Journal of Geophysical Research: Atmospheres

The roles of temperature and water vapor at different stages of the polar mesospheric cloud season



[1] Temperature, or alternatively, saturation vapor pressure (PSAT), dominantly controls the polar mesospheric cloud (PMC) seasonal onset and termination, characterized by a strong anticorrelated relationship between the Solar Occultation for Ice Experiment (SOFIE)-observed PMC frequency and PSAT on intraseasonal time scales. SOFIE is highly sensitive to weak clouds and can obtain a nearly full spectrum of PMCs. Both the SOFIE PMC frequency and PSAT indicate a rapid onset and termination of the season. Compared to PSAT, the water vapor partial pressure (PH2O) exhibits only a slight increase from before to after the start of the season. We are able to use the PSAT daily minimum and two averaged PH2O levels taken before and after the solstice, respectively, to estimate the start and end days of the PMC season within 1–2 days uncertainty. SOFIE ice mass density and its relationship to PH2O and PSAT are examined on intraseasonal scales and for two extreme conditions, i.e., strong and weak cloud cases. In the strong cloud case, such as those bright clouds that occur during the core of the season, PH2O far exceeds PSAT and dominantly controls the ice mass density variation, while in the weak cloud case, such as those clouds that occur at the start and end of the season, PH2O and PSAThave comparable magnitudes, vary in concert, and have similar effects on the ice mass density variation. These results suggest that the long-term brightness trends reported by DeLand et al. (2007) are primarily driven by changes in water vapor (H2O), not temperature.

1. Introduction

[2] Polar mesospheric clouds (PMCs), also called noctilucent clouds (NLCs) when observed from the ground, form under the prevailing conditions of a cold mesopause (e.g., <150 K) and enhanced mesospheric water vapor (H2O) in the high-latitude summer (poleward of 60°) [e.g.,Garcia and Solomon, 1985]. Accordingly a low saturation vapor pressure (e.g., PSAT < 1.0 × 10−8 hPa) and relatively high water vapor partial pressure (PH2O > 1.0 × 10−8 hPa) coexist. When summer starts, PSAT, which is predominantly dependent on e−1/T, drops rapidly as T decreases in the summer mesosphere. Meanwhile, PH2O experiences a moderate increase that is caused by the upward transport of H2O from the wetter lower atmosphere. Both occurrences contribute to achieve a supersaturated state (S > 1, where S = PH2O/PSAT) and are linked to a global scale mesospheric residual circulation that has an upwelling branch in the polar summer mesosphere [Garcia and Solomon, 1985]. As a prominent seasonal phenomenon, PMC variability on a series of time scales, for example, hourly to daily, intraseasonal, interannual, and decadal, has attracted intense research interest over the years. It has been difficult to readily obtain PMC variability on all desired time scales because historically the PMC/NLC measurements lack temporal continuity and also are sparse in spatial coverage. Despite such a limitation, using a collection of NLC observations since 1964 in northwest Europe (∼54–61°N), Gadsden [1998a]performed a comprehensive study on the secular change of NLC brightness, frequency, southern edge, seasonal length, and preferred local time of appearance. It was concluded that after removing the solar cycle modulation there is an upward trend in the NLC frequency, while other aspects of the morphology remained fairly constant over the years. For example, there is no apparent brightness increase, and the length of the season also remained unchanged over the years. However, there are other studies that yield different conclusions on the long-term trend of the NLC frequency. For example,Kirkwood and Stebel [2003] analyzed the NLC appearance frequency in part of northern Europe and did not find any notable trend for the last 40 years [also see Thomas, 2003]. Consistent with this finding, an analysis of the NLCs in Moscow for the last 40 years also indicated that there was no apparent long-term trend in the cloud frequency [Romejko et al., 2003]. Results of Romejko et al. [2003] did however imply a slight upward trend in the cloud brightness. Using the SBUV series of satellites data sets, Shettle et al. [2009] found an upward trend in the PMC frequency for the last 30 years since 1979, which is consistent with the SBUV cloud brightness trend found by DeLand et al. [2007]. Determination of the cloud frequency or brightness trends can be affected by a number of factors such as, data sampling techniques, local time (LT), latitude, and the brightness threshold used in the cloud detection [Stevens et al., 2007]. For example, the conclusions drawn from analyzing the SBUV data sets may have been affected by the fact that the SBUV instruments only detect bright clouds. To diagnose whether the long-term trend or any other PMC variability is properly determined, we must understand the mechanisms that control the cloud frequency and brightness.

[3] It is well known that temperature and H2O are two key factors that control the PMC formation and variation, and this subject has been extensively studied through model simulations. Within a well-established theoretical framework a number of PMC models were proposed to study how temperature, H2O, and other factors such as nucleation and dynamics control PMC formation and variation [e.g., Jensen and Thomas, 1988; Gadsden, 1998b; von Zahn and Berger, 2003; Lübken et al., 2007; Hervig et al., 2009b], but unequivocal observational evidence has, up to now, been lacking. The SOFIE data set can serve to provide such evidence and to better define our understanding of the roles of temperature and H2O in controlling PMCs.

[4] The Solar Occultation for Ice Experiment (SOFIE) aboard the Aeronomy of Ice in the Mesosphere (AIM) satellite (2007–present) [Russell et al., 2009] measures PMCs, temperature, and H2O simultaneously on fine vertical grids, which provides a better opportunity to clarify how temperature and H2O control PMCs than existing and past satellites. First, SOFIE can detect weak clouds, including the faintest ice layers [Hervig et al., 2009a]. This capability is essential in studying the existence of PMCs. The high sensitivity of SOFIE is related to its high signal-to-noise ratio (∼106at ∼83 km). Although higher signal-to-noise ratio is a known advantage of the solar occultation technique, the band pairs used in SOFIE further reduces the noise level [Gordley et al., 2009]. Second, SOFIE measurements have high vertical resolution. SOFIE measures all parameters with a ∼2 km vertical resolution throughout its altitude range (∼15–100 km). The high vertical resolution makes the PMCs and the corresponding temperature and H2O more precisely matched so that the correlation between the PMCs and their environmental variables can be readily obtained. Third, SOFIE measurement latitude remains poleward of 65°, and therefore a continuous intraseasonal time series can be obtained in a polar regional averaged sense. Nevertheless, since the SOFIE measurement latitude varies significantly throughout the PMC season we must consider the possible effect of this latitude migration on the results. Last, SOFIE infrared observations are a direct measure of ice mass density. Accordingly, a macrophysical relationship with the environmental variables can be obtained without the necessity of looking into the microphysics.

