Download - Wiley Online Library

JOURNAL OF GEOPHYSICAL RESEARCH: SPACE PHYSICS, VOL. 118, 3898–3908, doi:10.1002/jgra.50370, 2013
Is thermospheric global cooling caused by gravity waves?
W. L. Oliver,1,2 S.-R. Zhang,3 and L. P. Goncharenko3
Received 15 February 2013; revised 17 April 2013; accepted 30 May 2013; published 27 June 2013.
[1] We analyze ion temperature data near 350 km altitude over the years 1966–2012 to
seek explanations for three outstanding questions concerning the long-term cooling
observed in the upper thermosphere: (1) Why is the cooling so much larger than
expected, (2) why has the cooling lasted so long, and (3) why is the thermospheric
density response to the cooling so small? We speculate that gravity waves may cause this
cooling and provide answers to these questions. Recent simulations have shown that
gravity waves are expected to cool the upper thermosphere by an amount comparable to
that observed over our data timeline. A gravity wave proxy formed from the nontidal
fluctuations in temperature shows a positive long-term trend throughout its timeline,
consistent with the increasing cooling observed. The time scales of the long-term trend
and its decadal fluctuations are characteristic of the ocean, not the atmosphere. We
suggest that the following scenario may explain these behaviors: (a) the climate regime
shift of 1976–1977 launched slow Rossby waves across the oceans which continue to
propagate to this day, (b) winds over this increasingly corrugated ocean have launched
increasing fluxes of gravity waves into the atmosphere, and (c) these increasing fluxes of
gravity waves have propagated to the thermosphere to produce increasing amounts of
cooling. The strong thermospheric cooling seen would be expected to produce
thermospheric density declines much larger than those observed via satellite drag. These
temperature and density results would be compatible if the turbopause were lowered 4 km
over the time span of observations.
Citation: Oliver, W. L., S.-R. Zhang, and L. P. Goncharenko (2013), Is thermospheric global cooling caused by gravity waves?,
J. Geophys. Res. Space Physics, 118, 3898–3908, doi:10.1002/jgra.50370.
1. Introduction
[2] Manabe and Wetherald [1967] first showed that an
increase in CO2 content in the atmosphere would heat the
atmosphere below about the tropopause (by absorbing IR
radiations from the ground) and cool the atmosphere above
that level (by radiating thermal energy to space). Roble and
Dickinson [1989] calculated the degree of cooling above
60 km altitude that one would expect for a doubling of
greenhouse gas content at that altitude. They estimated a
global mean cooling of 50 K near 350 km altitude. Holt
and Zhang [2008], however, in considering the 1978–2007
database of incoherent scatter radar measurements of ion
temperature collected at 375 km above Millstone Hill (43ı N,
289ı E), found a noontime cooling rate of 47 K per decade,
or a 141 K decline over the span of their measurements.
1
Department of Electrical and Computer Engineering, Boston
University, Boston, Massachusetts, USA.
2
Center for Space Physics, Boston University, Boston, Massachusetts,
USA.
3
Haystack Observatory, Massachusetts Institute of Technology,
Westford, Massachusetts, USA.
Corresponding author: W. L. Oliver, Department of Electrical and
Computer Engineering, Boston University, 8 Saint Mary’s St., Boston,
MA 02215, USA. ([email protected])
©2013. American Geophysical Union. All Rights Reserved.
2169-9380/13/10.1002/jgra.50370
As the CO2 concentration increased only 12% during this
time period, the simulation would estimate only an 6 K
decline. Donaldson et al. [2010] conducted an independent
analysis of the 1966–1987 Saint Santin (45ı N, 2ı E) radar
database and confirmed the Holt and Zhang [2008] finding. In light of this stark factor-of-20 disagreement between
theory and observation, processes other than CO2 cooling
have been sought to explain the observations. Based on the
Saint Santin data, Walsh and Oliver [2011] suggested some
agency of O3 as the cooling source, based on the coincidence
in time between the beginning of the temperature decline
and the beginning of a strong decrease in O3 content in the
lower atmosphere. Laštovička [2012] noted, however, that
the longer 1978–2007 Millstone Hill data showed that the
temperature continued to decline beyond 1994, when the O3
content began a long recovery.
