Size distribution of microparticles in the ice cores

The ice-core samples span a range of three orders of magnitude in C _m (10²–10⁵ μg kg⁻¹; Figs 2 and 3). In Figure 5, the low-concentration samples exhibit abnormal volume (mass) distributions and show no systematic characteristics, mostly exhibiting a low positive slope with particles existing sparsely in higher channels. A few of the clearer samples show that most of the volume consists of fine particles ranging between 1 and 3 μm. This phenomenon has also been reported for particles between 0.63 and 16 μm in other ice cores from the Tibetan Plateau (Reference DavisDavis, 2002). The irregularity in size distribution for microparticles from the low-C _m samples is common in mountain ice cores, as indicated by reports from Hispar Glacier, Karakoram, and Ngozumpa Glacier, near Everest (Reference Wake, Mayewski, Li, Han and QinWake and others, 1994). This appears to be partly because a few, or even single, coarse particles comprise a large fraction of the total volume, although the coarse particles in the clear samples have a considerable uncertainty in volume measurement, as discussed above.

Fig. 5.

Volume size distributions (in 50 channels on logarithmic scale) of microparticles in low- and high-Cm samples (black curves) and the log-normal fitting curves (grey curves) from the MA6350 and Dunde ice cores.

Almost all the dust particle size distributions of high C _m samples fit a log-normal size distribution well. The coarse particles (>15 μm) in these ice cores can constitute a large proportion of up to 40% of the total mass (volume). However, the average mass fraction of the >15 μm particles is generally minor, with about 2.5–7.6% of the total (Table 1). This indicates that the total mass is predominately dependent on the medium-size particles. In the Dunde ice core, the 3–15 μm particles contribute an average of 76 ± 6% (range 54–92%) to the total mass in the log-normal samples, while in the Muztagata 6250 m (MA6250), 6350 m (MA6350) and 7010 m (MA7010) ice cores, this proportion is 74 ± 6% (range 59–83%), 74 ± 6% (range 60–89%) and 71 ± 7% (range 53–84%), respectively. In the log-normal distribution, 68% of the total mass is in the 1σ size interval [exp(In D _m−In σ), exp(In D _m+ In σ)]. The log-normal modes are mainly located in this 3–15 μm size range, and the 3–15 μm mass fraction (average ∼71–76%) in these samples is similar to this value (68%). We conclude that the lognormal distribution is predominantly attributed to the midsized particles between 3 and 15 μm, which contribute most of the total volume.

The larger size of the mode of some samples arises from the effect of large particles that appear infrequently but make up a large fraction of the volume, instead of a continuous distribution of particles across all the measurement channels. Furthermore, though the log-normal distribution is widely used, it is neither exclusively suitable nor, in some cases, the best for describing atmospheric aerosol size distributions. The normal (Gaussian) distribution fits some high-concentration samples in those ice cores well (Reference WuWu, 2004).

The origin of coarse and giant particles in the ice cores (as shown in Fig. 4) is not clear. Dust-storm events on the Tibetan Plateau are frequent, though they are less frequent and less intense than those in the peripheral deserts. Satellite data show that the dust layers over the Tibetan Plateau appear most frequently around 4–7 km a.s.l. (Reference HuangHuang and others, 2007). The relative elevation between the bare ground surface and the ice-core drilling site is ∼3 km. Therefore, transport of local coarse material to the ice-core drilling site seems quite possible.

