Spring Aleutian Low Weakening and Surface Cooling Trend in Northwest North America During Recent Decades
Abstract
State-of-the-art climate models predict an enhanced warming over global land areas in response to the greenhouse gas (GHG) forcing. In nature, however, the observed land surface warming is not spatially uniform in recent decades as GHG concentration has risen rapidly. There is no warming and even cooling trend in spring northwest North America (NWNA) surface temperature since 1980. Here, we link the observed multidecadal surface cooling trend in NWNA to the weakening of spring Aleutian low (AL), which is ecologically important during the growing season. The AL weakening leads to reductions of southwesterly wind and warm air transport from eastern North Pacific to NWNA. The spring AL weakening and associated NWNA surface cooling are reasonably reproduced by AMIP models, driven by the historical external forcing and imposed sea ice concentration (SIC) and sea surface temperature (SST) from observations. Further investigations into the roles of different forcing components suggest that the changes in the AL are most likely driven by the tropical SST changes in recent decades while the external forcing and SIC play minor roles. There is a zonal dipole pattern of trends in tropical SSTs since 1980, with central-eastern Pacific cooling and warming in other tropical basins, and the atmospheric responses to this SST change show a meridionally propagating Rossby wave train from tropics into subarctic North Pacific, which mediates the SST forcing and the AL changes.
Key Points
- Spring Aleutian low is weakening in recent decades causing unexpected surface cooling over Northwest North America
- The recent spring Aleutian low weakening and Northwest North America cooling are reproduced in a large number of AMIP model simulations
- Multidecadal tropical SST change, rather than external forcing or sea ice loss, is responsible for the recent spring Aleutian low weakening
1 Introduction
The Aleutian low (AL) is one of the two semipermanent low-pressure systems (the other is the Icelandic low) in the Northern Hemisphere (NH), centered over the Aleutian Islands, southwest of Alaska in the North Pacific (Namias, 1969). It is active during the late fall to late spring, most pronounced and dominant over most of the North Pacific during the winter (Pickart et al., 2009). During the summer, the subtropical high pressure system dominates over the North Pacific basin, leading to a retreat of AL toward the North Pole, and during this time of year, the AL becomes almost nonexistent. The AL generates many storms that travel eastward along the polar front during its active season, and relatively higher humidity and air temperatures prevail across the Gulf of Alaska and northwest North America (NWNA).
Variability in the strength and location of the winter AL has been studied extensively. The interannual variability of winter AL strength is closely related to the tropical El Niño-Southern Oscillation (ENSO) phenomenon, which excites a meridional Rossby wave train spanning from the equatorial Pacific to the NWNA, known as the Pacific and North America (PNA) teleconnection pattern (Niebauer, 1988; O'Reilly, 2018; Trenberth, 1990; Wallace & Gutzler, 1981). The El Niño (La niña) forces a positive (negative) phase of PNA pattern and strengthens (weakens) the AL, leading to a warmer-than-normal (colder-than-normal) winter for the Bering Sea and the NWNA region (Liu et al., 2007; Niebauer, 1988).
The decadal variation of the winter AL is known to be coupled with the oceanic Pacific Decadal Oscillation (PDO, also named as Interdecadal Pacific Oscillation, in short IPO, in the whole Pacific), which is an important driver of decadal winter climate variability of western North America (Mantua et al., 1997; Minobe, 1997). The winter AL was significantly enhanced from 1976 to 1988 (Overland et al., 1999) when the PDO showed a remarkable upward trend. This decadal-scale change corresponds well to the documented climate and ocean regime shift of 1976 (Trenberth, 1990; Trenberth & Hurrell, 1994). The strengthened AL was accompanied by stronger surface westerlies across the midlatitude North Pacific and intensified southerly to southeasterly flow along the northwest coast of North America, leading to a decadal-scale warming of the NWNA (Mantua et al., 1997). The changes in the intensity of AL are often associated with the shift of the center of action of the AL, and during the 1976–1988 strengthening of winter AL, an eastward shift of AL center was observed (Trenberth, 1990; Trenberth & Hurrell, 1994). Some studies also linked the location of the AL to the sunspot numbers and found that the AL migrates eastward when solar activity is at a minimum during its 11-year cycle (Christoforou & Hameed, 1997; van Loon et al., 2007).
