The interaction of aerosol particles with clouds and radiation is the largest source of uncertainty for estimating the total anthropogenic forcing since pre-industrial times . Anthropogenic aerosol particles such as sulfate and carbonaceous aerosols have substantially increased the global mean burden of aerosol particles from pre-industrial times to the present day. While the largest increases in aerosol emissions in the 20th century were located in Europe and North America, nowadays anthropogenic aerosol emissions are highest in south-eastern Asia .

Aerosol particles affect the vertical extent, lifetime, phase and optical properties of clouds by acting as cloud condensation or ice nuclei. Whether aerosol particles also impact precipitation and, if so, how, is controversial. The scientific review of aerosol pollution impact on precipitation by concluded that clear causal relationships between an increase in aerosol particles and changes in precipitation are difficult to identify and even the sign of the change in precipitation is uncertain. Based on Monte-Carlo simulations showed that, due to the uncertainties in representing cloud microphysics in models, it is a challenge to quantify the impact of aerosols on clouds. Most studies (both observational and numerical) have analysed the possible impact of aerosol particles on low-level non-precipitating or slightly precipitating liquid water clouds. Some studies suggest that air pollution delays the onset of orographic precipitation whereas the effect of cities on precipitation cannot be unambiguously related to air pollution .

Studies evaluating a possible aerosol effect on precipitation can be categorized among others into those that examine the weekly cycle of aerosols and precipitation, those analysing aerosol effects on precipitation from convective clouds and those from cyclones (extratropical cyclones and hurricanes). They can be summarized as follows.

Weekly cycles in cloud properties and precipitation have been reported as evidence for an aerosol effect on precipitation because of the weekly cycle in aerosols and their precursor emissions e.g.. However, many of these studies were disputed because of weaknesses in their applied statistical methods, their methodology or because the results could have been obtained by a simultaneous response of aerosol particles and precipitation to meteorological conditions e.g..

Aerosol effects on convective clouds also found contradictory results. In the case of pyrocumulus, which are characterized by extreme air pollution and extremely high aerosol concentrations, found that the onset of precipitation is delayed and its intensity is reduced with increasing aerosol concentration. On the contrary, proposed an invigoration of convective clouds due to pollution. The growth of cloud droplets into raindrops is slower in polluted clouds, which consist of more but smaller cloud droplets, than in clean clouds. This delays the formation of warm rain and more cloud water is available for freezing in polluted deep convective clouds. The increased latent heat release may invigorate polluted convective clouds and lead to overall more precipitation from them.

analysed the impact of aerosol particles on precipitation in the vicinity of a warm front. They found a shift in location caused by a delay of the onset of precipitation with increasing aerosol concentration, but the total precipitation from the warm front remained relatively constant. They attributed the rather constant total precipitation to a buffering effect. Smaller cloud droplets due to increased aerosol concentrations in the mixed-phase region of the cloud, where cloud droplets and ice crystals coexist, caused a decreased riming efficiency but led to enhanced growth by diffusion. studied the impact of aerosol particles on precipitation from a large winter cyclone. They also observed a delay in warm-phase precipitation formation, but an increase in snow to the north of the warm front. They concluded that aerosol impacts were much stronger in areas with light precipitation rates than in those with higher precipitation rates.

The present study extends the above studies on the aerosol impact on one extratropical cyclone to a climatology of precipitation in extratropical cyclones in the North Pacific using the ECHAM6-HAM global climate model. We chose the North Pacific because of the increase in anthropogenic emissions over south-eastern Asia and because the prevailing westerly winds carry these anthropogenic aerosols and their precursors over the North Pacific. Therefore, we expect to see large differences in aerosol burden and aerosol optical depth between present-day and pre-industrial conditions in this region. In addition, low-pressure systems frequently form here e.g.. Thus if an impact of anthropogenic aerosol particles on low-pressure systems can be identified somewhere, then the North Pacific is the region to study this.