[5] This paper investigates how temperature and H2O control the PMC existence and strength on intraseasonal scales using the SOFIE measured PMCs, temperature, and H2O, and a 0-D model proposed byHervig et al. [2009b]. Only the Northern Hemisphere (NH) clouds are examined in this study because they exhibit less variability and their controlling mechanisms are presumably less complex than their southern counterpart [e.g., Gumbel and Karlsson, 2011]. We present two main parts of research in this paper. In part one we use the PMC daily occurrence frequency to examine what controls the start and end of the PMC season; in part two we examine what controls the PMC ice mass density variation during the core of the cloud season. In the analysis of the cloud seasonal start and end, the Microwave Limb Sounder (MLS) [Waters et al., 2006] temperature and H2O are also used to support the SOFIE results. The 0-D model results are used in both parts of the analysis to compare with the SOFIE observations. The 0-D model assumes that ice forms as long as the supersaturation ratio (S) is greater than one and that H2O in excess of PSATexists as ice. In the 0-D model the nucleation processes and the effect from the atmospheric flow field are ignored. Although highly simplified, the 0-D model has proven to be effective in revealing PMC variations on intraseasonal scales [Hervig et al., 2009b; Russell et al., 2010]. Instead of directly addressing temperature and H2O we use PSAT and PH2O as intermediate variables to reflect the temperature and H2O in this study. The PSAT and PH2Oare chosen because their relationship with the equilibrium ice mass density is quasi-linear, and because the PSAT variation can be used to effectively interpret the rapid onset and termination of the PMC season.

2. Data Sets and 0-D Model

2.1. SOFIE Level2 Temperature, H2O, and PMC Data Sets

[6] SOFIE is one of the two instruments that are operating aboard the AIM satellite [Russell et al., 2009], and it has been collecting scientific data since 14 May 2007. The publically released level2 data began on 28 May 2007. SOFIE measures the atmospheric limb transmission using eight channels centered between 0.292 μm and 5.316 μm. Each channel consists of two broadband radiometer measurements, one in a strong absorption band, and the other in a spectrally adjacent region of weak absorption [Gordley et al., 2009]. SOFIE level2 products include temperature, O3, H2O, CO2, CH4, and NO. Although the vertical range of SOFIE level2 products is ∼15–100 km, the altitude range with the highest data quality is in the mesosphere. As was mentioned above, the vertical resolution of SOFIE remains at ∼2 km for all of its retrieved products. The horizontal resolution corresponding to the vertical field of view at ∼83 km along the tangent path and about ∼7 km perpendicular to the path. SOFIE temperature is retrieved from the two CO2 channels, channel 4 (2.785 and 2.939 μm) and channel 7 (4.324 and 4.646 μm). The H2O product is retrieved from channel 3 (2.462 μm and 2.618 μm) signals. The level 2 SOFIE H2O throughout the upper stratosphere and mesosphere has been validated by Rong et al. [2010], and has been shown to have high precision and accuracy based on the analysis of its instrument properties and the comparisons with ACE/Sci-Sat1 (Atmospheric Chemistry Experiment) [Bernath et al., 2005] and MLS/Aura measured H2O [Lambert et al., 2007]. The SOFIE H2O random error is ∼0.25–1.0% below ∼85 km, which is the highest among the currently existing mesospheric data sets. The SOFIE H2O systematic error is within ∼3–12% below ∼85 km, as compared to the ∼9–34% systematic error of MLS H2O for the same altitude range. SOFIE H2O in the NH shows overall excellent agreement (∼2–5% mean percent difference) with both ACE and MLS data except for differences caused by the enhancement layer at ∼80–82 km. This layer is formed at the bottom of the PMC region due to recycling of H2O from the ice to vapor form [Summers et al., 2001]. SOFIE detects this feature more distinctly owing to its high vertical resolution. The high resolution also enables a more precise determination of the mesopause and leads to serendipitous findings in the seasonal development of the mesopause region, such as, a double-mesopause detected at the end of the summer (seesection 3.3). A sequence of temperature validation studies were conducted at different stages of data release and a paper describing these results has been written and will be submitted in the near future (M. H. Stevens et al., Validation of upper mesospheric and lower thermospheric temperatures measured by the solar occultation for ice experiment, manuscript in preparation, 2012). These unpublished studies indicate that in the NH polar summer mesosphere for near-coincident locations and timeframes (1° in latitude and 1 h in time) the SOFIE temperature agrees well with Sounding of the Atmosphere using Broadband Emission Radiometer aboard the TIMED satellite (SABER/TIMED) [Russell et al., 1999; Remsberg et al., 2008] and ACE/Sci-Sat1 [Sica et al., 2008] measurements, with the mean differences being ∼2–5 K. Especially at the mesopause there is no cold or warm bias shown in these comparisons based on the coincidences. However, the average of all events north of 65°N suggests a few degrees warmer mesopause region in SOFIE than in SABER. This is because SOFIE did not capture some very low temperatures (<130 K at mesopause) that are present in SABER [Russell et al., 2010]. When compared to the falling sphere measurements [Lübken et al., 1996], the SOFIE mesopause is warmer by ∼15–20 K and lower in height based on the comparisons of SABER and falling sphere climatology shown in the work of Remsberg et al. [2008].

[7] SOFIE channels 2 and 5 are dedicated to PMC measurements. Hervig et al. [2009a] described detailed theoretical frameworks and algorithms for the retrieval of several key PMC variables such as ice mass density, ice particle number density, ice particle axial ratio, effective radius, PMC top and bottom heights, and the height where maximum ice mass density occurs (Zmax, also called cloud peak height hereinafter).

2.2. MLS Level2 Temperature and Water Vapor

[8] MLS/Aura level2 temperature [Schwartz et al., 2008] and H2O [Lambert et al., 2007] are used to conduct parallel analyses to verify the SOFIE results. The MLS temperature and H2O vertical resolutions degrade to 14–16 km in the mesopause region and therefore the mesopause height cannot be precisely determined. In this case simply the coldest mesospheric temperature is taken as the mesopause temperature to compare with the SOFIE results. The coarser vertical resolution of MLS is not a problem in this study because mesopause height is not essential in our analyses.

2.3. SOFIE Latitudes and Local Times

[9] Prior to the main analysis we first address two major considerations that should be taken into account when interpreting the results of the analysis. The first consideration is the SOFIE latitude migration throughout the PMC season shown in Figure 1. From mid-May to late August the SOFIE latitude varies between ∼66° and ∼80°; before and after the summer solstice, SOFIE measurements extend into lower and higher latitudes, respectively. Temperature is known to exhibit a significant latitudinal gradient in the polar summer mesopause region, i.e., a ∼10 K decrease on average from 66° to 80° around the summer solstice [e.g.,Garcia and Solomon, 1985]. This can affect the cloud frequency in the core of the season. However, the temperature gradient is generally smaller at the start and end than in the middle of the summer. In particular, the onset and termination of the PMC season will not be significantly affected. This will be discussed further in a following section. Water vapor (e.g., volume mixing ratio) shows an overall less distinct latitudinal gradient, and therefore less impact is expected from the latitude change.

Figure 1.

SOFIE Northern Hemisphere (NH) latitude coverage during summer and early fall. SOFIE latitude coverage repeats every 6 months. The horizontal axis is days from summer solstice (DFS).