At longer time scales, the winter AL intensity was significantly enhanced during the last 120 years (Gan et al., 2017), particularly for the second half of the twentieth century (Deser & Phillips, 2009; Gillett et al., 2003; Lu et al., 2004). This long-term strengthening trend corresponds to an intensification of basin-scale cyclonic flow, which brings warmer and wetter air masses to the eastern coast of North Pacific, contributing to the winter warming and moistening in the NWNA (Deser & Phillips, 2009). Several previous studies have linked the long-term strengthening trend of winter AL to the influence of anthropogenic greenhouse gases (Gan et al., 2017; Gillett et al., 2003), and some other studies emphasized the role of SST forcings over tropical and extratropical regions (Deser et al., 2004; Sugimoto & Hanawa, 2009).
Although the climatological AL is most pronounced in winter, the spring AL is still active and ecologically important. High marine primary production in the subarctic North Pacific region (including the Bering Sea and Gulf of Alaska) and vegetation growth in middle-high latitude North America occur during the spring growing season (Beamish & Bouillon, 1993; Mantua et al., 1997; Ueshima et al., 2006; Wang et al., 2011), and these productivities are sensitive to surface temperature over the pan-North Pacific region that is closely coupled with the intensity of the AL (Mantua et al., 1997). However, the variability and changes in the intensity of spring AL is still poorly understood relative to the winter counterpart. It is still unknown whether the spring AL shows a similar multidecdal trend to that observed for the winter AL, and the driving force behind the multidecadal change in the intensity of spring AL remains elusive. In this study, we attempt to address these questions by performing comprehensive analyses of multisource observational data sets and multimodel simulation outputs and to reveal the spatiotemporal features of multidecadal changes in spring AL and the driving mechanism. The remainder of this paper is organized as follows. The data sets and methodology are described in section 2. The spatiotemporal characteristics of multidecadal changes in spring AL during recent decades are investigated in section 3, and the multimodel simulation of this change in multiple models are thoroughly analyzed. The driving forces of the recent change of spring AL are also discussed. Finally, the main conclusions and a discussion are given in section 4.
2 Data and Methodology
Multisource observation/reanalysis data of sea level pressure (SLP) and land surface air temperature (LSAT) are employed to analyze the spatiotemporal features of the multidecdal changes in the spring AL. The characteristics of the data sets and their sources are completely listed in Table 1. We focus on the recent multidecadal change in AL and the impact on the LSAT of its surrounding continents, so the present analysis is conducted for the period from 1980 to 2013. The analysis period for the observation/reanalysis data ends in 2013 not only for comparing the multidecadal trend among the different observational/reanalysis data sets (20CR2 reanalysis ends in 2012) but also for assessment of the CMIP5 multimodel outputs, which end before 2010 (details are given in Table 2). In fact, the results are largely similar if we extend the trend analysis over a longer time period (1980–2016, shown in Figures S1 and S2 in the supporting information). A spring (March-April-May) AL index is calculated as the area-weighted average of SLP anomalies over the region 45°–65° N and 160°E to 160°W based on the position of the center of action of climatological spring AL (the climatological location of AL is shown in Figure 1 as contour), which displays a northward retreat compared to the winter counterpart (Gan et al., 2017; Namias, 1969).
