The noctilucent clouds (NLCs), also known as polar mesospheric clouds (PMCs), are the highest and coldest clouds in the terrestrial atmosphere, forming in the high-latitude summer mesosphere at ∼83 km altitude, where the temperature can drop to ∼140 K or lower. The long-term trends in NLCs are thought to be associated with global climate change. NLCs are susceptible to perturbations from lower-atmospheric activities such as gravity waves (Gao et al., 2018) and planetary waves (France et al., 2018). NLCs are strongly influenced by both solar and lunar tides, with diurnal and semidiurnal variations observed in the NLC properties (Fiedler and Baumgarten, 2018; Stevens et al., 2017; von Savigny et al., 2017). NLCs can also be affected by solar activities on various timescales, including solar proton events (Bardeen et al., 2016; Winkler et al., 2012), the 27 d solar rotation (Robert et al., 2010; Thomas et al., 2015; Thurairajah et al., 2017) and the 11-year solar cycle (Dalin et al., 2018; DeLand and Thomas, 2019; Hervig et al., 2019). To distinguish the contribution of solar activity to polar mesospheric clouds from that of climate change, it is important to clarify the mechanisms of the solar influence on NLCs. Based on the observed modest anti-correlation in NLCs with the 27 d and 11-year solar variations, both photodissociation and dynamic origins have been proposed in which the solar ultraviolet irradiance as characterized by the Lyman alpha (Ly-α) index is supposed to play a key role by altering the water vapor and temperature in the NLC region (Dalin et al., 2018; Thomas et al., 2015), while in general the exact mechanism is still unclear. In this paper, the interplanetary magnetic field (IMF) By rather than the Ly-α is applied as the solar activity index to explore the solar wind–NLC link, and new hypotheses are discussed in the next section.

The main IMF By-related process is the change in ionospheric potential in polar cap regions, which determines the flow of the regional downward ionosphere–earth current density JZ. The current flow is part of the global atmospheric electric circuit (GEC), with ionospheric potential being ∼250 kV positive relative to earth's surface, maintained by the global thunderstorms and electrified clouds (Slyunyaev et al., 2019; Williams and Mareev, 2014). The earth experiences a Lorentz electric field applied by the cross-product of solar wind magnetic field and velocity, which is mainly northward (southward) for positive (negative) IMF By, and observations have shown that the IMF By-dependent daily averaged perturbation of ionospheric potential ranges from -30 to 30 kV at high geomagnetic latitudes and is opposite in the SH and NH (Tinsley and Heelis, 1993).

A possible link may exist between the solar wind By variations and polar surface meteorology through the ionospheric potential, which has been supported by a variety of observations, in terms of polar surface pressure (Lam et al., 2013), geopotential height (Lam et al., 2014), temperature (Freeman and Lam, 2019; Lam et al., 2018) and below-cloud irradiance (Frederick et al., 2019; Frederick and Tinsley, 2018; Tinsley et al., 2021). It should be noted that these observations are characterized by two features: the responses in the SH and NH are opposite, in line with the opposite IMF By-induced ionospheric potential in the SH and NH; the delay time is short, lasting only a few days or less. A hypothesis has been proposed to explain the above observations: firstly, solar wind By induces changes in the ionospheric electric potential, as well as the downward current density JZ in the GEC; secondly, the microphysical processes inside clouds are sufficiently sensitive to space charge generated by JZ so that the cloud properties such as infrared opacity and albedo will consequently be affected. Finally, polar surface meteorology will be influenced by cloud radiative forcing (Lam and Tinsley, 2016). The invoked cloud microphysical changes have been detailed for individual aerosol–droplet collisions (Zhang et al., 2018, 2019; Zhang and Tinsley, 2017, 2018), but direct measurements in clouds and modeling are required to test this hypothesis.

