Upwelling and methane oxidation are primary H₂O sources in the upper mesosphere (Lübken et al., 2018; Shi et al., 2023). The rapid decrease of H₂O with altitude in the upper mesosphere (Fig. 1a) is mainly caused by the photolysis of solar ultraviolet (UV) radiation, particularly in the summer polar region under continuous illumination. Figure 1 demonstrates how summer polar upwelling creates concurrently temperature minima and H₂O maxima in the summer hemisphere. These conditions promote PMC formation between 80 and 90 km (Rapp and Thomas, 2006), which subsequently redistributes H₂O through the well-established freeze-drying effect (Hervig et al., 2015; von Zahn and Berger, 2003). This top-down process, involving ice particle nucleation, growth, sedimentation, and sublimation, produces dehydration above PMCs and subsequently hydration below (Hultgren and Gumbel, 2014), with clear H₂O depletion visible above 60–80° S PMCs in Fig. 1a. Siskind et al. (2018) successfully simulated the dehydration and hydration above and below PMCs, which extend to midlatitudes due to meridional circulations (Fig. 1 therein). Current PMC models systematically overestimate the dehydration/hydration magnitudes (Bardeen et al., 2010; Lübken et al., 2009), suggesting an incomplete understanding of PMC formation. PMC variability is modulated by various dynamical processes, including gravity waves (Gao et al., 2018), planetary waves (Liu et al., 2015), tides (Fiedler and Baumgarten, 2018), and inter-hemispheric coupling (Gumbel and Karlsson, 2011). While solar UV theoretically influences H₂O and thus PMCs (Remsberg et al., 2018), the 11-year solar signal is insignificant over the past two decades (Hervig et al., 2019; Vellalassery et al., 2023), highlighting the need to identify alternative drivers.

In the upper mesosphere, the O concentrations increase sharply with altitude due to solar photolysis of O₂ (Mlynczak et al., 2013), and the three-body recombination reaction of O produces a secondary ozone layer near 90 km (O+O2+M→O3+M) (Smith et al., 2013). The O exhibits a long photochemical lifetime near the mesopause (days to months), consequently, its daily and seasonal variability is governed by vertical transport processes, including tides and downwelling of the mean circulation (Smith et al., 2010b). O₃ is primarily destroyed by photolysis (O3+hν→O(¹D,³P)+O₂) during daylight with ∼ 2 min lifetime, and by reaction with hydrogen (H+O3→OH+O2) and oxygen (O+O3→2O₂) at night (Huang et al., 2008; Smith and Marsh, 2005). The ozone photochemical equilibrium assumption, i.e., ozone loss is balanced by ozone production, is crucial for retrieving O and H concentrations (Kulikov et al., 2018; Mlynczak et al., 2007). The secondary ozone layer exhibits strong seasonal dependence, with peak O₃ mixing ratios occurring in winter (Fig. 1b) when reduced sunlight minimizes photolysis and lower H₂O limits H availability. Temperature further modulates ozone kinetics, with colder conditions enhancing O₃ by accelerating production rate and inhibiting loss rates (Smith et al., 2018).

Despite these advances in understanding ozone chemistry, key questions remain regarding ozone variability drivers. Solar activity appears to suppress ozone via production of odd hydrogen and odd nitrogen species (Jia et al., 2024), with solar cycle signals and long-term trends occasionally documented (Huang et al., 2014; Lee and Wu, 2020). PMCs may enhance ozone through dehydration, which is expected to reduce the H above PMCs (Siskind et al., 2008, 2018; Siskind and Stevens, 2006). The secondary ozone layer impacts mesospheric energy budgets through competing processes, including solar radiative heating, chemical heating, and infrared cooling (Ramesh et al., 2015), and the total (radiative and chemical) heating of O and O₃ is of great importance for the energy budget of upper mesosphere (Mlynczak et al., 2018; Mlynczak and Solomon, 1993).

Mesospheric temperatures display complex variability patterns across multiple timescales. Interannual variations (4–5 K), outweigh the 11-year solar signal (1–3 K/100 solar flux unit) and the long-term cooling trend (-1 to -2 K/decade) (French et al., 2020a; French et al., 2020b). In the summer polar mesopause, solar signals are unexpectedly absent in temperature, presumably due to a compensating dynamical cooling effect (Karlsson and Kuilman, 2018; Qian et al., 2019). A well-established CO₂-induced cooling trend prevails throughout most of the middle atmosphere, with a rate exceeding tropospheric warming magnitudes (Feofilov and Kutepov, 2012; Laštovička, 2017; Roble and Dickinson, 1989). This strong cooling is regarded as a potential indicator of climate change (Liu et al., 2024; Solomon et al., 2018). However, the summer polar mesopause presents a notable exception, exhibiting an unexpected warming trend that contradicts the dominant CO₂ cooling. Proposed explanations include shrinking effect (Bailey et al., 2021; Dawkins et al., 2023; Lübken et al., 2013) and long-period vertical oscillations of temperature profile (Kalicinsky et al., 2018; Offermann et al., 2021), though the primary mechanism remains elusive. Stratospheric ozone recovery further complicates this picture by modulating gravity wave propagation, which subsequently affects mesospheric winds and temperatures (Smith et al., 2010a; Venkateswara Rao et al., 2015).