| Abbreviation | Name | Analysis period | Variable used | Reference |
|---|---|---|---|---|
| 20CR2 | NOAA-CIRES 20th Century Reanalysis Version 2 | 1980–2012 | SLP, SAT, surface wind | (Compo et al., 2011) |
| MERRA | The Modern Era Retrospective-Analysis for Research and Applications from NASA | 1980–2013 | SLP, LSAT, surface wind | (Rienecker et al., 2011) |
| ERA-Interim | ERA-interim reanalysis from ECMWF | 1980–2013 | SLP, LSAT, surface wind | (Dee et al., 2011) |
| HadSLP2 | Hadley Centre Sea Level Pressure data set | 1980–2013 | SLP | (Allan & Ansell, 2006) |
| NCARSLP | Monthly gridded Northern Hemisphere sea level pressure data from NCAR | 1980–2013 | SLP | (Trenberth & Paolino, 1980) |
| CRU_TS | Climatic Research Unit (CRU) TS (time series) data sets version 4.01 | 1980–2013 | LSAT | (Harris et al., 2014) |
| GHCN_CAMS | CPC GHCN_CAMS Monthly Global Surface Air Temperature (SAT) Data Set | 1980–2013 | LSAT | (Fan & van den Dool, 2008) |
| Modeling center | Model name | Simulation period | Ensemble size |
|---|---|---|---|
| Canadian Centre for Climate Modelling and Analysis | CanAM4 | 1950–2009 | 4 |
| CSIRO in collaboration with Queensland Climate Change Centre of Excellence | CSIRO-Mk3-6-0 | 1979–2009 | 10 |
| NOAA/Geophysical Fluid Dynamics Laboratory (GFDL) | GFDL-CM3 | 1979–2008 | 5 |
| National Aeronautics and Space Administration Goddard Institute for Space Studies (GISS) | GISS-E2-R | 1951–2010 | 6 |
| L'Institut Pierre-Simon Laplace (IPSL) | IPSL-CM5A-LR | 1979–2009 | 5 |
| Max Planck Institute for Meteorology (MPI-M) | ECHAM5 | 1979–2018 | 30 |
A set of the Atmospheric Model Intercomparison Project (AMIP; Gates et al., 1999) model outputs as part of the CMIP5 (Taylor et al., 2012) are used, and the details of the models are summarized in Table 2 (including model name, the associated center name, ensemble member number, and simulation period). The simulation data are available from the Earth System Grid Federation (ESGF) through the Program for Climate Model Diagnosis and Intercomparison (PCMDI). The AMIP models are uncoupled models forced by observed monthly SSTs and sea ice concentrations during the historical period, as well as other time-varying external conditions (various forcing agents including CO2 concentrations and aerosols) same as in the CMIP5 historical experiment of fully coupled models (Taylor et al., 2012). Only five models that provided at least four ensemble members (some have as many as 10) are analyzed, and the ensemble means for each model are derived and used in the analysis. This restriction ensures a large enough ensemble size to isolate the simulated forced component of the variations in atmospheric circulation due to the variability of forcings (i.e., SST/sea ice boundary conditions and radiative forcings). Another 30-member ensemble of the same AMIP-type experiment using the ECHAM5 model is also employed, and the results from this experiment are compared with the other AMIP simulations to show the impact of ensemble size. A similar AMIP experiment using the ECHAM5 model but with the sea ice boundary conditions set to a repeating seasonal cycle of 1979–1989 is also performed, and this experiment is referred to as AMIP_Clim_Polar. Thus, the difference between the AMIP and AMIP_Clim_Polar experiments largely represents the effect of sea ice melting on the atmospheric circulation. The ECHAM5 model simulation data (AMIP and AMIP_Clim_Polar) are provided by the NOAA ESRL Physical Sciences Division, Boulder, Colorado, USA, from their website (at http://www.esrl.noaa.gov/psd/). Compared with the observation/reanalysis, the AMIP models simulate a similar position of climatological spring AL with low pressures prevailing over the Aleutian Islands (not shown here). Therefore, the definition of AL index in the AMIP simulations is similar to the observation/reanalysis for a comparison of the results and assessment of the AMIP model outputs.
The SST/sea ice boundary conditions for the AMIP experiment were derived from the HadISST data set (Hurrell et al., 2008; Rayner et al., 2003), and they are also available through the PCMDI (https://pcmdi.llnl.gov/). To estimate the SLP and LSAT responses to the external radiative forcing, we use the fully coupled models from the CMIP5 historical experiment, of which a total of 59 ensemble members from 27 CMIP5 models are selected. Since the number of selected ensemble members is large enough, the multimodel ensemble means (MMEMs) of the CMIP5 fully coupled models reasonably represent the forced signal due to the time-varying radiative forcings in the models, as the forced signal in all ensemble members is common and the internal variability in different runs is uncorrelated and cancels out in the MMEMs (Dong & McPhaden, 2017a; Kravtsov et al., 2018).
