This study uses ten summers (June–July–August, JJA) of multi-satellite observations during 2006 to 2016, except for 2011 due to data gaps, to investigate the distribution characteristics and formation mechanism of ice crystal particles over the TP. The primary dataset is the DARDAR-Nice product, complemented by additional retrievals from CloudSat and CALIPSO observations.

The DARDAR-Nice PRO product provides high-vertical-resolution estimates of Ni retrieved along the A-Train satellite track. The retrievals are based on the VarCloud algorithm (Delanoë and Hogan, 2008, 2010), which combines observations from the Cloud Profiling Radar onboard CloudSat and the cloud-aerosol Lidar with Orthogonal Polarization (CALIOP) lidar on CALIPSO. DARDAR-Nice profiles are provided with a vertical resolution of 60 m. The product includes Ni values and corresponding uncertainty estimates for particle with size larger than 5, 25 and 100 µm. In this study, the standard error of Ni is derived directly from the icnc_5um_error variable, ensuring that the uncertainty associated with each profile is properly represented in the analysis. This production has been systematically and comprehensively evaluated based on theoretical considerations and a large body of in situ observations (Sourdeval et al., 2018). However, it tends to overestimate ice crystal number concentrations in cloud parcels warmer than -30 °C, due to the assumption of a monomodal particle size distribution in the retrieval algorithm. To ensure the reliability of the results, this study focuses exclusively on clouds with temperatures below -30 °C and discusses the Ni of ice crystals with sizes larger than 5 µm.

In addition to Ni (>5 µm), Ni (>25 µm) was also analyzed to evaluate the sensitivity of the results to the particle size threshold. Given that the DARDAR-Nice dataset predominantly samples mature or aged cirrus layers, larger ice crystals (>25 µm) may better represent the evolved stage of cloud microphysical development. Therefore, examining Ni (>25 µm) provides complementary information and helps indicate the plausible range of Ni values.

The spatial distribution and vertical structure derived from Ni (>5 µm) and Ni (>25 µm) are largely consistent, indicating that the main structural features of cirrus over the Tibetan Plateau are robust across size thresholds. The primary difference lies in the absolute magnitude: Ni (>5 µm) values are systematically higher by approximately a factor of 2–3 compared to Ni (>25 µm), reflecting the contribution of smaller ice crystals to the total population. Since Ni (>5 µm) captures a broader fraction of the total ice crystal population, it is retained as the primary variable in this study. The corresponding Ni (>25 µm) results are presented in the Supplement to illustrate the range of Ni values and to demonstrate that the main conclusions are not sensitive to the selected particle size threshold.

In addition, although ice water content (IWC) is also provided in the DARDAR-Nice product, it has not been specifically validated. Therefore, this study uses IWC data from the CloudSat 2B-CWC-RO product (https://www.icare.univ-lille.fr/asd-content/archive, last access: 2 April 2025), for which the retrieval quality and accuracy have been discussed in detail by Austin (2007) and Austin et al. (2009). Besides 2B-CWC-RO, the CloudSat 2B-CLDCLASS-LIDAR product is employed, which classifies clouds into eight types across ten vertical layers with a horizontal resolution of 2.5km×1.4km. Its classification algorithm integrates vertical and horizontal cloud structures, precipitation features, cloud temperature, and MODIS radiative measurements to enhance classification accuracy. These CloudSat products provide critical microphysical parameters and cloud classification necessary for understanding ice cloud properties.

CALIPSO can monitor the vertical distribution characteristics of clouds and aerosols, automatically identify aerosol types, and provide global aerosol horizontal distribution characteristics and vertical distribution information (Zheng et al., 2018). Liu et al. (2008) also conducted aerosol detection using CALIPSO, further confirming its effective aerosol detection capabilities.

Dust aerosols exhibit strong ice-nucleating activity and represent an important global source of INPs (Hoose and Möhler, 2012; Murray et al., 2012; Ladino Moreno et al., 2013). Meanwhile, sampling studies during biomass burning conducted by Prenni et al. (2012) and McCluskey et al. (2014) indicate that particles from biomass combustion constitute a significant regional source of INPs, particularly when other effective INPs are scarce. In addition, recent observational analyses by Mamouri et al. (2023) and Ansmann et al. (2025) suggest that smoke aerosols can exert a substantial influence on ice crystal formation at altitudes while temperatures fall below -45 °C. Therefore, this study primarily focuses on the role of dust and smoke aerosols. This study employs information from the Level-2 Version 5 kmCLay standard products of the CALIPSO satellite data spanning from 2006 to 2016 (except 2011) to assess the impact of dust and smoke aerosols on the formation of cirrus clouds.

