy of Sciences North Atlantic Atmospheric Downscaling (RAS-NAAD) project, which provides a 40-yr 3D hindcast of the North Atlantic (10°–80°N) atmosphere at 14-km spatial resolution with 50 levels in the vertical direction (up to 50 hPa), performed with a regional setting of the WRF-ARW 3.8.1 model for the period 1979–2018 and forced by ERA-Interim as a lateral boundary condition. The dataset provides a variety of surface and free-atmosphere parameters at sigma model levels and meets many demands of meteorologists, climate scientists, and oceanographers working in both research and operational domains. Three-dimensional model output at 3-hourly time resolution is freely available to the users. Our evaluation demonstrates a realistic representation of most characteristics in both datasets and also identifies biases mostly in the ice-covered regions. High-resolution and nonhydrostatic model settings in NAAD resolve mesoscale dynamics first of all in the subpolar latitudes. NAAD also provides a new view of the North Atlantic extratropical cyclone activity with a much larger number of cyclones as compared with most reanalyses. It also effectively captures highly localized mechanisms of atmospheric moisture transports. Applications of NAAD to ocean circulation and wave modeling are demonstrated.
Subsynoptic and mesoscale atmospheric dynamics over the North Atlantic Ocean are of great interest for understanding the mechanisms of highly localized precipitation, heat and moisture transports, and low-level baroclinicity in the atmosphere. Changes in the intensity and location of the North Atlantic storm tracks are critically important for the quantification of the impact of highly variable atmospheric processes onto air–sea fluxes and associated ocean signals and for understanding the responses of cyclone activity to those ocean signals (Minobe et al. 2008, 2010; Woollings et al. 2012; Tilinina et al. 2018). Many works hint at the critical role of mesoscale dynamics in air–sea interaction in the North Atlantic, first of all in forming cold-air outbreaks (Zolina and Gulev 2003; Bond and Cronin 2008; Papritz et al. 2015; Kim et al. 2016) over the Gulf Stream, the Labrador Sea, and the Greenland–Iceland–Norwegian (GIN) Seas and in generating polar lows characterized by small scales and extreme surface fluxes in the subpolar regions (Kolstad 2011; Condron and Renfrew 2013; Stoll et al. 2018, among others). Extremely high turbulent heat and momentum surface fluxes associated with these phenomena are highly localized in space and in time and require high temporal and spatial resolution for their adequate representation in models (Gulev and Belyaev 2012; Vihma et al. 2014).
Many lower-troposphere responses to the ocean signals are also associated with mesoscale processes, including the low-level baroclinicity over the western boundary currents (Nakamura et al. 2012; Ogawa et al. 2012; Small et al. 2014; Ma et al. 2017; DuVivier et al. 2016; Parfitt et al. 2016, 2017) and anomalous convective precipitation in warm seasons (Minobe et al. 2008; Hand et al. 2014). High-resolution regional model experiments demonstrated the responses of the lower atmosphere to the ocean signals at length scales of less than 30–50 km, suggesting ocean–atmosphere coupling at mesoscales and submesoscales (Small et al. 2008, 2014, 2019; Ma et al. 2017; Parfitt et al. 2016; Bishop et al. 2017). Subsynoptic and mesoscale processes are also crucial for better understanding the mechanisms of atmospheric moisture transports, first of all in the atmospheric rivers (ARs; Lavers et al. 2011; Lavers and Villarini 2015; Gimeno et al. 2014) providing strong ocean-to-continent moisture intrusions associated with abundant precipitation. All of these phenomena cannot always be adequately captured by global reanalyses, such as ERA-Interim, JRA-55, MERRA-2, and ERA5 (Dee et al. 2011; Kobayashi et al. 2015; Gelaro et al. 2017; Copernicus Climate Change Service 2017; for definitions of acronyms, see https://www.ametsoc.org/PubsAcronymList), partly because of their relatively coarse resolution but also because of the use of hydrostatic model configurations. Remarkably, the Arctic System Reanalysis (ASR; Bromwich et al. 2016, 2018) performed with the polar Weather Research and Forecasting (WRF) Model (Bromwich et al. 2009) demonstrated a considerable improvement of the representation of many phenomena in the Arctic atmosphere (Tilinina et al. 2014; Moore et al. 2015, 2016; Smirnova and Golubkin 2017; Boisvert et al. 2018; Justino et al. 2019, among others).
Ongoing ocean modeling activities also require high-resolution forcing functions accounting for mesoscale atmospheric features. Existing datasets used for forcing model experiments such as Drakkar forcing set (DFS4/5), Coordinated Ocean-Ice Reference Experiments (CORE), and JRA-55-do are based on global reanalyses (Large and Yeager 2004, 2009; Brodeau et al. 2010; Danabasoglu et al. 2014, 2016; Tsujino et al. 2018) with a spatial resolution of approximately 50–100 km. At the same time, modern eddy-resolving ocean general circulation models use resolutions finer than 1/10°, equivalent to a few kilometers at subpolar latitudes (Deshayes et al. 2013; Sérazin et al. 2015; Guo et al. 2014; Rudnick et al. 2015; Behrens et al. 2017), and up to 1/50°–1/60° in some regional simulations (Chassignet and Xu 2017; Fresnay et al. 2018; Fallmann et al. 2019). These experiments are focused on essentially small-scale ocean features. The role of small-scale atmospheric processes, however, remains unclear when using relatively coarse-resolution forcing. Similarly, modern spectral wave models account for highly nonlinear wave generation and development processes strongly dependent on the submesoscale wind structure (Ardhuin et al. 2012; Hanley et al. 2010; Semedo et al. 2011; Zieger et al. 2015; Markina et al. 2019). At the same time, in most cases these advanced configurations are forced with relatively coarse-resolution reanalysis winds.
