Drivers of Recent North Pacific Decadal Variability: The Role of Aerosol Forcing

Climate variability in the Pacific has an important influence on climate around the globe. In the period from 1981 to 2012, there was an observed large‐scale cooling in the Pacific. This cooling projected onto the negative phase of the Pacific Decadal Oscillation (PDO) and contributed to a slowdown in the rate of near‐surface temperature warming. However, this cooling pattern is not simulated well by the majority of coupled climate models and its cause is uncertain. We use large multi‐model ensembles from the sixth Climate Model Intercomparison Project, and an ensemble of simulations with HadGEM3‐GC3.1‐LL that is specifically designed to sample the range of uncertainty in historical anthropogenic aerosol forcing, to revisit the role of external forcings. We show that anthropogenic aerosols can drive an atmospheric circulation response via an increase in North Pacific sea level pressure and contribute to a negative PDO during this period in several global climate models. In HadGEM3, this increase in North Pacific sea‐level pressure is associated with an anomalous Rossby Wave train across the North Pacific, which is also seen in observations. Our results provide further evidence that anthropogenic aerosols may have contributed to observed cooling in the Pacific in this period. However, the simulated cooling in response to aerosol forcing is substantially weaker than the warming induced by greenhouse gases, resulting in simulations that are warming faster than observations, and further highlighting the need to understand whether models correctly simulate atmospheric circulation responses.


Introduction
Understanding what drives multi-decadal variations in the Pacific has been a topic of research for many decades. Pacific sea surface temperatures (SSTs) are an important contributor to setting the rate of warming in global mean surface temperatures (e.g., Kajtar et al., 2019;Meehl et al., 2016). Teleconnections from the Pacific make Pacific climate variability important for many regions of the globe. The most prominent mode of interannual climate variability, the El Niño Southern Oscillation, is hugely important for climate across the world (e.g., McPhaden et al., 2006). On longer timescales, the Pacific Decadal Oscillation (PDO) or Interdecadal PDO and their drivers have been widely studied (e.g., Mantua and Hare, 2002;Schneider & Cornuelle, 2005;Newman et al., 2016;Power et al., 1999) and shown to modulate climate variability in many regions of the world (e.g., Barnett et al., 1999;Dong and Dai, 2015;Salinger et al., 2001;Wang et al., 2008;Zanchettin et al., 2008).
In recent years, significant attention has been devoted to understanding the so-called "global warming hiatus" over the period from 1998 to 2012 and the role of Pacific SSTs in explaining slower warming during that period Kosaka & Xie, 2013;Maher et al., 2014;Medhaug et al., 2017;Meehl et al., 2013;Xie & Kosaka, 2017), though estimates of the importance of the Pacific for global temperatures vary across studies . Nevertheless, it seems clear that the Pacific has played an important role in setting the rate Abstract Climate variability in the Pacific has an important influence on climate around the globe. In the period from 1981 to 2012, there was an observed large-scale cooling in the Pacific. This cooling projected onto the negative phase of the Pacific Decadal Oscillation (PDO) and contributed to a slowdown in the rate of near-surface temperature warming. However, this cooling pattern is not simulated well by the majority of coupled climate models and its cause is uncertain. We use large multi-model ensembles from the sixth Climate Model Intercomparison Project, and an ensemble of simulations with HadGEM3-GC3.1-LL that is specifically designed to sample the range of uncertainty in historical anthropogenic aerosol forcing, to revisit the role of external forcings. We show that anthropogenic aerosols can drive an atmospheric circulation response via an increase in North Pacific sea level pressure and contribute to a negative PDO during this period in several global climate models. In HadGEM3, this increase in North Pacific sea-level pressure is associated with an anomalous Rossby Wave train across the North Pacific, which is also seen in observations. Our results provide further evidence that anthropogenic aerosols may have contributed to observed cooling in the Pacific in this period. However, the simulated cooling in response to aerosol forcing is substantially weaker than the warming induced by greenhouse gases, resulting in simulations that are warming faster than observations, and further highlighting the need to understand whether models correctly simulate atmospheric circulation responses.