The effect of Asian pollution on clouds in the North Pacific region has been investigated in different studies. showed that anthropogenic aerosols lead to changes in the cloud microphysical properties and the radiative forcing at the top of atmosphere. Furthermore, they stated that an increase in cloud top height indicated invigorated midlatitude cyclones connected to an overall increase in precipitation. Also, found that higher anthropogenic aerosol emissions lead to an increase in the amplitude of synoptic eddies and subsequently to an increase in surface precipitation. An intensification of the Pacific storm track is also found in . They argued that the wintertime Pacific is highly vulnerable to cloud–aerosol interactions because of the coupling between the Pacific storm track and Asian pollution outflow.

In our study, we specifically focus on the so-called warm conveyor belt (WCB) airstream, which is a typical feature of (intense) extratropical cyclones. By focusing on the WCB we investigate aerosol effects in a relatively well-defined flow setting, in contrast to other climate model studies that consider the aerosol effect on total precipitation without distinguishing different categories of weather systems that produce precipitation .

WCBs are coherent moist ascending airstreams in extratropical cyclones associated with the formation of elongated frontal cloud bands and intense precipitation e.g.. They can be objectively identified with the aid of trajectory calculations. showed that WCBs are formed by moist boundary layer air parcels that ascend by about 600 hPa or more within a time period of 2 days. Thereby intense cloud formation and latent heating occurs in WCB trajectories, leading to a typical loss of specific humidity of more than 10 g kg-1 and an increase of potential temperature of about 20 K . Initially, WCB air parcels are cloud free, then liquid water clouds form in the early part of the ascent leading to mixed-phase clouds in the mid-troposphere and pure ice clouds in the WCB outflow at upper-tropospheric levels (about 350 hPa with temperatures below -30 ∘C) . WCBs are intrinsic subsynoptic scale features of extratropical cyclones and therefore climatological frequency maxima of WCBs occur in the extratropical storm track regions . In these regions more than half of total precipitation and up to 90 % of extreme precipitation events are associated with WCBs . In particular, in the western North Pacific just to the east of Japan, more than 60 % of the climatological precipitation and more than 90 % of the precipitation extremes (defined as events above the 99th percentile) are collocated with WCBs . Furthermore, due to the strong ascent, WCBs connect the different tropospheric layers and are therefore important for the transport of pollution from the boundary layer to the middle/upper troposphere . By investigating precipitation formation in WCBs in the western North Pacific, this study examines potential effects of anthropogenic aerosol emissions within the highly relevant category of extratropical weather systems.

The paper is organized as follows. In Sect. , the data and methods are introduced. Section  gives an overview of the simulations and in Sect. , a case study of a WCB is briefly discussed. In Sect. , the influence of aerosol particles on precipitation in WCBs is examined statistically for the entire 10-year climatology. The conclusions follow in Sect. .

Before evaluating the WCBs, some general characteristics of the ECHAM6-HAM simulations are presented and compared with ERA-Interim reanalyses . Figure a–c shows the winter mean field of SLP and potential temperature at 850 hPa for the PI and PD simulations and ERA-Interim. The main features in the time mean are the general southward-oriented gradient in SLP and potential temperature, the Aleutian low with a central pressure of about 1000 hPa in the simulations and 995 hPa in ERA-Interim, and the strong horizontal temperature gradient in the region of Japan. The fields for the two simulations are almost identical, which reveals that the different aerosol emissions in the two simulations have no effect on the time mean SLP distribution and baroclinicity. Compared to ERA-Interim the ECHAM simulations underestimate the intensity of the Aleutian Low, which is most likely due to the well-known problem of fairly coarse global climate models to not resolve the peak intensity of extratropical cyclones . Another difference between the simulations and ERA-Interim appears for the tropical temperatures, which are about 5 K too low in the simulations, consistent with the underestimation of tropical precipitation. This shortcoming of climate models can be related to the parameterization of tropical convection and can be improved if the vertical resolution is increased .