Linear trends of anomaly time series for both observation and simulation data are calculated using a nonparametric estimation technique based Sen's slope estimator (Sen 1968), since this trend-estimation method makes no assumption of a distribution for the residuals and is much less sensitive to the effect of outliers in the time series than the least squares method. The Sen's slope method is a robust estimate of linear trend that has been used widely in the studies of hydrological and extreme climate change (Kosaka & Xie, 2013; Sun et al., 2018; Zhao et al., 2016). The statistical significance of the Sen's slope is tested by using nonparametric Mann-Kendall test method (Zhao et al., 2016).
3 Results
3.1 Spring AL Change in the Observations and Reanalysis
In order to identify the multidecadal variation of spring AL and its spatial features, trend analysis of both spatial distribution and time series based on the spring SLP anomalies are conducted. Figure 1 shows the spatial distributions of North Pacific SLP trends in three reanalysis data sets (20CR2, MERRA, and ERA-Interim) and two observational datasets (HadSLP2 and NCARSLP). Based on the data since 1980, the North Pacific SLP trends are calculated at each grid cell using the trend estimation method. The Mann-Kendall test is further applied to determine whether each grid cell has a statistically significant trend at the 90% confidence level. The five data sets show generally consistent pattern of SLP trends over the North Pacific basin since 1980. All data sets exhibit an increasing trend of SLP over the subarctic North Pacific, where the spring AL is most active, and this increasing trend is statistically significant in all the datasets. The differences are minor and the trend analysis results from different data sets only differ in the trend magnitude of SLP over the subarctic North Pacific, and the magnitudes of SLP increasing trend in two observational data sets are slightly weaker than those in the reanalysis data sets. Thus, the geographical patterns of North Pacific SLP trends in multisource reanalysis/observational data sets suggest a weakening trend of spring AL in recent three decades.
Figure 2 displays the spring AL index (area-weighted average of SLP anomalies over the subarctic North Pacific) based on different reanalysis and observational data sets. The variations of spring AL are highly consistent among different data sets, and the differences are negligible. The AL SLP shows large interannual variability, superimposed on an upward trend. All the data sets have increasing trend of AL SLP, ranging from 0.96 to 1.51 hPa (10 year)-1 (shown in Figure 2), and all these trends are statistically significant. The trend analysis of spring AL index is consistent with the spatial patterns of SLP trends in Figure 1, and both confirm a weakening trend of spring AL during recent decades. The winter AL shows a strong strengthening trend during the second half of the 20th century (Gan et al., 2017; Gillett et al., 2003) and reverses to a weakening trend after 2000 (Farneti et al., 2014), leading to no clear trend in the winter AL strength during the analysis period 1980–2013 (not shown here). Thus, the spring AL weakening is clearly different from the multidecadal variability of the winter counterpart. This suggests that the multidecadal changes of AL are seasonally dependent and the mechanisms responsible for the changes may be different for the two season.
The AL activity is closely linked with the climate over the NWNA region. We also analyze the trend of LSAT over NH continents during the same period as the SLP trend analysis using both reanalysis and observational datasets (Figure 3). In association with the recent weakening trend of spring AL, the NWNA region shows a remarkable cooling trend, in sharp contrast to the continental warming over other regions in NH. Meanwhile, the spatial pattern of the correlation between AL index and LSAT at decadal timescales shows that the surface cooling associated with spring AL weakening is most pronounced over the NWNA region, indicating a tight relationship between multidecadal changes in the AL strength and NWNA LSAT (Figure S3). The NWNA surface cooling trend is most pronounced and statistically significant in 20CR2 and ERA-Interim data sets. In the MERRA reanalysis data set, the NWNA cooling trend is weakest and statistically insignificant but also contrasting the NH continental warming. In the two observational data sets (CRU and GHCN_CAMS), the cooling trend is modest and statistically significant in large parts of the NWNA region. The cooling trend of spring NWNA LSAT revealed in present study is consistent with previous analysis of the NH LSAT trend in different seasons (Cohen et al., 2012), in which the NWNA LSAT experienced a clear cooling trend during recent decades only in the spring. The recent spring NWNA cooling is a unique feature of NH climate change in the context of land surface warming (Wang et al., 2011), and this striking cooling trend is closely associated with the weakening of spring AL. The AL weakening reduces the warm advections from North Pacific to the NWNA region and weakens the warm southerly wind over the Northwest coast of North America (Figure 4). Thus, the spring AL weakening is crucial for understanding the mechanism of the recent spring cooling of NWNA.