In summary, there is a high demand from different communities for long-term high-resolution atmospheric hindcasts performed with high-resolution model configurations for the North Atlantic where mesoscale and submesoscale processes are of high relevance. Facing this challenge, the P.P. Shirshov Institute of Oceanology of the Russian Academy of Sciences (IORAS) in cooperation with the Institut des Géosciences de l’Environnement (IGE) developed a high-resolution (14 km) atmospheric downscaling experiment for the North Atlantic Ocean [North Atlantic Atmospheric Downscaling (NAAD)]. In NAAD, the nonhydrostatic WRF Model was forced at the lateral boundaries by the ERA-Interim reanalysis over the 40-yr period (1979–2018). In the following, we describe the NAAD model configuration and production strategy in section 2 followed by a short description of NAAD products and data availability (section 3). Then we turn to the NAAD evaluation (section 4) and pilot applications (section 5). In the conclusive section 6 we discuss the NAAD added value and perspectives of further developments of the product.
2. NAAD model configuration and production strategy
In NAAD we used the nonhydrostatic WRF Model, version 3.8.1 (Skamarock et al. 2008; Powers et al. 2017). The domain covers the North Atlantic from 10° to 80°N and from 90°W to 5°E (Fig. 1), with the center at 45°N, 45°W. The initial and lateral boundary conditions [including sea surface temperature (SST)] were provided by the ERA-Interim reanalysis (Dee et al. 2011) at 0.7° × 0.7° spatial resolution and 60 levels in the vertical direction. The spatial resolution in the basic NAAD high-resolution experiment (HiRes) was 14 km (551 × 551 points) and 50 terrain-following, dry hydrostatic pressure levels, starting from around 10–12 m above the ocean surface to 50 hPa with ~15 levels in the boundary layer (http://www.naad.ocean.ru). Besides the HiRes experiment, we also conducted a moderately low resolution experiment (LoRes) with the hydrostatic setting of the WRF Model at 77-km resolution (110 × 110 points) with 50 vertical levels (as in HiRes). The LoRes experiment (with resolution comparable to ERA-Interim) will be used to quantify the added value of the HiRes experiment, which cannot be directly compared with ERA-Interim (constrained by data assimilation and using a very different model configuration). All experiments were run for the 40-yr period from January 1979 to December 2018. Details of the model settings for the HiRes and LoRes experiments are presented in Table 1.
Most parameterizations were used in both HiRes and LoRes NAAD experiments. We used the Kain–Fritsch (KF) convective parameterization scheme (Kain 2004). The RRTMG longwave and shortwave radiation schemes (Iacono et al. 2008) were used for terrestrial and solar radiation processes, which additionally utilize effective cloud water, ice and snow radii from the single-moment 6-class (WSM6) scheme for microphysics (Hong and Lim 2006) in HiRes, and subgrid convective cloud information from KF for a more accurate estimation of atmospheric optical depth. The surface layer was parameterized by the MM5 scheme of (Skamarock et al. 2008) based upon similarity theory, accounting for a viscous sublayer and incorporating the COARE 3.0 algorithm (Fairall et al. 2003) for calculating thermal and moisture roughness lengths (or exchange coefficients for heat and moisture) over the ocean surface. For the planetary boundary layer (PBL) we used the Yonsei University (YSU) nonlocal scheme (Hong et al. 2006) and the Noah land surface model (Chen and Dudhia 2001). An important issue is the number of vertical levels captured by the PBL. In NAAD (see the specification of vertical levels at http://www.naad.ocean.ru), 15 vertical levels are below 850 hPa. Computation of the PBL height (not shown) reveals the highest PBL exceeding 1000 m over the regions with active convection and the lowest PBL of less than 200 m in the polar regions. This implies that the number of vertical levels in PBL range from as few as 5–6 to as many as 15–16 over the NAAD domain. In this respect, for example, version 2 of ASR (ASRv2; Bromwich et al. 2018) with approximately 10–12 levels in the PBL (implied by 25 levels below 850 hPa) is more effective in polar latitudes; however, ASRv2 is based on the local Mellor–Yamada–Janjić PBL scheme, which may not necessarily be effective over the whole North Atlantic region.
Since the Noah scheme updates deep soil temperature, the skin sea surface temperature is calculated using the Zeng and Beljaars (2005) formulation. The PBL scheme is responsible for vertical subgrid-scale fluxes due to eddy transports in the whole atmospheric column, and not only in the boundary layer. Horizontal eddy viscosity coefficients are obtained in the WRF dynamic core independently using the Smagorinsky first-order closure approach. Parameterizations of microphysics were nevertheless different in HiRes and LoRes. Thus, the WSM6 scheme for microphysics (Hong and Lim 2006) was used in the NAAD-HiRes and WSM 5-class (WSM5; Hong et al. 2004). Additionally, employing entrainment information from KF was applied in the NAAD-LoRes case. For the long-term runs with WRF in both HiRes and LoRes experiments, the RRTMG scheme uses climatological ozone and aerosol data. The ozone data were adapted from the Community Atmospheric Model radiation scheme with latitudinal (2.82° step), height, and temporal (monthly) variation. The aerosol data were based on the Tegen et al. (1997) dataset with relatively coarse spatial (5° in longitude and 4° in latitude) and temporal (monthly) resolution.
The WRF settings used for NAAD HiRes, with some modifications, was applied in a number of applications. In the Polar WRF used in ASRv2 (Bromwich et al. 2018) the major difference was in the use of the Mellor–Yamada–Nakanishi–Niino (Nakanishi 2001; Nakanishi and Niino 2004, 2006) 2.5-level PBL parameterization. However, the nonlocal YSU scheme used in NAAD is effective to resolve strong convective processes in the midlatitudes and tropical regions. This parameterization was used in a number of RCM simulations (Bukovsky and Karoly 2009; Otte et al. 2012; Gao et al. 2015; Tang et al. 2017).
At the ocean surface, we used ERA-Interim SST and sea ice, which was updated every 6 h during the simulation. ERA-Interim SST is combined from different sources (Dee et al. 2011). Kumar et al. (2013) demonstrated that in different reanalyses, the intraseasonal SST–precipitation relationship is dependent on the SST used. In this respect we understand that the relatively coarse (with respect to HiRes) resolution of ERA-Interim SST may have an effect on the atmospheric surface layer and PBL. Currently, several high-resolution SST datasets, while limited in time coverage, are available (Chelton and Wentz 2005; Chao et al. 2009; Ricchi et al. 2016). Nevertheless, for the 40-yr-long NAAD experiments we used ERA-Interim SST, which is considered to be homogeneous and adequate for multidecadal scales.