Marked differences between the two simulations occur when considering aerosol optical depth (AOD) and liquid water path (LWP, vertically integrated liquid water content) (see Fig. d, e). By design, the PD simulation shows much higher values of AOD over China, Japan and the nearby oceanic regions than the PI simulation. In the WCB starting region defined in the previous section (black box in Fig. ) the winter mean values of AOD vary roughly between 0.1 and 0.15 in the PI simulation and between 0.15 and 0.2 in the PD simulation, indicating that the mean AOD in the WCB starting region of interest is increased by less than 50 % in PD compared to PI, but this increase is of course much larger over the main industrial areas. The increase in AOD is caused by an increase in accumulation and Aitken mode aerosol particles that serve as cloud condensation nuclei (CCN) and cause liquid water clouds to consist of more cloud droplets. Because the available water vapour remains the same, these cloud droplets do not grow as large. In a cloud consisting of more but smaller cloud droplets, the collision efficiency between cloud droplets is reduced and hence their growth to precipitation-sized drops is retarded . In order to produce rain in a polluted cloud, the liquid water content needs to adjust to higher values. This is visible in the elevated values of the liquid water path in the PD simulation.

Considering again the WCB starting region, the increase of the LWP amounts to about 30 %. Further downstream, the differences are small, illustrating the limited scale of the region impacted by the anthropogenic aerosol emissions. For the ice water path, the two simulations show very similar mean values (not shown). Here, no comparison is shown with ERA-Interim because different microphysical schemes lead to fairly large differences in liquid and ice water paths, which, however, cannot be interpreted as a model shortcoming. The ECHAM6-HAM microphysics is more sophisticated and complete compared to ERA-Interim and therefore, for these parameters, we cannot regard ERA-Interim as a reference.

Finally, Fig. f, g shows the winter average surface precipitation and 850 hPa horizontal wind vectors for the two simulations and ERA-Interim (Fig. h). Consistent with the differences discussed above, the simulations underestimate tropical convection in particular in the western Pacific, but north of 20∘ N the comparison with ERA-Interim shows only a weak underestimation. The lower values along the western flank of the Rocky Mountains can be explained by the lower topography in the coarser resolution model. The lower tropospheric wind fields show the correct pattern in the simulations; however, in agreement with the errors in the SLP field, the westerlies in the main storm track region are too weak. Differences between the two simulations are generally small. The main differences appear for precipitation over China (where PI is slightly wetter) and in the central North Pacific (where PD has slightly more precipitation on average). Table  shows the domain average values of AOD, LWP, CDNC and surface precipitation. In the mean it can be seen that the AOD, LWP and CDNC increase from PI to PD whereas the precipitation stays almost identical. This can be explained by the fact that due to the reduced efficiency in precipitation formation caused by the strong increase in CDNC, more condensate stays in the atmosphere and leads to an increase in LWP.

We first present an example WCB to illustrate the method and the evolution of SO2 along the pathway of an initially strongly polluted WCB. Figure  shows one WCB identified with the method described above that occurred in the PD simulation in February. Thus, at the shown point in time, 66 trajectories fulfill the criteria of an ascent exceeding 600 hPa in 48 h in the vicinity of an extratropical cyclone. All of the shown trajectories belong to one and the same WCB. The 66 WCB trajectories are shown as 66 individual lines in Fig. a, and b indicates their fast ascent between times 0 and 48 h from, on average, 950 to 350 hPa. During their ascent the WCB air parcels move from their starting region to the east of Japan (35∘ N) to their outflow region over the central North Pacific (50–60∘ N, Fig. a). During the 2 days prior to their ascent, i.e. from time -48 to 0 h, the WCB trajectories are fairly stationary and experience, on average, a slow descent from 900 to 950 hPa (Fig. b). The SO2 mass mixing ratio (see colouring of trajectories) shows very high values for most of the trajectories before the ascent. A substantial fraction (46 of the 66 trajectories, i.e. 70 %) exceeds the threshold to be classified as “polluted” (see Sect. ). During the ascent the SO2 values rapidly decrease for three reasons: SO2 can be oxidized to sulfate in the gas phase and subsequently serve as CCN, it can be dissolved in cloud droplets and oxidized in the aqueous phase , or it can condense on other preexisting aerosol particles.