Plain Language Summary Climate variability in the Pacific has an important influence on climate around the globe. In the period from 1981 to 2012, there was an observed large-scale cooling in the Pacific. The causes of this observed cooling are still uncertain. Climate models are able to simulate similar cooling patterns, but these are often weaker than observed. We show that anthropogenic aerosols, tiny particles emitted by human activity, have likely contributed to observed cooling in the Pacific since the 1980s via changes in atmospheric circulation. The simulated response to anthropogenic aerosols is weaker than the warming response to greenhouse gases, resulting in a majority of climate model simulations that are warming faster than observations over this period. We speculate that Pacific cooling induced by atmospheric circulation changes in response to aerosol may be underestimated in current global climate models, but other hypotheses exist, highlighting the need for further research in this area.  (Deser et al., 2020;Liguori et al., 2020;Qin et al., 2020). Anthropogenic aerosols (among other forcing agents such as volcanic eruptions) have previously been proposed as a driver of North Pacific SSTs (Boo et al., 2015;Smith et al., 2016). They have also been implicated for contributing to SST variability in the Pacific over the recent hiatus period (Smith et al., 2016), but other studies have found no role for anthropogenic aerosols on early 21st century temperature trends (Oudar et al., 2018).
Robustly quantifying the role of aerosols for recent Pacific Decadal Variability is difficult for several reasons. First, the magnitude of historical aerosol forcing itself is highly uncertain (Bellouin et al., 2020), and given the same emissions, different climate models will generate a wide range of historical aerosol forcing due to large diversity in aerosol schemes and their representation of aerosol processes (e.g., Wilcox et al., 2015). Beyond uncertainty in aerosol forcing and model diversity, small ensemble sizes in the presence of large internal variability and biases in the representation of processes important for Pacific climate variability (e.g., McGregor et al., 2018) render a robust quantification of the role of aerosol forcing for recent trends in Pacific temperatures difficult.
Nevertheless, understanding what drives observed variability in the Pacific is crucially important, especially in light of new challenges in the interpretation of climate projections from several climate models with high climate sensitivities in the most recent iteration of the Coupled Model Intercomparison Project (CMIP6, Eyring et al., 2016). Many of these models have warming trends inconsistent with observations since the 1980s, and it has been argued that this recent period can be used to constrain future climate projections (e.g., Brunner et al., 2020), but this approach relies on the assumption that models can simulate the effects of anthropogenic aerosols and volcanic eruptions accurately. Some of this rapid warming in climate models stems from the fact that Pacific SSTs are warming substantially faster than observations during this period (e.g., Olonscheck et al., 2020). Therefore, understanding the drivers of real-world Pacific SST variability is necessary in the context of interpreting this period of rapid warming in global climate models. It is therefore timely to reexamine the evidence for aerosol forcing contributing to historical Pacific SST variability in the most recent generation of climate models and in observations. Additionally, reductions in aerosol emissions across the globe in coming decades are likely to be important for regional climate change. It is therefore important to understand the implications of reducing aerosol burden for climate variability in the Pacific.
Here, we use CMIP6 data and a large ensemble of historical simulations with the HadGEM3-GC3.1-LL model to investigate the role of aerosol forcing for North Pacific climate variability since the 1980s. The HadGEM3 ensemble, known as the SMURPHS ensemble (Dittus et al., 2020), was run with varying levels of anthropogenic aerosol and precursor emissions across the ensemble, thus allowing the identification of responses to changes in anthropogenic aerosols. We assess the evidence for a role of anthropogenic aerosol forcing in explaining historical Pacific SSTs in CMIP6 and investigate the mechanisms responsible in the HadGEM3 global climate model. Existing hypotheses and remaining questions in reconciling observed and simulated Pacific temperature trends since the 1980s are discussed.