To reduce unrealistic atmospheric dynamics in the regional domain in both LoRes and HiRes experiments, we applied throughout the 40-yr period the procedure of spectral interior nudging (Jeuken et al. 1996; Miguez-Macho et al. 2004). The spectral nudging technique optimizes the adjustment of the large-scale dynamics inside the domain to that implied by the boundary conditions. The nudging procedure was applied to the zonal and meridional wind components, air temperature, and the perturbation of the geopotential height. We did not nudge the moisture fields because their variability is not always represented adequately in the coarse-resolution ERA-Interim (Miguez-Macho et al. 2004; Otte et al. 2012). Configuration of nudging was set according to the sensitivity study of Markina et al. (2018), which implied the optimal wavelength cutoff being 1100 km, applied only above the PBL. For determining the optimal nudging strength, we performed sensitivity experiments with the nudging strength coefficients increasing from 3 × 10−5 to 3 × 10−3 s−1. These experiments implied an optimal value of the nudging strength coefficient of 3 × 10−4 s−1 (equivalent to a damping scale of about 1 h). This value is also consistent with other studies (Miguez-Macho et al. 2004; Otte et al. 2012; Tang et al. 2017, among others). Note also that the ASRv2 (Bromwich et al. 2018) used spectral nudging with the strength 3 × 10−3 s−1 (i.e., an order of magnitude stronger relative to NAAD) for all levels in the outer domain and above 100 hPa in the inner domain. However, a direct comparison is unlikely to be possible here, as the ARSv2 assimilates a lot of observational data in the surface layer using the same technique (Newtonian relaxation, also called “observational nudging”; Jeuken et al. 1996).
The dynamical solver of WRF uses a Cartesian grid. The difference between the geographical and the model horizontal distance (map-scale factor) should not deviate significantly from unity in order to match the CFL criterion. The NAAD grid is based on a cylindrical equidistant projection (lat-lon in the namelist; http://www.naad.ocean.ru) with the rotated pole in order to locate the equator at the middle of the domain (pole_lat = 45; pole_lon = 180; stand_lon = −45). For this projection, the maximum map-scale factor amounts to 1.2 in the northernmost and southernmost regions of the domain (Fig. 1). This allowed for keeping a constant third-order Runge–Kutta time step of 30 s in HiRes and 240 s in LoRes. In NAAD we used the USGS topography data with 10 min spatial resolution.
3. NAAD products and data availability
NAAD products (see http://www.naad.ocean.ru for the parameter identification and namelist) include surface and upper-troposphere variables provided at the native NAAD grid for HiRes and LoRes products with resolutions of 14 and 77 km, respectively, for the period of 1979–2018. Since the NAAD core model (WRF-ARW) uses a staggered Arakawa C grid, all 3D fields are preinterpolated from the C grid on the mass points (unstaggered grid). No interpolation was applied for surface variables and fluxes, provided at the mass points. The entire archive of the NAAD data amounts to 150 terrabytes (TB) with individual annual files ranging from approximately 140 MB in LoRes to 3.3 GB in HiRes for surface variables on to 165 GB for HiRes 3D fields. The whole NAAD data output is organized as annual NetCDF files by variable and is available online (http://www.naad.ocean.ru) for download using FTP and Network Data Access Protocol (OPeNDAP) accesses.
4. NAAD evaluation
Surface-state variables and fluxes are particularly significant as they are used for forcing ocean models and for the diagnostics of extreme events. Figures 2a and 2b show that surface air temperature diagnosed by NAAD HiRes is colder than in LoRes. The differences between LoRes and HiRes are smaller than 0.2°C over most of the domain and increase to 0.4°C in the western midlatitude North Atlantic. Larger surface air temperatures in HiRes compared to LoRes, amounting to 1°C, are observed over the ice-covered regions and occur primarily in the winter months. This regional bias can be explained by using a coarse-resolution sea ice mask from ERA-Interim in both HiRes and LoRes NAAD simulations. Compared to ERA-Interim (Fig. 2c), HiRes shows 0.2° to 0.4°C lower surface air temperatures over most of the North Atlantic and considerably colder surface air temperatures in the ice-covered regions in the subpolar North Atlantic. A similar pattern of differences in surface air temperature is revealed for ASRv2 (Fig. 2d) for the area of overlap of the two domains. Surface relative humidity (Figs. 2e,f) over open ocean midlatitudes and subtropics is slightly smaller in HiRes compared to LoRes with the differences exceeding 1% identified in the western tropics. At the same time, somewhat more humid surface conditions in HiRes compared to LoRes are identified in the subpolar western North Atlantic and in the offshore region of the subtropical eastern Atlantic. Here the differences amount to 0.6%–0.8%. Compared to ERA-Interim and ASRv2 (Figs. 2g,h), NAAD HiRes shows a relative humidity higher by 2%–3% over the North Atlantic with the strongest differences (>6%) in the eastern North Atlantic subtropics.
Figure 3 shows the evaluation of climatological winter winds in NAAD HiRes and LoRes. NAAD HiRes and LoRes surface winds are consistent south of 45°N with the differences within ±0.15 m s−1. At the same time, in the Irminger Sea NAAD HiRes shows stronger mean winds by 0.2–0.6 m s−1, thus likely reflecting a better representation of the mesoscale features such as tip jets and katabatic winds in this area. The comparison with ERA-Interim (Fig. 3c) shows stronger trade winds in the NAAD HiRes experiment, somewhat lower wind speeds in the western midlatitude North Atlantic, and also stronger winds in the subpolar North Atlantic along the eastern Greenland coast. North of 40°N, differences between HiRes and ASRv2 (Fig. 3d) are consistent with those for ERA-Interim. Dukhovskoy et al. (2017) noted differences between high-resolution satellite wind products and reanalyses attributing them to the subsynoptic and mesoscale processes. The impact of high-resolution and nonhydrostatic setting onto the wind field is especially evident for extreme winds (Figs. 3e–h). We note much stronger katabatic winds and tip jets along the Greenland coast, with the differences in surface winds between HiRes and LoRes experiments being locally over 4 m s−1 (about 20% of mean values of 99th percentile of wind speed). Relative to ERA-Interim, the HiRes experiment shows extreme winds stronger by more than 7 m s−1. Important is that HiRes also shows much better localization of katabatic winds near the coast in HiRes compared to LoRes and ERA-Interim. Figure 3d shows the localization of maximum extreme winds much closer to the Greenland coast in agreement with the case studies (Moore and Renfrew 2005; Moore et al. 2015). Extreme wind speed differences between NAAD HiRes and ASRv2 in the Irminger Sea (Fig. 3h) amount to 1.5–2 m s−1. Comparison of NAAD winds with QuikSCAT winds (Ricciardulli and Wentz 2015) (Fig. 4) indicates that both NAAD HiRes and LoRes have considerably smaller errors with respect to QuikSCAT when compared with ERA-Interim, which clearly demonstrates a systematic negative bias of 0.5–2 m s−1.