It is important to note that such a configuration where the majority of WCB trajectories are strongly polluted is rare (see further analysis below). More typically, whenever a WCB (consisting of many trajectories) starts, a much smaller fraction of WCB trajectories is polluted due to the highly variable inflow of air parcels into a WCB.

In this section, first the general characteristics of the identified WCB trajectories starting from the region defined above are described, followed by a detailed comparison of so-called clean and polluted WCB trajectories in the simulations. It is our aim to compare, for PI WCB trajectories and for clean and polluted PD WCB trajectories, the evolution of LWC and IWC and the associated surface precipitation along the WCB trajectories in order to identify potential effects of the different initial aerosol concentrations on clouds and precipitation.

In Fig. , the average evolution of potential temperature, specific humidity, BC, CDNC, cloud condensate (sum of LWC and IWC) and surface precipitation is shown as a function of pressure, separately for the three categories of WCB trajectories. The evolution of potential temperature and specific humidity (Fig. a, b) is very similar for the PD clean, PD polluted and PI trajectories. The trajectories start their ascent between 290 and 295 K and reach the upper troposphere (i.e. 300 hPa) on the 315 K isentrope. The initial moisture is between 8 and 10 g kg-1 (by design; see Sect. ) and decreases rapidly along the ascending trajectories at a very similar rate. This nicely shows that the overall meteorological conditions and the large-scale ascent of the WCB trajectories are comparable between the three categories and should not be responsible for potential differences in the formation of precipitation.

The evolution of BC (Fig. c) and SO2 (not shown) along the considered trajectory samples, however, reveals huge differences. The polluted trajectories exhibit high BC values at low levels with values between 50 and 130 pg kg-1. Because the internally mixed BC aerosols serve as CCN and are activated to cloud droplets during the ascent, the BC mass mixing ratio strongly decreases with height. The PD clean and PI trajectories show much lower BC values around 10–20 pg kg-1 being 5 to 10 times smaller than the polluted PD trajectories and even their 25–75 % percentiles do not overlap below 750 hPa. The clean PD trajectories have twice as much BC as the ones in the PI simulation with a slight overlap of their 25–75 % percentiles. This difference indicates that clean present-day conditions cannot be taken as a surrogate for pre-industrial conditions. For SO2, PI and PD clean trajectories show a very similar evolution, whereas the polluted trajectories have much more SO2 in the lower troposphere – up to ∼ 700 hPa (not shown). The overall differences in SO2 between the PD polluted and PI trajectories are even larger than for BC (Table ).

The cloud droplet number concentration also decreases with decreasing pressure (Fig. d), partly because of the decrease of CCN with decreasing pressure and partly because the formation of precipitation increases with decreasing pressure due to the higher liquid and ice water contents (Fig. e) and the larger cloud droplets and ice crystals. The differences in CDNC between PD polluted and PD clean are much smaller than in BC, partly because in ECHAM-HAM we assume that each cloud has a minimum CDNC of 40 cm-3. Furthermore, other aerosol species can also act as CCN and therefore the CDNC cannot be determined by the BC concentration only. Apart from that we see the same differences as in BC, with the PI trajectories having the smallest CDNC and the PD polluted trajectories the highest CDNC.

Regarding the highest CDNC values in the PD polluted trajectories, the median value of total condensate, i.e. the sum of the liquid and ice water content (LWC + IWC) (see Fig. e), is larger in the PD simulation, but the differences between the three sets of WCB trajectories are much smaller and the 25–75 % percentiles strongly overlap. The total condensate peaks at 750 hPa in all simulations. At lower pressure levels the cloud condensate is reduced on the one hand due to precipitation formation becoming more efficient higher in the cloud and on the other hand due to the smaller specific humidity at colder temperatures, i.e. smaller condensation/deposition rates. While the PD clean trajectories were in between the PI and PD polluted trajectories in terms of BC and CDNC, they now fall on top of the PD polluted ones. Note that the case-to-case variability in every subsample is larger than the systematic difference between them as to be expected due to the high variability in the associated cyclone dynamics.

