Assessment of responses of North Atlantic winter SST to the NAO in 13 CMIP5 models on the interannual scale

Assessment of responses of North Atlantic winter SST to the NAO in 13 CMIP5 models on the interannual scale Yujie Jing, Yangchun Li, Yongfu Xu State Key Laboratory of Atmospheric Boundary Layer Physics and Atmospheric Chemistry, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029, China 5 Department of Atmospheric Chemistry and Environmental Sciences, College of Earth and Planetary Sciences, University of Chinese Academy of Sciences, Beijing 100049, China


Introduction
The most notable climatic event in the North Atlantic (NA) is a strong inverse relationship between Iceland's and the Azores' monthly mean sea level pressure (most significant in winter), which is called the North Atlantic Oscillation (NAO) (Walker, 1924). Studies have shown that the NAO has a significant impact on climate change in the Northern 30 Hemisphere, including the significant impact on temperature and precipitation in Europe and the NA (Trigo et al., 2002). Because of the internal atmospheric dynamic process, the NAO is closely related to the location and intensity of storm track in the NA (Riviè re and Orlanski, 2007). In addition, the NAO not only impacts the atmospheric field, but also the oceanic field through air-sea interactions, such as the sea surface temperature (SST) in the NA.
The influence of the atmospheric anomalies on SST is mainly reflected in the change of sea surface heat flux 35 models (AOGCMs), and pointed out that the spatial pattern of the NAO is more reasonable, but the action center of high pressure is west of the observation comparing with observation. In addition, Woollings. (2014) have simulated the 55 mechanism of change of the NAO with an atmospheric circulation model (HiGEM), and proposed the impact of jets in the upper troposphere on the change of the NAO. The Coupled Model Intercomparison Project Phase 5 (CMIP5) (Taylor et al., 2012) includes more Earth system models with higher spatial resolution, which helps to better understand ocean and atmospheric variability and their interaction. In recent years, more and more people have realized that the evaluation of the CMIP5 Earth System Models (CMIP5-ESMs) is the basis for study by these models. 60 For example, Wang et al. (2017) paid attention to the ability of the CMIP5-ESMs to simulate annual NAO and found that basically all models can reasonably reproduce the spatial distribution of the NAO. Meanwhile, Wang et al. (2014) evaluated the global SST simulated by the CMIP5-ESMs and found that the SST in the Northern Hemisphere, especially in the NA, is underestimated, and pointed out that it is mainly caused by the unreasonable simulation of AMOC. Liu et al. (2013) evaluated the SST variability in the NA warm pool simulated by 19 CMIP5-ESMs, and 65 considered that the deviation of radiation balance caused by the CMIP5-ESMs' unreasonable simulation of high-level cloud fraction can impact the SST variability. The relationship between the SST and NAO in the North Atlantic from the CMIP5-ESMs has not been systematically evaluated, but it is of great significance to study the North Atlantic variability and climate change in the entire Northern Hemisphere. Multiple observation-based studies have indicated that there is a close connection and strong interaction between the NAO and the tripolar pattern of winter SST 70 anomalies (Czaja and Frankignoul., 2002;Chen et al., 2015). Therefore, the purpose of this paper is to evaluate https://doi.org/10.5194/os-2020-16 Preprint. Discussion started: 11 May 2020 c Author(s) 2020. CC BY 4.0 License.
whether the CMIP5-ESMs can simulate the relationship between the NAO and SST in winter in the NA (0-65°N), to investigate the mechanism of the response of SST to the NAO, and to explore the deviation of models in simulating the response mechanism of SST to the NAO.

Data
The observation-based data in this study are monthly sea level pressure (SLP) from 1901 to 2010 from the reanalysis dataset of the 10-member ensemble of coupled climate reanalysis of the 20th century (CERA-20C) of the https://www.esrl.noaa.gov/psd/data/gridded/data.godas.html, Behringer and Xue, 2004). The 13 Earth System Models used for this work are from the historical experiment of CMIP5 (Table 1, cera-www.dkrz.de, Taylor et al., 2012). The 90 simulation results from these models provide the monthly average data of SLP, SST, latent/sensible heat net flux, and sea water Y velocity during 1850-2005. In order to make comparisons and analyses between the simulated and observed results, all variables are interpolated into a spatial resolution 1°×1° by linear interpolation, and the time range of all variables from observations and models is 1902-2010 and 1897-2005, respectively.