Data and Methods
Several types of simulations are used in this paper. First, simulations from the CMIP6 historical experiments are used to compare with observations, while simulations from the Detection and Attribution experiments (DAMIP, Gillett et al. (2016)) are used to identify which forcing agents are responsible for the forced response in the historical experiments. The DAMIP simulations are single-forcing experiments, where one forcing of interest is changing through time, and all other forcings are kept at constant pre-industrial levels. The experiments used in this paper are historical (all forcings), hist-aer (aerosol-only), hist-GHG (greenhouse gases only), and histnat (natural forcings only, i.e., solar and volcanic). We use all simulations available at the time of writing. The number of ensemble members available for each model varies with the experiment and variable considered, so we have chosen to use all ensemble members available for a given experiment, regardless of whether the same ensemble members are also available for other variables and experiments. As such, the number of members can vary from one experiment to the next. Note that multi-model averages are constructed from each model's ensemble mean, where all models with >= 3 ensemble members are considered. The CanESM5 contribution is 3 of 14 also shown separately to the CMIP6 multi-model ensemble, as this modeling center provides a large ensemble size for all experiments, including the single-forcing experiments. There are 50 ensemble members available for this model for the historical, hist-GHG, and hist-nat experiments, and 30 for the hist-aer experiments (Swart et al., 2019). All the CMIP6 models used and the number of ensemble members per experiment are listed in Table  S1 in Supporting Information S1.
Second, the SMURPHS ensemble is used to investigate the mechanisms associated with aerosol forcing in more detail. This ensemble consists of historical simulations with the HadGEM3-GC3.1-LL climate model, where anthropogenic aerosol emissions have been scaled by five factors (0.2, 0.4, 0.7, 1.0, and 1.5) to sample the wide uncertainty range in historical aerosol forcing (Dittus et al., 2020). These scaling factors have been applied to anthropogenic aerosol and precursor emissions (black and organic carbon and sulfur dioxide) from 1850, as illustrated for SO 2 in Figure 1a. Biomass burning emissions were not scaled in this experiment design. The resulting aerosol effective radiative forcings (ERF) for the year 2014 are −0.38, −0.60, −0.93, −1.17, and −1.50 W/m 2 , respectively. Five initial condition members have been run for each of these aerosol scaling experiments, resulting in an ensemble of 25 historical simulations in total (5 × 5). The version of HadGEM3 used in this ensemble corresponds to a pre-release of the HadGEM3 model used for the CMIP6 simulations and differs from the CMIP6 version only in its treatment of ozone. More detail can be found in the Supplementary  The observational data sets used in this study are the ensemble mean of the HadCRUT5 analysis for near-surface temperatures (Morice et al., 2021) and HadSLP2 for sea level pressure data (Allan & Ansell, 2006). The variance-adjusted data set was used for the HadSLP2 data set from 2005 to 2012. For meridional winds at 250 hPa, we compare the SMURPHS simulations to data from the ERA5 reanalysis (Hersbach et al., 2019).
To calculate multi-model quantities, the CMIP6 simulations were regridded to a common resolution of 2.5° × 2.5° using bilinear interpolation. Where pattern correlations between observations and model simulations were calculated, the simulations were regridded to the resolution of the observational data set for HadCRUT5 and HadSLP2, while for Figure 7 ERA5 was regridded to the resolution of the model simulations (regridding to the lowest resolution in all cases). To calculate SLP trends over the North Pacific, area averages were computed over a box from 30° to 65° N and 160°-220° E, using the definition of the NP pressure index used in Trenberth and Hurrell (1994). Stippling indicates areas where the observed trends are within the simulated range of trends in the SMURPHS experiments, with the exception of Figures 3 and 8. In these latter figures, stippling is used to indicate areas where trends are statistically significant according to a two-sided Student's t-test at the 5% level. Annual means are shown in all figures except Figure 7, where trends over the October to March extended season are shown instead.