Surface turbulent fluxes (sensible plus latent, Fig. 5) in NAAD HiRes show the structure to be consistent with reanalyses and blended climatologies such as OA-FLUX (e.g., Yu and Weller 2007) with the locally strong fluxes over the Gulf Stream (primarily due to latent heat) and in the Labrador Sea (mostly due to sensible heat). NAAD HiRes turbulent fluxes are generally larger than those of NAAD LoRes by 0–10 W m−2 in the open ocean regions and by ~30 W m−2 over the Gulf Stream and in the Labrador and Irminger Seas. In the subpolar latitudes, the stronger HiRes fluxes are explained by the differences in wind speed (Fig. 3) with a significant contribution from surface temperature and humidity gradients, especially for sensible heat flux (no figure shown). In the midlatitudes and subtropics, the differences in turbulent heat fluxes between HiRes and LoRes are associated with surface temperature and humidity vertical gradients, as wind speed differences here are close to zero (Fig. 3). Differences with ERA-Interim (Fig. 5c) are 30%–50% larger than those with NAAD LoRes, while the direct comparison here is difficult because of the differences between the ERA-Interim surface flux algorithm and the COARE 3.0 flux algorithm (Fairall et al. 2003) used in NAAD (Brodeau et al. 2017). Of interest is also the evaluation of extreme surface turbulent fluxes, which might be strongly dependent on mesoscale processes (Ma et al. 2015) and demonstrate differences between different products not consistent with those for mean values (Gulev and Belyaev 2012; Bentamy et al. 2017). Figures 5d–f presents extreme fluxes quantified by the 99th percentile of the modified Fisher–Tippett distribution (Gulev and Belyaev 2012). Relative to LoRes, HiRes shows stronger extreme fluxes over the Gulf Stream and the North Atlantic Current where differences may locally exceed 30 W m−2 (up to 5%–10% of the mean values). In the Irminger Sea the differences amount to 200 W m−2, being more than 20%–25% of the mean values of extreme turbulent fluxes. Extreme fluxes diagnosed by HiRes are also considerably stronger than in ERA-Interim (Fig. 5f), with maximum differences locally exceeding 300 W m−2.
b. Storm tracks and cyclone dynamics
NAAD opens a new avenue in the analysis of cyclone dynamics and storm tracks. Cyclone tracks were diagnosed using the IORAS algorithm (Zolina and Gulev 2002; Tilinina et al. 2013), tested within the IMILAST project (Neu et al. 2013). Tracking was performed on the NAAD HiRes grid. For tracking cyclones in the limited area over the North Atlantic we applied an approach that accounts for the so-called entry–exit uncertainties (Tilinina et al. 2014). Postprocessing was further applied to cutoff cyclones with migration distances smaller than 1000 km and lifetimes shorter than 24 h (Tilinina et al. 2013).
NAAD HiRes (Fig. 6a) captures well the main North Atlantic storm tracks that are consistent with cyclone climatologies based on the global reanalyses (e.g., Neu et al. 2013; Tilinina et al. 2013). NAAD HiRes shows 30%–60% larger local cyclone numbers compared to LoRes. Also, NAAD LoRes shows slightly larger number of cyclones compared to ERA-Interim while the differences are within 3%–5%. Over the North Atlantic storm track, the number of cyclones in NAAD HiRes is close to that in ERA5 (Fig. 6d) with slightly larger cyclone counts over the storm formation region in the Northwest Atlantic and slightly smaller counts in the central North Atlantic and the subpolar regions.
Figure 7 shows the winter [December–February (DJF)] time series of the domain integrated number of cyclones with different intensities (quantified by central pressure). NAAD HiRes allows for identification of ~2 times more cyclones compared to LoRes, which indicates a high consistency with the global reanalyses, except for ERA5, which reveals practically the same number of cyclones as NAAD HiRes (Fig. 7a). Importantly, these differences between HiRes and all other products (including LoRes) are formed mostly by moderately deep and shallow cyclones (Figs. 7b,d). At the same time, the number of deep cyclones in HiRes is more consistent with LoRes and reanalyses showing 10%–15% higher counts. Overall, the winter cyclone activity in NAAD is very close to that in ERA5 and considerably more intense compared to the other reanalyses. Summer results (not shown) are qualitatively similar with even higher differences especially for shallow cyclones dominating during the warm season.