These interconnected physical and chemical processes create substantial challenges for a comprehensive understanding of the climate in upper mesosphere. This paper elucidates a systematic bottom-up mechanism, linking summer polar upwelling to interannual variability in H₂O, H, O₃, O, and temperatures. Section 2 details the satellite datasets (MLS, SABER, SOFIE) and theoretical framework, Sect. 3 quantifies the hemispheric climate patterns during solstice months (December/June), Sect. 4 discusses variability mechanism, and Sect. 5 summarizes key conclusions.

This study utilizes measurements from three satellite instruments to investigate upper mesospheric climate variability. The Microwave Limb Sounder (MLS) aboard NASA's Aura satellite, launched in July 2004, provides global atmospheric measurements between 82° N and 82° S (Jiang et al., 2007). Operating in a sun-synchronous orbit with equatorial crossings at 01:30 (ascending) and 13:30 (descending) local time, MLS delivers vertical profiles of H₂O (0.001 hPa top level recommended, 7–9 km vertical resolution, 30 % precision), O₃ (0.001 hPa, 5–7 km, 35 %), and temperature (0.00046 hPa, 6–12 km, 3 K). Daily mean values were generated by averaging ascending and descending orbits, focusing on zonal (5° gridded) and monthly means for interannual variability studies. Figure 1 shows the December climatology of H₂O, O₃, and temperature observed by MLS.

The Sounding of the Atmosphere using Broadband Emission Radiometry (SABER) instrument aboard the Thermosphere, Ionosphere, Mesosphere Energetics, and Dynamics (TIMED) satellite was launched in December 2001. With an orbital inclination of 74.1°, SABER's limb scan provides latitude coverage alternating between 83° N to 52° S (during June) and 52 to 83° S (during December) due to its 60 d yaw cycle (Russell et al., 1999). This 60 d yaw cycle allows interannual comparison by maintaining stable latitude and local time coverage for a given month across different years. We used TIMED/SABER version 2.07 data between 0.01 and 0.003 hPa from 2002 to 2019, including temperature (Remsberg et al., 2008), ozone (Rong et al., 2009; Smith et al., 2013), atomic oxygen (Mlynczak et al., 2013, 2018), and atomic hydrogen (Mlynczak et al., 2014).

The Solar Occultation for Ice Experiment (SOFIE) instrument onboard the Aeronomy of Ice in the Mesosphere (AIM) satellite was launched on 25 April 2007 into a sun-synchronous polar orbit (Russell III et al., 2009). SOFIE covers latitudes between 65 and 82°, with particular focus on ∼ 70° latitude during PMC seasons. Using solar occultation measurements, SOFIE obtains vertical profiles of PMC properties, temperature, water vapor, and ozone (Gordley et al., 2009; Hervig et al., 2009). SOFIE provides datasets for 8 PMC seasons (from 2007 to 2014) in the northern hemisphere (NH) and 7 PMC seasons (from 2007/2008 to 2013/2014) in the southern hemisphere (SH), after which SOFIE measurements shifted to lower latitudes where PMCs typically do not form.

Our methodology examines the responses of mesospheric variables (H₂O, H, O₃, O, T90) to upwelling on interannual timescales. Since direct observations of summer polar upwelling are unavailable, we use temperature at ∼ 80 km (T80_S) as a proxy, based on its relationship with adiabatic cooling. Figure 2 illustrates the conceptual framework for how summer polar upwelling drives interannual climate variability during solstice months through the interconnected pathways in the “upwelling–H₂O(H)–O₃(O)–T90” chain:

H₂O/H variability. Upwelling produces both hydration that occurs below PMCs through direct transport of H₂O by upwelling, and dehydration above PMCs through a “cold-trap effect”. Explicitly, adiabatic cooling from upwelling lowers T80_S promotes water vapor condensation into ice particles, thereby reduces upward H₂O transport and causes dehydration above PMCs (Zhang et al., 2025a). While similar to the conventional freeze-drying effect, the cold-trap effect shows distinct characteristics that we discuss later in Sect. 4.1. Meridional winds then transport both the hydration and dehydration toward the winter hemisphere. Atomic hydrogen H, produced through H₂O photolysis, consequently varies in synchrony with H₂O in both hemispheres.