The coarse or giant particles can also come from a remote source hundreds or thousands of kilometers away. The giant particles (>62.5 μm), contrary to the traditional opinion that they will be preferentially removed by gravitational settling along the transport path, are found less often at greater distances (>1000 km) (Reference Middleton, Betzer and BullMiddleton and others, 2001), though the mechanism is not clear. Stokes gravitational settling overestimates the losses of large dust particles during atmospheric transport from North Africa over the tropical North Atlantic and Caribbean (Reference Maring, Savoie, Izaguirre, Custals and ReidMaring and others, 2003). On the Tibetan Plateau, the transport of coarse particles from local sources and areas remote from the glacier is therefore feasible. In an early dust-tracer transport model (GISS GCM) the calculated lifetime of silt was 179 hours for 2–3.6 μm and 28 hours for 12–20 μm particles (Reference Tegen and LacisTegen and Lacis, 1996). By contrast, in the late GISS GCM model, the silt fraction (8–16 μm) was supposed to have a gravitational deposition lifetime of 2.2 days (Reference MillerMiller and others, 2006). Dry deposition is less efficient in the current model, with a particle lifetime of 10.1 days (Reference MillerMiller and others, 2006), compared with the previous estimate of 8.9 days (Reference Miller, Tegen and PerlwitzMiller and others, 2004). The longer lifetime in the model allows dust particles to be transported over a longer range. The dust mass contributions from the lower, middle and upper troposphere in different size categories in the model show that most of the dust mass concentrates in the diameter range 1.2–6 μm (Reference Miller, Tegen and PerlwitzMiller and others, 2004). Our results show that for the high-concentration samples, which fit the log-normal law in volume size distribution, most of the mass is contained in the particle fraction of 3–15 μm diameter, which is much coarser than in the model. Our results also indicate that coarse particles might be common in the upper-level troposphere, at least over the Tibetan Plateau, and suggest that the lifetime of silt particles, especially the large ones, might still be underestimated, although we cannot discern the remote and local sources of these coarse/giant particles.

The complete size distribution for eolian dust may have three modes:

1. 1. mode A, centered between 1 and 10 μm radius, which consists mainly of soil-derived aerosols under all conditions

2. 2. mode B, centered between 10 and 100 μm radius, which may be the soil parent particles, and which appears only under heavy dust-loading conditions

3. 3. mode C, centered between 0.02 and 0.5 μm radius, which might be the background aerosol and might be ambient (Reference Patterson and GillettePatterson and Gillette, 1977).

We are aware that our measurements cover only particles in the range 1–30 μm, i.e. the mode A range, a small interval within the complete size range of atmospheric dust. Mode B particles, which are common in Chinese loess deposits, are less likely, but not impossible (as discussed above), to be found in the ice cores, owing to the difficulty of lifting and transporting parent soil particles at high elevation to the glacier. Mode C particles, the background aerosol (Reference Ram and GayleyRam and Gayley, 1994), are expected to exist in our samples, though they are too fine to be measured by Coulter Counter. Our results are strictly confined to the measured size range.

We need to pay attention to the possible solubility of some dust particles. If insoluble materials are internally mixed with a substantial amount of water-soluble materials, measurements of particle size by Coulter Counter will be biased to some degree, especially in Dasuopu and Everest where sea salts and anthropogenic substances might contribute a relatively higher proportion than at other sites. Aerosol collected in urban areas from a semi-arid region of western India indicates that the mineral dust component contributed to about 69% of the total suspended particles after neutralization by acidic pollutants (Reference Rastogi and SarinRastogi and Sarin, 2006). However, due to the high altitude of ice-core sites and relatively weak anthropogenic effects, we believe that this issue is insignificant at the remote Everest/Dasuopu and other sites. Therefore, we consider that the measured size distribution is reliable.

The impact of altitude on size distribution

The MA6250, MA6350 and MA7010 ice-core samples from Muztagata provide a comparison for the impact of altitude on the mass concentration and size distribution of eolian dust deposited in alpine glaciers with similar air-mass trajectories (through the HYSPLIT model). The upper 6 m of the MA6350 ice core is found to correspond to the 0.8–7.5 m section of MA6250 core by matching the δ¹⁸O profiles (Fig. 2, gray curves). These two sections show no significant difference in average C _m during the overlapping period (7466 μg kg⁻¹ for the MA6350 section and 8893 μg kg⁻¹ for the MA6250 section). However, many more samples that obey the log-normal size distribution were found in the MA6250 core than in the MA6350 core.

With increasing altitude, the log-normal distribution parameters also change, as the coarse particles settle out faster than finer particles. There is only a small difference in log-normal parameters between the MA6250 and MA6350 ice cores, with similar average mode (5.87 μm vs 5.74 μm), standard deviation (SD) (2.07 vs 2.06) and >15 μm mass percentage (4.78% vs 4.68%) (Table 1). We thus discuss the altitude impact focusing on the difference between the MA6250/MA6350 cores considered together and the MA7010 ice core (Table 1; Fig. 6):

1. 1. The mode is smaller and less variable in the high-altitude ice core than in the lower ice cores. Particles from the MA7010 ice core have a mean mode of 4.99 μm (range 3.02–7.52 μm), while particles from MA6250 and MA6350 ice cores have mean modes of 5.87 μm (range 3.68–8.87 μm) and 5.74 μm (range 3.23–10.49 μm) respectively. The t test verifies that there is a significant difference between the MA6350 and MA7010 modes (P = 2.8 × 10⁻⁶ ≪ 0.05).