[3] During our efforts to identify the cause of the great
temperature decline in the thermosphere, we have noticed a
correlation between the behavior of that temperature and the
behavior of ENSO (El Niño–Southern Oscillation) activity,
both in the timing of an onset of change, a subsequent linear
trend, and decadal variations about that trend. This recognition has led us to ask if gravity waves, produced by wind
action over the oceans and propagated to the upper atmosphere, may be responsible for cooling that upper region.
Recent simulations [e.g., Yiğit and Medvedev, 2009] show
that gravity waves are expected to cool the thermosphere
3898
OLIVER ET AL.: GLOBAL COOLING OF THE THERMOSPHERE
on the order of 100 K, the order of long-term cooling
observed. We know of no other agency capable of cooling
the thermosphere by that amount long term.
[4] Below we present multidecadal timelines of thermospheric temperature and ENSO behavior, discuss their
behaviors within the climate context of the times, pose the
possibility that gravity waves may serve as the teleconnection between the two regions, and develop a gravity wave
proxy from the thermospheric data that does indeed show the
long-term increase in magnitude needed to cause continuous
cooling.
2. Database
[5] Our database consists of ionospheric incoherent scatter radar ion temperature (Ti ) measurements collected in the
altitude range 325–375 km above Millstone Hill and at the
specific altitude 350 km above Saint Santin. We quote results
of the analyses of the Saint Santin data by Donaldson et al.
[2010] while we perform an extensive new analysis of the
Millstone Hill data.
[6] The Millstone Hill data set consists of 1,756,521 measurements made in the latitude range 42ı N–43ı N during
the period 1970 to mid-2012. The original Holt and Zhang
[2008] analysis of the 1978–2007 Millstone Hill data considered data near noon and found strong cooling throughout
the period. Donaldson et al. [2010] showed that the cooling
trend, while strong for daytime hours, largely disappeared at
night at 350 km above Saint Santin. We have reanalyzed the
entire Millstone Hill data set to look at the trend in the daily
mean temperature using the analysis method of Donaldson
et al. [2010].
[7] Our database has not been fully vetted, and we found
it beneficial to remove damaging entries and adopt robust
methods of analysis to mitigate the effects of bad data.
For example, in this database we find a Ti measurement of
8050 ˙ 7 K, while on the same day and time 4 km away, we
find the measurement of 1683 ˙ 3913 K. We find another
measurement of 3 ˙ 11 K, while on the same day and time
4 km away, we find the measurement of 2562 ˙ 1152 K.
Data with large uncertainties are not damaging as we weight
each data point inversely as the square of its uncertainty in
all fitting and averaging exercises. Data with erroneously
low uncertainties overemphasize the effects of those data in
analysis. Bad data with erroneously low uncertainties can
devastate a fit or average. To avoid the latter cases, we have
not allowed the weight given to any data point to exceed
the median of the weights calculated from the set of data
uncertainties, an application we call “clipped weighting.”
We identify these issues not to criticize the database but to
show why we have adopted these robust analysis methods.
[8] We have used Chauvenet’s criterion to identify and
remove outlier data. Chauvenet’s criterion identifies the minimum number nC of standard deviations from the mean
beyond which we can expect no data points to fall in a
normal distribution. There is a 50% probability that no data
point falls outside of the bounds ˙nC . The criterion seeks
to retain all data within the distribution and to identify only
the outliers lying beyond all expectation. For
p disp a normal
tribution of n data values, nC has the value 2 erf –1 ( n 1/2).