[10] Local time (LT) variation is another consideration. SOFIE always measures the NH at the local sunset time (2200–2300 LT) while the chosen MLS data points experience a wider range of local time variation, i.e., typically from about 0200 LT to 1300 LT. Stevens et al. [2010]suggested that the polar summer mesosphere temperature at 69°N in June varies with LT and the magnitude reaches ∼8 K around the mesopause. On the basis of their results, SOFIE and MLS LTs are in the warm and cold periods of the diurnal cycle, respectively, and this could lead to a temperature bias between the two data sets and accordingly could have some effect on the 0-D model determined start and end of the PMC season.

2.4. The 0-D Model

[11] The 0-D model [Hervig et al., 2009b] assumes that all H2O in excess of saturation is instantaneously transferred into the ice phase. Although highly simplified, the 0-D model is a suitable framework to describe PMC variations in a developed stage. More specifically, the 0-D model has been shown to be useful in revealing PMC variations with time scales longer than a day. Many previous modeling studies [e.g.,Jensen and Thomas, 1988; Rapp and Thomas, 2006] have suggested that it takes about a few hours to a day for PMCs to form and grow into a mature stage in which there is a strong freeze-dried region below the clouds and near-steady Zmaxand ice mass density. Since the PMC measurements are mostly taken at the developed stage of the clouds, the results of the 0-D model can agree well with the observations. For example,Russell et al. [2010]has shown that the daily averaged difference of cloud height and mesopause height remains at ∼3.5 km throughout the entire PMC season, and this height difference can be well reproduced by the 0-D model. In the 0-D model a given temperature vertical profile and the corresponding H2O profile are used to produce one ice mass density profile. Since the fall velocity and vertical transport are ignored, the ice mass density peak height is the altitude where the H2O in excess of the saturation value (i.e., S > 1) is the largest, which is generally 3.5 km below the mesopause. This indicates that on a daily scale the fall velocity and vertical transport has little effect on adjusting the peak cloud height up or down. The PSATformula used in the 0-D model is fromMurphy and Koop [2005], written as:

display math

The 0-D ice mass density is written as:

display math

where mice(PH2OPSAT)represents the 0-D ice mass density,Mww = 18.0 g/mol is the molecular weight of H2O, and R = 8.314 J/mol/K.

3. Start and End of the PMC Season

[12] The causes for the start and end of the PMC season is one of the main issues of PMC science that needs to be addressed and in particular it is of great importance in serving as an indicator of mesospheric climate change. This topic was implicitly embedded in the AIM prelaunch microphysics objective [Russell et al., 2009], and with a continuous time series of many key variables provided by SOFIE throughout the summer and especially with SOFIE's ability to detect weak clouds, new light can be shed on this problem. Since the start and end of the PMC season addresses the cloud existence rather than its brightness, we use the PMC “appearance” frequency to attack this problem.

3.1. Rapid Onset and Termination of the PMC Season Observed by SOFIE

[13] Figure 2a shows the daily SOFIE PMC frequencies in the 2007–2010 northern seasons. The cloud frequency here refers to the percentage of PMCs detected within the daily measured 15 events. A striking feature in Figure 2a is that for all 4 years there are three fairly distinct stages for the PMC season, starting period, main period, and ending period. At the two ends the cloud frequency rises to nearly 100% or drops to nearly zero within ∼10 days. It should be noted however that in 2007 the start of the season was not entirely revealed because reliable SOFIE data collection did not begin until after the season started. Nevertheless, the rapid increase of the cloud frequency is partly captured. For all 4 years, after the solstice and toward the end of the season, the cloud frequency is notably declined from 100%, although the speed of the decline shows interannual variability unlike the variation at the start of the season which is similar for all 4 years. The rapid onset and termination of the PMC appearance was reported as early as 1972 by Donahue et al. [1972] but only in the latitude range north of 80°N. Sudden onset and termination of the PMC appearance at high polar latitudes was also discussed by Thomas [1984] and Lübken et al. [1996], but never before SOFIE was this occurrence found at such low polar latitudes as 65–75°N. For most PMC data sets retrieved from satellite measurements, the cloud frequency north of 60°N exhibits a rather gradual start and end of the season [Bailey et al., 2005; DeLand et al., 2006; Petelina et al., 2007; Robert et al., 2009]. This is because when the instrument is not sensitive enough to the weak clouds, the brightness threshold has to be set higher, and in such cases the cloud strength (i.e., brightness or albedo) is involved in determining the cloud existence. This can be clarified by examining the SOFIE frequency intraseasonal variation using larger threshold values, shown in Figure 2b. The SOFIE default threshold is approximately ∼0.15 ng/m3 in terms of ice mass density. As the threshold is increased, the overall frequency is reduced, the seasonal start and end become more gradual, and the frequency variation increasingly resembles the ice mass density variation. This last point can be clarified by comparing Figure 2b and Figure 8 in section 4.1 that discusses the ice mass density intraseasonal variations. Figure 2b clearly indicates that the detection of weak clouds can substantially affect the results.

Figure 2.

(a) SOFIE observed daily PMC frequency for four consecutive NH seasons. (b) SOFIE PMC daily frequency variations based on a series of different threshold ice mass density values, which are 5, 10, 25, 50, and 100 times of the default threshold, i.e., 0.15 ng/m3, respectively.

[14] We next examine the 0-D modeled PMC frequency. For a given pair of SOFIE or MLS temperature and H2O profiles, as long as S > 1 is met for any altitude range, mostly around the coldest point, we claim the existence of a cloud. Figure 3shows the 0-D modeled and the SOFIE observed PMC frequencies. There are two 0-D modeled frequencies; one using SOFIE v1.022 data and the other using MLS v2.2 data. MLS data is chosen daily at the SOFIE latitude ±1.0° range. Excellent qualitative agreement exists between the 0-D modeled and the observed frequencies, i.e., both show rapid onset and termination, and the shorter time scale variations during the main period also agree well. This suggests that the criterion S > 1 is a sufficient condition to reproduce the main characteristics of observed PMC frequency intraseasonal variation. In a quantitative sense, the SOFIE 0-D cloud frequency marks a slightly earlier ending period than the observation but a similar starting period. The MLS 0-D cloud frequency on the other hand marks a significantly earlier starting period than the observation but a similar ending period. This finding is consistent for all years of analyses. Since the 0-D cloud frequency depends on both PSAT and PH2O, the difference in either could have caused the discrepancies. We however can quickly rule out the PH2O for two reasons. First, a validation study has verified overall good agreement between MLS and SOFIE water vapor in the NH mesosphere [Rong et al., 2010]. Second, we will later find that the start or end of the PMC season is not sensitive to a small change in the water vapor partial pressure. It is highly probable that the temperature difference between the two data sets is causing these differences. The more extended cloud season shown in the MLS 0-D frequencies suggests that the mesopause temperature in MLS is systematically lower than in SOFIE. But a striking issue here is not the overall temperature bias, but the fact that the start and end appear asymmetric when compared to the observed PMC frequency. If both SOFIE PMC measurements and the 0-D assumption are valid, it points to a conclusion that SOFIE is biased warm at the end while MLS is biased cold at the start. This, however, may not be the only interpretation. For example,Petelina and Zasetsky [2009] (also see Hervig and Gordley [2010]) argued that when ice is present, the solid-phase temperature, i.e., ice temperature, is more appropriate to describe the thermal state of the PMC region. There is a possibility that toward the end of the season, gas-phase temperature alone is not sufficient to fully describe the thermal state of the cloud region. Although ice temperature cannot be used to predict the start and end of PMC season because its retrieval requires the knowledge of ice existence, when exiting the PMC season the thermal state may have been affected by a long history of ice presence, which could have contributed to the discrepancy at the end. Another noteworthy difference between the 0-D modeled and the observed PMC frequencies is that the latter is generally higher than the former, especially after the solstice when the mesosphere temperature begins to rise. The nearly 100% cloud frequency, which is higher than the maximum ice production (S > 1) allows, suggests that the clouds are ubiquitous and exist at locations where the temperature is warmer than the frost point. Horizontal transport is one possible cause. After the solstice the PMCs are fully developed and become increasingly stronger, and during this time the clouds cover all longitudes. Under such a prevailing condition, horizontal transport can further spread them over a more extended spatial range in the presence of a fairly strong easterly wind of about ∼30–50 m/s at the PMC altitudes.Baumgarten et al. [2011] suggested that the ice particles could behave like a passive tracer for up to an hour and travel several hundred kilometers downstream if no wave structures are involved. Numerical modeling studies are required to test this hypothesis and to pursue other possible causes.