In comparison with the tropospheric clouds within which many factors are involved, the polar mesospheric clouds provide a relatively pure scenario to study the role played by electric charges in the microphysical process of clouds. By extending the above “solar–GEC–cloud microphysics–tropospheric meteorology” hypothesis, it is straightforward to propose the “IMF By–ionospheric potential–NLC microphysics–NLC brightness” hypothesis for the polar mesospheric clouds: IMF By induces changes in polar ionospheric potential, which will modulate the charge distribution on meteoric smoke particles (MSPs) with major implications for the nucleation rate and ice particle formation processes in NLCs and ultimately affect the macroscopic properties of NLCs.

The formation of ice particles in NLCs is still not well understood, as a variety of factors are involved in the microphysical process, among which the nucleate rate and number density of ice nuclei contribute the most important uncertainties (Rapp and Thomas, 2006). Although the homogeneous nucleation has been considered feasible (Murray and Jensen, 2010), the extreme conditions required make the homogeneous nucleation unlikely to occur at the typical mesospheric supersaturation level (Tanaka et al., 2022). The heterogeneous nucleation instead is thought to be more effective by providing pre-existing ice nuclei, for which candidates such as ion clusters, soot, sulfate aerosols and meteoric smoke particles have been proposed (Rapp and Thomas, 2006). MSPs are abundant in the mesosphere and considered to be most likely; evidence that ice particles contain small numbers of MSPs has been provided by observations (Hervig et al., 2012). The exact nucleation process of MSPs is still poorly known, due to the lack of laboratory measurements at the mesospheric condition.

The MSPs are generated by meteor ablation in the upper mesosphere and lower thermosphere, with the radius ranging from sub-nanometer to nanometer size. The 2-D simulations involving the middle-atmospheric circulation revealed that the MSPs will move upward along with the strong updrafts in the summer mesosphere and are then transported to winter mesosphere by the meridional winds and finally sink down into the stratosphere by the downwelling (Megner et al., 2008a, b). The global mass redistribution of MSPs results in a pronounced reduction in MSP concentration and lifetime in the summer mesosphere, and thus the conventional idea of nucleation on MSPs is challenged.

The above dilemma can be resolved when the charged MSPs are taken into consideration because the MSPs' charge can effectively reduce the critical radius of ice nuclei at low temperature, allowing the charged MSPs to act as ice nuclei (Gumbel and Megner, 2009; Megner and Gumbel, 2009). It should be noted that the galactic cosmic rays can generate continuous ions throughout the atmosphere, and the charged molecular clusters are found to grow much faster than neutral clusters. The so-called ion-mediated nucleation (IMN) is of great important for the formation of cloud condensation nuclei in the atmosphere and has been studied for decades (Yu and Turco, 2000; Yu et al., 2008). The distribution of charges on MSPs becomes important with regards to the above assumption, while the efficiency of MSPs collecting electrons in the mesosphere is still unclear. Because the mobility of electrons is much greater than that of positive ions, negative charging is supposed to be dominant in the upper mesosphere, and rocket-borne measurements show that about 10 % of MSPs are negatively charged (Plane et al., 2014; Robertson et al., 2014). The NLCs are located in the D-region ionosphere, where the electric environment is sensitive to disturbances from solar winds. This provides a possible way through which solar activity may impact the NLCs through an electrically related mechanism.

The Cloud Imaging and Particle Size (CIPS) experiment onboard the Aeronomy of Ice in the Mesosphere (AIM) satellite began observing the NLCs in 2007, and 20-summer-season data in the SH and NH from 2007 to 2017 are available now. Therefore, we investigated the hypothetical IMF By-driven solar–NLC link in this study. The paper is structured as follows: Sect. 2 provides a brief description of the CIPS data and solar wind data. Section 3 presents the results of NLC correlation with IMF By during the 20 NLC seasons on the day-to-day scale, as well as the superposed epoch analysis for NLCs' response to IMF By reversals. Section 4 discusses the results, and Sect. 5 summarizes our main conclusions.

