Journal of Atmospheric and Solar-Terrestrial Physics 73 (2011) 2073–2081

Contents lists available at ScienceDirect

Journal of Atmospheric and Solar-Terrestrial Physics

1364-68

doi:10.1

� Corr

E-m

journal homepage: www.elsevier.com/locate/jastp

Impact of solar proton events on noctilucent clouds

Nabiz Rahpoe a,�, Christian von Savigny a, Charles E. Robert a,b, Matthew T. DeLand c, John P. Burrows a

a Institute of Environmental Physics, University of Bremen, Germanyb Belgian Institute Space Aeronomy (BIRA), Brussels, Belgiumc Science Systems and Applications Inc (SSAI), Maryland, USA

a r t i c l e i n f o

Article history:

Received 7 January 2010

Received in revised form

22 June 2010

Accepted 19 July 2010Available online 12 August 2010

Keywords:

Noctilucent clouds

Polar mesospheric clouds

Summer mesopause

SBUV

Solar proton events

26/$ - see front matter & 2010 Elsevier Ltd. A

016/j.jastp.2010.07.017

esponding author. Tel.: +49 42121862089.

ail address: nabiz@iup.physik.uni-bremen.de

a b s t r a c t

The impact of solar proton events (SPEs) on noctilucent clouds (NLCs) is studied using the 30-year NLC

data set based on measurements with SBUV instruments on NIMBUS 7 and the NOAA 9-17 satellites.

We performed a cross-correlation analysis between GOES proton flux time series and NLC occurrence

rate (albedo) time series to investigate, whether a depletion of NLCs is frequently observed during SPEs.

We focused on the analysis of the years when SPEs occurred during the core NLC season (days 0–40

with respect to solstice). For energetic protons in the 5–10 MeV energy range we find for the majority of

SPEs statistically significant anti-correlations between NLC occurrence rate (albedo) and proton flux

time series with time lags of up to 8 days. For the two strong SPEs in July 2000 and January 2005 the

sensitivity of the results to the proton energy range was investigated. We found that the main results do

not depend on proton energy. Together with a recently proposed physical mechanism leading to a

SPE-induced warming of the polar summer mesopause region, our results may indicate that NLCs are

frequently depleted during SPEs.

& 2010 Elsevier Ltd. All rights reserved.

1. Introduction

Noctilucent clouds (NLCs) occur during summertime at latitudeshigher than about 503 (e.g., Thomas and Olivero, 1989) and altitudesof about 82–85 km. The temperatures near the polar summermesopause frequently drop below 150 K, and low temperatures aswell as sufficient H2O are the two principal requirements for theformation of NLC particles, mainly consisting of H2O ice (Hervig et al.,2001). NLCs are discussed as being early indicators of global change,because they react very sensitively to temperature and water vapourvariations (e.g., Thomas and Olivero, 1989, 2001; von Zahn, 2003; vonZahn et al., 2004). Satellite observations of NLC albedo (DeLand et al.,2007) and occurrence rate (Shettle et al., 2009) throughout the lastthree decades indeed show evidence for a long-term increase, inaddition to a pronounced 11-year solar cycle signature. The NLCoccurrence rate during individual NLC seasons is highly variable,particularly in the southern hemisphere. This variability is, however,not well understood. The following physical processes are known tocontribute in part to the observed variability: gravity waves (e.g., Witt,1962), tides (von Zahn et al., 1998), planetary wave perturbations(e.g., Merkel et al., 2003; von Savigny et al., 2007a), the solar 27-daycycle (Robert et al., 2010) and interhemispheric coupling (Becker andFritts, 2006; Karlsson et al., 2007).

ll rights reserved.

(N. Rahpoe).

von Savigny et al. (2007b) recently presented evidence for apossible depletion of NLCs in the southern hemisphere during theJanuary 2005 solar proton event (SPE). For the [703S, 803S] latituderange NLC occurrence rates dropped from about 80% to less than20% within the period with enhanced proton precipitation. Theobserved severe reduction in NLC occurrence rates – measuredwith SCIAMACHY (Scanning Imaging Absorption spectroMeter forAtmospheric CHartographY) (Bovensmann et al., 1999) on Envisat– was shown to coincide with a temperature increase of morethan 10 K – measured with MLS (Microwave Limb Sounder)(Waters et al., 2006) on the Aura satellite. Solar proton eventsassociated with coronal mass ejections (CMEs) lead to theprecipitation of highly energetic solar protons into the Earth’spolar cap regions. The precipitating protons ionize the atmo-sphere and initiate a series of complicated ion-chemical reactionsterminated by recombination reactions. Several consequences ofthese chemical perturbations are well understood. The producedions lead to the formation of NOx and to the conversion of H2O toHOx. Both the HOx and NOx families participate in catalytic O3

destruction cycles. Catalytic O3 losses exceeding 50% in themesosphere and upper stratosphere are frequently observedduring major SPEs (e.g., Porter et al., 1976; Solomon et al., 1981,1983; Rohen et al., 2006).

The impact of highly energetic solar particles on atmospherictemperature and dynamics is less well understood. Severalstudies indicate temperature and dynamical effects caused bySPEs (Banks, 1979; Roble et al., 1987; Johnson and Luhmann,1993; Krivolutsky et al., 2006; Jackman et al., 2007). The more

Fig. 1. Zonally and daily averaged NLC occurrence rate for the 2004/05 NLC season

in the southern hemisphere and the 602903S latitude range derived from

measurements with SBUV on NOAA-16 (black solid line). The red line shows

GOES proton fluxes for channel 2 (E¼5–10 MeV, proton flux unit: counts

cm�2 s�1 sr�1). The curves have been smoothed with a 5-day running mean

filter. (For interpretation of the references to colour in this figure legend, the

reader is referred to the web version of this article.).