2. 2. The three Muztagata ice cores show very similar SD (t test: P = 0.43 ≫ 0.05) for the log-normal fit, indicating that elevation has no clear impact on this parameter.

3. 3. The mass contribution of coarse particles in the MA7010 ice core is lower than in the MA6250/MA6350 cores, indicating that coarser particles make up a relatively smaller fraction of the total mass at higher altitude than at lower altitude.

4. 4. Samples with lower C _m exhibit log-normal size distributions at greater altitude than those at lower elevations. In MA7010, dust concentrations as low as 4000 μg kg⁻¹ can exhibit strong log-normal size distributions, while in MA6250 the C _m must be >7000 μg kgcm^{− 1} to be well described as log-normal. This also indicates more extensive sorting at higher-altitude sites.

Fig. 6.

Log-normal fitting parameters (the mode, SD, >15 um mass fraction and determination coefficient R ²) vs C _m (x axis; unit: μg kg⁻¹) for microparticles from ice cores on the Tibetan Plateau. The result from the GRIP ice core (Reference SteffensenSteffensen, 1997) is also shown for comparison. The line in each panel indicates the trend.

These log-normal parameter characteristics indicate that with increasing elevation of deposition, microparticles at higher altitude are more extensively sorted during transport and their size distribution more closely mimics a log-normal distribution. In addition, the size of the distribution mode trends smaller, with less variable SD and higher R ² (Fig. 6). Our measurement is consistent with the modeled dust size distribution with height, which showed that modal size decreased quickly from 3.0 μm at 970 hPa to 1.6 μm at 470 hPa altitude (Reference Tegen and LacisTegen and Lacis, 1996).

Log-normal size distribution parameters

The parameters describing the log-normal fitting of dust size distributions provide additional insight into the characteristics of eolian dust. Particles in the Greenland ice cores (Reference Zielinski and MershonZielinski and Mershon, 1997), as the end members of Asian dust, show a distinct log-normal distribution with a mode size commonly of ∼2 μm. In the Greenland Icecore Project (GRIP) ice core (Fig. 6), microparticles show a log-normal distribution in the size interval 0.4–6 μm with nearly identical shapes, in spite of the large variability in C _m ranging from 10 to almost 10⁴ μg kg⁻¹ (Reference SteffensenSteffensen, 1997). Samples with very low C _m (only tens of μg kg⁻¹) have unimodal log-normal distributions similar to those with high dust concentration (thousands of μg kg⁻¹). This characteristic is also found in the NorthGRIP (Reference Ruth, Wagenbach, Steffensen and BiglerRuth and others, 2003) and Antarctic (Reference Delmonte, Petit and MaggiDelmonte and others, 2002) ice cores, indicating that, after long-range transport, atmospheric dust is well sorted and can reach an equilibrium state with uniform and log-normal size distributions in some cases. In the GRIP ice core, as shown in Figures 6 and 7, the mode increases with increasing C _m, while the SD shows no correlation with C _m and remains relatively constant at about 1.6 μm. In addition, for those samples with C _m greater than 100 μg kg⁻¹, the SD and coarse (>4 μm) mass fraction seem to be constant. Consequently, a relatively constant SD and increasing mode size would result in a broader log-normal distribution with increasing C _m.

Fig. 7.

The relation between the mode and SD in the ice-core samples from the Tibetan Plateau. The result from the GRIP ice core (Reference SteffensenSteffensen, 1997) is also shown for comparison. The line in (a–d) indicates the trend.

However, the observed size distribution of microparticles from the Tibetan Plateau ice cores does not demonstrate the same trend. The dust particle size distribution parameters in Figure 6 show that the microparticle size distributions become more log-normal when C _m increases in samples from all the sites, and especially those in the Dunde ice core. In contrast to the end members of the Asian dust in Greenland, the standard deviation of dust over the Tibetan Plateau decreases with increasing mass concentration. This suggests that the size distributions of microparticles in mountain ice cores differ from those in Greenland due to different effects of altitude and distance. R ² becomes higher and the SD of the log-normal fitting generally decreases with increasing C _m. In all the Tibetan ice-core samples, the SD decreases with increasing mode size (Fig. 7), indicating that samples with larger mode sizes have volume distributions more concentrated in the log-normal part of the dust size distribution than those with smaller modes. This trend is much clearer in the Dunde ice core, a location surrounded by Asian dust-source areas. The correlation between the lognormal distribution parameters and mass concentration would help to simplify the characterization of dust used in general climate models.