3.2 Simulated AL Change in the AMIP Models
In order to explore the mechanism responsible for the spring AL weakening, we investigate the simulation results from the AMIP models (see section 2 for details about the models). Figure 5 shows the simulated spatial patterns of North Pacific spring SLP trend since 1980 in six AMIP models. The models consistently reproduce the observed increasing trend of SLP over the subarctic North Pacific. Four of the six models (Figures 5a, 5b, 5e, and 5f) correctly simulate the location of SLP increasing center and the magnitude of SLP increasing trend in comparison with the observations, and the trends are statistically significant. The GISS-E2-R model simulated SLP increasing trend is the weakest among all the AMIP models, and the GFDL-CM3 model simulate the SLP increasing center slightly shifting to the east compared with the other models and the observations. The time series of the simulated spring AL index are further shown in Figure 6. Different from the observations, the interannual variations of the AL indices in the six models differ substantially from each other. Although all the AMIP models have the same forcing configurations, the internal variability due to intrinsic dynamical processes in each model is possibly strong enough to generate different atmospheric circulation variability. On the other hand, the correlations of the interannual AMIP AL indices with the Niño 3.4 SST index are all negative and ranging from −0.61 to −0.81 among different models, suggesting that AL interannual variations in the AMIP simulations are closely related to the ENSO phenomena (Figure S4). Nevertheless, all the simulated AL indices show an increasing trend of SLP since 1980, and five out of the six models reproduce a statistically significant trend of AL SLP (shown in Figure 6). Considering both the spatial patterns and time series analysis, in general, the AMIP models have considerable skill in simulating the spring AL weakening during the recent decades.
The trends of spring NH LSAT simulated in the AMIP models are also analyzed (Figure 7). The simulated spring LSAT trend exhibits a similar pattern to that in the observations, showing warming trend over the entire Eurasian continent and eastern and southern North America, and the NWNA is the only region in the NH where is characterized by a LSAT cooling trend. Nevertheless, the simulated cooling trend of NWNA LSAT is weaker than the observations and statistically insignificant in all the six models. One possible explanation for the weak cooling trend in the models is that the model simulated spring AL weakening trend is less pronounced than that observed. The AL weakening trend ranges from 0.49 to 1.08 hPa (10 year)-1 (Figure 6), while in the observations the trend ranges are between 0.96 and 1.51 hPa (10 year)-1. In addition, there is a good linear relationship between spring AL trend and NWNA LSAT trend among different AMIP members (totally 60) with the correlation reaching −0.65 (shown in Figure S5), further indicating that the bias in AMIP simulated NWNA cooling trend can be better explained by the AL trend bias (about 42% of the bias according to the correlation). Thus, in the AMIP models, the cooling effect induced by the AL weakening is not so strong that it can significantly offset the warming effect by the radiative forcing of greenhouse gas increase. Nevertheless, we also examine the trend of spring LSAT averaged over the NWNA region (45°–65°N and 160–100°W) in each ensemble member of the AMIP models (totally 60 members as listed in Table 2) and find that a large majority of the ensemble members (46 of 60) can capture the observed surface cooling in NWNA and that more than half (33 of 60) can reproduce a statistically significant cooling trend of spring NWNA LSAT. Meanwhile, as shown in Figure S4, 55 members simulate a weakening of spring AL, and 80% of these members (44 of 55) can capture a concurrent surface cooling trend over the NWNA region.