For our database of 1,756,521 data, nC = 5.1. The outliers
can corrupt standard-definition estimates of and . We
have chosen to use the median as a robust estimate of and to base our estimate of on the robust “median absolute deviation (mad)”, which robustly identifies the inner
50% of the p
data about the median. For a normal distribution,
mad = 2 erf –1 (1/2) = 0.6745 , so is approximated by 1.4826 (mad). The criterion should be applied
to the patternless data-minus-fit residuals obtained by fitting
and removing all solar-geophysical patterns of behavior. We
have applied this filtering in an iterative data-fitting/dataremoval manner, with the first filtering assuming zero fit.
Subsequent filterings worked on the “flattened” residuals
obtained by removing the fit from the previous iteration.
We applied the filtering to the logarithm of the Ti values as
the Ti distribution is bounded by zero on the low side but
unbounded on the high side and is more symmetric about
its central value on a logarithm scale. Applied to log Ti , the
process removed the following number of data in successive
iterations: 19,493; 51,443; 6262; 770; 99; 13; 3; and 0, summing to 4% of the data in all. The first iteration removed the
grossest outliers. The second iteration, the first to act with the
fit removed, removed the majority of the outliers. The process then converged rapidly and did not experience any of
the continual erosion of the “high-energy” tail of the distribution that can result when such a process employs a design
that removes data lying within that tail.
[9] A summary of our experience with these robust analysis techniques is that weighting of the data was essential,
that the usage of clipped weighting was essential, and that
the outlier elimination tightened the scatter of the data about
the trendline did but not change any essential feature or
interpretation of any pattern of behavior.
3. Thermospheric Temperature
and ENSO Timelines
[10] Our yearly data trends are shown in Figure 1. The
top panel shows the thermospheric temperatures. The thick
straight lines show fitted linear trendlines. The thin straight
lines show two-segment fits to be discussed later. The Saint
Santin results are the yearly averages of Donaldson et al.
[2010] with linear trendline superposed. The Millstone Hill
results come from a new analysis. We have determined by
least squares fit to the data set the diurnal, seasonal, solaractivity, magnetic-activity, and long-term trends, removed
all but the long-term trend, and then bin-averaged the result
into years to give the results shown. Details of this fitting
procedure may be found in Donaldson et al. [2010]. The
only modification that we have made for application to the
Millstone Hill data set is to add a linear height gradient as
these data cover a 50 km height range whereas the Saint
Santin data were all located at 350 km altitude. Error bars
are included on the temperature data points in Figure 1.
Each error bar is computed as the standard deviation of the
bin values divided by the square root of the bin population and so would represent the uncertainty in the mean if
the values in a bin were independent. The original data are
experimentally independent but they are not geophysically
independent. Owing to the campaign style of data collection,
we lack the complete time series necessary to investigate
that dependence. Emmert and Picone [2011] did investigate
this dependence for satellite density data, for which they
did have complete time series. Their conclusion was that
3899
OLIVER ET AL.: GLOBAL COOLING OF THE THERMOSPHERE
1200
El Chichon Pinatubo
Ti (K)
1150
1100
1050
Millstone Hill
∇
∇
1980
1990
Saint Santin
ENSO MEI Index
1000
950
1
0
−1
−2
1950
1960
1970
2000
2010
Year
Figure 1. Timelines of ion temperature at (top) 350 km altitude and (bottom) ENSO activity. In the top panel, the thick
straight lines show linear cooling trends while the thin twosegment line fits indicate years of possible cooling onset. In
the bottom panel, the dashed line shows a low-pass representation of the ENSO timeline chosen to eliminate El Niño
events while the discontinuous two-segment line fit identifies the climate shift of 1976. The top panel shows the times
of the eruptions of two volcanoes.
the dependence increased trend uncertainty by about 50%.
If we should increase the error bars in Figure 1 by 50%,
they would still be small compared with the year-to-year
fluctuations seen, and we may believe that these year-toyear fluctuations are real geophysical changes and not just
measurement uncertainty.