N. Rahpoe et al. / Journal of Atmospheric and Solar-Terrestrial Physics 73 (2011) 2073–20812074

recent model simulations by Krivolutsky et al. (2006) andJackman et al. (2007) show a positive temperature response atthe polar summer mesopause associated with solar proton events.These studies indicated that dynamical effects play a major rolefor the warming of the polar summer mesopause, but they didneither focus on this specific atmospheric region, nor provide adetailed description of the underlying physical mechanism.

Becker and von Savigny (2010) recently established a physicalmechanism causing a warming of the polar summer mesopauseduring and after SPEs. Briefly summarized, the chain of effects isas follows: the catalytic O3 loss driven by the SPE-induced HOx

leads to a reduced diabatic heating in the middle polar meso-sphere, which in turn leads to a reduced meridional temperaturegradient. This reduction causes reduced zonal winds according tothe thermal wind balance, leading to altered gravity wave filteringand eventually to a dynamically induced warming of the polarsummer mesopause due to a changed meridional circulation. Thishypothetical chain of events was tested by Becker and vonSavigny (2010) using the Kuhlungsborn Mechanistic GeneralCirculation Model (Becker, 2009) and the proposed mechanismwas essentially confirmed. Therefore, one may expect a polarsummer mesopause warming – and consequently an impact onNLCs – to occur regularly during SPEs.

The main aim of this paper is to investigate whether the NLCdepletion observed by SCIAMACHY during the January 2005 SPE (vonSavigny et al., 2007a) is exceptional, or whether a NLC occurrence ratedecrease is frequently observed during or after SPEs. For this purposewe use the 30 year SBUV (Solar Backscatter Ultraviolet) NLC data setdescribed by DeLand et al. (2003) and DeLand et al. (2007) togetherwith proton flux measurements with instruments on the GOES(Geostationary Operational Environmental Satellites) platform(Chesters, 1996). As NLCs react very sensitively to temperaturechanges this investigation may also allow drawing conclusions on thetemperature effects associated with SPEs. Another relevant aspect isthe apparent solar cycle variation observed in NLCs (e.g., DeLand et al.,2003). If NLCs are always depleted by SPEs to a certain extent, thismay contribute to the apparent solar cycle variation in NLCs, becauseSPEs occur more frequently during solar maximum.

2. Instrumentation

The NLC detection algorithm applied to SBUV observations wasfirst developed by Thomas et al. (1989), who used severalwavelengths between 252 and 292 nm to determine the dailycloud-free background albedo as a function of solar zenith angle(SZA). This background is then subtracted from all data tocalculate a residual. To detect a NLC, four different tests areapplied to the wavelength-dependent residual albedo in order toavoid spurious NLC detections (DeLand et al., 2003). If all the testsare positive, then the NLC quality flag is set to its maximum valueof 31. For the present study we assume that an NLC is detected ifthe quality flag exceeds a value of 29. Note that in the work ofDeLand et al. (2003, 2006, 2007) NLCs are only detected if thequality flag is 31. We tested different quality flag threshold valuesand found that the derived cloud occurrence frequencies dependon the chosen threshold, as expected. However, the choice of thequality flag threshold does not affect the main conclusions of thisstudy, because they mainly depend on the relative variation ofNLC occurrence rate and albedo during and after SPEs. For moreinformation on the cloud detection scheme and the definition ofthe quality flag, see DeLand et al. (2003).

The NLC occurrence rate is defined as the ratio of the numberof detected NLCs and the number of all SBUV measurements for agiven day and a certain latitude range. Here, we determine thezonally averaged occurrence rates for all latitudes poleward of 603

in both hemispheres. As an example, the NLC occurrence rate timeseries for the 2004/05 NLC season in the southern hemisphere isshown in Fig. 1 (black solid line) together with the GOES protonflux measurements (red line).

The proton flux measurements used in this study wereperformed by SEM (Space Environment Monitor) instruments onthe GOES series (GOES 5 to GOES 12). SEM is a subsystem onboardthe GOES satellites performing in situ measurements of X-rayintensity, energetic particle fluxes and the magnetic field. For ourpurpose we use the proton energy channels 2–7. Channel 1,corresponding to proton energies between 1 and 5 MeV can becontaminated with trapped, magnetospheric protons, and there-fore is not useful for our applications. Measurements withchannels 2–7 are corrected for the impact of trapped protonsand are called ‘corrected integral proton fluxes’. These channelscover proton energies ranging from 5 to 100+ MeV, correspondingto energy deposition altitudes between about 75 and 30 km,respectively. The use of proton fluxes for energy ranges withenergy deposition altitudes well below NLC altitude is mainlydriven by the consideration that the physical mechanism causingthe NLC depletion is likely initiated by the catalytic O3 depletionin the mesosphere (Becker and von Savigny, 2010) and not causedby a direct effect at NLC altitude. The effect of using different SEMchannels on the results of this analysis is discussed in detail inSection 3. SEM measurements are available since 1986, andtherefore the SBUV measurements during the 1978–1985 NLCseasons are not considered in this study.

Note that a sudden and strong ozone depletion may lead toproblems with NLC detections in the SBUV nadir observations.Since precipitating solar protons also lead to ozone depletionin the middle atmosphere, the SBUV NLC data are not usedduring the SPE peak days (2 days at most). The resulting gapsare linearly interpolated. After the interpolation we can nowbegin to compare the two time series for every single NLC season(Fig. 2).

3. Data analysis

For this work we perform a cross-correlation analysis betweenthe temporal derivatives of NLC and proton flux time series withina finite time window (with widths between 15 and 30 days)

Fig. 2. A–H: Zonally and daily averaged NLC occurrence rate time series smoothed with a 5-day box-car (black line) for NLC seasons with SPEs. The solar proton flux

(channel 2: 5–10 MeV) is plotted for the same season (red line) using a logarithmic scale. (For interpretation of the references to colour in this figure legend, the reader is

referred to the web version of this article.).