Figure 6 demonstrates relatively strong differences between cyclone counts in NAAD and in all reanalyses over the North American continent. This is likely associated with the fact that the finer-resolution NAAD HiRes is capable of identifying cyclone generation events at an earlier stage than the global reanalyses. Our analysis of cyclogenesis events (not shown) demonstrates considerably larger counts of cyclone generation events over the North American storm track in NAAD when compared with even ERA5, while the total number of tracks is close in both products (Fig. 7). Also, we note that the cyclone effective radius, characterizing cyclone size (Rudeva and Gulev 2007) is smaller in NAAD HiRes by about 50–100 km as compared with NAAD LoRes and is also smaller by 100–150 km relative to ERA-Interim (no figure shown). NAAD HiRes also demonstrates capabilities in capturing characteristics of extreme cyclones. Thus, our analysis of extremely deep cyclones shows that the 100 deepest cyclones in NAAD HiRes have a central pressure about 4 hPa lower than those in LoRes and ERA-Interim. Similar conclusions were drawn from WRF-based high-resolution ASR in comparison with global reanalyses (Tilinina et al. 2014).
c. Hydrological cycle
Evaluation of the North Atlantic hydrological cycle in the NAAD is of special interest. Figure 8a shows time series of annual mean precipitation over the NAAD domain as computed from NAAD HiRes and LoRes experiments, as well as from ERA-Interim, the Global Precipitation Climatology Project (GPCP1DD) (Huffman et al. 2001), and, starting from 2013, the Global Precipitation Measurement (GPM) mission (Skofronick-Jackson et al. 2017). NAAD HiRes shows slightly higher domain integrated total precipitation values compared to NAAD LoRes before the early 2000s and slightly smaller precipitation during the last 15 years. Both HiRes and LoRes domain integrated values are 4%–7% less than ERA-Interim and 5%–10% less than GPCP. Starting from 2013, NAAD HiRes is in a very good agreement with GPM, demonstrating very small (relative to the other products) differences for all seasons (Fig. 8b). The consistency with GPM is, however, somewhat better in autumn–winter than in spring–summer, likely because of a stronger contribution of the convective precipitation (requiring even higher resolution than that in HiRes) during the warm season.
Figures 9a–c shows annual mean total precipitation in NAAD HiRes and the differences with LoRes and ERA-Interim. NAAD HiRes when compared with LoRes shows stronger precipitation in the western Atlantic tropics and over the Gulf Stream by up to 1 mm day−1 and lower, by 0.3–0.5 mm day−1, precipitation in the North Atlantic subpolar gyre. Relative to ERA-Interim (Fig. 9c), NAAD HiRes shows stronger precipitation over the Gulf Stream (up to 2 mm day−1) and weaker precipitation over the subpolar latitudes. The precipitable water content (PWC) (Figs. 9d,e) is lower in NAAD HiRes than in LoRes by 4%–6%, with the strongest absolute differences in the western tropics. ERA-Interim, however, shows a slightly higher PWC than NAAD HiRes over most of the domain except for the western tropics and subtropics (Fig. 9e). Differences in precipitation and PWC suggest stronger tropical water recycling in NAAD HiRes and somewhat weaker recycling in mid- and subpolar latitudes relative to LoRes and ERA-Interim.
Representation of summer precipitation over the western North Atlantic is important for quantifying the lower-atmosphere responses to the ocean frontal signals in the Gulf Stream (Minobe et al. 2008, 2010; Parfitt et al. 2016). Precipitation responses in summer are typically associated with convective processes and mostly located over the westernmost part of the Gulf Stream. In winter, precipitation responses are associated with the atmospheric frontal activity and enhanced precipitation over the central and eastern Gulf Stream proper (Minobe et al. 2010). Figure 10 demonstrates precipitation for July 2015 as revealed by NAAD HiRes and LoRes as well as by ERA-Interim, ERA5, GPM, and GPCP. NAAD HiRes with its capability to capture convective processes shows the best agreement with GPCP in the structure of precipitation pattern and in magnitude. NAAD LoRes and ERA-Interim tend to underestimate precipitation rates by 2–4 and 3–6 mm day−1, respectively. ERA5 is most consistent with HiRes in spatial structure but shows smaller precipitation rates of 2–3 mm day−1. Comparison with ASRv2 (no figure shown) is somewhat difficult as the ASR domain boundary cuts a considerable part of the region analyzed. However, for the overlapping part of the domain the precipitation pattern in ASRv2 is comparable in structure to that of NAAD HiRes.
Capabilities of the high-resolution NAAD in representing moisture transports can be analyzed through the diagnostics of ARs (Lavers and Villarini 2015; Ralph et al. 2017; Shields et al. 2018), representing narrow synoptic-scale jets transporting an abundant amount of water vapor from the ocean to the continents. ARs may be responsible for more than 90% of poleward moisture transport (Zhu and Newell 1998) and also result in extreme precipitation events over continental coastal areas (Viale and Nuñez 2011; Lavers and Villarini 2015; Gershunov et al. 2017; Waliser and Guan 2017).
For the detection of ARs, we used the 85th percentile of the integrated water vapor transport (IVT) along with a fixed lower IVT limit of 100 kg m−1 s−1 (Guan and Waliser 2015). The 85th percentile of IVT was computed from the 20-day sliding windows in the NAAD-HiRes and LoRes outputs, and moisture transports were computed according to Dufour et al. (2016). Figure 11 shows the case study for 5 December 2015, with an AR associated with the deep cyclone east of Iceland and clearly seen in IVT fields. Associated daily accumulated precipitation in the NAAD HiRes closely matches that diagnosed by GPM, while, for example, ERA-Interim shows the shift in the location of the AR with respect to GPM. NAAD HiRes precipitation in the AR landfall areas over the U.K. and Norway coasts is over 66 mm day−1. This is greater than the values diagnosed by NAAD LoRes and ERA-Interim by up to 30 mm day−1 (Fig. 11f). We also note an accurate location of the coastal precipitation maxima in the landfall areas in NAAD HiRes (Fig. 11f). The maximum of water vapor transport in the AR (Fig. 12a) at 2.5-km height is associated with locally strong winds amounting to more than 40 m s−1. Figure 12b shows that the AR core in HiRes is characterized by 10%–15% stronger transports relative to LoRes, reflecting the fact that ARs in the HiRes experiment are narrower. In climatological context, this reduces the time during which western Europe coasts are exposed to ARs; however, this makes the impact of individual ARs stronger and highly localized. Our estimate for 2015, performed using a method similar to Guan and Waliser (2015), shows that AR landfalls in HiRes happen during 10%–15% of the time, which is less than in LoRes and MERRA (15%–20%).
d. Mesoscale features
To demonstrate the representation of mesoscale features in the NAAD, we use kinetic energy (KE) wavenumber spectra (SKE) derived from the wind speed data and averaged over the layer between 3- and 5-km height over the whole domain (Figs. 13a,b). These spectra characterize the atmospheric turbulence at different scales by the power law SKE(k) ~ k−γ, where k is the wavenumber and γ is changing from approximately −3 to approximately −5/3 between the ranges of space scales larger and smaller than 500 km (Waite and Snyder 2009; Condron and Renfrew 2013; Dukhovskoy et al. 2017). Figure 13 demonstrates marked differences in the wavenumber spectra for HiRes and LoRes experiments. For the total and geostrophic KE, the HiRes spectrum is better matching k−3 than LoRes is. It is important, however, that the spectral decay rate for both geostrophic and ageostrophic components at smaller scales (<500 km) is considerably smaller in HiRes than in LoRes. This reflects stronger pressure gradients (and, thus, stronger winds) in synoptic and mesoscale transients in the HiRes experiment. Dukhovskoy et al. (2017) noted a similar tendency in ASR relative to global coarser-resolution reanalyses in the subpolar North Atlantic.