[11] The cooling trends shown in this top panel have several interesting features. The linear trend of 2.0 K/yr is much
stronger than the 2 K/decade trend expected from greenhouse cooling. The two-segment lines fit to the data identify
breakpoint years at which a cooling onset may have begun
or intensified for both locations. The pre-onset segment of
Millstone Hill data is so short that we were forced to give
that segment zero slope and make the segments discontinuous. The Saint Santin breakpoint year is 1979.1 ˙ 5 years.
The Millstone Hill breakpoint year is 1981.3 ˙ 2 years.
These uncertainties assume independence of the yearly average data. We will see below that these results are consistent
with the timing of a change in the broader climate. There
are decadal-scale fluctuations about the trendlines (seen in
the sufficiently long Millstone Hill time series). Finally, we
have noted the times of two major volcanic eruptions on the
figure. Volcanic eruptions have been shown to affect atmospheric temperature up to at least the mesopause [e.g., She
et al., 1998]. Both Saint Santin and Millstone Hill measured
temperature drops in the thermosphere on the order of 50 K
during 1982, the year of the El Chichón eruption. Unfortunately, Saint Santin did not operate the following year, but
the temperature above Millstone Hill had recovered. Saint
Santin had ceased operation by the time Pinatubo erupted.
The temperature above Millstone Hill was a few tens of
degrees below the trendline for 3 years following this eruption, though it is not clear that the response was due to the
volcanic effects. Delayed responses to volcanic activity have
been noted in the past [e.g., She et al., 1998].
[12] The bottom panel of Figure 1 shows El Niño–
Southern Oscillation (ENSO) activity from 1950 to the
current time as depicted by the Multivariate ENSO Index
(MEI), available at www.esrl.noaa.gov/psd/enso/mei/table.
html [Wolter and Timlin, 1993, 1998]. The monthly values listed online have been averaged into the yearly bins
and shown as data points connected by straight lines. The
index gives departures from the mean, with positive values corresponding to warm El Niño periods and negative
values corresponding to cool La Niña periods. Prominent
features of this time series include a “switch” in long-term
trend from largely negative values to largely positive values in the late 1970s and then a slow relaxation back toward
zero ever since. The discontinuous two-segment line shown
has the time of discontinuity and slopes of the segments
chosen for best least squares fit to the monthly data. The discontinuity occurs in 1976. The 1982–1983 and 1997–1998
El Niño events were the strongest (in peak monthly amplitude) experienced in the historical record that extends back
to 1871 [Wolter and Timlin, 2011]. The dashed line is a
low-pass representation of the monthly data, which we will
discuss below.
[13] There are interesting correlations between the time
series in the two panels of the figure. The beginning of true
thermospheric cooling seems to have begun soon after the
time at which the ENSO index trend switched to positive
values, in the late 1970s. The ensuing linear cooling trend
in thermospheric temperature seems to correspond to the
slow long-term relaxation of the ENSO index (we later question the cause and effect for this correlation). Furthermore,
excursions of the ENSO index seem to anticorrelate with
excursions of the Millstone Hill temperature about its trendline in decadal fashion. But there are shorter-term changes in
the ENSO index, like the large El Niño event of 1997–1998
(a 16 month feature in the monthly data), that show no corresponding fluctuations in temperature. These ocean effects
have large spatial and temporal scales. Latif and Barnett
[1994] note that the ocean response involves the propagation of ocean planetary-scale waves. Trenberth and Hurrell
[1994] note (1) that climate is molded by tropical influences with the sluggish ocean bringing out the long time
scales and planetary waves setting very large spatial scales
and (2) that atmospheric circulation forms the main link
between regional changes in wind, temperatures, precipitation, and other climatic variables. Accordingly, we have
passed the ENSO MEI monthly time series through a lowpass filter, simply by Fourier transforming the time series,
lopping off frequencies above a certain threshold, and transforming back to a time series. The dashed curve in the figure
shows the result of setting that threshold period at 85 13
months, chosen subjectively to eliminate the 1997–1998 feature. This low-pass curve has peaks and valleys that seem
to correspond well, respectively, with the decadal excursions below and above the Millstone Hill trendline. We may
also note that the El Chichón eruption occurred just as the
largest El Niño of the century (as of 1982) was beginning.