The AIM satellite was launched on 25 April 2007 to a sun-synchronous polar orbit whose local time is mainly midday–midnight at high-latitude regions. The CIPS experiment onboard AIM comprises a panoramic UV nadir imager, consisting of four cameras operating at 265 nm, with a field of view of 120∘×80∘ and a horizontal spatial resolution of 5km×5 km. This platform observes the scattered radiance from NLCs and images the NLCs of the ∼40–85∘ latitude zone for the summer hemisphere ∼15 times per day. The CIPS has provided NLC data from the 2007 summer season until now, in terms of ice particle radius, albedo and ice water content (IWC), and detailed descriptions of the CIPS data products, calibration, retrieval algorithms, and retrieval uncertainties have been published (Carstens et al., 2013; Lumpe et al., 2013). The CIPS level 2 orbit data provide rectangular images of NLC properties for each of the 15 orbit strips per day, in which a single pixel represents a 25 km2 (5km×5 km) area anywhere on the globe, and a 5800km×1000 km strip region is covered; thus the cloud cover as well as the frequency of occurrence (FO) of NLCs can be obtained by counting the number of pixels showing them in the images. This study applied the version 5.20 CIPS polar mesospheric cloud level 2 data to investigate the response of NLCs to solar variations during 10 NLC seasons (from 2007 to 2016) in the NH and 10 NLC seasons (from 2007/2008 to 2016/2017) in the SH.

The solar wind By data in geocentric solar magnetospheric (GSM) format were downloaded online from the GSFC/SPDF OMNI web interface (https://omniweb.gsfc.nasa.gov/form/dx1.html, last access: 25 FEbruary 2022). In the GSM coordinate system, the origin is located at the center of the earth; X points towards the sun; Z lies in the plane of the X and geomagnetic dipole and is perpendicular to X (roughly northward); and Y completes the right-hand coordinate system, stretching toward the dusk. The solar wind structures are fairly complex, varying from two-sector to four-sector and sometimes irregularly; therefore, during a 27 d solar rotation period, the IMF By can reverse two or four or more times, unlike other solar indexes such as Ly-α or F10.7, which show a regular 27 d period. In order to apply the widely used superposed epoch analysis (SEA) method, the key days of By reversals are listed in Table 1, which have been selected to ensure that during the 5 d period before and after the key day there are no NLC data missed and that the direction of IMF By is relatively stable. The IMF By changing from positive to negative (from negative to positive) is denoted by p2n (n2p). Four groups of dates during 2007 and 2017 are listed in Table 1, corresponding to the n2p (28 cases) and p2n (29 cases) reversals during NH summer and the n2p (23 cases) and p2n (18 cases) reversals during SH summer, respectively.

Figure 1 shows the variations in the daily averaged solar wind magnetic field and NLC properties during the NLC seasons from 2007 through 2017. The daily averaged IMF By varies between -5 and 5 nT, as shown in Fig. 1a and b, and the periods of IMF By variations are complex, as noted above. Figure 1c–h show the intensity of NLCs in terms of mean ice particle radius (rm), mean albedo (Albm) and mean ice water content (IWCm), while Fig. 1i and j show the cloud cover of NLCs, as calculated by counting of pixels, which is linearly proportional to FO. In order to diminish noise, the NLC data in the latitude bands 65–85∘ are used because the NLCs are rarely observed by CIPS below 65∘ latitude, and an albedo threshold of 5×10-6 sr-1 was applied. The intensity and coverage of NLC peak ∼20 d after the solstice and show strong seasonal variations, with the exception of the mean ice particle radius, rm.