In this study, both daytime and nighttime satellite observations are included, the aerosol information is used to characterize climatological, grid-cell-averaged aerosol occurrence rather than instantaneous cloud-aerosol collocation.

To investigate meteorological conditions for the satellite observations, this study utilizes ERA5 reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF). ERA5 provides hourly global data at a spatial resolution of 0.25°×0.25° across 37 vertical pressure levels, covering the period from 1979 to the present (Xie et al., 2021). The key variables used in this study are specific humidity and vertical wind velocity from ERA5, as well as temperature from satellite observations. Together, these three variables are essential for analyzing meteorological conditions related to cirrus cloud formation and deep convective vertical transport. All datasets and corresponding variables used in this study are summarized in Table 1.

The focus area of this study is the TP and its surrounding regions, spanning from 66 to 106° E and 24 to 40° N. The original orbital data of the DARDAR-Nice PRO product, 2B-CLDCLASS-LIDAR classification product and 2B-CWC-RO cloud product are interpolated into grid point data with a resolution of 2°×2° based on the method outlined in Wang et al. (2023). The overall data processing workflow adopted in this study is illustrated in Figs. S1 and S2 in the Supplement. Figure S1 outlines the procedure used to derive Ni statistics, including calculations in both the horizontal and vertical directions. Figure S2 illustrates how aerosol classification data from CALIPSO are combined with cirrus cloud properties retrieved from the DARDAR-Nice product. These schematics provide a transparent overview of the integration and processing of the various satellite datasets used in this study.

The 2B-CLDCLASS-LIDAR deep convective cloud product was used to quantify deep convection. For each grid cell and time interval, the presence of one or more deep convective clouds was counted as a single event, irrespective of the number of profiles exhibiting convection. These events were then summed over all intervals to yield the total number of deep convection occurrences per grid cell. For the investigation of Ni, statistical analysis was conducted at intervals of 60 m, based on the vertical resolution of the DARDAR-Nice PRO product. Data points with large uncertainties were set to NaN to minimize bias in the statistics.

The horizontal distribution of Ni may be influenced by uneven sample distribution resulting from the varying occurrence frequency of ice particles across different layers. To address this, we normalized Ni for each grid during the ten summer seasons over the TP, to obtain the horizontal distribution of Ni for cirrus clouds. The normalization process is presented in Eq. (1): 1y=∑i=1Numxi∑i=1Nummi where xi is the sum of Ni where the temperature is below -30 °C. Num is the total number of profiles included in the analysis. mi is the effective layers within the corresponding grid cell for which Ni is greater than 0. y is the normalized Ni in the corresponding gird.

To compute the vertical distribution of Ni, each profile is analyzed layer by layer. For each profile, if Ni is present in any layer, the profile is counted as 1; this count is used for normalizing the total number of profiles. For each layer, the Ni from all profiles are summed and then divided by the total number of counted profiles, yielding the normalized Ni for that layer. The detailed calculation method is given in Eq. (2): 2an‾=∑i=1Numai,n∑i=1NumCi where an‾ represents the normalized Ni for layer n, ai,j is the Ni in layer n of profile i, Ci is the profile count (Ci=1 if Ni is present in at least one layer of profile i, and Ci=0 otherwise).

In the absence of INPs in the atmosphere, ice crystal formation occurs primarily through homogeneous nucleation.It is generally acknowledged that temperatures near -38 °C represent the threshold for homogeneous freezing of supercooled water droplets and aqueous aerosol particles under sufficiently high ice supersaturation (Duft and Leisner, 2004; Murray et al., 2010a; Koop and Murray, 2016). Traditionally, the identification of homogeneous nucleation has relied primarily on temperature thresholds. However, due to the continuous dynamic growth of ice particles through condensation, accurate simulation remains challenging.