N. Rahpoe et al. / Journal of Atmospheric and Solar-Terrestrial Physics 73 (2011) 2073–2081 2075

centered at the SPE investigated. The analysis is applied to boththe NLC occurrence rate and the albedo time series. The mainreason for using the derivatives of the NLC (dNLC) and the protonflux time series (dSPE) (Fig. 3) is that the time lag between apossible cause in the proton flux time series and a possible

consequence in the NLC time series can be estimated after areduction during or shortly after an SPE then more reliably. Ifthere is no recovery in the NLCs time series, the time lag for whichthe extreme cross-correlation value is assumed depends on thewidth of the window used for the cross-correlation analysis.

N. Rahpoe et al. / Journal of Atmospheric and Solar-Terrestrial Physics 73 (2011) 2073–20812076

In Fig. 4, the cross-correlation function for a given scanningwindow (here 25 days) versus the time lag is drawn. Theextremum in the correlation coefficients rc and its time lag tc

are the two parameters being used in the following analysis.These parameters have been calculated by applying a Gaussian fit(dashed line) to the correlation function. The height and positionof the fitted Gaussian curve give the correlation coefficient rc andtime lag tc. Time lags of tc o�1 day and tc 415 days for theextremum of the curve are rejected by this procedure. Thepositive and negative correlation curves are fitted separately, andtherefore both positive and negative correlations are considered.A cross-correlation analysis was also applied to the original timeseries, without taking derivatives, leading to a strong dependenceof the obtained time lags for the extreme correlation coefficientson the length of the time series samples used for analysis. Thismeans that the time lag between two signals will increase withthe width of the sliding window (see Fig. 5). In order to avoid thisartificial feature, we found that the use of derivatives is moreappropriate.

In the next step the extreme correlation coefficients rc andtime lags tc obtained for different window widths wi are averagedto yield the mean values rc and tc for a given solar proton event.

Fig. 3. Sample normalized dALB and dSPE time series for a window of width 25

days and for the July 2000 SPE. One can see a maximum in dSPE (increase of solar

proton flux) and a minimum of the NLC albedo derivative time series (decrease of

NLC).

Fig. 4. Sample cross-correlations of proton flux and albedo time series (derivatives dSP

2005 SPE (panel B), and a sliding window with a width of 25 days. The black dashed l

If SBUV measurements are available from multiple satellites for asingle season, then the correlation analysis is performed on anyindividual time series, and the obtained correlation values andtime lags are then averaged giving mean values for each season.The computed correlation coefficients (rc) for all SPEs thatoccurred during the northern and southern NLC seasons coveredby SBUV(/2) observations are used to sort those seasons into twomain categories, i.e., cases with positive (r þc ¼ rc 40) andnegative (r�c ¼ rc o0) correlation coefficients. Since the correla-tion function can have negative and positive correlation coeffi-cients with different time lags for the same event, we consideronly those parameters, where the condition tcðr

7c Þotcðr

8c Þ is

fulfilled.In addition a statistical test has been performed by applying

the Ebisuzaki method (Ebisuzaki, 1997). Using this algorithm,which correlates resampled time series, we obtain a criticalcorrelation coefficient rcrit. The resampling is done by adding arandom-phase in Fourier space. Fig. 6 shows an individual samplerandom-phase time series. The random-phase resampling isperformed 100 times in order to establish rcrit, which is the maincriterion for randomness of the time series at a given confidencelevel. Here we assume a confidence level of 95%. If the calculated

E & dALB) as a function of time lag for the July 2000 SPE (panel A) and the January

ine is the Gaussian fit applied to the minima of the curve.

Fig. 5. Dependence of the obtained time lag of the negative correlation extremum

on window widths (here 15–30 days) for the January 2005 SPE and no derivation

of the time series. As we can see, there is a strong increase of the time lag from

tc¼1 day for a 15-day window to tc¼8 days for a 30-day window.

Fig. 6. Example of resampled dALB and dSPE time series for a window of width 25

days. The phase of each sample is randomized between R � 2p�p where R is a

random number RA ½0,1�. An ensemble of 100 random time series has been used to

derive the critical correlation coefficient rcrit.

Fig. 7. Comparison of negative rc and rcrit for different widths of the scanning

window for the January 2005.

Fig. 8. Same as Fig. 5 except for the derivative of the NLC albedo time series and

for a window width range from 15 to 30 days.

N. Rahpoe et al. / Journal of Atmospheric and Solar-Terrestrial Physics 73 (2011) 2073–2081 2077

correlation coefficient rc is lower than rcrit (jrcjo jrcritj) then thenull hypothesis is accepted at a given confidence level, i.e., ourderived correlation coefficient is then considered a purely randomresult. If, on the other hand jrcj4 jrcritj, then the null hypothesis isrejected and we conclude that our correlation coefficient isstatistically significant at the 95% confidence level.

Fig. 7 shows the dependence of rc and rcrit as a function ofwindow width for the January 2005 SPE. There is a slight decreasein rc with increasing window width, but the condition jrcj4 jrcritj

is fulfilled for all window widths. The dependence of the time lagtc on window width is shown in Fig. 8. In contrast to the previouscalculation (Fig. 5), the time lag remains constant for all windowwidths. As already mentioned above, this is the main reasonfor using the derivatives of the proton flux and NLC time seriesrather than the original time series themselves. In the followingwe restrict the analysis to a window width range between 20and 30 days, since it was found that this range delivers reliableresults.

One has to be aware of the fact that the random-phase test isinappropriate for time series with lag-1 autocorrelation coeffi-cients (b) close to 1, because then the test becomes too liberal(Ebisuzaki, 1997). Since in our case the low frequencies are poorlyresolved, we only use time series with b lower than a certain

threshold (here: jbjo0:75). As demonstrated by Ebisuzaki (1997)the random-phase test allows for reliable significance testing for bvalues below 0.75.

We now summarize the steps taken to determine the finalparameters rc and tc:

1.

The time series are daily and zonally averaged, latitudinallybinned, as well as optionally smoothed (with a 3-day box-car).

2.

Derivatives of proton flux, NLC albedo and occurrence ratetime series are determined.