Pacific Decadal Variability in CMIP6
In the period from 1981 until 2012, observed trends in surface temperatures show regions of warming and cooling in the Pacific and the Southern Ocean (Figure 2b), in a pattern that projects onto the negative phase of the PDO. During this period, most CMIP6 climate models simulate trends exceeding the observed rate of warming (Figure 2a), though there are realizations that simulate similar or lower rates of warming to those observed. The CMIP6 ensemble mean trend pattern shows spatially varying rates of warming, but importantly there are no areas of cooling. This lack of cooling suggests that any forced cooling (e.g., in response to anthropogenic aerosols) is outweighed by warming from greenhouse gases and natural forcings over this period in the multi-model average (Figures 2a and 2b). Indeed, the forced response to anthropogenic aerosols is relatively small over this time period and diverse across climate models (Figure 2a). While aerosol forcing has been shown to be important in driving variations in surface temperatures in numerous studies, including in the SMURPHS ensemble (Dittus et al., 2020), the largest variations in GMST due to aerosol forcing are found in the period between around 1945 and 1975, when emissions increased rapidly across the globe. In the 1981-2012 period, global emissions of SO 2 are relatively constant, due to compensating trends between North America and Europe combined, where emissions have been decreasing since the late 1970s, and trends over Asia where emissions have continued to increase over this period (Figures 1a and 1b). Globally averaged, black and organic carbon emissions on the other hand have continued to increase during this time period. Consequently, model responses to aerosol over this period can be very different (Figure 2a), as different aerosol schemes are more or less sensitive to regional aerosol emissions. Additionally, internal variability may also explain some of the differences between models, as the residual internal variability in three-member ensemble means may still be quite large, especially in this period of relatively weak global aerosol forcing.
There are individual realizations in the CMIP6 ensemble that have a cooling pattern of Pacific SATs similar to those observed. Two individual ensemble members are shown in Figure 2b that look the most and the least like observations, based on pattern correlations over the Pacific, respectively. This is done to illustrate the range of patterns seen across CMIP6 only and is not indicative of model performance. However, members that have similar SAT patterns to observations are infrequent and, in most cases, the cooling is weaker than observed (not shown). This discrepancy could suggest a bias in the forced response or errors in the representation of internal variability in many CMIP6 models. However, it is also possible that the real world was in an unusual phase of internal variability during this time, and the models are consistent with observations when internal variability is considered (Olonscheck et al., 2020). In the following sections, the role of anthropogenic aerosols in driving this The PDO has been widely studied and has previously been described as a superposition of various processes (Newman et al., 2016). One of the many drivers of North Pacific SSTs is sea level pressure in the North Pacific, in the region of the Aleutian Low. While other mechanisms may also be important and respond to aerosol forcing, the primary pathway that has been proposed for aerosols to affect the PDO is via changes in the Aleutian Low, so the remainder of the paper will focus on this mechanism.
In CMIP6, most models (14 out of 15 with >= 3 ensemble members) simulate increases in annual mean North Pacific sea level pressure in their historical simulations over the period from 1981 to 2012 (Figure 3a). Results from the DAMIP simulations suggest that this increase is primarily driven by increases in anthropogenic aerosols, as all but two models with more than three ensemble members simulate an increase in SLP during that period. In contrast, neither GHGs nor natural forcings can explain these trends in historical SLP (Figure 3a). It is unclear whether the two models showing a decrease in SLP in response to anthropogenic aerosols have a different forced response to anthropogenic aerosols to the other models or whether this is due to an insufficient number of ensemble members to isolate the forced response in NP SLP. Indeed, results from the CanESM5 large ensemble show that internal variability in NP SLP could be large and that more than three ensemble members may be needed to reliably determine the forced response. However, at the time of writing, only 5 models had more than three ensemble members available for the DAMIP simulations, so the 3-ensemble member threshold was chosen as a compromise between sampling internal variability and model diversity. Nevertheless, the majority of models do simulate an increase in NP SLP in response to anthropogenic aerosols. Figures 3b and 3c show the pattern of surface air temperatures (SATs) associated with variations in the magnitude of trends in NP SLP across CMIP6 models and within the CanESM5 historical ensemble. Both CMIP6 and CanESM5 suggest that larger increases in NP SLP are associated with a negative PDO. For CMIP6, the temperature trend patterns from each of 14 CMIP6 models with at least three ensemble members are regressed onto the NP SLP trends from the same 14 models. Hence, the CMIP6 regression is also sensitive to differences in factors such as climate sensitivity in addition to internal variability assessed in CanESM5, which may account for its lower significance. Nevertheless, both patterns provide further evidence that NP SLP trends affect Pacific SAT trends.