3.3 The Forcing of Spring AL Change
In the AMIP experiment, the models are forced by observed monthly-varying SSTs and sea ice concentrations and external forcing conditions as in the CMIP5 historical experiment. The skillful simulation of the recent spring AL weakening by the AMIP models indicates that we can use the simulations to investigate the driving mechanism and which forcing might be appropriate to explain the spring AL weakening. First, we examine the effects of the external forcing on the SLP and LSAT during the recent decades, which can be estimated as the ensemble mean responses of the CMIP5 historical experiment with a large ensemble size (see section 2 for details). Figure 8 shows the simulated externally forced changes of spring North Pacific SLP and NH LSAT since 1980. The SLP trend pattern shows a rather weak response of spring AL to the external forcing, and no significant change of SLP can be found over the subarctic North Pacific. Previous studies have identified a significant effect of external forcing on the winter AL on both multidecadal and centurial time scales (Gan et al., 2017; Gillett et al., 2003), but the spring AL seems insensitive to the changes in external forcing conditions.
Another interesting feature of the externally forced change of spring North Pacific SLP is that it shows a significant increasing trend over the eastern North Pacific, where the climatological subtropical high is active. This suggests that under the external greenhouse forcing, the spring subtropical high could significantly enhance, and this is consistent with a recent finding of the seasonally dependent response of subtropical high to the anthropogenic warming effect (Song et al., 2018). The simulated response of NH LSAT to the external greenhouse forcing is easily understood, showing a uniform warming over the entire NH landmass (Figure 8b), and the observed surface cooling trend of NWNA is missing in the CMIP5 historical simulation, in association with the absence of spring AL weakening. This may also confirm that the spring AL weakening plays the key role in the NWNA surface cooling. Therefore, the weakening of spring AL and associated NWNA cooling in the observations is probably not driven by the external forcing and cannot be explained by the anthropogenic warming effect.
The role of SST boundary conditions is further investigated. Figure 9 shows the three-decade trend of spring global SST north of 20°S since 1980. Strong and significant SST warming is observed over most basins except for the tropical central-eastern Pacific. The SST cooling trend of central-eastern Pacific is evident and particularly significant for the region north of the equator. The tropical central-eastern Pacific cooling and warming in other tropical basins lead to emergence of a zonal dipole pattern of SST trend in the tropics. A similar SST trend dipole during the recent decades has been revealed by using the annual mean SST (Li et al., 2016; McGregor et al., 2014), and our results suggest the dipole of SST trend is also evident in spring. Previous studies also suggested that this SST trend dipole is largely internal climate variability and closely related to the slowdown of global warming (Kosaka & Xie, 2013; Li et al., 2016; McGregor et al., 2014) and the interbasin warming contrast (Zhang & Karnauskas, 2017). We also examine the CMIP5 ensemble mean simulation of SST trend during recent decades and find a uniform SST warming pattern as a response to the external forcing (Figure 9b). This further confirms the SST trend dipole is largely internally generated. The SST cooling trend over the tropical central-eastern Pacific largely projects on the negative IPO phase, and it is known that the IPO shifted from a positive phase to a negative phase in the late 1990s (Dong and McPhaden, 2017). In addition, due to the effects of trans-basin atmospheric teleconnections and coupled air-sea interaction, the tropical SST trend dipole can be linked to the Atlantic multidecadal variability (Li et al., 2016; McGregor et al., 2014; Sun et al., 2017; Sun, Li, Kucharski, Kang, et al., 2019), which is to a large extent a result of internal climate variability and induced by the ocean circulation variations (i.e., Atlantic meridional overturning circulation; Sun et al., 2015; Sun, Li, Kucharski, Xue, & Li, 2019; Zhang et al., 2019). Besides the Atlantic Ocean, the Indian Ocean warming in recent decades may also have an effect in shaping the tropical SST trend dipole pattern, leading to a cooling trend over the tropical eastern Pacific region (Luo et al., 2012; Lee et al., 2015; Nieves et al., 2015; Dong & McPhaden, 2017b). The simulation results from the ensemble means of CMIP5 historical experiment suggest that the recent weakening of spring AL is unlikely a result of external forcing but most likely a consequence of internal climate variability, and thus, the tropical SST trend dipole could be the key to understanding the cause of the recent changes over subarctic North Pacific and NWNA in spring.