Our analysis of MLS/Aura data reveals distinct interannual variability patterns during December (Fig. 3). Using T80_S as an upwelling proxy, we observe opposing H₂O responses between hemispheres: dehydration in the summer polar mesopause (75–82° S) indicated by positive correlations with T80_S (R=0.93, Fig. 3a), and hydration at low latitudes (15–20° N) in the NH indicated by negative correlations (R=-0.84, Fig. 3e). The H₂O patterns drive corresponding O₃ variations through chemical interaction, with T80_S showing negative correlations (R=-0.91) with O₃ at 0.001 hPa (65–75° S, Fig. 3b) but positive correlations (R=0.94) at 0.002 hPa (15–20° N, Fig. 3f). Temperature responses exhibit clear hemispheric anti-symmetry, with T90_S at 0.0046 hPa (65–75° S) negatively correlated (R=-0.92, Fig. 3d) and T90_W at 0.001 hPa (15–20° N) positive correlated (R=0.80, Fig. 3h) with T80_S. In addition, the standard deviations of these variables significantly exceed their linear trends, potentially masking long-term signals.

Figure 4 shows the latitudinal extent of these relationships, with sensitivities derived from linear regression. Meridional transport extends dehydration signature to 35° S (beyond the PMC coverage) and hydration to 35° N (Fig. 4a), while O₃ responses show comparable latitudinal ranges (Fig. 4b). SABER observations corroborate these findings (Fig. 5), though atomic hydrogen data gaps at summer high latitudes (Fig. 5a) prevent complete verification of dehydration effects. The ozone photochemical equilibrium assumption holds well, with atomic oxygen O varying in pace with O₃ (Fig. 5b, c). Mlynczak et al. (2018) showed that the total (radiative and chemical) heating rate of O and O₃ is ∼ 10 K d⁻¹ on global and annual scale (Fig. 3 therein). The relative variations of O and O₃ to T80_S in Fig. 5 are both ∼ 4 % K⁻¹, therefore the sensitivity of O/O₃ total heating rate to T80_S is roughly estimated to be ∼ 0.4 K d⁻¹ K⁻¹, which partly explain the ∼ 1 K K⁻¹ (Fig. 5d) response of T90 to T80_S.

Daily-scale SOFIE observations provide further insights (Figs. 6–8). Figure 6 shows that O₃ at 90 km is bottom-up controlled by the T80_S at 79 km. It is noteworthy that in November, during which PMCs are weak and dehydration is absent, the negative correlation between O₃ and T80_S remains significant, possibly due to the temperature-dependent ozone kinetics. Summer polar O₃ near 0.01 hPa (Fig. 5b) or 80 km (Fig. 7c) is negatively correlated to T80_S, which should be attributed to ozone kinetics. As shown in Fig. 8b, O₃ below 90 km lags T80_S by zero days possibly through ozone kinetics, while O₃ above 90 km lags T80_S by ∼ 3 d, suggesting combined kinetic and dehydration influences. However, for upper-level O₃, the relative contribution of ozone kinetics and dehydration (depletion of H) is unclear.

Figure 7a, b shows that the H₂O above and below PMCs is both bottom-up controlled by T80_S rather than the temperature above PMC, which supports the cold-trap effect. Figure 8a further shows that the hydration and dehydration both lag behind T80_S by about zero days, consistent with the cold-trap effect. In contrast, according to the conventional freeze-drying effect, the dehydration tends to occur prior to the hydration, with the time interval roughly equal to the lifetime of ice particle which is up to two days in simulations (not consider transport and diffusion of ice particles). In addition, T90 is positively correlated to O₃ (Fig. 7c) and negatively correlated to T80_S (Fig. 7d), with lag time of about ∼ 3 d, matching the radiative and chemical heating of O₃/O.

The climate patterns in June (Figs. 9–12) mirror those of December, but with reduced amplitudes. The T80_S variability is weaker (1.5 K vs. 5.1 K in December), yielding smaller constituent variations and lower significance (Figs. 3, 9). SABER H responses are insignificant (Fig. 11a), while T90_S correlations weaken or disappear (Figs. 11d, 12d). MLS data show a localized hydration signal at 0.001 hPa between 70–80° N (Fig. 10a), possibly from PMC sublimation due to higher T90_S.

Interhemispheric transports are more pronounced in June than in December due to differences in gravity wave forcing of the circulation (Smith et al., 2011). The enhanced meridional transport extends H₂O influences to 40–50° S, affecting O₃ and T90_W in the winter high-latitudes (Fig. 10). Downward winds in the winter polar region generally result in the lowest H₂O levels near solstice, although vertical winds and diffusion can increase winter polar H₂O above 90 km (Lossow et al., 2009). Whether the summer polar upwelling in June could influence the winter high-latitude climate through interhemispheric H₂O transport is an interesting question.