Seasonal differentiation of dust particle volume size distribution

The mean number diameter and mean mass (volume) diameter of dust particles in ice cores on the Tibetan Plateau vary significantly on a seasonal basis (e.g. Reference WuWu and others, 2006). The seasonality of size distribution variability is also important in evaluating the radiative properties of atmospheric dust, especially for those sites where the prevailing circulations change seasonally. Differences in the lognormal fitting of samples from the spring–summer and autumn–winter seasons are clear in recent snows from the GRIP site (Reference SteffensenSteffensen, 1997). Though the Dasuopu ice core demonstrates the most obvious seasonal variation in δ¹⁸O and dust concentration profile (Fig. 3), only a few samples are well modeled by a log-normal size distribution. Here we investigate the seasonal variability in dust size distribution parameters using the well-dated MA6250 ice-core record, which has clear seasonality in δ¹⁸O values and finely resolved vertical sampling.

Samples well modeled by a log-normal distribution can occur during both the winter and summer seasons in the MA6250 ice core (Fig. 8). Although this record is subject to some uncertainties owing to limited data from only one site and a relatively short time-span, the results indicate that the seasonal differences in the log-normal fitting parameters appear to be minor and no systematic trend is obvious. Generally, in all the cores studied in this paper, no obvious systematic correlations between the log-normal parameters and oxygen isotope ratios were evident. Therefore lognormal fitting parameters appear more dependent on dust-storm events rather than season. Dust-storm events occur mainly during the dust season, but they can also occur, though less frequently, throughout the year. We suspect that weak dust-storm events loft only a small amount of dust particles into the atmosphere and do not produce the lognormal distribution. However, the detailed physical processes that lead to a strong dust storm creating a log-normal distribution still elude us.

Fig. 8.

The time series of oxygen isotope ratios (δ¹⁸O) and log-normal parameters from the MA6350 ice core. W and S indicate winter season (grey area) and summer season respectively.

Regional differences of log-normal characteristics

Reference Wake, Mayewski, Li, Han and QinWake and others (1994) pointed out that the observed size distribution of microparticles from ice cores in the Tibetan Plateau is dependent upon physiographic location with respect to dust source regions and atmospheric transport pathways. For example, the size distribution of the dust in the northern and western Tibetan Plateau, such as that from Muztagata and nearby Congce, appears to be log-normal, while the dust in the southern Himalaya shows a slightly positive slope (volume increases with increasing diameter) in volume-size distribution.

Our results indicate the volume size distribution tends toward log-normal depending on dust mass concentration but is independent of physiographical location. The sites of the Everest and Dasuopu ice cores located in the southern part of the Tibetan Plateau have similar climate and landscape. We designated the combination of these two ice cores with different symbols (Figs 6 and 7) to represent the atmospheric dust over the central Himalaya. The three sites with different climate regimes on the Tibetan Plateau, namely Dunde, Muztagata and Dasuopu/Everest, provide a regional view of the dust particle size distribution. In the northern (Dunde) and western (Muztagata) Tibetan Plateau, adjacent to the Asian dust-source areas, dust particles have a clear monomodal size distribution, and no bi- or multimodal distributions were found in the range of measurement (1–30 μm). In the southern Tibetan Plateau, high-concentration samples from Everest and Dasuopu also have a monomodal log-normal distribution, while some samples have clear bi- or polymodal size distributions, though the C _m is low. Aerosols collected in Nepal indicate that desert dust can be transported regionally as well as long distances (possibly from the Sahara) to the Himalaya (Reference Carrico, Bergin, Shrestha, Dibb, Gomes and HarrisCarrico and others, 2003). In the Himalaya, the mixture of dust particles from both local and remote sources may combine to produce a multimodal size distribution. Bimodal size distribution of Asian dust can be found in some ice cores (Reference Wake, Mayewski, Li, Han and QinWake and others, 1994) and loess sediments (Reference SunSun and others, 2002). The dust concentration and polymodal size distribution for microparticles in the Everest and Dasuopu ice cores will be discussed in further work.