The dynamical mechanisms that links the tropical SST trend pattern to North Pacific atmospheric circulation changes are further explored. Figure 10 shows the simulated 30-year trends of upper-level velocity potential since 1980 in the six AMIP models. In all the models, the simulated trend patterns of 200-hPa velocity potential are similar and show a zonal dipole structure most pronounced and statistically significant over the tropical region, with enhanced upper-level divergence over the tropical Atlantic and Indo-Pacific warm pool and converging trend over the tropical central-eastern Pacific. This simulated feature is consistent with the dipole trend pattern of SST forcing and can be physically explained. The SST warming over tropical Atlantic and Indo-Pacific warm pool strengthens the ascending motion and upper-level divergence, while the SST cooling over tropical central-eastern Pacific could lead to an intensification of descending motion there, in association with converging trend of upper-level atmospheric flow. These simulated tropical atmospheric circulation changes are also in accord with the observed intensification of Walker circulation in recent decades as reported in previous studies (Chung et al., 2019; Li et al., 2016).
The tropical upper-level divergence/convergence could act as an important Rossby wave source for the poleward propagation of Rossby waves into the extratropics. Figure 11 shows the simulated trends of recent three-decade upper-level zonal wind in the AMIP experiment. Under the SST forcing, the models simulate significant changes in the NH upper-level zonal wind according to the trend pattern, which is characterized by a clear wave train structure originating from the tropical central-eastern Pacific toward the subarctic North Pacific. The locations of the centers of action and the wave phases and amplitudes along the wave train are generally consistent among different AMIP models. The wave amplitudes are most pronounced in the tropics and decrease as latitude increases, corresponding to a poleward propagation path. The wave activity fluxes associated with the upper-level zonal wind trend pattern also confirm a poleward propagation of wave energy from the tropics (Figure 11). The propagation path of the simulated wave train follows the classical “Great Circle” theory (Hoskins & Karoly, 1981; Yang et al., 2015), which predicts poleward propagation of Rossby waves on a two-dimensional sphere. Numerous modeling studies have also confirmed that the tropical-SST-induced upper-level divergence anomaly can excite a Rossby wave train propagating poleward into the extratropics, causing significant remote atmospheric responses (Ding et al., 2014; Lin et al., 2007; O'Reilly, 2018). The SST forcing and associated response of upper-level atmospheric circulation changes are also consistent with the recent findings that focus on the implications of atmospheric circulation/SST trends on snowpack changes in North America (Mudryk et al., 2018; Siler et al., 2019). In the present case, the AMIP models simulate a remote atmospheric response of AL weakening to the tropical SST forcing, and the northward propagating wave train acts as the important dynamical bridge linking the extratropical AL weakening to the changes in the tropical SST forcing.
The observed time-varying arctic sea ice concentrations are prescribed as lower-boundary forcing of the AMIP models, and its role in the spring AL changes is also examined and discussed. It is known that the arctic sea ice has been melting due to the greenhouse gas warming effect, but the melting of sea ice is seasonally dependent (Stroeve & Notz, 2018). Figure 12 shows the three-decade trends of arctic sea ice concentrations since 1980 for both the spring and annual mean. The spring arctic sea ice loss is rather weak and statically insignificant over most of the arctic region, particularly for the subarctic North Pacific and adjacent regions. In contrast the annual mean arctic sea ice melting is significantly strong. Previous studies have identified that the arctic sea loss is most pronounced in summer and autumn and weakest in spring (Deser & Teng, 2013; Semenov et al., 2019; Stroeve & Notz, 2018), and thus, our results are consistent with the previous findings. It is obviously less convincing that an insignificant sea ice loss could force a significant North Pacific atmospheric circulation response.
On the other hand, the results from the AMIP_Clim_Polar experiment are also analyzed and compared with the AMIP experiment to show the effects of arctic sea ice loss (Figure 13). The simulated SLP trend during the three decades over the North Pacific is similar to that in the AMIP experiment, consistently showing a significant weakening trend of spring AL, and the difference in the SLP trend between the two experiments is insignificant. This indicates that the role of arctic sea ice loss in the recent weakening of spring AL may be rather minor, and this is also consistent with previous modeling studies that showed strong responses of atmospheric circulation to sea ice melting in summer and autumn while rather weak responses in other seasons (Semmler et al., 2016). Thus, the significant increase of SLP over the subarctic North Pacific is unlikely to be explained by the weak sea ice loss in spring, and the SST forcing could be the most likely driving mechanism responsible for the recent weakening of spring AL.