Figure 2 (left) shows the relationship between daily IMF By with the NLC intensity and covers the anomaly in the 65–85∘ latitude zone for the 2008/2009 season in the SH, with the anomaly obtained by removing the 40 d running mean. The corresponding correlation coefficients are present in the right panel, indicating a clear positive relationship between the NLC brightness and the By component of solar wind magnetic fields in the SH. Figure 3 further shows the correlation coefficients of NLC intensity and coverage anomaly in the 65–85∘ latitude zone with IMF By for each of the 20 summertime seasons, from 2007 to 2016 in the NH and from 2007/2008 to 2016/2017 in the SH. To remove the seasonal variation, the solar signals are extracted by subtracting the 40 d smoothed CIPS data. To avoid the no-cloud days, only the CIPS data during the period of 10 d before and 50 d after the solstice day are used (Fig. 1). The link between the anomalous mean ice particle radius rm with IMF By is conspicuous, with positive correlation coefficients in all of the SH summer seasons and negative correlations in most of the NH summer seasons (Fig. 3a). These opposite responses in the SH and NH are consistent with the opposite ionospheric potential changes in the SH and NH caused by IMF By. Further, the response was stronger in the SH, with the correlation coefficient being about twice that in the NH. In NLCs, the larger the ice particle size is, the greater the albedo and IWC are; namely, the mean ice particle radius is normally positively correlated with the albedo and IWC (Lumpe et al., 2013), and the 20-season CIPS data show a correlation coefficient of ∼0.52 between rm and Albm and of ∼0.35 for rm and IWCm. Conversely, the cloud cover of NLCs will also change in pace with the formation and growth process of ice particle radius, and the 20-season CIPS data also show a correlation coefficient of ∼0.48 between rm and FO. It is thus reasonable to speculate that the albedo, IWC and FO will respond to IMF By in concert with ice particle radius, and Fig. 3b–d show that the correlation coefficients between the anomaly of Albm, IWCm and FO with IMF By are pronounced in the SH, but not in the NH.

We have also tried to roughly estimate the column number density of ice particles, Nice, based on the CIPS data of IWC and ice particle radius r. Assuming the mass of ice particle mice to be ρice4πr3/3, where ρice=0.92 gcm-3, the ice particle concentration Nice will be approximately equal to the ice water content divided by the mass of the ice particle, IWC/mice. It is of great interest to study the correlation of ice particle concentration with IMF By, since it can reveal the microphysical process during the NLCs' responses to solar wind magnetic fields. The results show that the correlation coefficient between ice particle concentration with IMF By is -0.14 ± 0.06 in the SH and 0.09 ± 0.04 in the NH, which are surprisingly opposite to that of rm and IWCm shown in Fig. 3. In the dry NLC region, ice particles compete for the limited water vapor, resulting in an anticorrelation between the ice particle concentration and ice particle radius, which have been verified by observation and simulation (Hervig et al., 2009; Wilms et al., 2016). Our above results support this anticorrelation again, implying that the solar wind may firstly increase or decrease the nucleate rate and ice particle number density in NLCs, then decrease or increase the ice particle radius.

NLCs are dominantly influenced by the solar tides with the diurnal variation, and the NLC occurrences are usually more frequent at the local time of morning (Fiedler and Baumgarten, 2018; Stevens et al., 2017). In addition, the NLCs can also be affected by the lunar tides, and the longitudinal variations in NLCs attributed to the non-migrating lunar tides have been found (Liu et al., 2016; von Savigny et al., 2017). To check whether the local time differences between the descending and ascending branches of the AIM satellite will affect the results in Fig. 3, we separate the CIPS data of the descending and ascending branches into two groups. Similarly, in order to check the longitudinal variations, the CIPS data are divided into two groups in terms of the longitude ranges of (-180∘,0∘) and (0∘,180∘). The correlation coefficients for the above two scenarios have been calculated and listed in Table 2, and the results for all of them are consistent with the results shown in Fig. 3. In summary, the correlation coefficients are found to not be affected by the local time variations and longitudinal variations in the CIPS data caused by the tide effects. This further proves that our results are robust.