Figure 13c shows the wavenumber kinetic energy spectra near the surface using data from NAAD HiRes and ASRv2 over the overlapping part of the two domains. As in the case of the free troposphere (Figs. 13a,b), the NAAD HiRes spectrum built from the surface data closely matches k−5/3 decay rate in the range of scales less than 700 km. The ASRv2 spectrum demonstrates a very similar behavior in terms of change in γ from ~−3 to ~−5/3 and the decay rate in the range from 700 to at least 100–150 km.
In this respect it is of interest to consider the representation of polar lows in NAAD. Polar lows represent highly localized maritime atmospheric phenomena associated with extreme winds and advection of very dry cold air, playing an important role in high-latitude atmospheric dynamics and air–sea interaction processes (Zahn and von Storch 2008; Condron et al. 2008; Condron and Renfrew 2013, among others). They are hardly detectable in global reanalyses primarily due to their small size (Zappa et al. 2014; Stoll et al. 2018). Many publications, however, report the capability of the WRF Model even without data assimilation to effectively simulate polar lows (Wagner et al. 2011; Wu et al. 2011; Føre et al. 2012; Kolstad et al. 2016). Kolstad (2011) developed a unique database of 63 polar lows (1999–2009) in the subpolar and subarctic North Atlantic and Arctic using AVHRR and QuikSCAT winds. Of the 21 events identified by Kolstad (2011) in the NAAD domain, NAAD HiRes was able to successfully detect 20 polar lows.
Figure 14 shows a case study for the polar low that developed on 2 March 2008 in the Irminger Sea. NAAD HiRes detects well the location of the pressure minimum identifying a 978-hPa central pressure, which is deeper than that in ERA-Interim (986 hPa) and even in ERA5 (984 hPa). Also, NAAD HiRes demonstrates the well-detectable comma-type structure not present in ERA-Interim and ERA5 and less evident in ASRv2. Note that NAAD LoRes results (no figure shown) are very close to ERA-Interim for this case. The associated maximum surface wind speed in the polar low amounts in HiRes to 32 m s−1, with 26–27 m s−1 in ERA-Interim and ERA5, and nearly 30 m s−1 in ASRv2 (Fig. 14). Note that Gutjahr and Heinemann (2018) clearly demonstrated that an accurate representation of the tip jets and associated extreme winds around Greenland requires resolutions of at least 15 km. Remarkably, it is also the case that surface turbulent fluxes (sensible plus latent) are considerably stronger in HiRes (up to 900 W m−2) than in ERA-Interim (600 W m−2) and are comparable to both ERA5 and ASRv2. We note that the direct comparison of surface fluxes between these products should be taken with caution because of the use of somewhat different flux algorithms. However, the analysis of surface flux PDFs and the surface flux relative extremeness (Gulev and Belyaev 2012; Tilinina et al. 2018) show that the strong flux event during 2 March 2008 in NAAD HiRes contributed more to the total monthly flux as compared with ERA5 and ASRv2.
The capability of the NAAD to capture the mesoscale dynamics in the tropics is evaluated in Fig. 13d showing wavenumber spectra of the kinetic energy near the surface (10 m) and at 1500 m in HiRes and LoRes simulations. Remarkably, the spectra near the surface and at 1500 m are qualitatively close to each other for both simulations and are also close to the spectra for the free troposphere (Figs. 13a,b) and to the surface spectra in the subpolar region (Fig. 13c). At the same time, tropical spectra for HiRes demonstrate a k−5/3 decay rate for the wave lengths from 200 to 1000 km, while LoRes spectra follow k−3 and a slightly stronger decay rate in this range. This implies more energetic mesoscale features of the same size in the tropics in HiRes than in LoRes.
NAAD is also capable of identifying tropical cyclones generated and propagating north of the southern margin of the domain. Figure 15 shows the diagnostics of Hurricane Gaston, which developed over the North Atlantic between 22 August and 2 September 2016. At the moment of maximum development, NAAD HiRes diagnoses the lowest central pressure (6 hPa deeper when compared with ERA5 and more than 15 hPa deeper than NAAD LoRes and ERA-Interim) as well as winds stronger by 5–8 m s−1 than in ERA5. Associated precipitation in NAAD HiRes is considerably stronger than in LoRes and ERA-Interim and consistent in magnitude with ERA5. However, the precipitation pattern in HiRes more accurately captures the shape implied by GPM than ERA5.
5. Pilot ocean applications
a. Effects in modeled ocean circulation
To demonstrate the NAAD capabilities for driving long-term experiments with regional configurations of ocean general circulation models, we developed surface forcing functions for the ocean based upon NAAD HiRes and NAAD LoRes and used them to drive a northern North Atlantic regional configuration of the NEMO (version 3.6) ocean and sea ice general circulation model (Madec et al. 2016). This configuration (referred to as NNATL12) covers the subpolar gyre of the North Atlantic (Verezemskaya et al. 2020, manuscript submitted to J. Geophys. Res.) at a resolution of approximately 4.5 km. The model configuration setup including configuration geometry (1/12° grid, 75 vertical z levels with fine separation (1 m) near the surface, coastlines, and bathymetry), numerical schemes and physical process parameterizations are those commonly used for the global 1/12° eddy-resolving global ocean circulation model ORCA12 for the operational forecasts (Lellouche et al. 2018) as well as for climate-oriented long-term simulations (e.g., Sérazin et al. 2018; Hewitt et al. 2016) and process studies (e.g., Akuetevi et al. 2016). At the open northern and southern boundaries as well as at the western boundary of Hudson Bay, the model is driven by monthly mean temperature, salinity, velocity, and sea ice from the Global Ocean Reanalysis Simulation (GLORYS12), version 4 (v4) (Garric and Parent 2018). The model was tested in a set of sensitivity experiments and validated against the high-resolution the 1/12° GLORYS12 reanalysis of the Copernicus Marine Environment Monitoring Service (Fernandez and Lellouche 2018), satellite observations, and repeated full-depth hydrographic sections at 60°N (Sarafanov et al. 2012; Gladyshev et al. 2018; Verezemskaya et al. 2020, manuscript submitted to J. Geophys. Res.).