Figure 3.

The 0-D frequencies for SOFIE and MLS and the SOFIE PMC frequency. MLS data points are chosen daily within a 2.0° range that is centered at SOFIE latitudes. A 4-day smoothing is applied to each time series.

3.2. Latitude Dependence of the 0-D Frequency

[15] We have so far examined the cloud frequencies at SOFIE latitudes using both SOFIE and MLS data and compared the timings of rapid increase and decrease at the start and end of the PMC season. One must then wonder how the SOFIE measurement latitude change affects these timings and the 0-D frequency variations in general.

[16] Figure 4shows the MLS 0-D frequency in different latitude bands (65–70°N, 70–75°N, 75–80°N, and 80–82°N). For all 4 years, we find that the 0-D modeled start and end days would have been highly consistent between different latitude bands if <5% cloud presence is chosen as a threshold. This is because the condition S > 1 is simultaneously met in all latitude bands. We have mentioned above that the temperature latitudinal gradient is smaller at the two ends than in the core of the season, and this is especially true for the daily minimum temperature that is used to claim the first 0-D cloud. We also note that in all latitude bands the maximum frequency can reach ∼90–100% although the frequency variation is clearly latitude dependent. In the starting period the 0-D frequencies in all latitude bands are relatively consistent while after the summer solstice toward the end the frequencies on the lower latitudes are more notably declined. Although there is no exact observational evidence to support this latitude dependence of the cloud frequency, it is basically consistent with whatFigure 2a shows, that is, there is more variability at the end than at the start of the season. SOFIE PMC frequency did not show severe decline toward the end of the season because the SOFIE latitude reaches 75–80°N after August 15, i.e., ∼55 days after the solstice. Direct observational evidence does not currently exist because no data set covering the polar cap region has been obtained with the SOFIE sensitivity.

Figure 4.

The 0-D frequencies in the different latitudinal bands calculated from MLS temperature and H2O. A 4-day smoothing is applied to each time series.

3.3. Temperature Controls the Onset and Termination of the PMC Season

[17] In Figure 5 we separate the roles of PSAT and PH2O in determining the start and end of the cloud season. Among the 4 years used in our analysis 2010 is shown as an example. The analysis starts by showing all the PSAT and PH2O values from the chosen events. For SOFIE all 15 events per day are included, while for MLS the events are chosen at the SOFIE latitude ± 1.0° as mentioned above. For each pair of temperature and H2O profiles the PSAT and PH2Ovalues are taken at the altitude where S maximizes. This altitude is very close to the mesopause in the core of the season. The left-hand vertical axis is logarithmic so that we can see the full ∼8–10 order span of PSAT variation from May to September. The blue curves are the daily minimum and median PSATwith a 4-day smoothing applied. The PH2Ovariation, on the other hand, is nearly flat for the same scale, i.e., before the solstice it exhibits a gradual increase until it reaches a near-constant but slightly increasing level and continues after the solstice. The far more rapid change in PSAT than in PH2O at the start and end of the PMC season supports the argument that temperature is in primary control when entering or exiting the PMC season. Knowing that PH2O is less variable, we simplify its development into a stepwise function, jumping from a lower level before the solstice to a higher level after the solstice. Either before or after the solstice, a geometric mean of all the Log(PH2O) values is calculated and a mean PH2O level is obtained accordingly. The two PH2Olevels will be used later to obtain a set of 0-D determined start and end days of the cloud season. In the postsolstice stage the PH2Orange marked by the dashed lines are determined by the 1-σ standard deviation of all Log(PH2O) values. The dashed lines of SOFIE and MLS PH2O mark a very similar scatter, but it is worth mentioning that the overall MLS PH2O magnitude is about half of SOFIE because the mesopause pressure is lower in MLS. This, however, will not qualitatively affect the result because PH2O does not dominantly control the onset and termination of the cloud season. Through comparing SOFIE and MLS, we note that the intraseasonal variations of SOFIE and MLS PSAT and PH2Oare very similar; clearly for both data sets, the coldest day is close to the solstice and the rates of cooling before and the warming after the solstice are very similar. We do, however, note that throughout the PMC season the MLS temperature is on average lower than SOFIE. This temperature difference has been reflected in the 0-D frequencies shown inFigure 3. The daily minima of the two data sets show a large difference, i.e., ∼20 K, suggesting that a significant number of MLS measurements indicate a much colder mesopause than what is measured by SOFIE. The daily median difference between the two data sets however is much smaller, being ∼10 K or less. The daily maxima of the two data sets are close, and some MLS data points show even warmer temperature than observed by SOFIE. Nevertheless one should note that the daily warmest points, which are mostly above the frost point, are not relevant to PMC formation. Measurements that indicate a colder mesopause than in the SOFIE temperature are not unprecedented before MLS. For example, mesopause temperatures measured by the falling sphere at 69°N can be as low as ∼120 K in late July to August [Lübken et al., 1996], pointing to even colder condition than in MLS. The significant discrepancies in the upper mesospheric temperature between falling sphere, MLS, and several other satellite data sets such as ACE, SABER, and SOFIE remain unresolved [e.g., Schwartz et al., 2008]. We have just argued that part of the differences may be attributed to the LT difference. But the LT difference cannot account for the full magnitude of the temperature difference. MLS temperature may actually be biased cold. For example, at the start of the PMC season the SOFIE temperature accuracy is better supported by the SOFIE PMC frequency, while the MLS 0-D frequency indicates a start time for which no PMCs were ever reported. We do, however, note that the 0-D assumption is highly simplified and therefore may have limitations. For example, the seasonal onset may appear earlier in the 0-D model than in the real atmosphere owing to the omission of the nucleation. Further research is required to define the extent of these limitations and to better understand the process of ice particle formation in general.