The temperatures above Saint Santin and Millstone Hill
fell 100 K during the period 1979–1982, culminating in
this eruption.
[14] Can we argue cause and effect between the behaviors
seen at ground level and at 350 km altitude? We suggest in
the following discussions that gravity waves, caused by air
3900
OLIVER ET AL.: GLOBAL COOLING OF THE THERMOSPHERE
flow over corrugated ocean surfaces, be considered as the
possible teleconnection between these regions. We know of
no means other than gravity waves by which such a large
long-term cooling of the thermospheric may be produced. To
provide possible support for this suggestion, we will develop
and compute a gravity wave proxy from the Millstone Hill
data themselves and show its long-term trend. First, we wish
to review the evidence that gravity waves abundantly propagate from the surface to the thermosphere and cause cooling
there and to discuss the important climate context in which
the cooling took place.
4. Gravity Waves
4.1. Gravity Waves in the Thermosphere
[15] Gravity waves are seen in the ionosphere essentially
anytime an experiment is run to see them [Fritts and Lund,
2011]. While they exhibit a large variety of spatial and temporal scales and can be excited by a variety of sources, like
topography, convection, or wind shear near the surface, it
is generally agreed that primary gravity wave sources are
located in the lower atmosphere. Oliver et al. [1997] say that
the thermosphere over Shigaraki, Japan, seems to be continuously swept by waves that come for hours on end from
a consistent or slowly varying direction. These waves had
periods from 40 to 130 min and horizontal wavelengths from
500 to 1800 km. Djuth et al. [2010] say that these are ever
present above Arecibo, Puerto Rico, and traced their origin back to the Mid-Atlantic Ridge in the Atlantic Ocean.
These waves had a period of about an hour and horizontal
wavelengths of 200 to 500 km. Presumably, since these are
atmospheric bouyancy waves, they are generated by air flow
over structured surfaces, such as might be created over the
Mid-Atlantic Ridge.
[16] While gravity waves have long been known to have
controlling effects on the momentum budget of the middle
atmosphere, they have sometimes been considered to be just
interesting phenomena passing through the thermosphere,
insufficient to modify the background gas appreciably.
Walterscheid [1981] wrote a seminal paper that showed that
while gravity waves may heat the atmosphere below about
110 km altitude, they actually “refrigerate” the ambient gas
at higher altitudes by inducing a divergent downward heat
flux. Yiğit and Medvedev [2009] estimated heating and cooling in the thermosphere due to dissipating and breaking
gravity waves of tropospheric origin in a comprehensive
general circulation model. They found that the net effect
of gravity waves above the turbopause was cooling, conservatively estimated to be 100–170 K at midlatitudes near
their upper boundary of 240 km altitude. So we gain a possible picture of gravity waves launched from the surface,
growing as they propagate to the thermosphere, and draining energy to lower altitudes as they dissipate, producing
thermal effects certainly capable of producing the long-term
cooling observed.
[17] If gravity wave cooling should be responsible for the
thermospheric temperature decline shown in Figure 1, the
wave flux should have the major characteristics exhibited
by the temperature timeline, inverted, of course, as greater
wave activity should cause greater cooling. These characteristics include an onset of increased fluxes in the late 1970s,
increasing fluxes from this onset to the present time, and
decadal modulation of these fluxes following the onset.