3.

Only increasing parts (dSPE40) of the proton flux time seriesare used.

4.

Cross-correlations of dSPE and dALB as well as dSPE and dNLCtime series as a function of time lag are determined fordifferent window widths wi.

5.

Application of the Ebisuzaki method to obtain rcrit. 6. Determination of rc and tc for a given window wi based on

Gaussian fits for r40 and ro0 separately.

7. Cases with time lags tc o�1 and tc 415 days are not

considered.

8. Calculation of mean values (rc ,tc) by averaging over results

for different window widths wi.

9. Only cases with jrcj4 jrcritj are considered further.

10.

Of the values r þc 40 and r�c o0 for the same event only theone with shorter time lag is considered.

11.

If rc and tc values are available from NLC measurementswith different satellites for a given season, these areaveraged yielding all-instrument means, where the standarddeviation is an estimator for differences between individualsatellite.

This procedure is applied to the times series for all northernand southern hemisphere NLC seasons – between 1986 and 2008– in which at least one SPE occurred.

4. Results

Fig. 9 shows scatter plots of the correlation coefficients and thetime lags obtained by applying the analysis procedure discussedin the last section. The left panel shows results for the5–10 MeV proton energy channel and the right panel for the10–30 MeV proton energy range. For the higher proton energiesthe number of data points is smaller, because these energies arereached by less SPEs. There are also cases with positive correlation

Fig. 9. Scatter plot of rc (correlation coefficient) and t c (time lag) for all the SPEs during the NLC core season (days 0–40 with respect to solstice), for the NLC albedo time

series and for energy channels 2 (panel A) and 3 (panel B).

Fig. 10. Scatter plot of r c (correlation coefficient) and proton flux Evprot for all the SPEs during the NLC core season, for the NLC albedo time series and for proton energy

channels 2 (left panel) and 3 (right panel).

N. Rahpoe et al. / Journal of Atmospheric and Solar-Terrestrial Physics 73 (2011) 2073–20812078

coefficients—which is to be expected, given that NLCs are alsoaffected by other processes. However, the majority (470%) ofcorrelation coefficients is negative. We can see that for most of theSPEs the correlation coefficients are around rc ¼�0:5. There is noobvious dependency of time lag tc on rc. The time lags in Fig. 9vary between 0 and 8 days. Important questions are, whether theobserved range of time lags is consistent with an SPE-inducedeffect, and whether it is conceivable that a time lags as long as 5–8days can be caused by an SPE.

First we have to note that the durations of the individual SPEsvary between about 3 and about 8 days. Secondly, the temporalevolution of the ozone depletion and the subsequent dynamicaleffects associated with the precipitating particles during the SPEwill strongly depend on the temporal variation in proton fluxes,which may also differ significantly from SPE to SPE. Some SPEsshow pronounced double peak or even triple peak structures inthe proton fluxes. Due to these reasons we expect that SPE-induced effects are associated with a certain range of time lags.The model simulations by Becker and von Savigny (2010) showthat the summer mesopause heating starts essentially with theproton flux enhancements, but the maximum heating is observedat the end of the anomalous mesospheric cooling period (assumed

to last 5 days) caused by reduced diabatic heating due to catalyticozone losses.

Fig. 10 shows scatter plots of the calculated cross-correlationcoefficients rc as a function of Evprot for the NLC albedo time seriesand the 5–10 MeV proton energy channel (left panel) and the 10–30 MeV energy channel (right panel). Evprot is the total integratedproton flux for a given energy channel during the SPE. As can beseen in Fig. 10, for the 10–30 MeV proton channels, the negativecross-correlation coefficients appear to decrease with increasingproton flux. For the 5–10 MeV channel this behaviour is not aspronounced.

We note that there are not only cases with significant negativecorrelation coefficients, but also a few cases with positivecorrelation coefficients significant at the 95% confidence level.All other cases with positive correlation coefficients fail thestatistical test.

In order to examine the possible influence of smoothing andthe choice of energy channel on the derived cross-correlationsand time lags, we performed the analysis for all individualproton channels (2–7) corresponding to the energy rangesE¼ ½5210,10230,30250,50260,602100,4100 MeV� with andwithout smoothing. The smoothing applied was a convolution

Table 3Analysis results for the SPEs that occurred during the northern hemisphere NLC

seasons for proton energy channel 2 (5–10 MeV) and smoothing with a 3-day box-

car, which have passed the statistical test.

Day (drs) Year Data Ech. rc src tc (d) stc rcrit b ID

40 1990 NLC 2 +0.65 0.04 5.60 0.01 0.44 0.43 –

40 1990 ALB 2 +0.58 0.03 4.61 0.02 0.38 0.44 –

35 1990 ALB 2 �0.51 0.08 0.39 0.02 �0.43 0.41 –

10 1991 NLC 2 �0.56 0.04 1.12 0.01 �0.44 0.54 A

10 1991 ALB 2 +0.55 0.23 6.59 8.20 0.37 0.54 A

17 1991 NLC 2 �0.84 0.11 8.48 0.01 �0.40 0.56 A

17 1991 ALB 2 �0.51 0.02 3.27 0.02 �0.44 0.52 A

4 1992 ALB 2 �0.58 0.04 4.26 0.08 �0.50 0.70 B

25 2000 NLC 2 �0.66 0.03 �0.50 0.01 �0.47 0.60 C

25 2000 ALB 2 �0.70 0.04 0.31 0.04 �0.42 0.65 C

22 2002 NLC 2 �0.55 0.07 5.24 0.15 �0.42 0.68 D

22 2002 ALB 2 �0.58 0.04 3.18 0.08 �0.47 0.74 D

33 2002 NLC 2 �0.59 0.06 1.26 0.12 �0.42 0.34 D

33 2002 ALB 2 +0.54 0.21 2.00 7.36 0.40 0.35 D

34 2004 NLC 2 �0.59 0.10 7.69 0.42 �0.35 0.49 E

34 2004 ALB 2 �0.65 0.12 7.81 0.52 �0.36 0.49 E

23 2005 NLC 2 �0.57 0.13 9.79 0.16 �0.42 0.64 F

23 2005 ALB 2 �0.58 0.18 3.68 5.07 �0.43 0.51 F

38 2005 NLC 2 +0.63 0.19 10.12 2.10 0.39 0.66 F

38 2005 ALB 2 �0.73 0.02 1.44 0.02 �0.50 0.56 F

drs stands for day relative to solstice. NLC (ALB) refers to the NLC occurrence rate

(albedo) time series.