95
Two definitions of the observation-based winter NAO are used in this work. The index proposed by Gong and Wang (2000) is expressed as where P* represents the normalized SLP. A three-point (10°W, 0°E, and 10°E for the high-pressure area and 10°W, 20°W, and 30°W for the low-pressure area) spatial arithmetic average of P* differences between the high-and 100 low-pressure area is used. The other NAO index is a site-based one from the Climate Analysis Section of the National Center for Atmospheric Research (NCAR; www.climatedataguide.ucar.edu/climate-data/hurrell-north-atlantic-oscillation-nao-index-station-based, Hurrell and Deser, 2009).
Because the locations of the NAO action centers are different between the model and the observation, and 105 https://doi.org/10.5194/os-2020-16 Preprint. Discussion started: 11 May 2020 c Author(s) 2020. CC BY 4.0 License. between the different models, we use the method proposed by Wang et al. (2017) and Zheng et al. (2013) to define the NAO index based on the results of the models. This method is basically similar to the one proposed by Gong and Wang (2000), in which the winter NAO index is defined as the difference in the normalized SLP, zonally averaged over the North Atlantic sector (30-80°N; 80°W-30°E), between the two latitudes that have the strongest negative correlation in SLP variability. 110 The winter duration used to define the winter NAO index / SST / wind speed / turbulent heat flux is December-January-February (DJF). For the variables of DJF, the January in the given year is used as the reference to obtain the winter variables. In other words, the variables for DJF of 1980 are obtained based on data in December of 1979 and January and February of 1980. For the winter variables, when calculating the seasonal average NAO index (DJF), the winter season average of SLP is firstly calculated, and then the NAO index is obtained. 115 Because the main cycles characterized by the interannual and decadal signs of the NAO are within 2-6 years and above 8 years, respectively (Jing et al., 2019), the interannual scale is extracted using a 2-6 year Lanczos band-pass filter. For the regression analysis between the NAO and ocean physical variables, the effective degree of freedom (DOF) is calculated following Bretherton et al. (1999): (2) 120 Where N is the sample size, r1 and r2 are the lag-one autocorrelations of the two time series, respectively.
In order to understand the mechanisms of the impact of the NAO on SST, we need to know the main factors leading to the change of SST. The variability of SST is described by: https://doi.org/10.5194/os-2020-16 Preprint. Discussion started: 11 May 2020 c Author(s) 2020. CC BY 4.0 License.
Here, C0 is the thermal capacity of the upper mixed layer of the ocean, which is approximately constant, A is the divergence of ocean heat transport, Q is the air-sea heat flux and has both radiative (QR) and turbulent (QB) components. The turbulent heat fluxes are the sensible (QS, SHF) and latent (QL, LHF) heat fluxes (the positive value indicates the flux from the sea surface to the atmosphere). Among them, the sensible and latent heat fluxes are mainly related to wind speed (U) and SST, which are usually calculated by the following equations: 130 whereρis a near surface air density, Cp is the specific heat of the air, Lp is the latent heat of evaporation, CS and CL are the transfer coefficients of sensible and latent heat fluxes, respectively, Ta is the temperature of the atmosphere near the sea surface,  ⃗ ⃗ = ⃗ ⃗ -⃗ ⃗ is the vector difference between the wind speed at the sea surface and the sea 135 surface current speed, in which the current speed is often neglected, qa and qs correspond to saturation specific humidity of air over sea surface and sea surface temperature, respectively. qs is usually calculated by the saturation humidit qsat, for pure water at SST: where a multiplier factor of 0.98 is used to take into account reduction in vapor pressure caused by a typical 140 salinity of 34 psu. https://doi.org/10.5194/os-2020-16 Preprint. Discussion started: 11 May 2020 c Author(s) 2020. CC BY 4.0 License.