Furthermore, Fig. 4 shows the mean correlation coefficients for time lags varying from -7 to 7 d. The error bars illustrate the standard deviation of the mean, which is calculated from the 10 seasonal correlation coefficients and is also given in Fig. 3 at 0 d lag time. A very short delay time was observed (Fig. 4), with the maximum correlations occurring near 0 d, implying a microphysical response in NLCs to IMF By similar to the short delay time that has also been observed in the solar–troposphere studies. In previous studies of the link between Ly-α and NLCs, the proposed mechanisms involving photodissociation, heating or circulation all required longer time. The photodissociation process accounts for a negative correlation for the H2O in the mesosphere and the 27 d solar irradiance variations, with a phase lag of about 6–7 d, which can be attributed to the lifetime of H2O at those altitudes (Shapiro et al., 2012). Satellite observations showed the time lag for the water response 27 d solar rotation of about 0–3 d and for the temperature response of about 0–8 d, depending on altitudes, and the time lag between NLC property variations and solar Ly-α ranges from 0 to 3 d in the NH and from 6 to 7 d in the SH, depending on instruments and properties (Thomas et al., 2015; Thurairajah et al., 2017). In contrast, the IMF By-related processes of ionospheric potential changes respond quickly to solar wind magnetic field reversals. In summary, the nearly zero lag time of NLC properties responding to IMF By variations implies a mechanism of electrodynamic origin rather than thermodynamic origin.

In order to further verify the response of NLCs to solar wind at different latitudes, the approaches in Fig. 3 were repeated for the five latitude bands of 80–85, 75–80, 70–75, 65–70 and 60–65∘, respectively. The correlation coefficients of the anomaly of NLC properties with IMF By are shown in Fig. 5, and the slope of the anomaly of NLC properties versus IMF By is given in Fig. 6. Figures 5a and 6a show that in the SH, the correlation and sensitivity of ice particle radius rm to IMF By are both greater at higher latitudes, in agreement with the ionospheric potential perturbations caused by IMF By changes, while in the NH the correlation and sensitivity are just about half of that in the SH but still significant at latitudes higher than 65∘. For the 60–65∘ latitude region, the results are not significant; this may be because at lower latitudes the IMF By-induced processes are too weak and because the rare NLC occurrences at lower latitudes entail weaker signal-to-noise ratio. Figures 5b–d and 6b–d show that the responses of the anomaly of Albm, IWCm and FO to IMF By are noticeable for high latitudes in the SH and obvious for Albm only at latitudes higher than 75∘ in the NH but are not obvious for IWCm and FO in the NH. Dividing the slope given in Fig. 6 by the yearly averaged NLC properties in 65–85∘ latitude bands, the relative slope can be obtained: (0.71 % ± 0.16 %) nT-1 in the SH and (-0.28 % ± 0.08 %) nT-1 in the NH for rm, (1.36 % ± 0.59 %) nT-1 in the SH and (-0.52 % ± 0.32 %) nT-1 in the NH for Albm, (0.74 % ± 0.48 %) nT-1 in the SH and (-0.26 % ± 0.28 %) nT-1 in the NH for IWCm and (2.28 % ± 1.73 %) nT-1 in the SH and (-0.38 % ± 0.60 %) nT-1 in the NH for FO. In consideration of the ∼5 nT amplitude of IMF By variation during solar wind reversals, the responses of NLC intensity and coverage to IMF By are not negligible. The correlation coefficient of ice particle column number density Nice with IMF By can also be obtained for different latitudes varying from 85 to 60∘: -0.14 ± 0.06, -0.13 ± 0.05, -0.09 ± 0.03, -0.03 ± 0.04 and -0.004 ± 0.07 in the SH and 0.06 ± 0.05, 0.09 ± 0.05, 0.12 ± 0.04, 0.04 ± 0.04 and 0.01 ± 0.04 in the NH. Again, the correlation coefficient of ice particle concentration with solar wind magnetic field is opposite to that of mean ice particle radius and ice water content. However, it should be noted that due to the detection threshold of the CIPS instrument for ice particles with radii greater than 10–15 nm, the variation in the invisible smaller ice particles' concentration is unknown.