N. Rahpoe et al. / Journal of Atmospheric and Solar-Terrestrial Physics 73 (2011) 2073–2081 2079

with a 3-day box-car function. Tables 1 and 2 present the analysisresults for the July 2000 SPE, and the January 2005 SPE,respectively. As expected, the correlation coefficients as well asthe lag-1 autocorrelation values b increase if the smoothing isapplied. Note that the correlation coefficients are statisticallysignificant at the 95% confidence level even for the unsmoothedtime series. For the July 2000 SPE (Table 1) the time lag tc isalmost independent of the energy channel used and not stronglyaffected by the smoothing, either. For the January 2005 SPE(Table 2) the smoothing and the choice of energy channel have astronger influence on the derived time lags. For the unsmoothedcase the time lag decreases with increasing energies. With the3-day smoothing, the time lag remains nearly constant (tc � 526days) throughout all energy ranges. For the January 2005 SPEthere is a general tendency of decreasing correlation coefficientsfor increasing energy.

In Tables 3 and 4, the derived parameters for the cases thatpassed the statistical test are presented. Note that only correla-tions with proton energy channel 2 (5–10 MeV) are shown. Formost SPEs the proton fluxes in the higher energy channels areonly weakly enhanced or not enhanced at all during the SPE,rendering a correlation analysis impossible. Only for the twostrongest SPEs (July 2000 and January 2005, see Tables 1 and 2)the proton fluxes were elevated for all proton energy channels.The time lag seems to differ for NLC albedo and NLC occurrencerate data for some cases, in particular the seasons 1991 and 2002

Table 1Correlation results of NLC albedo time series and proton fluxes for different proton

energies and for the July 2000 SPE.

E (MeV) rc src tc (d) stc rcrit b Smoothing

5–10 �0.59 0.01 0.51 0.01 �0.33 0.30 N

5–10 �0.70 0.02 0.31 0.01 �0.42 0.64 Y

10–30 �0.59 0.01 0.51 0.01 �0.33 0.30 N

10–30 �0.70 0.02 0.34 0.01 �0.44 0.64 Y

30–50 �0.59 0.01 0.51 0.01 �0.33 0.30 N

30–50 �0.74 0.03 0.51 0.01 �0.46 0.66 Y

50–60 �0.74 0.02 0.64 0.01 �0.38 0.30 N

50–60 �0.74 0.03 0.63 0.01 �0.47 0.66 Y

60–100 �0.62 0.01 0.68 0.02 �0.38 0.30 N

60–100 �0.74 0.03 0.74 0.01 �0.47 0.66 Y

4100 �0.59 0.01 1.51 0.01 �0.34 0.30 N

4100 �0.73 0.02 0.93 0.01 �0.46 0.66 Y

r c is the cross-correlation coefficient, src is standard error of r c , t c is the time lag,

stc is its standard error. rcrit denotes the critical correlation coefficient, and bcorresponds to the lag-1 autocorrelation value of the NLC albedo time series. The

last column indicates whether the NLC albedo and the proton flux time series were

smoothed (by a 3-day box-car function) or not.

Table 2Same as Table 1 but for the January 2005 SPE.

E (MeV) rc src tc (d) stc rcrit b Smoothing

5–10 �0.65 0.03 6.27 0.02 �0.35 0.29 N

5–10 �0.81 0.03 6.86 0.02 �0.36 0.56 Y

10–30 �0.64 0.01 4.45 0.01 �0.36 0.29 N

10–30 �0.81 0.03 6.82 0.02 �0.36 0.56 Y

30–50 �0.53 0.01 4.52 0.01 �0.35 0.29 N

30–50 �0.81 0.03 6.82 0.02 �0.36 0.56 Y

50–60 �0.42 0.02 2.97 0.71 �0.33 0.29 N

50–60 �0.76 0.03 6.37 0.03 �0.40 0.56 Y

60–100 +0.57 0.01 0.48 0.01 0.30 0.29 N

60–100 �0.63 0.01 5.18 0.03 �0.39 0.56 Y

4100 �0.46 0.02 3.59 0.01 �0.35 0.29 N

4100 �0.70 0.01 4.33 0.01 �0.42 0.56 Y

Table 4Same as Table 3 but for the SPEs that occurred during the southern hemisphere

NLC seasons.

Day (drs) Year Data Ech. rc src tc (d) stc rcrit b ID

13 1987/88 NLC 2 �0.78 0.03 �0.50 0.01 �0.44 0.61 G

13 1987/88 ALB 2 �0.68 0.02 2.94 0.01 �0.41 0.56 G

40 1990/91 ALB 2 �0.56 0.06 5.80 0.04 �0.37 0.61 –

5 2001/02 NLC 2 �0.66 0.20 8.27 2.13 �0.40 0.56 –

5 2001/02 ALB 2 +0.80 0.09 10.03 2.01 0.42 0.45 –

10 2001/02 NLC 2 �0.53 0.12 7.58 0.02 �0.39 0.44 –

10 2001/02 ALB 2 +0.67 0.08 10.94 0.18 0.18 0.53 –

21 2001/02 NLC 2 �0.45 0.07 5.14 7.07 �0.39 0.58 –

21 2001/02 ALB 2 �0.55 0.13 7.87 6.29 �0.42 0.55 –

29 2004/05 NLC 2 �0.70 0.12 5.98 1.12 �0.39 0.56 H

29 2004/05 ALB 2 �0.75 0.09 6.06 1.08 �0.36 0.56 H

in the NH and 1987/88 in the SH. For the January 2005 and July2000 SPEs the calculated parameters show no deviation.