Figure 5.

The NH intraseasonal variations of PSAT and PH2O on logarithm scales. The dots are for individual events. For each event the pair of data points is chosen at altitude where S maximizes. The blue curves are daily minimum and median PSATtime series obtained using a 4-day smoothing. The stepwise black lines are mean PH2O before and after summer solstice. The dashed lines (after the solstice) bound the PH2Orange that is determined by 1-σ standard deviation of the PH2O values. (a) SOFIE analysis. (b) MLS analysis. The data points are chosen in a 2.0° latitude range that is centered at SOFIE latitudes. Only the analysis for 2010 is shown as an example.

[18] Another noteworthy feature shown in Figure 5a is a bimodal behavior of both PSAT and the corresponding PH2O in September. More specifically, the bimodal behavior refers to the fact that at the end of the summer the data points of PSAT or PH2Oare distributed at two levels, respectively. This occurrence is associated with a double-mesopause feature that appears from late August to September in SOFIE temperature. This is a transitional time period when the lower summer mesopause rises to a higher winter mesopause. During this time two temperature local minima, which are about 10–12 km apart, coexist. The double-mesopause does not necessarily result in a bimodal behavior inFigure 5abecause the criterion of S being maximized naturally selects the lower one when it is cold enough. In September, however, the lower mesopause warms up considerably and the chosen data point jumps up and down. Although to the best of our knowledge, no in depth study has been conducted so far to clarify the mechanism of the double-mesopause, the presence of a double-layer mesopause was in fact documented in several previous studies such asvon Zahn et al. [1996] and States and Gardner [1999]. In these studies, however, mostly the midlatitude to low-latitude region was the focus of the analyses. As for the cause of the double-mesopause structure,States and Gardner [1999]argued that incomplete sampling of the diurnal cycle, for example, nighttime measurements being chosen exclusively, made these disturbances stand out because the double-mesopause occurs preferably during the nighttime. Overall speaking, since the bimodal behavior occurs after the PMC season ends, it is not an immediate concern of this study. But the fact that it appears strikingly clear at the end but not at the start of the summer shares some resemblance to the finding ofNielsen et al. [2010], that is, an enhanced 5-day wave activity exists in August but not in May. The MLS temperature does not detect the double mesopause mainly because the MLS vertical resolution is 14–16 km in the summer mesopause altitude region. Furthermore, when a double-mesopause does occur, the upper branch usually exceeds the pressure range of the recommended MLS data usage.

[19] Figure 6 shows the PSAT, PH2O, and cloud frequencies on a linear scale to examine their correlation. The shaded area is between the minimum and twice the median (2× median) PSAT, and the curve in the middle is the median PSAT. Three curves are shown to represent the collective behavior of PSAT for which the supersaturated condition (S > 1) is satisfied. Figure 6 indicates that all PSAT curves experience a rapid decrease at the start and a rapid increase at the end of the PMC season, which is just opposite of the behavior of the PMC frequency. This clearly shows that the rapid onset and termination of the PMC season is caused by the PSAT variation. In addition, PSAT variations in the main period, especially those of the median and 2× median PSAT for SOFIE, are roughly anticorrelated with the variations of the PMC frequency. The anticorrelated relationship holds particularly well for those years that had distinct frequency decline toward the end of the season, i.e., 2007 and 2010. In the MLS analysis, the anticorrelation also holds very well for 2007 and 2010. The decline of the frequency reflects a systematic warming that is substantial enough to make a notable increase in PSAT. On the contrary, if temperature gets increasingly lower and necessarily PSAT becomes indefinitely small, the cloud frequency remains at ∼100%. Overall, the analysis above suggests that temperature controls the cloud frequency variation in all stages of the cloud season. This controlling role of temperature on the PMC frequency was also shown by Fiedler et al. [2011] who analyzed the diurnal variation of the cloud frequency.

Figure 6.

The intraseasonal variations of PSAT and PH2O on linear scales. The shaded area and the thin black line in between are minimum, median, and twice the median of PSAT daily, calculated from (top) SOFIE and (bottom) MLS data. The stepwise lines are the mean PH2O levels shown in Figure 5. The thick black curves are SOFIE observed PMC daily frequencies shown in Figure 3. The cross signs are the closest possible start and end days that are determined by the crossings between the PSAT minimum and the stepwise PH2O lines.

3.4. Start and End Days of the PMC Season

[20] So far we have not discussed the actual days on which the PMCs are first or last detected; instead, we have been more focused on the timings when the cloud frequency substantially increases or decreases. The first and last cloud detection can be affected by a number of factors. For example, low sensitivity to the weak clouds can make the start day appear delayed although SOFIE should not have this problem. Also, an occasional detection of one or two clouds can be due to the fluctuations of temperature or H2O that favor PMC formation before the systematic onset of the PMC season. In addition, a satellite instrument can only scan a given location at discrete LTs while clouds that occur at other LTs will be missed. All these factors combined can lead to several days of difference in the determined start or end days. For example, in the work of Bailey et al. [2005] and Petelina et al. [2006]the NH start days are on average between 20 and 25 days before the summer solstice, which are at least 5 days delayed compared to the days shown in this analysis. Owing to these uncertainties, the observed PMC start and end days are not fully robust characteristic. In this study we define a set of 0-D start and end days to compare with the observations. These days are used to describe the timing of a systematic increase or decrease of the 0-D cloud frequency. The combination of the observed and the 0-D start and end days will describe the onset and termination of the PMC season more concretely. The 0-D start and end days are determined by the first and last crossing points (from left to right) between the minimum PSAT and the stepwise PH2O mean levels shown in Figures 5 and 6. The minimum PSATis used since the 0-D assumption requires only one event meeting the condition of S > 1 to form the first cloud.

[21] The SOFIE observed and the 0-D determined PMC start and end days, in terms of days from solstice (DFS), are given inTables 1a and 1b. We have seen from above that in 2007 the cloud season had already started when SOFIE data became available. So with 2007 being taken out, the 3-year statistics indicate that the SOFIE 0-D start day is on average 1.6 days earlier than the observed start date with a scatter of 4.2 days. The mean difference is smaller than the scatter, roughly indicating that the SOFIE 0-D and observed start days agree well and do not have a significant bias. The MLS 0-D start day, on the other hand, is about 18.5 days earlier than the observation with a scatter of 4.3 days, suggesting a large bias in temperature. The comparisons of the start days reflect whatFigure 3 shows on the frequency development at the start. The SOFIE derived end day is on average 1.1 days later than the observed end day with a scatter of 4.3 days, which also supports a good agreement. The reason why we did not see a SOFIE warm bias as suggested in Figure 3is because in 2009 there is a return of coldness after a substantial warming. If we remove the year 2009, the SOFIE 0-D end day will be 3.2 days earlier than the observation, with a scatter of 1.1 days, which would suggest a warm bias in SOFIE temperature. Similarly, we found that the MLS derived end day is on average 3.9 days later than the observation with a scatter of 3.0 days. If 2009 is excluded, the end day is 2.7 days later than the observation with a scatter of 2.2 days. Both cases indicate that the bias is insignificant. This is roughly consistent with the previous finding (Figure 3) that the MLS 0-D frequency shows strong agreement with the SOFIE PMC frequency during their rapid decrease that marks the termination of the season. As a final point, one should note that although 4 years are far too short to yield any reliable statistics, we do not entirely rely on the statistics in this case.Figure 3 indicates that years 2007–2010 show highly consistent results. The statistics in Tables 1a and 1bmainly tests whether the approach to determine the 0-D start and end days works efficiently and produces results consistent withFigure 3.