In summary, Figure 3b shows that trends in NP SLP affect SAT trends in the Pacific across CMIP6 models, while Figure 3a shows that anthropogenic aerosol forcing can drive an increase in NP SLP. Together, these results suggest that changes in anthropogenic aerosol emissions in the period from 1981 to 2012 have contributed to Pacific variability in at least some CMIP6 models. However, it is not simply the case that models with the strongest aerosol forcing have the strongest response in SLP and Pacific SATs and vice-versa (not shown). Differences in model representation of aerosol and their processes combined with the limited number of ensemble members in the presence of large internal variability make it difficult to identify the reasons between the different responses across CMIP6 models and to provide a robust quantification of the fraction of Pacific variability driven by anthropogenic aerosols. Therefore, dedicated experiments to investigate the response to aerosol forcing in the presence of large internal variability are required to better understand the processes involved.

Mechanisms for a Response of North Pacific SATs to Anthropogenic Aerosols in the SMURPHS Ensemble
To investigate the mechanisms of a Pacific SAT response to anthropogenic aerosols further, we now focus on the SMURPHS ensemble of climate model simulations with HadGEM3-GC3.1, where anthropogenic aerosols have been scaled to sample a wide range of uncertainty in historical aerosol forcing (Dittus et al., 2020). Since the only difference between the experiments is the level of anthropogenic aerosol emissions, significant differences between the experiments can be directly attributed to differences in anthropogenic aerosol forcing.
Much like the CMIP6 multi-model mean, the five-member ensemble means for each aerosol-scaling experiment in the SMURPHS simulations show large-scale warming regardless of aerosol forcing, although there are also differences between the experiments (Figure 4). For example, the rate of warming is reduced in the high aerosol simulations, particularly in the polar regions and the eastern North Pacific. The 0.4, 0.7, 1.0, and 1.5 ensemble means show increasing areas of cooling in the South Pacific and the Southern Ocean, indicating that part of this DITTUS ET AL.

10.1029/2021EF002249
7 of 14 SAT pattern may be externally forced, with differences between the experiments attributable to anthropogenic aerosols. The effect of increasing aerosol forcing across scaling experiments is clearer when the greenhouse gas-induced warming trend has been removed from all experiments in the SMURPHS ensemble ( Figure 5). Overall, there are individual ensemble members in the SMURPHS ensemble that resemble the observed negative PDO pattern, but the cooling in those members is much weaker than observed in all cases. This is illustrated in Figure 4 by showing the three members with the highest pattern correlation in the Pacific. This is likely related to the finding that all ensemble members warm more rapidly than observations over this period (Dittus et al. (2020), also shown in Figure 2) and may be a feature common to other CMIP6 generation models (see Discussion). Nevertheless, the differences between the experiments suggest a role for anthropogenic aerosols in driving these SAT patterns, which we now examine further.