The overall results for both the NLC albedo and occurrence ratetime series are presented in Table 5 and are as follows: In thenorthern hemisphere a total of 20 SPEs occurred during the NLCseasons, 12 of which occurred during the core NLC season (days0–40 with respect to solstice). Of the 13 SPEs occurring duringsouthern hemisphere NLC seasons, only seven occurred during thecore NLC season. Here we took all the relevant SPEs that occurredbetween the years 1986–2008. The correlation coefficients for theNLC albedo time series are statistically significant at the 95%confidence level for eight out of 12 cases for energy channel 2 andthe smoothed time series in the northern hemisphere and four outof seven cases for channel 2 and the smoothed case in thesouthern hemisphere. Using unsmoothed time series generallyleads to a smaller number of significant cases. If the NLCoccurrence rate time series are used, then similar numbers ofsignificant cases are obtained, as presented in Table 5. We alsonote that the number of statistically significant cases decreaseswith increasing proton energy. This is likely related to the factthat only for the strong SPEs the proton fluxes are enhanced forthe higher energy channels.

Table 6The correlation coefficients Rc and time lags Tc are averages for all cases with

statistically significant negative correlation coefficients.

Hemisphere Data Rc Tc (d)

North NLC �0.62 70.08 4.73 74.11

North ALB �0.61 70.08 3.04 72.44

South NLC �0.62 70.13 5.84 73.65

South ALB �0.64 70.10 4.99 71.42

NLC and ALB refers to NLC occurrence rate time series and NLC albedo time series,

respectively.

Table 5Results of the NLC data analysis.

Energy range (MeV) Channel NSPE Ncore NuNLC ðN�NLC Þ NuALBðN

�ALBÞ

North. hem.

E: 5–10 2 20 12 4 (7) 5 (8)

E: 10–30 3 20 9 4 (6) 4 (6)

E: 30–50 4 20 1 �(1) 1 (1)

E: 50–60 5 20 1 1 (1) 1 (1)

E: 60–100 6 20 1 1 (1) 1 (1)

E: 100+ 7 20 1 1 (1) 1 (1)

South. hem.

E: 5–10 2 13 7 4 (5) 4 (4)

E: 10–30 3 13 6 3 (3) 4 (2)

E: 30–50 4 13 2 1 (1) 2 (1)

E: 50–60 5 13 1 1 (1) 1 (1)

E: 60–100 6 13 1 1 (1) �(1)

E: 100+ 7 13 1 1 (1) 1()

NSPE is the total number of SPEs occurring during all NLC seasons considered

(channel 2); Ncore is the number of SPEs occurring during the all core NLC season

(with proton flux Evprot Z50 counts cm�2 s�1 sr�1) and NuNLC (N*NLC) corresponds to

the number of statistically significant cases with negative correlation coefficients

for unsmoothed (smoothed) NLC occurrence rate time series. The corresponding

numbers for the NLC albedo time series are NuALB (N�ALB).

N. Rahpoe et al. / Journal of Atmospheric and Solar-Terrestrial Physics 73 (2011) 2073–20812080

Table 6 lists the average correlation coefficients Rc and thetime lags Tc for all the SPE cases of NLC occurrence rate and albedotime series. The time lag Tc and the time lag range Tc 7sTc

aresmaller for the albedo time series compared to NLC occurrencerate results. Thus, the time lag is sensitive to the data sets used.

5. Discussion

We now discuss the general implications of the resultspresented so far. The observed behaviour in terms of the January2005 SPE is in line with the observations described by von Savignyet al. (2007b). We also note that the NLC observations with allSBUV instruments operational during the July 2000 SPE (NOAA 11and NOAA 14), and January 2005 SPE (NOAA 16 and NOAA 17),respectively, show similar reductions in NLC occurrence rate andalbedo, making instrumental problems or measurement anoma-lies of an individual instrument unlikely.

As shown in Fig. 2, the NLC occurrence rates are quite variableeven during times without SPEs, which makes the establishmentof a causal relationship between proton precipitation andsignatures in the NLC time series difficult in general. Any singlepoint shown in Fig. 10 may be spuriously introduced by the highvariability of NLCs observed during most seasons – particularly inthe southern hemisphere – and may not be related to SPEs at all.However, this variability should also lead to positive cross-correlations. The fact that only few statistically significantpositive correlation coefficients are observed for all SPEs that

occurred during the core NLC season of the years 1986–2008 mayindicate that the depletion of NLCs by an SPE is not an unusualphenomenon, but occurs frequently. Therefore, the effectdescribed in von Savigny et al. (2007a) may not be an exception.

The depletion of NLCs during and after SPEs is also indirectlysuggested by the available model simulations of the effects ofprecipitating and highly energetic solar protons (Krivolutsky et al.,2006; Jackman et al., 2007; Becker and von Savigny, 2010), whichshow an increase in the summer mesopause temperatures as aconsequence of an induced dynamical effect. Krivolutsky et al.(2006) investigated the impact of the July 2000 SPE on atmo-spheric temperature and dynamics. Their model produced atemperature increase near the polar summer mesopause of morethan 15 K. Near 60 km, where the catalytic O3 depletion due toSPE-produced HOx is strongest a temperature decrease of up to8 K is modelled. Although Jackman et al. (2007) and Krivolutskyet al. (2006) presented model simulations for different SPEs, andthe absolute values of the modelled summer mesopause tem-perature increases are different, both model studies are inqualitative agreement, i.e., both show essentially no temperaturechanges in the winter hemisphere, but (a) a temperature decreasein the mesosphere caused by a reduction in solar heating due tocatalytic ozone destruction, (b) a dynamically induced apparentheating of the summer mesopause region caused by a reduction inadiabatic cooling due to smaller vertical (upward) motion.Unfortunately, there are no published model simulations of thetemperature response of the polar summer mesopause during theJanuary 2005 SPE.