Precipitation formation depends more strongly on the liquid and ice water contents than inversely on the number concentrations of cloud droplets and ice crystals. Therefore its maximum at 700 hPa (Fig. f) is more determined by the maximum in liquid and ice water content than by CDNC and ice crystal number concentration. At 900 hPa the median precipitation rate is higher in the PD clean and the PI trajectories than in the PD polluted trajectories, although the variability between individual trajectories is again rather large and the 25–75 % percentiles strongly overlap. At pressures below 470 hPa, the median precipitation rate of the PD polluted trajectories crosses that of the PI trajectories, pointing to a delay in precipitation formation in the PD polluted WCBs. As summarized in Table  most of the precipitation falls as stratiform large-scale precipitation. It is largest in the PD clean trajectories where the average specific humidity is smallest, i.e. where more water exists in the liquid and ice phase. One could also argue that the initial BC concentration is determined by the initial precipitation. Less precipitation in the PD polluted trajectories leads to a reduced scavenging and therefore to an increased BC concentration. In order to exclude this effect, we compared all trajectories that start with similar initial precipitation. However, the results (not shown) confirm the behaviour shown in Fig. . We therefore conclude that it is the aerosol concentration that influences the precipitation and not the other way round.

In order to investigate the effect of the differences in the aerosol loadings on the precipitation formation, phase space diagrams of q and total condensate (LWC + IWC) as a function of the total precipitation are shown in Fig. , which makes it possible to directly link moisture and total condensate to the associated precipitation.

The amount of condensate that can be formed is determined by the large-scale ascent, the availability of moisture and the conversion efficiency from condensate to precipitation. Figure a, nicely shows that for all three cases the evolution of q is very similar. All trajectories start with a specific humidity of ∼ 9 g kg-1 followed by a strong decrease at very similar rates. However, the precipitation falling out of these trajectories is reduced for the PD polluted subsample and the peak value reaches less than 50 mm day-1 in the PD polluted case whereas it reaches 55 mm day-1 in the PD clean and PI case. This can also be seen in Fig. b, where the total condensate is shown as a function of total precipitation. However, the total amount of total condensate also varies between PD and PI cases. With decreasing pressure, total condensate increases up to a height of 750 hPa. For the PD clean and PD polluted trajectories, this increase is steeper and the total condensate peaks around 140 mg kg-1, whereas it increases more slowly and only reaches ∼ 120 mg kg-1 for the PI case.

Because of the much lower availability of CCN in the PI run, fewer cloud droplets form that can grow to larger sizes. These larger and fewer cloud droplets form precipitation quite efficiently, leading to a fast removal of condensate from the atmosphere. This effect can be seen in Fig. b (black line) where the increase in total condensate is less steep and the maximum value is smaller compared to PD polluted/clean. On the contrary, in the PD clean case, the mass mixing ratio of BC is already twice as large compared to PI. This increase in the availability of CCN leads to more and smaller cloud droplets and reduces the efficiency of precipitation formation. Therefore, less precipitation is reaching the ground and more condensate remains in the atmosphere (see brown and blue line). In the PD polluted subsample, where even more CCNs are available, precipitation is more strongly reduced. Thus, Fig. b nicely shows that for the same liquid and ice water content, less can be converted to precipitation shifting the phase curve in the PD polluted sample to the left. An increase in liquid and ice water content due to a reduced precipitation formation can be observed when going from PI to PD clean conditions.

The slight delay in the precipitation formation between PD clean and PD polluted trajectories could cause a spatial shift in the location of precipitation if a WCB consisted entirely of clean or polluted trajectories. In order to investigate, how many trajectories of one WCB are polluted, we calculate, for every point in time when a WCB starts with more than 50 trajectories from the black box shown in Fig. , the fraction of all trajectories that are classified as polluted. The results are shown in Fig. . As an example, for the majority of points in time (∼ 145), only 0–5 % of all trajectories belonging to one WCB are classified as polluted. Thus, it can be seen that for the majority of time steps in which WCB trajectories start, less than 20 % of all starting trajectories are classified as “polluted”. Given that very few strongly polluted WCBs exist, it is not surprising that hardly any differences are observed in the DJF mean precipitation between pre-industrial and present-day conditions in Fig. . In other words, the systematic effect of pollution observed in Fig.  is blurred by the high variability of inflow conditions for WCBs.