Space state
An empirical orthogonal function (EOF) analysis is performed on the normalized winter-averaged North Atlantic 145 sea level pressure to obtain the first mode (EOF1) of the sea level pressure (SLP) field, that is, the NAO mode (Hurrell and Deser, 2009). Figure 1 shows the NAO modes of the observation and CMIP5-ESMs simulations. The NAO mode calculated with the observed SLP is significant, which explains 40.4% of the total variance. The explanation variances of the NAO mode by the models are close to those from the observations, ranged from 37.1% -53.4%. The result of observation shows that the low-pressure action center of the NAO is at about 65 ° N and from 30 ° W to 20 ° E, and 150 that the high-pressure action center is at 40 ° N and from 32 ° W to 20 ° E ( Fig. 1). Compared with the location of the NAO action centers obtained with EOF1 of the same set of observed SLP from 1950 to 2010 by Jing et al. (2019), the location of the high-pressure action center is more eastward during 1902-2010, indicating that the location of the NAO action centers at different time periods will be slightly biased. This is consistent with previous results (Jung et al., 2003;Moore et al., 2013;Jing et al., 2019). The simulated locations of the NAO action centers by the CMIP5-ESMs are 155 basically reasonable, although there are some slight differences of the NAO action centers between different models and between the models and observation. Because the locations of the NAO action centers simulated by most of the CMIP5-ESMs in different NAO phases do not show the movements illustrated by the observation (the figure is https://doi.org/10.5194/os-2020-16 Preprint. Discussion started: 11 May 2020 c Author(s) 2020. CC BY 4.0 License. omitted), the differences between the models are not caused by the NAO period or the phase of the initial sign, but are only related to the structures of models. 160 Figure 2 shows the observed and simulated multi-year average of winter SST in the NA (0-65°N). The CMIP5-ESMs can basically reproduce low-temperature center near the Labrador Sea. The spatial correlation coefficients with the observations are all above 0.98, reaching a significance level of 99%. Nevertheless, CMIP5-ESMs underestimate the SST. Wang et al. (2014) pointed out that the underestimation of the multi-year mean SST in the NA by the CMIP5-ESMs may be due to the weaker Atlantic meridional overturning circulation (AMOC) and the 165 unreasonable vertical structure of it (the weaker northward heat transport caused by shallower AMOC cell). In the term of interannual variability of winter SST in the NA (0-65°N) (Fig. S1), all CMIP5-ESMs can reproduce the strong interannual variability of SST in the Gulf Stream extension, but the simulated strong interannual variability of SST by most models is more easterly than the observations. In addition, some models also simulate strong interannual variability at higher latitudes that is not observed. 170 Figure 3a shows the periods of the NAO indexes calculated with the method proposed by Gong and Wang (2000) based on the observation and with the method proposed by Zheng et al. (2013) based on the models. The significant periods (at a 90% confidence level) of the observed NAO index are 2.3-2.7, 4.7-5.8, and 8. Based on the above analysis, simulated periods of the NAO indexes and area-averaged SST anomalies on the 185 decadal scale are different from the results of observations. In addition, the impact of the atmospheric anomalies (NAO) on SST in the NA is mainly reflected in the impact of local change of wind stress on the sea-air heat flux on the interannual scales (Eden and Jung, 2001;Chen et al., 2015;Han et al., 2016). Therefore, we will extract the interannual signal of 2-6 years by band-pass filter based on the periods of the NAO and area-averaged SST anomalies to evaluate the relationship between the simulated NAO and SST in the CMIP5-ESMs on the interannual scale.

The role of wind speed 200
Since the influence of the NAO on the SST is mainly through the wind field in the NA (Zhou et al., 2006;Deser et al., 2010), in order to evaluate the mechanism of the influence of the simulated NAO on the SST in the NA (0-65°N), the response of the wind speed to the NAO should be firstly considered. Figure  that during the positive phase of the NAO, the deepening of the low pressure in Iceland causes the abnormal east wind 210 superimposed on the mid-latitude westerly wind, which weakens the mid-latitude wind speed (Deser et al., 2010;Chen et al., 2015).
According to the Eq. (3-6), the wind speed anomalies impact the SST by affecting the turbulent heat flux, but the wind speed only affects the magnitude of the turbulent heat flux. Therefore, when analyzing the effect of wind speed anomalies on the turbulent heat flux, it is necessary to consider the direction of the turbulent heat flux that is 215 determined by the difference of temperature and specific humidity between the atmosphere and the sea surface. From the results of the multi-year averaged winter sensible / latent heat fluxes (SHF / LHF, Fig. S2), the observed and simulated SHF / LHF are all from the sea to the atmosphere. The main difference between the observations and models is that all models overestimate the SHF in the region north of 50° N, and some models such as CESM1-BGC, GFDL-ESM2M / 2G, MPI-ESM1 and NorESM1-ME simulate the strong LHF in the subpolar NA that is not observed. 220