NP SLP trends are broadly larger with each higher aerosol scaling experiment, thus clearly demonstrating the sensitivity of North Pacific SLP to changes in aerosol over this period (Figures 3 and 6). Note that the trends in the 1.0 and 1.5 experiments are statistically significant at the 5% level and that the zonal winds at 250 hPa show changes consistent with these NP SLP trends (not shown). Observations from HadSLP2 over the same time period also show an increase in NP SLP, indicating that simulated aerosol-induced trends in SLP are consistent with observations over this time period. However, it is also important to note that Figure 6 shows 5-member ensemble means for the modeled trends, while the observations are by definition a single realization. While the higher aerosol simulated patterns appear more similar to observations than those from the low aerosol experiments, due to the large contribution from internal variability trends such as those seen in observations could also arise from internal variability and occur in individual realizations in some of the lower aerosol experiments (Figure 3a). Substantial internal variability across ensemble members can also be seen in the time series of 20-year running means in NP SLP (Figure 6b). Additionally, these time series also show that there is substantial low-frequency variability in these time series, likely due to different forcings in different time periods, including but not limited to aerosols and greenhouse gases. The end of these time series also shows the sensitivity of NP SLP to anthropogenic aerosol forcing since the 1980s shown in Figures 3 and 6a, with NP SLP decreasing in the low aerosol simulations and increasing in the higher aerosol simulations.
Meridional wind trends at 250 hPa show a pattern of alternating positive and negative wind anomalies emerging in the higher aerosol scaling experiments (Figure 7). These patterns suggest that a Rossby Wave train across the Pacific may be being triggered in the higher aerosol simulations, consistent with previous studies that showed a Rossby Wave response to changes in Asian aerosol emissions (Smith et al., 2016;Wilcox et al., 2019). A similar pattern in meridional wind trends is seen in the ERA5 reanalysis over the Pacific in the same time period (Figure 7), but the trends in ERA5 are much stronger. Since this is predominately a winter signal, the extended winter season is shown in Figure 7. With only one realization of the real historical climate, it is not possible to say whether the stronger trends in ERA5 are due to an underestimation of the forced response in climate models or simply due to a particular realization of internal variability in the real world. Figures 5-7 is summarized in Figure 8 and compared to earlier time periods. Figure 8 shows the regression of temperature (left), SLP (middle) and meridional winds (right) trend patterns from each realization onto the absolute value of the 2014 aerosol effective radiative forcing for each experiment (−0.38, −0.60, −0.93, −1.17, and −1.50 W/m 2 ). In other words, it shows how the trend patterns change as a function of strengthening aerosol forcing across all 25 ensemble members of the SMURPHS ensemble. Figure 8 shows the patterns associated with an increase in anthropogenic aerosol forcing for each variable in each time period, in the absence of responses to other forcings (since these are identical across all experiments, they do not influence the regression).

The information shown in
As expected, a stronger aerosol forcing is associated with a larger relative decrease in global temperatures in all time periods. The strongest cooling occurs in the period from 1951 to 1980, corresponding to the period where changes in global aerosol burden are largest across the ensemble, while the changes are weakest in the period from 1921 to 1950. Importantly, however, this figure highlights that the responses to anthropogenic aerosol forcing during the 1981-2012 period differ from earlier periods. In particular, it is the period with the strongest (and most significant) atmospheric circulation responses to anthropogenic aerosol over the Pacific, while the 10.1029/2021EF002249 9 of 14 preceding periods are dominated by global cooling but seemingly weaker circulation responses. As the globally averaged aerosol forcing over the most recent time period is approximately constant, it is likely the regional changes in aerosol emissions that are important in explaining the changes over 1981-2012. Indeed, while European and North American aerosol emissions have been declining over this period, Asian emissions have been at their highest levels and with the most rapid growth since 1850 (Figure 1), suggesting that Asian aerosol emissions are a plausible driver of the changes seen in this period. This hypothesis is consistent with results from a previous study that has shown a Rossby-Wave response with a similar structure to increases in Asian aerosol emissions over the same time period in a predecessor of this model, HadGEM3-GC2 (Wilcox et al., 2019). Furthermore, the 1981-2012 period is also the period with the strongest black carbon emission increases (Figure 1). A recent study has suggested that the increase in NP SLP is primarily a response to black carbon aerosols rather than sulfate aerosols (Dow et al., 2021), but we do not differentiate between different species of anthropogenic aerosols here.