4 Conclusions and Discussion
The spring AL is ecologically important since it is a prevailing atmospheric circulation pattern in the growing season over the subarctic North Pacific and surrounding region, a biologically productive region around the globe. The multidecadal change in the spring AL is much less known than its winter counterpart. Using multisource observational and simulation data, the present study identifies a substantial multidecadal weakening of spring AL since 1980, different from the multidecadal variability of winter AL. The weakening of spring AL causes a contemporaneous cooling trend of surface air temperature over NWNA, due to reductions of warm westerly and southerly winds toward the NWNA. The NWNA is the only region in NH land characterized by a temperature cooling trend in spring. The AMIP models show considerable skill in simulating the recent spring AL weakening and NWNA cooling. Various forcing components, including both the external forcing and lower-boundary forcings of sea ice concentration and SST, that drive the AMIP models are analyzed, and the weakening of AL is most likely attributable to the change in the SST forcing condition. The SST change exhibits a zonal dipole pattern in the tropics, showing SST warming over tropical Atlantic and Indo-Pacific warm pool and cooling over the tropical central-eastern Pacific. This pattern of SST change excites a Rossby wave train propagating toward the subarctic North Pacific, which acts as a dynamical bridge linking the AL weakening to the SST forcing.
The present study reveals the multidecadal weakening of spring AL since 1980 and attributes this weakening to the changes in the tropical SST forcing, which shows a dipole pattern between central-eastern Pacific and other parts of tropical oceans. Previous studies have suggested that besides interannual variability the tropical oceans also have pronounced interdecadal/multidecadal variability in all basins (McGregor et al., 2014; Parker et al., 2007) and the multidecadal variations in different basins are tightly connected due to trans-basin atmospheric teleconnections and air-sea interactions (Li et al., 2016; Sun et al., 2017). These trans-basin multidecadal variability are present in both instrumental and proxy data (Fang et al., 2019; Sun et al., 2017). It is thus necessary to extend the present work using more data covering a longer period of time and to examine whether the multidecadal tie between the variations of spring AL and tropical SST forcing still holds. A preliminary examination of the relationship between multidecadal time series of IPO and spring AL indices since 1920 shows a modest negative correlation (r = −0.43), indicating that the phase shifts of IPO SST variability may be related to the multidecadal strengthening and weakening of AL, and this could be helpful for future prediction of AL strength. Besides the tropical SST changes, the SST trend pattern in recent decades also shows a significant warming over subtropical North Pacific (Figure 9a), and this SST warming could possibly be explained by the wind-evaporation-SST effect associated with the AL weakening (Sun et al., 2017). Nevertheless, the present analysis may not be quite sufficient to completely rule out the possible contribution from the feedback of subtropical SST warming to the AL weakening, since the AMIP experiments are forced with global SST. Thus, further modeling analysis to determine the separate effects of tropical and subtropical SST changes warrants future study. On the other hand, the present study shows that the recent multidecadal change of spring AL is different from the winter counterpart, and the possible cause of this difference remains to be explored. Previous studies have suggested that there are many other factors, such as ocean-land thermal contrast (Gan et al., 2017; He et al., 2014), planetary wave activity and internal atmospheric variability (Chen et al., 2005; Molteni et al., 2017), interactions with the North Pacific SST variability (Qiu et al., 2007; Wu et al., 2005), and solar activity (van Loon et al., 2007) influencing the decadal variations of winter AL. These influencing factors need to be considered in the comparison of the driving mechanisms of AL changes in the two seasons.
Acknowledgment
This work was jointly supported by the National Natural Science Foundation of China (41775038, 41975082, 41975047 and 41790474 ) and the National Program on Global Change and Air–Sea Interaction (GASI-IPOVAI-03 and GASI-IPOVAI-06). All observation/reanalysis and AMIP simulation data sets publicly available and can be downloaded from the corresponding websites. The sources of the data sets are introduced in detail in section 2.