A novel approach is proposed to identify homogeneous nucleation by leveraging aerosol classification data from the CALIPSO satellite over the TP during the summer from 2006 to 2016. Specifically, when aerosol types are classified as “clean”, it indicates a low concentration of INPs, favoring the dominance of homogeneous nucleation in ice crystal formation. Kim et al. (2018) performed a statistical analysis of different aerosol types in this product and found that “clean” aerosols account for only about 1 % of occurrences in the CALIPSO column-level aerosol product, representing background aerosol with very low concentration, which further supports the validity of this assumption. Grid points identified exclusively with “clean” aerosol conditions are therefore interpreted as environments in which homogeneous freezing may be more likely to dominate. However, clean conditions do not imply purely homogeneous nucleation, and uncertainties remain regarding the relative contributions of different ice formation mechanisms.

Although CALIPSO provides detailed vertical profiles of aerosols, this study does not explicitly use the height-resolved information. Instead, the aerosol occurrence is analyzed at the grid-cell level without distinguishing altitude. This approach is adopted for two main reasons. First, CALIPSO's aerosol detection is most reliable in the lower troposphere, while its sensitivity decreases significantly at higher altitudes due to signal attenuation and the difficulty of distinguishing aerosols from thin cirrus clouds (Mao et al., 2022). Therefore, focusing on overall aerosol occurrence within each grid ensures better data consistency and avoids potential misclassification errors. Second, the Ni analyzed in this study corresponds to temperatures below -30 °C, the relevant aerosols are those that can influence cloud formation through vertical transport or large-scale dynamical processes, rather than being co-located at the same altitude. Hence, by integrating aerosol occurrence over the entire column within the same grid, the analysis effectively captures the overall influence of low-level dust or smoke aerosols on upper-tropospheric ice clouds, without introducing additional uncertainty from vertical matching. Therfore, the overall comparison, statistical results, and main conclusions remain robust.

Based on the DARDAR-Nice PRO product, this study analyzes the spatial variation of Ni across all layers where the temperature is below -30 °C during the study period. The horizontal distribution of Ni (>5 µm) in Fig. 1 demonstrates that the average concentration over the TP is 187 L-1 during the study period. The corresponding average concentration for Ni (>25 µm) is 87 L-1, which is less than half of Ni (>5 µm) (Fig. S3). The average Ni (>5 µm) reported here is higher than the approximately 150 L-1 over the TP reported by Gryspeerdt et al. (2018), who used DARDAR-Nice data from 2006 to 2013 to study global Ni (>5 µm) but focused only on cloud-top statistics. Considering that our analysis includes all layers below -30 °C, the higher Ni (>5 µm) is reasonable and consistent with physical expectations, which also indirectly supports the reliability of our results. The average concentration in the south (24–30° N, 66–106° E) is significantly higher than other areas, reaching 213 L-1, with the maximum value located in the north-central region of India (24–26° N, 78–80° E), reaching 253 L-1. Over the north, including the Xinjiang, Inner Mongolia, the north of the Qilian Mountains and the Kunlun Mountains, Ni is only 143 L-1, only two-thirds of Ni compared with that in the southern region. For Ni (>25 µm), the corresponding average concentrations are 94 L-1 in the southern region, 112 L-1 in the north-central region of India, and 70 L-1 in the northern areas. Compared with Ni (>5 µm), the spatial distribution remains generally consistent, although the absolute concentrations are systematically lower.

We investigated the relationship between the incidence of deep convective clouds, INPs, and Ni at all grid points over the TP based on the different formation mechanisms of cirrus clouds. As shown in Fig. 2a, deep convection occurrences (DCO) is significantly higher in the southern and southeastern regions of the Plateau, where the Ni tend to be elevated (Fig. 1). Also, Ni revealed a positive nonlinear relationship with DCO, with a coefficient of determination (R2) of 0.55 (p<0.01) and a root mean square error (RMSE) of 24.03 L-1, indicating that deep convective activity plays a significant role in modulating Ni over the TP. During the ASM, frequent deep convection in the southern TP facilitates the transport of warm, moist air and water droplets from the Indian Ocean and the Bay of Bengal to higher altitudes (He et al., 2019). Under moist conditions, ascending air parcels are more likely to experience a prolonged period of ice supersaturation, thereby increasing the probability of exceeding the supersaturation threshold required for homogeneous ice nucleation. In humid environments, air parcels can therefore maintain supersaturated conditions for a longer duration, making homogeneous nucleation more likely to dominate under such circumstances (Zhao et al., 2018). By contrast, heterogeneous nucleation is initiated by INPs and typically requires a lower ice supersaturation threshold, allowing it to occur earlier during ascent (DeMott et al., 2010). As a result, in environments with abundant water vapor, homogeneous nucleation may gain a relative advantage in competition with heterogeneous nucleation, favouring the formation of higher Ni.