Discussion and Broader Implications
The results presented in this paper provide further evidence that anthropogenic aerosols likely contributed to driving variations in North Pacific SATs since the 1980s in at least some CMIP6 models. This is associated with an increase in NP SLP and, in HadGEM3, a Rossby Wave train across the Pacific. Consistency of the response pattern shown here with previous studies suggests that the response is likely to be driven by changes in Asian aerosol emissions, which induce local circulation changes and initiate a Rossby wave train from the midlatitudes as shown in Smith et al. (2016) and Wilcox et al. (2019). Assessment of the role of different aerosol species by Dow et al. (2021) suggests an important role for black carbon emissions over Asia in particular, with emissions from  et al., 2014). It has separately been suggested that Atlantic Multi-decadal Variability could be influenced by anthropogenic aerosols (Booth et al., 2012), though research into that topic is ongoing (e.g., Andrews et al., 2020;Kim et al., 2018;Zhang et al., 2013). It is therefore plausible that other  pathways exist in which anthropogenic aerosols could influence Pacific Decadal Variability, but further analysis of these is left for future studies.
A key uncertainty in interpreting these results is how simulated Pacific Decadal Variability relates to real-world Pacific Decadal Variability. Indeed, several previous studies have identified issues in how the previous generation of climate models represent tropical Pacific variability, with suggestions that models underestimate variability at multi-decadal scales and longer (e.g., Ault et al., 2013;Capotondi et al., 2015;Kajtar et al., 2019). In addition, difficulties in capturing ENSO characteristics accurately and biases in the connection between the Atlantic and Pacific basins mean that it is difficult to quantify the fraction of the observed trends that is attributable to anthropogenic aerosols (e.g., Capotondi et al., 2015;Deser et al., 2018;McGregor et al., 2018;Zhao et al., 2021;Power et al., 2021 and references therein). These issues, among others, likely contribute to the apparent discrepancy between observed and simulated Pacific SAT trend patterns over this time period. Although there are some ensemble members across CMIP6 that appear to be able to simulate patterns that resemble the SAT pattern seen in observations, these patterns are, on average, much weaker than observed. Only very few ensemble members in all of CMIP6 show a basin-wide cooling as strong as the one seen in observations.
One possible contributing factor to explain this difference is that climate models may underestimate the atmospheric circulation responses to external forcing (Scaife & Smith, 2018), potentially related to an underestimation of the eddy feedback (Scaife et al., 2019). A recent study has shown that the predictable signal for the North Atlantic Oscillation is underestimated in climate models by an order of magnitude . It is plausible that this may also be the case for the response of other aspects of the atmospheric circulation to external forcing, and by extension, affect the response of the North Pacific climate system to aerosol forcing. If this were the case, it is possible the circulation-mediated temperature response to aerosol forcing seen in the models may be too weak.
However, this is only one of several hypotheses that have been proposed in the literature to explain the difference between observed and simulated temperature trends during this period, both in the Pacific and globally. It is well documented that the Pacific plays an important role in setting the rate of GMST warming. The remaining hypotheses to reconcile observed and simulated temperature trends broadly fall into the following categories: current GCMs are too sensitive to greenhouse-gas forcing (Dittus et al., 2020;Nijsse et al., 2020;Tokarska et al., 2020) and/or have errors in the forced response (Seager et al., 2019); Pacific cooling is underestimated due to an incorrect response to volcanic eruptions (Gregory et al., 2019); the simulated internal variability is too weak (Capotondi et al., 2015;Gregory et al., 2019), or the real-world was experiencing a really unusual phase of internal variability during this time and the models are in fact consistent with observations (Olonscheck et al., 2020). We refer the reader to the papers cited above for more discussion of these possible causes and suggest that further research is needed to better understand this issue.