We must also point out that satellite observations of NLCsduring the July 2000 SPE were also performed by SNOE (StudentNitric Oxide Explorer) (Bailey et al., 2005) and the SNOE NLCoccurrence rates do not exhibit a very strong signature in mid-July2000 that may be a consequence of the energetic particleprecipitation. This may be explained by the higher sensitivity ofSNOE in comparison to the SBUV/2 instruments, which can detecteven faint NLCs, whereas SBUV/2 is only capable of detecting thebrightest NLC only. It cannot be excluded, though, that a decline ofthe NLC occurrence rate at 70753N around July 14 may be relatedto the SPE. Interestingly, Bailey et al. (2005) find a discontinuity inNLC altitude around the time of the July 2000 SPE, with a suddendecrease of NLC altitudes of about 1 km, but it is not clear howSPE-induced effects may introduce a discontinuity in NLC altitude.This effect is possibly due to an SPE-induced effect on the tangentheight retrievals based on the Rayleigh-knee in the limb radianceprofiles at 215 nm (Merkel et al., 2001).

6. Concluding remarks

The effect of energetic solar protons associated with SPEs onNLCs was investigated using the SBUV NLC data set (covering theperiod from 1978 to 2008) and GOES SEM proton flux measure-ments available since 1986. Unfortunately, SPEs are rare phenom-ena and there were a total of only 12 SPEs occurring during thecore NLC season in the northern hemisphere and seven SPE casesin the southern hemisphere for the solar proton energy range of5–10 MeV. Most of these SPEs are associated with a decrease inthe NLC occurrence rate and albedo time series. The meancorrelation coefficient for all cases with statistically significantnegative correlations for the NLC occurrence rate and albedo timeseries is Rc¼�0.62 and the time lag is in the range of Tc¼0–8 days(Table 6). With few exceptions, all of the statistically significantcross-correlation coefficients are negative. But the overall numberof significant cases is limited, i.e., 67% for the northern hemi-sphere and 57% for the southern hemisphere.

N. Rahpoe et al. / Journal of Atmospheric and Solar-Terrestrial Physics 73 (2011) 2073–2081 2081

The evidence provided by the SBUV observations and the physicalmechanism leading to a mesopause temperature increase during SPEsestablished with model simulations (see Becker and von Savigny,2010) may suggest that a depletion of NLCs during SPEs is likely afrequent phenomenon. However, considering (a) the high intrinsicvariability of the thermal conditions at the polar summer mesopause(see Fig. 2); (b) the fact that for most of the SPEs the observedreduction in NLC occurrence rate or albedo is not very strong, wecannot claim that our study shows that NLCs are affected by everySPE. The reasons for why this effect has not been observed muchearlier are probably the limited number of SPEs during NLC seasons,and the relatively large intrinsic variability of the summer mesopauseregion, particularly in the southern hemisphere. Further and detailedmodel simulations are required in order to quantify the differentcontributions of direct heating by dissipating gravity waves, heating/cooling by advection and adiabatic heating due to a dynamicallyinduced reduction in adiabatic cooling near the summer mesopause.NLCs may then be employed to study dynamically inducedtemperature responses to energetic particle precipitation.

Acknowledgements

We would like to thank Drs. C. Jackman and E. Becker forhelpful discussions, and the GOES team for providing the protonflux data. This work was supported by the Deutsche Forschungs-gemeinschaft (DFG) within the priority program CAWSES (Climateand Weather of the Sun-Earth System), the German AerospaceCenter (DLR) and the University of Bremen.

References

Banks, P.M., 1979. Joule heating in the high-latitude mesosphere. Journal ofGeophysical Research 74, 6709–6712.

Bailey, S.M., Merkel, A.W., Thomas, G.E., Carstens, J.N., 2005. Observations of polarmesospheric clouds by the Student Nitric Oxide Explorer. Journal ofGeophysical Research 110, D13203. doi:10.1029/2004JD005422.

Becker, E., Fritts, D.C., 2006. Enhanced gravity-wave activity and interhemisphericcoupling during the MaCWAVE/MIDAS northern summer program 2002.Annales Geophysicae 24, 1175–1188.

Becker, E., 2009. Sensitivity of the upper mesosphere to the Lorenz energy cycle ofthe troposphere. Journal of Atmospheric Science 66, 1434–1449.

Becker, E., von Savigny, C., 2010. Dynamical heating of the polar summermesopause induced by solar proton events. Journal of Geophysical Research115, D00I18 . doi:10.1029/2009JD012561.

Bovensmann, H., Burrows, J.P., Buchwitz, M., Fredrick, J., Noel, S., Rozanov, V.V.,Chance, K.V., Goede, A.P.H., 1999. SCIAMACHY: mission objectives andmeasurement modes. Journal of the Atmospheric Sciences 56, 127–150.

Chesters, D., 1996. GOES I/M databook. /http://rsd.gsfc.nasa.gov/goes/text/goes.databook.htmlS.

DeLand, M.T., Shettle, E.P., Thomas, G.E., Olivero, J.J., 2003. Solar backscatterultraviolet (SBUV) observations of polar mesospheric clouds (PMCs) over twosolar cycles. Journal of Geophysical Research 108 (D8), 8445.

DeLand, M.T., Shettle, E.P., Thomas, G.E., Olivero, J.J., 2006. A quarter-century ofsatellite polar mesospheric cloud observations. Journal of Atmospheric andSolar-Terrestrial Physics 68, 9–29.

DeLand, M.T., Shettle, E.P., Thomas, G.E., Olivero, J.J., 2007. Latitude-dependentlong-term variations in polar mesospheric clouds from SBUV version3 PMC data. Journal of Geophysical Research 112, D10315. doi:10.1029/2006JD007857.

Ebisuzaki, W., 1997. A method to estimate the statistical significance of acorrelation when the data are serially correlated. Journal of Climate 10 (9),2147–2153.