Table 1a. Observed and 0-D Determined Start and End Days
SOFIE 0-D65.7−32.565.5−27.872.3−34.567.2
SOFIE Observed68−2870−3167−3170
MLS 0-D73.2−48.571.4−48.074.6−49.171.5
Table 1b. Mean Differences Between the 0-D Determined and Observed Dates Over Different Years and the 1-σ Standard Deviations of These Differencesa
 Start (Without 2007)End (All Years)End (Without 2009)
  • a

    The two rows are calculated using either SOFIE 0-D observation or MLS 0-D observation.

SOFIE 0-D−1.6 ± 4.2−1.1 ± 4.3−3.2 ± 1.1
MLS 0-D−18.5 ± 1.83.9 ± 3.02.7 ± 2.2

[22] Given the approach proposed above to derive the start and end days, one would wonder how the temperature or H2O change can affect these derived days. A sensitivity study has been performed and the results are summarized as follows. If the temperature is shifted by plus or minus 10 K the start or end date will vary approximately plus or minus 10 days. If the H2O is changed, it takes about a 10 times wetter or drier mesosphere to make the same difference in the two dates. In practice, a highly concerned issue is whether there is a long-term trend in the cloud seasonal start or end days. The limited number of trend analyses in the past decade or so suggest that neither the H2O trend (e.g., ∼0.05 ppmv/year from 1996 to 2000 in the work of von Zahn et al. [2004]) nor the temperature trend (e.g., −0.24 K/decade from 1964 to 1996 in the work of Lübken [2000; see also Beig et al., 2003]) in the polar summer mesosphere is large enough to significantly change the start and end of PMC season, especially when the reliability of these trends is still unclear. Nevertheless, assume that the extremely small temperature trend shown by Lübken [2000] is present it will take about 4 decades to make the start day 1 day earlier, which is far less than the interannual scale fluctuations. This is consistent with the findings of Gadsden [1998a] who suggested that the length of the NH PMC season remains fairly constant over the years.

4. PMC Ice Mass Density

[23] In the following sections we examine how PH2O and PSAT control the ice mass density. Figure 7shows a scatterplot of 0-D ice mass density values at Zmax, denoted by mice, on the PSAT versus PH2O-PSAT plane. It should be noted that all analyses below are performed at Zmax. We use PH2O-PSATas one dependent variable simply because the 0-D ice mass density is proportional to (PH2O-PSAT)/T. The temperature in the denominator is a negligible factor compared to PH2O-PSATin controlling the 0-D ice mass density variation. Since PH2O divided by temperature is proportional to the H2O number density, we take PH2O to be in a very similar role as H2O number density. The contours reflect all the mice values using any possible combination of temperature and H2O while the individual data points are calculated using the SOFIE temperature and H2O profiles from 2007 to 2010. The same color scheme is applied for both the contours and the dots. It is noted that the contours are parallel to the horizontal axis in most cases, suggesting that mice does not vary much with PSAT; instead, PH2O-PSAT is in nearly full control of the mice variation except for the very large mice (>100 ng/m3) at which the effect of the temperature becomes notable. But apparently these large values are rarely attained among all SOFIE 0-D modeledmice values. Another striking feature is that the orientation of the cluster of dots maintains a very small angle to the axis of PH2O-PSAT, indicating that PSAT remains fairly low with respect to PH2O-PSAT. This suggests that as the mice increases, PH2O experiences a more drastic change than PSAT and therefore PH2O takes a dominant role in the mice variation.

Figure 7.

The 0-D ice mass densities on the PSAT versus PH2O-PSAT plane. The dots are ice mass densities calculated using SOFIE temperature and H2O. The color of any given contour or dot represents the magnitude of ice mass density, in units of ng/m3.

4.1. Comparison of the 0-D and the SOFIE Observed Ice Mass Densities on Intraseasonal Scales

[24] In this subsection we compare the 0-D modeled and the SOFIE observed ice mass densities on intraseasonal scales and further examine their relationship with PSAT and PH2O. The observed ice mass density at the observed Zmax is denoted by mice_obs. Although Figure 7 suggests that PH2O is in control of mice variation in an overall sense, more detailed analyses are needed to further separate the roles of PH2O and PSAT. It is also necessary to separate different time scales or cloud strengths since the relative importance of PH2O and PSAT may vary with these factors. In this paper we are particularly interested in studying what controls the mice variation at different cloud strengths. In order to define the strong and weak cloud cases, we sort the mice or mice_obs values daily from smallest to largest. The strong cloud case is defined as the time series using the daily maximum ice mass density values. In defining the weak cloud case, two steps are required. First, the first 20% of the daily sorted events are chosen; second, among the chosen events the maximum value is selected to represent the weak cloud case. We did not simply use the daily minimum ice mass density because the daily weakest cloud is most severely affected by the cloud detection uncertainty. The medium cloud case, which uses 50% threshold instead of 20%, is included in some analyses but is not the focus of the discussion. It should be pointed out here that the above definitions only pertain to their meanings in a relative sense and are only appropriate for the main period of the season during which the daily cloud frequencies are persistently high. While at the seasonal start or end, the clouds are fewer and generally weaker, and therefore all clouds should be considered weak.

[25] Figure 8 shows the intraseasonal time series of mice_obs, mice, and the corresponding PH2O and PSAT in the strong cloud case. PSAT_min at Zmax is also overplotted. Although PSAT_min is not a key variable in the ice mass density investigation, it is shown here to confirm a rapid start and end of the cloud season (see Figure 5). A 4-day smoothing is applied to each time series to remove any random variability and to highlight the variation on longer intraseasonal scales. InFigure 8 we first note that the mice divided by a factor of 1.6 follows a very similar intraseasonal variation to the mice_obs, indicating that the 0-D model reproduces the intraseasonal variation very well. The factor 1.6 is empirically determined based on the analysis of ice mass density at the cloud peak height. The fact that the 0-D model systematically overestimates the ice production is well expected because it omits the nucleation barrier and the ice particle growth [Hervig et al., 2009b; Hervig and Gordley, 2010]. The correlation coefficients (see Table 2) between the mice and mice_obs reach 0.9 on average, with the lowest and the highest coefficients being 0.81 in 2007 and 0.96 in 2010, respectively. We also note extremely high correlation between mice and PH2O, with the coefficient varying from 0.95 to 0.98. The corresponding PSAT, on the other hand, shows relatively poor correlation with mice or PH2O; the PSAT and PH2O correlation coefficient varies between 0.41 and 0.62. Although substantially lower than the other coefficients in the strong cloud case, these coefficients reflect an inherently significant correlation between PH2O and PSAT, i.e., their confidence levels remain at 99.9%. This is because, in some cases, the variations in PH2O and PSAT are primarily caused by the cloud height variation instead of a fundamental change in the environmental temperature or H2O. In the cloud region below the mesopause, log(PH2O) and log(PSAT) both decrease monotonically with altitude, so the variations of PH2O and PSAT are not entirely independent. The facts that PH2O far exceeds the PSAT in magnitude in the core of the season and that PH2O and mice are strongly correlated both support a conclusion that PH2O is in dominant control of the mice variation in the strong cloud case. At this point we can immediately apply this conclusion to what DeLand et al. [2007]have found about a long-term upward trend of the SBUV cloud albedo. Since SBUV instruments detected only bright clouds, we assume that the SBUV measured PMCs match the strong cloud case. Accordingly, we expect a similar upward trend in the mesospheric H2O over the last 30 years.