Our results demonstrate it is the summer polar upwelling that fundamentally drives the interannual variability of both hydration and dehydration through two distinct pathways: (1) direct upward transport of H₂O by upwelling combined with meridional wind transport creates hydration in the winter hemisphere; and (2) adiabatic cooling associated with upwelling enhances ice particle growth that blocks upward H₂O transport, causing dehydration above PMC in the summer hemisphere.

The cold-trap effect differs from the conventional freeze-drying effect in several ways. While freeze-drying effect describes a top-down process dominated by ice particle sedimentation, the cold-trap effect represents a bottom-up process driven primarily by upwelling dynamics. Sedimentation of ice particles plays a dominant role in the freeze-drying effect, however, it is unnecessary in the cold-trap effect (Zhang et al., 2025a). Importantly, the cold-trap effect produces simultaneous hydration and dehydration. In contrast, the freeze-drying effect involves dehydration prior to hydration, with a time lag (estimated as the ice particle lifetime) of up to two days. Particularly, if PMCs are weak or absent and ice particles could not block upward H₂O transport, hydration could even occur in the absence of dehydration, inconsistent with the freeze-drying effect.

It should be emphasized that we are not attempting to demonstrate that the freeze-drying effect is incorrect. Instead, the two mechanisms should be complementary rather than contradictory. The freeze-drying effect has been verified by simulations for describing PMC microphysical processes at local scales, while the cold-trap effect better explains the global-scale H₂O redistribution patterns, especially the H₂O in the winter hemisphere. Additionally, the parameterization of freeze-drying effect is unavailable until now, due to the complexity in PMC simulations. In contrast, the cold-trap effect is much simpler, the interannual variability of hydration/dehydration is only dependent on the T80_S. The limited H variability detected in SABER data (Figs. 5a, 11a) may reflect both measurement uncertainties in H retrieval and gaps in spatial coverage in high-latitude regions where dehydration effects are strongest.

Upwelling forcing drives distinct hemispheric patterns in O₃ variations. In the winter hemisphere, hydration-driven increases in H abundance lead to enhanced O₃ destruction. In the summer hemisphere, however, O₃ increases via two pathways: reduced H abundance due to dehydration and temperature-dependent ozone kinetics caused by adiabatic cooling.

Our vertical analysis shows that these processes operate at different altitudes (Figs. 5b, 7c). Below 85 km, where dehydration is absent, temperature-dependent ozone kinetics dominate O₃ abundance, while above 85 km both dehydration and ozone kinetics contribute, with their relative contribution remaining unclear. In addition, the lag time of O₃ to T80_S differs for varying altitudes (Fig. 8b), which may be useful for identifying the two pathways.

The anti-phase temperature responses between hemispheres (T90_W and T90_S) reflect the combined effect of dynamical processes and O/O₃ heating. In the winter hemisphere, based on the known radiative and chemical heating rate (∼ 10 K d⁻¹ on global and annual scale) of O/O₃, the 3∼ 4 % K⁻¹ sensitivity of O₃ and O in low latitudes to T80_S (Fig. 5b, c) approximately explains the observed ∼ 0.5 K K⁻¹ sensitivity of T90_W to T80_S (Fig. 5d).

The summer hemisphere presents a more complex scenario involving competing processes. The O₃ concentration in summer polar region is very low due to the destruction by sunlight and H₂O (Fig. 1b), while the O concentrations do not differ much between the summer and winter hemispheres. The high sensitivity of T90_S to O₃ (51.7 K ppmv⁻¹, Fig. 3c) in December is likely unrealistic. O-heating likely dominates the interannual variability of T90_S, with O₃-heating playing a minor role.

Another explanation is that a potential downwelling may result in higher T90_S by adiabatic heating and higher O concentration by downward transports, which further increase O₃. The zonal winds change direction above the mesopause height, and gravity waves that induce upwelling near 80 km tend to cause downwelling at altitudes above the mesopause. In this scenario, the positive correlation between T90_S and O/O₃ results from vertical winds rather than solar or chemical heating. However, Fig. 8c, d shows that the T90_S lags behind O₃ and T80_S by ∼ 3 d, ruling out pure dynamical explanations. Figures 4c and 5d show that the T90_S at latitudes as low as 30° S are negatively correlated with T80_S, which should not be attributed to the adiabatic heating of downwelling. Moreover, shrinking effect may explain the negative correlation between T90_S and T80_S through the vertical shift of the temperature profile. However, the MLS and SABER results are presented on pressure levels rather than geometric heights, not supporting a shrinking effect as the major driver for the interannual variability of T90_S. In addition, T90_S lags behind T80_S by ∼ 3 d (Fig. 8d), inconsistent with the shrinking effect on the daily scale.