Hervig, M., Thompson, R.E., McHugh, M., Gordley, L.L., Russell III, J.M.,Summers, M.E., 2001. First confirmation that water ice is the primary

component of polar mesospheric clouds. Geophysical Research Letters 28,971–974.

Jackman, C.H., Roble, R.G., Fleming, E.L., 2007. Mesospheric dynamical changesinduced by the solar proton events in October–November 2003. GeophysicalResearch Letters 34, L04812.

Johnson, R.M., Luhmann, J.G., 1993. Poker Flat MST radar observations ofhigh latitude neutral winds at the mesopause during and after solarproton events. Journal of Atmospheric and Solar-Terrestrial Physics 55,1203–1218.

Karlsson, B., Kornich, H., Gumbel, J., 2007. Evidence for interhemispheric strato-sphere–mesosphere coupling derived from noctilucent cloud properties.Geophysical Research Letters 34, L16806. doi:10.1029/2007GL030282.

Krivolutsky, A.A., Klyuchnikova, A.V., Zakharov, G.R., Vyushkova, T.Yu., Kuminov, A.A.,2006. Dynamical response of the middle atmosphere to solar proton event of July2000: three-dimensional model simulation. Advances in Space Research 37,1602–1613.

Merkel, A.W., Barth, C., Bailey, S., 2001. Altitude determination of ultravioletmeasurements made by the Student Nitric Oxide Explorer. Journal ofGeophysical Research 106 (A12). doi:10.1029/2001JA001111.

Merkel, A.W., Thomas, G.E., Palo, S.E., Bailey, S.M., 2003. Observations of the 5-dayplanetary wave in PMC measurements from the SNOE satellite. GeophysicalResearch Letters 30 (4), 1194–1197.

Porter, H.S., Jackman, C.H., Green, A.E.S., 1976. Efficiencies for production of atomicnitrogen and oxygen by relativistic proton impact in air. Journal of ChemicalPhysics 65, 154.

Rohen, G.J., von Savigny, C., Sinnhuber, M., Eichmann, K-U., Kaiser, J.W., Llewellyn, E.J.,Rozanov, A., Bovensmann, H., Burrows, J.P., 2006. Impact of the october/november2003 solar proton events on mesospheric ozone: SCIAMACHY measurements andmodel results. Journal of Geophysical Research 110, A09S39. doi:10.1029/2004JA010984.

Robert, C.E., von Savigny, C., Rahpoe, N., Bovensmann, H., Burrows, J.P., DeLand, M.T.,Schwartz, M.J., 2010. First evidence of a 27-day signature in noctilucent cloudoccurrence frequency. Journal of Geophysical Research 115, D00I12.doi:10.1029/2009JD012359.

Roble, R.G., Emery, B.A., Garcia, R.R., Killeen, T.L., Hays, P.B., Reid, G.C., Solomon, S.,Evans, D.S., Spencer, N.W., Brace, L.H., 1987. Joule heating in the mesosphereand thermosphere during the 13 July 1982 solar proton event. Journal ofGeophysical Research 92 (A6), 6083–6090.

Shettle, E.P., DeLand, M.T., Thomas, G.E., Olivero, J.J., 2009. Long term variations inthe frequency of polar mesospheric clouds in the northern hemisphere fromSBUV. Geophysical Research Letters 36, L02803. doi:10.1029/2008GL036048.

Solomon, S., Rusch, D.W., Gerard, J.-D., Reid, G.C., Crutzen, P.J., 1981. The effect ofparticle precipitation events on the neutral and ion chemistry of the middleatmosphere, II, odd hydrogen. Space Science 29, 885–892.

Solomon, S., Reid, G.C., Rusch, D.W., Thomas, R.J., 1983. Mesospheric ozonedepletion during the solar proton event July 13, 1982, part II, comparisonbetween theory and experiment. Geophysical Research Letters 10, 257–260.

Thomas, G.E., Olivero, J.J., 1989. Climatology of polar mesospheric clouds 2. Furtheranalysis of solar mesosphere explorer data. Journal of Geophysical Research 94(D12), 673–681.

Thomas, G.E., Olivero, J.J., Jensen, E.J., Schroder, W., Toon, O.B., 1989. Relationbetween increasing methane and the presence of ice at the mesopause. Nature338, 490–492.

Thomas, G.E., Olivero, J.J., 2001. Noctilucent clouds as possible indicators of globalchange in the mesosphere. Advances in Space Research 28, 937–946.

von Savigny, C., Robert, C., Bovensmann, H., Burrows, J.P., Schwartz, M.J., 2007a.Satellite observations of the quasi 5-day wave in noctilucent clouds andmesopause temperatures. Geophysical Research Letters 34, L24808.doi:10.1029/2007GL.

von Savigny, C., Sinnhuber, M., Bovensmann, H., Burrows, J.P., Kallenrode, M.-B.,Schwartz, M., 2007b. On the disappearance of noctilucent clouds during theJanuary 2005 solar proton events. Geophysical Research Letters 34, L02805.

von Zahn, U., Baumgarten, G., Berger, U., Fiedler, J., Hartogh, P., 2004. Noctilucentclouds and the mesospheric water vapour: the past decade. AtmosphericChemistry and Physics 4, 2449–2464.

von Zahn, U., von Cossart, G., Fiedler, J., Rees, D., 1998. Tidal variations ofnoctilucent clouds measured at 693N latitude by groundbased lidar. Geophy-sical Research Letters 25 (9), 1289–1292.

von Zahn, U., 2003. Are noctilucent clouds truly a miners canary of global change?EOS Transactions American Geophysical Union 84 (28), 261–268.

Waters, J.W., et al., 2006. The earth observing system microwave limb sounder EOSMLS on the Aura satellite. IEEE Transactions on Geoscience and RemoteSensing 44, 1075–1092.

Witt, G., 1962. Height, structure and displacements of noctilucent clouds. TellusXIV (1), 118.