Figure 8.

The intraseasonal variations of PH2O, 0-D ice mass density, and SOFIE observed ice mass density for the strong cloud case. Daily values are taken at the cloud peak height. For all the daily values the maximum ice mass density is used and then the corresponding PH2O and PSAT are chosen. The daily minimum PSATat cloud peak height is also plotted. A 4-day smoothing is applied for all the time series.

Table 2. Correlation Coefficients Between PSAT, PH2O, mice, and mice_obs in Strong, Medium, and Weak Cloud Cases and Their Confidence Levels (in the Parentheses)a
Correlation Coefficient (Confidence Level %)2007200820092010
  • a

    Note that the time series of the strong and weak cloud cases are shown in Figures 8 and 9.

PH2O Versus PSAT
Strong0.62 (99.9)0.56 (99.9)0.41 (99.9)0.43 (99.9)
Median0.88 (99.9)0.64 (99.9)0.76 (99.9)0.70 (99.9)
Weak0.94 (99.9)0.78 (99.9)0.88 (99.9)0.90 (99.9)
PH2O Versus mice
Strong0.95 (99.9)0.97 (99.9)0.98 (99.9)0.96 (99.9)
Median0.84 (99.9)0.78 (99.9)0.81 (99.9)0.89 (99.9)
Weak0.08 (50.0)0.69 (99.9)0.29 (99.0)0.56 (99.9)
mice Versus mice_obs
Strong0.81 (99.9)0.93 (99.9)0.88 (99.9)0.96 (99.9)
Median0.87 (99.9)0.86 (99.9)0.88 (99.9)0.91 (99.9)
Weak0.37 (99.9)0.78 (99.9)0.71 (99.9)0.75 (99.9)

[26] Figure 9 shows the weak cloud case. Similar to the strong cloud case, we see fairly good agreement between the mice/1.6 and the mice_obs in magnitude as well as the overall seasonal variation. The correlation coefficient of the mice and mice_obs varies from 0.72 to 0.79 in 2008–2010, but in 2007 it only reaches 0.39. In the 2007 case we see fairly good agreement in the magnitude, and yet a low correlation coefficient is obtained because during the two short periods centered at DFS 15 and after DFS 60 there are anticorrelations that would reduce the correlation coefficient. However, even with the 2007 taken out, the overall correlation between mice and mice_obs is still lower than in the strong cloud case. This is mostly caused by a larger uncertainty induced in the weak cloud case, i.e., as PH2O and PSAT are comparable in magnitude and the uncertainties in both variables can contribute greatly to the PH2O-PSAT. A clearly notable characteristic in the weak cloud case is that the PH2O and PSAT roughly vary in concert on intraseasonal scales. As a result the correlation coefficient between the PH2O and PSAT reaches ∼0.88 on average. This is very different from the strong cloud case that shows a substantially poorer correlation between PH2O and PSAT. Another difference from the strong cloud case is that the correlation coefficient between the PH2O and mice is highly variable, varying drastically from 0.08 in 2007 to 0.69 in 2008. This should be expected since we have known from the above that neither PH2O nor PSAT alone dominantly controls the mice variation. Rather, their roles are basically equal and a given change in either can have a significant effect on mice. Figure 10 shows the same set of plots as Figures 8 and 9 except that daily averages of all clouds are shown. Examining the daily average is a necessary step because one would wonder whether it is a strong or weak cloud case in average sense. By viewing Figure 10, we can argue conclusively that the daily average of all clouds behaves more like the strong cloud case.

Figure 9.

Same as Figure 8 except for the weak cloud case. The definition of the weak cloud case is given in section 4.1.

Figure 10.

Same as Figure 8 except that the ice mass density is the daily mean of all the calculated or observed clouds. This represents the daily averaged case.

4.2. Relative Importance of the PH2O and PSAT in Controlling the Ice Mass Density

[27] Although the 0-D model has no sensitivity to time, it has proven to be highly effective in reproducing the intraseasonal variation of the observed ice mass density in both strong and weak cloud cases. These results suggest that the 0-D model should also accurately represent longer time scale variations such as those reported byDeLand et al. [2007] that are based on seasonal averages. As a further step, it is worthwhile to quantify the relative importance of PH2O and PSATto the 0-D ice mass density in a general sense. A scatterplot of 0-D ice mass density on the plane of log(PSAT) versus log(PH2O) is shown in Figure 11. The daily PSAT and PH2O values are taken from Figures 8 and 9, representing the strong and weak cloud cases, respectively. The data points in between (the green dots) represent the medium cloud case. The logarithm scale is used to reveal the temperature dependence. Figure 11 suggests that as the environment gets colder and wetter, and necessarily micegets larger, i.e., toward the right-lower corner, themice contours become increasingly parallel to the axis of log(PSAT), suggesting that the mice variation occurs increasingly predominantly along the log(PH2O) axis. In other words, at the lower right corner of Figure 11 PH2O is in nearly full control of the mice variation. In reality however mice may not actually reach the large values in the lower right corner given geophysically reasonable temperature and H2O, such as those mice values calculated from the SOFIE temperature and H2O. To actually quantify the relative importance between PH2O and PSAT, we calculate the mice contour slopes, shown by the pink lines. Along each pink line the relative importance of PH2O and PSAT is considered constant, and the importance level of PH2O rises as the slope reduces. Observing the three sets of data points, we note that as the clouds get stronger, the clusters move toward lower slope values, suggesting that the PH2O becomes increasingly more in control. For example, the cluster of dots for the strong cloud case is riding on the 0.2 line, suggesting that PH2O is approximately 5 times more important than PSAT in controlling the mice variation. It is also noted that the dots are less clustered and more oriented in the weaker cloud cases. For example, in the weak cloud case the dots spread along the diagonal line, i.e., PSAT = PH2O line, which explains the correlated changes of the two variables shown in Figure 9; the mice values can spread more extensively because weaker clouds can form for a much broader range of PH2O, i.e., drier and colder conditions combined, unlike the stronger clouds that can only form under higher PH2O, or wetter condition, regardless of PSAT.