Comparative model experiments were performed with NAAD HiRes and NAAD LoRes atmospheric forcings; the experiments are referred to as NAAD-OHR and NAAD-OLR, respectively. Significant differences in characteristics of turbulent and radiative heat fluxes as well as momentum fluxes between the two forcings at the ocean surface result in large differences in the simulated ocean mean state. Thus, the domain-averaged simulated SST is approximately 0.6°C lower in NAAD-OHR, with summer differences amounting to more than 1.5°C (Fig. 16), in close agreement with European Space Agency (ESA) SST (http://www.esa-sst-cci.org). Note that trends in SST are highly consistent in both NAAD-OHR and NAAD-OLR. The strongest SST negative differences of 1°–1.5°C and lower sea surface salinity (0.15–0.2 psu) in NAAD-OHR relative to NAAD-OLR are observed in the Labrador and Irminger Seas. Consistent with SST, NAAD-OHR shows a lower ocean heat content for both the upper (0–700 m) and intermediate ocean (700–1500 m), suggesting a more intense ventilation of the ocean by convective processes in NAAD-OHR (Fig. 16). Since almost all ocean models in noncoupled experiments have a tendency toward a warmer and saltier ocean (Tréguier et al. 2005; Rattan et al. 2010), the colder upper-ocean temperatures in NAAD-OHR should be considered as an improvement. The OHR forcing also appears to drive significant changes in the boundary currents around Greenland and in the different branches of the central North Atlantic Current, which were found to be more intense and more variable in NAAD-OHR than in NAAD-OLR, as revealed by the partition of eddy kinetic energy (not shown). NAAD-OHR also shows a somewhat deeper mixed layer depth (MLD) relative to NAAD-OLR (not shown) in regions known for being strongly ventilated by winter convection (the southwest sector of the Labrador Sea and the central Irminger Sea). A lesser ventilation is noticed in the areas where ocean eddies are known to counterbalance the effects of strong surface fluxes onto MLD (e.g., Chanut et al. 2008).
b. Effects in ocean wind-wave modeling
NAAD can be also effectively used for forcing spectral wave models whose solutions are critically dependent on the quality and spatial resolution of atmospheric forcing (Cavaleri 2009; Ardhuin et al. 2012). In this respect, the mesoscale activity in the lower atmosphere might be of critical importance for capturing extreme wind waves (Condron et al. 2006; Zappa et al. 2014; Markina et al. 2019).
We used NAAD HiRes and LoRes outputs for the WAVEWATCH III spectral wave model (WW3DG 2016) over the NAAD domain. Experiments (referred to as NAAD-WHR and NAAD-WLR for HiRes and LoRes forcing, respectively) were performed with the spatial resolution of 0.2° and spectral model resolution being 36 directions and 25 frequencies spanning from 0.04 Hz with an increment of 1.1. The WAVEWATCH–III configuration included ST6 parameterization for wave energy input and dissipation (Zieger et al. 2015; Liu et al. 2019) calibrated for WRF winds (Markina et al. 2018) and the ice source term package (IC0; Tolman 2003) implying the exponential attenuation of waves in partially sea ice–covered regions. To account for the ocean surface current impact on growing waves we used daily surface current velocities from the NEMO-based Ocean Reanalysis System 5 (ORAS5; Zuo et al. 2019). Cavaleri (2009) argued that the increase in spatial resolution of forcing per se does not necessarily result in the increase in significant wave height (SWH); rather, the formation of high waves is associated with higher winds, changes in the duration of wind action, and the length of fetch. In this sense, NAAD HiRes with its stronger winds (Fig. 3), smaller cyclone sizes and larger number of synoptic transients (Figs. 5 and 6) likely acts locally rather than on a larger scale.
Figures 17a and 17b show winter (January–March) climatological SWH for 1979–2018 in NAAD-WHR and the differences between NAAD-WHR and NAAD-WLR. In NAAD-WHR the highest SWH amounts to 5.4 m in the eastern subpolar North Atlantic in very close agreement with VOS observations (Gulev et al. 2003; Gulev and Grigorieva 2006). Differences between NAAD-WHR and NAAD-WLR SWH over most of the North Atlantic midlatitudes and subtropics are generally within 0.3 m but strongly increase in the subpolar North Atlantic where they amount to 0.8 m in the Irminger Sea. The highest extreme strong waves quantified as 95th percentile amount in winter in NAAD-WHR to 9 m (Fig. 17c), being higher than in NAAD-WLR by 0.2–0.5 m in the central and eastern subpolar North Atlantic (Fig. 17d). At the same time the most distinctive differences between the NAAD-WHR and NAAD-WLR experiments are identified along the southeast Greenland coast where the extreme SWH in NAAD-WHR is higher than that in NAAD-WLR by more than 1 m (up to 20% of mean values). This likely reflects a more accurate representation of katabatic winds and tip jets in this area in NAAD-WHR.
6. Summary and outlook
We presented NAAD—a new 3D multidecadal atmospheric dataset for the North Atlantic produced with a WRF nonhydrostatic model at mesoscale resolution (NAAD HiRes). In parallel, a coarser-resolution dataset (NAAD LoRes) was produced with a spatial resolution close to ERA-Interim used in both experiments as a lateral boundary condition.