In this study we have analysed the possible impact of aerosol particles on precipitation formation in WCBs of winter time extratropical cyclones in the North Pacific using the ECHAM6 global climate model coupled to the aerosol module HAM. We chose the North Pacific because here the difference in aerosol burden and aerosol optical depth between present-day and pre-industrial times is among the largest due to the rise in emissions in south-eastern Asia. When combined with the prevailing westerly winds, if differences in aerosol load upwind of the genesis regions of extratropical cyclones have an effect on them, we expect to see an effect in this region.

To investigate in detail the possible impact of aerosols on precipitation, we selected the most polluted and cleanest trajectories occurring in the PD simulations. The polluted trajectories start with black carbon aerosol concentrations that are 5–10 times higher. The comparison between the most polluted and cleanest trajectories shows that for the most polluted cases, CDNC is clearly increased and the precipitation formation is reduced for a given total water content.

Our main findings are that despite these pronounced differences in the PD clean and PD polluted trajectories, the overall amount of precipitation in the North Pacific has hardly changed between pre-industrial and present-day conditions. As has been shown in Fig. , WCBs consist of a mixture of clean and polluted trajectories. Typically, there are many more clean trajectories in one WCB than polluted ones. This means that the effect of precipitation suppression in the most polluted WCB trajectories is damped by the more numerous clean trajectories. This combined with the large variability of total condensate and total precipitation in the different WCBs explains why no signal in precipitation can be detected in extratropical cyclones due to anthropogenic aerosol particles.

Our study confirms the findings by for a single warm front where no change in overall precipitation was found. A shift in precipitation as found by is not inconsistent with our results because we showed that the amounts of precipitation are very variable within different WCB trajectories. The results shown in , however, show a clear impact of anthropogenic aerosols on the North Pacific storm track with an invigoration of winter cyclones and an associated increase in precipitation. Whereas the increase in LWP and CDNC in the mean fields can also be seen in our study, we use a very different method for the evaluation of the impact of aerosols on precipitation formation. In the impact of anthropogenic aerosols is investigated by comparing simulations with different aerosol loadings. In our approach, we use a feature-based method in order to only investigate the impact on precipitation formation along the strongly ascending WCB trajectories. The Lagrangian approach with following the air parcels enables us to directly assess the impact of aerosols on the most cloud-producing airstream in extratropical cyclones in isolation, i.e. independently of other precipitating cloud systems. As shown in , WCBs lead to ∼ 60 % of the total precipitation in the North Pacific. Our results suggest that precipitation formation in the WCB trajectories is not strongly influenced by aerosols. However, we cannot make any statement about the precipitation formation in cloud systems which are not as strongly dynamically driven as a WCB. Therefore, our results are not necessarily contradictory to . Also the fact that we use another global climate model and that the microphysical parameterizations are a large source of uncertainty explains why the impact of aerosols on clouds and precipitation is not directly comparable and further research is needed.

This study, however, has several caveats. First of all, it is a pure model study because no observational climatological data of the impact of aerosol particles on extratropical cyclones exist. Aerosol impacts on clouds and precipitation are generally hard to detect in observational studies because remote sensing studies suffer from not being able to detect aerosols and clouds simultaneously. Moreover, they present an Eulerian view whereas we analysed the possible impact of aerosols on WCBs in a Lagrangian way.

Another caveat is that WCBs are less well resolved in a global climate model than they are in the regional model studies cited above. Given that the regional model study by also did not find an effect on total precipitation, the resolution may not be a major issue. A more important shortcoming could be that we used climatological sea surface temperatures which, to a large extent, control the global mean evaporation and hence precipitation rates. Thus, changes in the overall amount of precipitation could be larger if we had coupled the atmospheric GCM to a mixed-layer ocean or a full dynamic ocean model.