Figure 11.

The rainbow-colored contours represent all 0-D ice mass density values on the log10(PH2O) versus log10(PSAT) plane. Along each pink line the slopes of the contours remain constant. On the basis of the equal-spacing of logarithm, the contour values are chosen as 0.04, 0.06, 0.09, 0.14, 0.20, 0.30, 0.46, and 0.68, respectively. The three sets of dots with different colors are daily PH2O and PSAT pairs for strong (red), median (green), and weak cloud (black) cases in the SOFIE related calculations. Note that the strong and weak cloud cases are shown in Figures 8 and 9.

4.3. Implications for Long-Term PMC Trends

[28] What Figure 11shows may hold for PMC variations on many different time or spatial scales, and further research is needed to explore these possibilities. But the most important issue for the scientific community at present is the long-term PMC trend. If such a trend does exist as suggested by recent papers, it is essential to find out whether it is H2O or temperature that is driving it. Studies using cloud observations taken during the core of PMC season [e.g., Romejko et al., 2003; DeLand et al., 2007] have indicated that there has been a multidecadal upward trend in the PMC (or NLC) brightness in the last 30–40 years. Results presented earlier in this paper suggest that the reason for these long-term PMC increases are due to H2O changes and therefore corresponding H2O observations should show consistent trends. However, there is not yet any conclusive finding regarding long-term H2O changes since no reliable long-term H2O records currently exist for the PMC region during summer. However, as mentioned above, an upward H2O trend was found in the period 1996–2000 at ∼80 km altitude at the Alomar observatory (69°N) in polar summer [von Zahn et al., 2004]. It was argued cautiously by these authors that inclusion of such a trend in a NLC model results in an upward trend in the cloud albedo that roughly agrees with the SBUV observations. Nevertheless, a 5-year record is far too short to support a definitive long-term trend owing to the intervention of a series of shorter or longer time scale variations, among which the most prominent is the solar cycle effect. A Halogen Occultation Experiment instrument on UARS satellite (HALOE) [Russell et al., 1993] H2O analysis performed by Hervig and Siskind [2006] has shown a clear solar cycle effect but no significant trend in the polar summer mesosphere. However, HALOE latitude coverage in summer is poor and varies with year, and therefore the result may change when different sampling approaches are used. Overall, more reliable and longer time series are required to verify the consistency between the PMC brightness and H2O variations.

5. Summary and Conclusions

[29] The rapid onset and termination of the PMC season, and the intraseasonal variation of the ice mass density, were investigated using both the SOFIE data set [Gordley et al., 2009; Hervig et al., 2009a] and a 0-D model [Hervig et al., 2009b]. The SOFIE PMC data set provides unprecedented observational support for the 0-D modeled cloud frequency and ice mass density so that the respective roles taken by temperature and H2O can be evaluated. The 0-D model assumes ice being produced whenever the supersaturation ratio (S) is greater than one. In such an optimal condition, the ice production is only dependent on temperature and H2O, i.e., in the form of PSAT and PH2Oin this paper. The 0-D assumption was never used as a primary framework to interpret PMC changes in studies of the last few decades because a number of factors that are ignored in the 0-D assumption, such as nucleation, transport of cloud particles via the atmospheric flow field, and ice particle fall velocity, were considered critically important in affecting the PMC variation. The 0-D assumption was brought to light after the SOFIE data analyses indicated that PSAT and PH2O take on controlling roles in the PMC frequency and ice mass density variations. The SOFIE data set provides a natural platform to study the relationship between PMCs, PSAT, and PH2O because the related variables are simultaneously measured and also because better temporal and spatial details are resolved in SOFIE than in other satellite data sets.

[30] The intraseasonal variation of the SOFIE observed PMC frequency indicates a rapid onset and termination of the PMC season. During the main period of the season the frequency remains at ∼80–100%. The SOFIE and MLS 0-D modeled frequencies also indicate a rapid seasonal onset and termination, showing excellent qualitative agreement with the SOFIE observed PMC frequency. Furthermore, the SOFIE 0-D frequency indicates a starting period that agrees well with the observations but an ending period that is a few days earlier. The MLS 0-D frequency indicates an ending period that agrees well with the observations but a starting period that is ∼15–20 days earlier. It was argued that such discrepancies between the 0-D and the SOFIE observed PMC frequencies could have been caused by a temperature bias between SOFIE and MLS data. We conclude that it is the temperature variation rather than H2O variation that dominantly controls the PMC seasonal onset and termination. When interpreting the asymmetry between the 0-D determined start and end in relative to the PMC observation, we argue that on exiting the cloud season, owing to a long history of ice existence, gas-phase temperature alone may appear warmer than desired in determining the end of the season; rather, ice-temperature should also be considered. These assertions, while logical, remain unproven and unresolved.

[31] Throughout the summer, PSAT experiences over 8 orders of magnitude variation, whereas the PH2O variation is nearly flat in comparison. The much smaller variation of PH2O is represented by a stepwise function using two PH2O levels, one averaged before and the other after the solstice, respectively. The collective behavior of all PSAT values that meet the condition S > 1 shows a rapid decrease at the start and rapid increase at the end of the season, which is exactly opposite to the variation of the cloud frequency. PSAT and the cloud frequency are also anticorrelated in detailed intraseasonal variations. This suggests that PSAT, or temperature variation, controls the cloud frequency variation throughout the cloud season. The estimated start (or end) day is defined as the day on which the daily minimum PSAT goes below (or above) the presolstice (or postsolstice) level of the PH2O. The 0-D determined start and end days are in good agreement with what the observed cloud frequencies suggest regarding the timing of the onset and termination of the cloud season.

[32] SOFIE observed and 0-D modeled ice mass densities, i.e.,mice_obs and mice, are highly correlated on intraseasonal scales, with the correlation coefficient being ∼0.9 and ∼0.7 in the strong and weak cloud cases, respectively. As a further step it is important to clarify how PH2O and PSAT control mice. PH2O is in dominant control of the mice variation in the strong cloud case, while in the weak cloud case PH2O and PSAT vary in concert and both have similar and significant effects on the micevariation. On the basis of the SOFIE 0-D model results we conclude that the PH2O is about 5 times more important than PSATin the strong cloud case. However, this factor may very well change if the mesospheric temperature is systematically warmer or colder than observed by SOFIE and it will not be definitive before a better consensus about the mesospheric temperature is reached. Finally, we point out that for both the observed and modeled PMCs, the daily average in the core of the season resembles a strong cloud case. As a result, we conclude that the long-term upward trend of cloud brightness reported byDeLand et al. [2007] should be accompanied by an upward trend of H2O.


[33] Funding of the AIM mission was provided by NASA's Small Explorers Program under the contract NAS5–03132. Many thanks are given to the SOFIE/AIM and other AIM team members for constant support, encouragement, and valuable advice for improvement on this work. We thank the SOFIE retrieval team for providing the SOFIE PMC data set and SOFIE level2 data set. We also thank the MLS/Aura retrieval team for making the MLS level2 data available online.