The role of SHF
The variability of SHF and SST is related. According to the calculation formula of SST and SHF, the increase of SHF can decrease SST (Eq. 3-4), while the decreased SST can further decrease the SHF (Eq. 5). Therefore, when the 240 variations of SST and SHF are negatively correlated, it can be inferred that the change of SHF influences SST, which means that the atmosphere forces the ocean; when the variations of SST and SHF are positively correlated, the change of SST leads to the change of SHF, which means that the ocean forces the atmosphere. have an impact on the variations of the SST in these regions beyond the SHF. When the change of SHF is synchronized with the change in SST, most models show that there is no effect of SHF on SST in the subpolar NA according to the positive covariance. This may be caused by faster feedback of the SST on atmospheric forcing in the models than that in the observation, or the overestimated influence of other factors in those models. 275

The role of LHF
The LHF is calculated by wind speed and the difference between the saturation specific humidity of lower air and sea surface. Because the saturation specific humidity of sea surface is a function of SST (Eq. 7), according to the calculation formulas of SST and LHF (Eq. 3-4, 6-7), the relationship between LHF and SST is similar to the one https://doi.org/10.5194/os-2020-16 Preprint. Discussion started: 11 May 2020 c Author(s) 2020. CC BY 4.0 License. between SHF and SST. It means that when the variations of the SST and LHF are negatively correlated, the 280 atmosphere forces the ocean through the LHF, and that when the variations of the SST and LHF are positively correlated, the ocean forces the atmosphere. Figure 9a is the RCs of the observed and simulated winter-averaged SST anomalies against the LHF anomalies.
The distributions of the RCs in the observation-based and modeled result are very similar to those of the SST anomalies against SHF anomalies. A main difference between these two sets of RCs is that the large range of negative 285 values of the SST anomalies against LHF anomalies in the observation-based result and three models, namely CanESM2, CESM1-BGC, and NorESM1-ME, appear to be more southward than those of the SST anomalies against SHF anomalies. The simulated RCs of the SST anomalies against LHF anomalies are still positive in large regions of the subpolar NA in most of models, which can enhance the unrealistic response of the SST anomalies to the NAO in these regions. 290 Figure 9b shows the lagged (leaded) covariance between the anomalies of the LHF and SST. It can be seen that the change of the LHF is strongly related to the change of SST in most regions of the NA (0-65°N). When the observed SST anomalies lag (lead) those of the LHF by 2 months, there is an obviously negative (positive) covariance in a large region of NA, which indicates that the change of LHF (SST) can influent the change of SST (LHF) after two month.
Compared with the relationship between the SHF anomalies and the lagged SST anomalies, the area where the LHF 295 plays an impact on the lagged SST is larger, and covers most regions of the NA except for the western NA from 30°-40°N with positive RCs. When the changes of SST and LHF are synchronized, there is an obviously positive https://doi.org/10.5194/os-2020-16 Preprint. Discussion started: 11 May 2020 c Author(s) 2020. CC BY 4.0 License. covariance in the large region except in the subpolar NA, which is also different from the relationship of the synchronized SHF anomalies and SST anomalies. This demonstrates that the time-scale of the LHF affecting SST is shorter than that of the SHF affecting SST, and the ocean plays an important role in the interaction of LHF and SST. 300 The CMIP5-ESMs basically reproduce the lagged or leaded relationship between the SST anomalies and the LHF anomalies, whereas the role of the ocean is overestimated. When the SST anomalies lag the LHF anomalies, most models except for CESM1-BGC and NorESM1-ME simulate a large region of positive covariance in the subtropical and subpolar NA, which only occurs in a small region in the observation-based results. When the two variables in the models are synchronized, their positive covariance simulated by models is significantly stronger than that in the 305 observation-based results. It can be concluded that the oceanic forcing on the atmosphere through the LHF variation is enhanced in the models, which results in a significant difference between the observed and simulated response of the winter-averaged SST anomalies to the LHF anomalies (Fig. 9a).