This interpretation is consistent with the relatively high Ni observed over the southern TP during summer. However, within an observational framework alone, the respective contributions of dynamical conditions, aerosol properties, and thermodynamic processes cannot be fully disentangled. The interpretation presented here should therefore be regarded as a qualitative explanation based on physical consistency rather than a definitive attribution.

In addition to convective activity, the presence of INPs also plays a critical role in modulating Ni over the TP. Zhao et al. (2018), using nine years of satellite observations, demonstrated that ice crystal formation is regulated not only by the availability of INPs but also by ambient water vapor conditions. This highlights the important role of moisture as a prerequisite for cirrus cloud evolution, while emphasizing that high water vapour availability alone is not sufficient to guarantee ice formation. Ice nucleation can only occur when the ice saturation ratio exceeds the threshold required for freezing; without reaching this threshold, no ice formation is possible (Gettelman et al., 2010). As a result, moisture should be regarded as a necessary background condition rather than a direct or sufficient driver of ice crystal formation.

Consequently, when investigating the relationship between INPs and Ni, directly comparing INPs and Ni across all grid cells may lead to misleading interpretations. This is because differing atmospheric conditions, particularly variations in moisture and the development of ice supersaturation, can strongly influence whether ice formation occurs. For example, high Ni in one grid cell may primarily reflect favourable moisture conditions that allow supersaturation to be achieved, rather than an enhanced influence of INPs, whereas in another grid cell the potential effect of INPs may be masked if the supersaturation threshold is not exceeded.

In principle, restricting the analysis to grid cells with broadly similar atmospheric conditions would allow a more direct comparison. However, the TP exhibits pronounced spatial heterogeneity, especially between its northern and southern regions. To partially account for differences in moisture-related thermodynamic conditions, this study introduces the IWC confined INPs concentration (ICIC), defined as the logarithm of the ratio between the occurrence number of smoke (or dust) particles and IWC within each grid cell (Eq. 3). This formulation aims to reduce the confounding influence of IWC on the apparent Ni-INPs relationship. To further demonstrate the robustness of this normalization, we compute the partial correlation between INPs and Ni after removing the effect of IWC. Specifically, the partial correlation coefficient is calculated by statistically removing the linear effect of IWC from both Ni and ICIC across all grid cells and then computing the correlation between the residuals. This approach does not eliminate the physical influence of IWC on Ni, but allows an evaluation of the Ni-INPs relationship independent of the first-order linear contribution of IWC. The resulting partial correlation coefficient (r=-0.38) indicates that a relationship between ICIC and Ni remains after accounting for IWC, although IWC itself continues to exert a strong influence. 3ICIC(type)=log⁡typeeventIWC where type is dust or smoke, type_event is the number of events of that aerosol type, and ICIC(type) is the ICIC value corresponding to that aerosol type in the corresponding gird.

Figure 3a shows the spatial distribution of this metric, revealing that ICIC is predominantly concentrated in the northern part of the Plateau, with significantly lower values in the south. Moreover, an inverse nonlinear relationship is observed between ICIC and Ni, with a coefficient of determination (R2) of 0.61 (p<0.01) and a root mean square error (RMSE) of 22 L-1, indicates that the quantity of ICIC has a significant impact on the Ni over the TP. While the Ni mainly arises from homogenization nucleation, heterogeneous nucleation of INPs promotes the formation of ice crystals (Khvorostyanov et al., 2006; DeMott et al., 2010) by absorbing a large amount of water vapor and destroying the conditions for homogeneous nucleation. The inhibitory effect of heterogeneous nucleation on homogeneous nucleation becomes more pronounced with an increase in INPs content, leading to a lower Ni. These two factors, namely the increased convective cloud frequency in the south and the elevated INPs levels in the north, are the primary contributors to the observed spatial pattern of Ni, which tends to be higher in the south and lower in the north.