Recent papers have argued that climate projections from high climate sensitivity models are unrealistic, as the rapid warming seen in those models is inconsistent with recent historical trends (Brunner et al., 2020). Studies have also suggested that the transient climate response (TCR) to greenhouse gases is unrealistically high by using the period since the 1980s to constrain the transient climate response (Dittus et al., 2020;Nijsse et al., 2020;Tokarska et al., 2020). These conclusions rely on the assumption that internal variability and the responses to anthropogenic aerosols and volcanoes are all simulated realistically over this period. Therefore, it is imperative that future work address this question, as the interpretation of projections from high sensitivity models relies upon knowing which of the above hypotheses is correct. Regardless of the cause of the observed negative PDO in the real world, a reversal of this trend would be expected to lead to enhanced near-term warming, potentially amplified by cloud and lapse-rate feedbacks becoming more positive (Andrews & Webb, 2018).

Conclusions
We analyze large multi-model ensembles of CMIP6 historical and DAMIP simulations, together with the SMURPHS ensemble of scaled aerosol-forcing experiments (Dittus et al., 2020). We provide robust evidence that anthropogenic aerosols likely contributed to an increase in North Pacific (NP) SLP over the period from 1981 to 2012 in multiple climate models, in agreement with previous analysis of CMIP5 simulations (Oudar et al., 2018;Smith et al., 2016). The SMURPHS ensemble provides an opportunity to assess the physical processes in more detail by correlating the forced response with the strength of the aerosol forcing. This analysis 12 of 14 reveals a meridional wind anomaly pattern in response to stronger aerosol emissions that is consistent with a Rossby wave train over the North Pacific. These results are consistent with previous studies that showed a Rossby wave response to Asian aerosol emissions since the 1980s (Wilcox et al., 2019) and aerosol-forced changes in Rossby wave source in the north-west Pacific (Smith et al., 2016). A similar pattern of meridional wind anomalies resembling a Rossby wave train, along with an increase in NP SLP, is also seen in reanalysis and observations, respectively, suggesting that anthropogenic aerosols may have plausibly played a role in driving real-world changes over this period.
NP SLP can influence the PDO (Newman et al., 2016) and has previously been proposed as a possible mechanism through which anthropogenic aerosol can affect its phase (Smith et al., 2016). We provide two new lines of evidence that support this connection over the period 1981 to 2012. First, in the CMIP6 simulations stronger aerosol-forced trends in NP SLP are associated with a more negative PDO. Second, regressing across the SMURPHS ensemble shows that increased aerosol forcing projects onto the negative phase of the PDO. However, this aerosol-forced negative PDO pattern is overwhelmed by warming from greenhouse gases in the CMIP6 and SMURPHS simulations such that their ensemble means show little resemblance to the observations, and the majority of model simulations overestimate the observed GMST trends.
The reason for differences between model simulations and observations over the 1981-2012 period remains open. It is possible that unusually strong internal variability coincidentally aligned with the aerosol-forced response in the real world and that models are consistent with observations, even if the observations lie toward one end of the distribution. However, errors in the simulated characteristics of internal variability in the Pacific and/or biases in either the response to greenhouse gases, the response to volcanic eruptions or the circulation-mediated response to anthropogenic aerosols in current climate models are also plausible sources of differences between models and observations. Given the role of atmospheric circulation in the aerosol-forced negative PDO, the latter would be consistent with evidence from seasonal and decadal predictions that models underestimate atmospheric circulation changes (Scaife & Smith, 2018;Smith et al., 2020). Of course, it is possible (and perhaps even likely) that a combination of these may be applicable. Which of these hypotheses are correct has important implications for whether climate projections from high climate sensitivity models are considered plausible or not, and it is, therefore, crucial that future research addresses these questions.

Data Availability Statement
Output from the SMURPHS climate model ensemble is archived at the Centre for Environmental Data Analysis https://catalogue.ceda.ac.uk/uuid/5808b237bdb5485d9bc3595f39ce85e3. CMIP6 data can be accessed via the Earth System Grid Federation.