The role of advection
The response of the SST anomalies to the NAO-driven SHF / LHF anomalies in the subtropical NA simulated by 310 CMIP5-ESMs can partly explain the unrealistic response of the SST anomalies to the NAO in some models, but cannot explain this response in the other models that generate an incorrect response of the SST anomalies to the SHF / LHF anomalies but a relatively reasonable response of the SST anomalies to the NAO. In the subtropical NA, it is also worth noting that some models such as HadGEM2-CC / ES and MPI-ESM-LR / MR cannot reproduce the significant negative RCs of the winter-averaged SST anomalies against the LHF anomalies that is observed in some areas of the 315 subtropical NA (Fig. 9a), but these models can reproduce the RCs of the anomalies of winter-averaged SST against the NAO in these areas (Fig. 4). This suggests that there are other factors that are driven by the NAO, which may influence SST in the subtropical NA.
In addition to the turbulent heat flux, the changes of long / short wave radiation and the ocean circulation also have effects on the change of SST. The long-wave radiation on the sea surface is mainly determined by SST, while the 320 change of short-wave radiation does not have a strong relationship with NAO (the figure is omitted). The simulated relationship between SST and the NAO by the CMIP5-ESMs may be also related to the NAO-driven horizontal heat advection. Some studies have pointed out that during the positive phase of the NAO, the northeast advection is strengthened and the heat transfer from south to north is also enhanced. Compared with other seasons, this phenomenon is more obvious in winter (Flatau et al., 2003;Bellucci et al., 2006). Nevertheless, some other studies 325 have argued that the impact of ocean heat advection on the change of SST in the subtropical NA is mainly on the decadal scale (Delworth et al., 1998, Krahmann et al., 2001. From the observation-based RCs of surface sea meridional velocity anomalies against the NAO in winter (Fig. 10), it can be seen that on the interannual scale, in the subtropical NA the observed meridional velocity (positive values indicate poleward advection) anomalies has a significant positive response to the NAO, and in the subpolar NA it has a significant negative response to the NAO. reproduce the response of the SST anomalies to the LHF anomalies, and thus also reproduce the positive response of the SST anomalies to the NAO in the subtropical NA. In the subpolar NA, the response of the heat meridianal transport to the NAO also can explain partly that some models (such as GFDL-ESM2G / 2M) can reproduce a reasonable response of the SST anomalies to the NAO with an unrealistic response of the SST anomalies to the NAO-driven turbulent heat flux. 345

Conclusion
We evaluated the influence mechanism of the NAO on the SST in the NA (0-65°N) simulated by CMIP5-ESMs.
Because there is a deviation between the simulated and observed periods of the NAO indexes / area-averaged SST on the decadal scale, we mainly evaluated the simulation of the relationship between the winter-averaged SST and NAO by 13 CMIP5-ESMs on the interannual scale. 350 Based on the observations, the response of winter-averaged SST anomalies to the NAO shows a significant tripolar distribution along the meridian in the NA. Only 7 models can reproduce an observed tripolar pattern of the response of SST anomalies to the NAO, and other 6 models produce a significant positive response of the SST anomalies to the NAO that is not observed in some areas of the subpolar NA (45-65°N). In the subtropical NA heat transport leads to the increase of SST in the subtropical NA and the decrease of meridianal heat transport leads to the decrease of SST in the subpolar NA during the positive phase of the NAO. This mechanism probably conceals the unreasonable impact of the heat turbulent flux anomalies on the SST anomalies in some models, so that the simulated 370 response of the SST anomalies to the NAO by all models is positive in the subtropical NA, which is also reflected in the observation-based results, although the response intensity in the models is much weaker than that in the observation-based results.
Data availability.

375
The addresses of downloading all data used in this study have been described in Sect. 2.1. All data for results are available by contacting the corresponding author.