4.2. Ocean Surface Structure
[18] Virtually, all of the world ocean is baroclincally
unstable and prone to develop a structured surface [Chelton
et al., 2011]. Chelton and Schlax [1996] discuss the largescale corrugation of the ocean surface and the interpretation
of that structure in terms of Rossby waves. Rossby waves
in the ocean are the large-scale response to wind forcing
and buoyancy forcing at eastern boundaries and over the
ocean interior. The wave restoring force results from the latitudinal variation of the Earth’s angular momentum, so the
waves owe their existence to the spherical shape and rotation
of the Earth. The lowest-order baroclinic wave, believed to
be of primary importance, requires months (near the equator) to decades (at midlatitudes) to cross a large ocean basin
[Chelton and Schlax, 1996]. The TOPEX/Poseidon satellite altimeter showed such wave-like structures to exist over
<
much of the world’s oceans with amplitudes 10 cm and
>
wavelengths 500 km, amplified by major topographic features outside of the tropics. These structures appeared to
originate at eastern boundaries and to be amplified over (and
maybe generated by) ocean topography, such as the Emperor
Seamounts, the Hawaiian Ridge, and the Hess Rise in the
Pacific Ocean (compare with Djuth et al. [2010] identification of the Mid-Atlantic Ridge as the source of atmospheric
gravity waves over Puerto Rico).
[19] While the TOPEX/Poseidon data set alone showed
ocean structure with convenient interpretation in terms of
Rossby waves, the higher resolution of the subsequently
merged TOPEX/Poseidon and ERS-1/2 data sets showed
more compact structures and gave a revised interpretation in
terms of the westward tracks of isolated eddy-like structures
of amplitude 5–25 cm and diameter 100–200 km [Chelton
et al., 2007, 2011]. These structures showed meridional
deflection, dispersion (the lack of), and latitudinal gradient
of scale size more characteristic of eddies than of Rossby
waves. On the other hand, the presence of large-scale, westward propagating curved crests and troughs of sea surface
height indicated that most of the propagating energy at latitudes below 20ı is in the form of Rossby waves. Such telltale
patterns were seen arguably as high as 50ı latitude. Also, the
observed scale-size dependence of propagation speed suggested dispersion and a population of larger, faster Rossby
waves distinct from the population of eddies. This effect
may be evidence of “wave drag” caused by the shedding
of Rossby waves that slows eddy propagation speed. Furthermore, the instability of Rossby waves themselves is a
viable mechanism for the generation of the eddies. So there
is significant evidence that the Rossby waves exist, though
their amplitudes would be small in comparison with that of
the eddies.
[20] Maximenko et al. [2008] used the World Ocean
Database 2005 to identify striations in the Pacific Ocean
having wavelengths of some 400 km and wave amplitudes
of 0.5–1.5 cm. Schlax and Chelton [2008] note that such
striations may be produced by ocean eddies produced at
preferential locations and by flow past marine topography.
Hertzog et al. [2008] find that though flow over terrestrial features may create the most notable waves, ocean
wave momentum fluxes match continental fluxes in zonal
3901
OLIVER ET AL.: GLOBAL COOLING OF THE THERMOSPHERE
averages and are less sporadic. There seems to be plenty
of ocean appropriately structured to produce atmospheric
gravity waves as air flows over it.
[21] We know of no studies of atmospheric gravity waves
produced by airflow over ocean Rossby waves, but studies of
atmospheric gravity waves produced by ocean waves of similar characteristics do exist. Chelton and Schlax [1996] note
that ocean Rossby waves are expected to have wavelengths
of 100s to 1000s of km; this range overlaps that of gravity waves seen in the thermosphere. Lognonne et al. [2006a,
2006b] note that tsunami waves with typical wavelengths
of 200 km and amplitudes as small as 1–2 cm can launch
gravity waves at 50 m/s speed that reach the ionosphere. We
have already noted the Djuth et al. [2010] study that finds
that ocean flow over the Mid-Atlantic Ridge is expected to
produce atmospheric gravity waves of the proper temporal
and spatial scales to explain waves seen in the thermosphere
over Arecibo. That case, as well as the case of air flow over
ocean Rossby waves, involves air flow over nearly stationary ocean surface structure. Djuth et al. [2010] note that
waves so generated may well break in the mesosphere, while
Vadas et al. [2003] show that such breaking is expected
to regenerate waves that propagate into the thermosphere.