In addition, other solar wind parameters such as IMF Bz, Ap index and Ly-α irradiance have also been examined by the same processes; however, no correlations were found for them at 0 d lag time. The solar wind magnetic field line has an Archimedes spiral pattern (i.e., IMF Bx is negatively proportional to IMF By), and a correlation coefficient of about -0.67 between them was obtained during the period of 2007 to 2017; thus similar correlations also exist between IMF Bx and NLC properties, but with the opposite sign. The IMF Bz corresponds to a dawn–dusk solar wind electric field and thus can generate a dawn–dusk ionospheric potential drop for both hemispheres, while the sun-synchronous orbit of AIM is designed to be midday–midnight with a rare opportunity to pass the dawn–dusk regions; thus the zero correlations with IMF Bz observed for NLCs are just as expected.

The superposed epoch analysis is frequently applied in the studies of atmospheric responses to short-term solar variations, in which solar signals are more obvious and easier to be extracted than for decadal or longer-term variations. Although the NLCs only occur in summer, during the 20-season period of CIPS data enough IMF By reversal cases have been accumulated, as listed in Table 1, allowing the SEA method to be used to explore the NLCs' responses. In the SEA method, the ice particle radius distribution is denoted by f(r), where the distribution is of the values of r over the array of pixels on a given day. The averages of f(r) during 3 d before and 3 d after the key day are denoted by f3-pre and f3-aft, respectively, then the changes in ice particle radius distribution δf during IMF By reversals are given by δf=f3-aft-f3-pre. The results of δf for the n2p and p2n IMF By reversals in the SH given in Table 1 are illustrated in Fig. 7, with an albedo threshold of 5×10-6 sr-1. The mean ice particle radius rm can be calculated by integrating the product of the radius and its distribution, rm=∑rf(r); thus the changes in rm during IMF By reversals are obtained by δrm=rm,3_aft-rm,3_pre=∑rδf, and the values of δrm are given in each panel of Fig. 7. For n2p and p2n IMF By reversals, the polar ionospheric electric potential will increase or decrease in the SH, and the rm increases (decreases) by about 0.88 nm (1.07 nm) in the SH, as shown in Fig. 7. Similarly, the results of the NH are illustrated in Fig. 8. For n2p and p2n IMF By reversals, the polar ionospheric electric potential will decrease or increase in the NH, and the rm decreases (increases) by about 0.25 nm (0.71 nm) in the NH, as shown in Fig. 8. Generally, the ice particle average radius changes by about 0.73 nm during IMF By reversals, and the responses in the SH are stronger than those in the NH. The results in Figs. 7 and 8 were subject to Monte Carlo sensitivity tests, in which the same number of key days as in Table 1 were randomly generated, and δrm can be calculated by SEA. By repeating this process 1000 times, the distribution of δrm is obtained, showing that the results in Figs. 7 and 8 are significant at the 90 % confidence level.

In addition, we also investigate the responses of NLCs to IMF By reversals for different brightness of noctilucent clouds. The NLCs were placed into five groups according to albedo: 5–10×10-6, 10–15×10-6, 15–20×10-6, 20–25×10-6 and 25–30×10-6 sr-1, respectively. It should be noted that the NLCs with albedo less than 5×10-6 sr-1 are viewed as noise, and the proportion of NLCs with albedo greater than 30×10-6 sr-1 is negligible. Figure 9 shows that for varying NLC albedos, the particle radius rm changes during IMF By reversals are consistent with the results in Figs. 7 and 8, verifying that both the dark and the light NLCs are sensitive to IMF By reversals. On the other hand, the NLCs with greater albedo usually have greater mean ice particle radius; thus the results in Fig. 9 also indicate that both the small and large ice particles are sensitive to IMF By reversals. In addition, the results in Fig. 9 also support the idea that the responses of NLCs to IMF By are stronger in the SH than that in the NH.