Our evaluation demonstrates reasonably realistic representations of most climatological characteristics in both NAAD HiRes and NAAD LoRes datasets. The main differences are identified in the ice-covered subarctic regions, especially for surface air temperature and partly for surface humidity. At the same time, atmospheric dynamics was quite adequately represented. The major purpose of the NAAD at this stage was not to provide extremely close comparability of NAAD LoRes with, for example, ERA-Interim. This is hardly achievable because ERA-Interim (as well as the other reanalyses) is largely constrained by data assimilation. The major NAAD focus was rather to develop a high-resolution atmospheric dataset that allows a better analysis of subsynoptic and mesoscale features—the task still not resolved by global reanalyses over the North Atlantic. In this respect, the objectives for NAAD are similar to those posed for the other regional reanalyses (e.g., ASR; Bromwich et al. 2018). The NAAD model configuration was quite close to the one used in ASR, the lateral conditions are the same (ERA-Interim) and the resolutions of the two products are similar. At the same time ASR (both ASRv1 and ASRv2) used extensive data assimilation input that exceeds data assimilation in the global reanalysis (e.g., ERA-Interim). Our capabilities for the direct comparisons of NAAD with ASR were limited to the subpolar latitudes. However, the analysis of kinetic energy spectra and extreme winds and fluxes associated with polar lows (Figs. 13 and 14) clearly demonstrated that the differences between NAAD HiRes and ASRv2 are the smallest relative to the other reanalyses, and this is reassuring. The comparative analysis of surface fluxes, winds, and precipitation in the North Atlantic midlatitudes is not representative, as these regions are close to the boundary of the ASR domain.
Extensive evaluation of ASR (e.g., Moore et al. 2015; Bromwich et al. 2016; Tilinina et al. 2014) demonstrated the added value of high-resolution nonhydrostatic model settings in improving the representation of polar lows and tip jets, as well as extratropical cyclones. In this respect, our comparison of HiRes and LoRes simulations confirms the conclusions drawn from the ASR evaluation. It also clearly demonstrates the added value of high-resolution and nonhydrostatic model settings in NAAD over the whole North Atlantic. Specifically, NAAD HiRes provides the possibility to resolve mesoscale dynamics associated with high winds, first of all in the North Atlantic subpolar latitudes characterized by small-scale polar lows and tip jets. Here NAAD HiRes demonstrated stronger extreme winds and their better localization compared to the LoRes version and modern reanalyses. Much higher resolution of the WRF Model in NAAD HiRes provides a new view of the North Atlantic extratropical cyclone activity with 2 times as large a total number of cyclones counted in NAAD HiRes relative to LoRes and most reanalyses. This difference was primarily due to cyclones that are smaller in size and relatively shallow, which were poorly simulated in the LoRes experiment and in global reanalyses.
Higher extreme turbulent fluxes in NAAD HiRes and a better representation of the convective precipitation over the Gulf Stream make NAAD potentially useful for quantifying ocean–atmosphere interactions at mesoscales and depicting ocean impacts on the lower atmosphere and associated responses in the dynamics of midlatitude storm tracks. NAAD is also capable of capturing highly localized mechanisms of atmospheric moisture transports such as ARs, more accurately locating them, and quantifying their intensity and impacts. In the tropics, NAAD HiRes also effectively captures mesoscale features, including an improved representation of tropical cyclones, especially in terms of central pressure, wind, and precipitation. Applications of NAAD to ocean modeling demonstrated the effect of HiRes onto the modeled ocean state and eddy kinetic energy distribution, specifically showing smaller surface temperature and upper-ocean heat content consistently with observations. Being applied to wind-wave modeling, NAAD HiRes resulted in higher simulated extreme wind waves in the eastern subpolar North Atlantic, reflecting stronger extremeness of surface winds.
Further near-time developments of the NAAD will include the adaptation of ERA5 as a source for lateral boundary conditions and changing to a finer spatial resolution of at least ~3 km with a higher number of vertical layers. This will also include at least for the period after 2000 the use of high-resolution SST and sea ice data, available from the Operational Sea Surface Temperature and Sea Ice Analysis (OSTIA) system (Roberts-Jones et al. 2012; Donlon et al. 2012) as well as from Multisensor Analyzed Sea Ice Extent (MASIE) AMSR2 (MASAM2; Fetterer et al. 2015). Improved representation of sea ice will help to minimize biases in temperature and humidity in the northern North Atlantic. Also planned is the domain extension to the full coverage of the North Atlantic tropics that will make it possible to provide accurate diagnostics of tropical cyclone dynamics including the poleward shift in the trajectories (Studholme and Gulev 2018; Sharmila and Walsh 2018). Finer spatial and vertical resolution will also provide a better representation of mesoscale features and will better demonstrate the value added by nonhydrostatic model configurations. For selected years (provisionally in the 2010s), we are also planning to develop ensemble simulations (up to 10 members). In parallel, on a midterm scale we will work on the transition of NAAD to the North Atlantic regional reanalysis with assimilating all available information over the domain that will make it possible to develop the product for assessing the impact of mesoscale processes onto longer-term climate variability in the atmosphere and the ocean.
We acknowledge the support by the Russian Academy of Sciences (ST-ASS-0149-2018-0001) and the Ministry of Science and Higher Education of Russian Federation under the Agreement 14.W0331.0006. Authors Tilinina and Markina were also supported by Grant 17-77-20112 from the Russian Science Foundation. The analysis of the hydrological cycle (authors Zolina, Dufour, and Krinitskiy) benefited from Helmholtz-RSF Grant 18-47-06202. We greatly appreciate critically important comments of the two anonymous reviewers and the editor. Their insightful suggestions helped to largely improve the first version of the paper. We thank Peter Koltermann of MSU for useful comments on the paper text. Our analysis of mesoscale dynamics was strongly motivated by Peter Arseev of PhIRAS and Alexey Morozov of ITEP. Discussions on the model setting with Andrey Glazunov of INMRAS are appreciated. We acknowledge the access to the facilities of HPC computing resources at Lomonosov Moscow State University as well as support of the IORAS administration and RSC-Technologies Group in maintaining supercomputers and data storage systems at IORAS.
Denotes content that is immediately available upon publication as open access.