Given these results for ocean waves with characteristics not
dissimilar from those of Rossby waves, we suggest that air
flow over ocean Rossby waves is a tenable mechanism for
the production of atmospheric gravity waves that appear in
the thermosphere.
5. The Climate Context
5.1. The Climate Regime Shift of 1976–1977
[22] Differing climate regimes separated by relatively
short transition periods have occurred episodically throughout the climate record. Minobe [1997] lists shifts occurring
in 1889–1890, 1925–1926, 1947–1948, and 1976–1977.
Mantua et al. [1997] independently identified 1925, 1947,
and 1977 as the years for the latter three transitions while
Hidalgo and Dracup [2003] identified years 1924–1925,
1946–1947, and 1976–1977. Of course, shift identification
is somewhat subjective and may identify regional effects.
Namias [1969] identified the inception of a regime in summer/autumn 1961. Dickson and Namias [1976] identified
a shift in winter 1970–1971 when the Greenland anomaly
showed almost total collapse. The shift in 1976–1977 is the
most studied but is by no means unique in the climate record.
Graham [1994] and Latif and Barnett [1996] note that the
cause of the 1976–1977 shift may be associated with the rise
in sea surface temperature (SST) but the cause of the SST
rise is unknown. Trenberth and Stepaniak [2001] go further and say that the shift appears to relate to the change in
the thermocline, the sharp decrease in sea water temperature
below the surface. Timmermann et al. [1999] simulated the
effects of increasing CO2 content in a global climate model,
finding that the equatorial thermocline becomes stronger in
response to greenhouse warming, with SST rising and deepsea temperature falling (owing to a greater inflow of cool
water as atmospheric Hadley circulation intensifies). The
stronger thermocline was found to be responsible for the
most important changes in mean state and variability of the
tropical Pacific Ocean-atmosphere system. Some conclude
that these shifts are aspects of natural internal oscillations
of the ocean [e.g., Minobe, 1997]. Others say that the
1976–1977 shift may have been driven by global warming [e.g., Trenberth and Stepaniak, 2001]. Others say they
do not know [e.g., Graham, 1994; Trenberth and Hurrell,
1994]. Trenberth and Hoar [1996] note that the 1990–
1995 ENSO event that followed the 1976–1977 regime shift
was the longest on record, so long that it was a statistical once-in-2000-years event, giving a small probability
that it occurred naturally and a greater likelihood that
it was forced by global warming. Graham [1995] notes
that the climate record following the climate shift bears a
disquieting resemblance to climatic response to increasing
greenhouse gases in model simulations. Merzylakov et al.
[2009] note that complex nonlinear systems may well transition from one quasi-stable state to another, and the climate
system, certainly nonlinear, has indeed shown these sudden shifts from one state to another. This informed opinion
continues to be debated.
5.2. Decadal Variations
[23] Figure 1 shows clear decadal-scale fluctuations both
at ground level and in the thermosphere. The idea of climate
anomaly “persistence” goes back at least as far as Namias
[1952]. Namias et al. [1988] identified low-frequency fluctuations in SST superposed on linear trends with periods
lasting years to decades. Trenberth [1990], Trenberth and
Hurrell [1994], and Graham [1994] noted the 1977–1988
period of frequent El Niños and no La Niñas and pointed
out that the long-term memory of the sluggish oceans can
act as a low-pass filter and produce such low-frequency
oscillations. Latif and Barnett [1994, 1996] say that the
decadal oscillation is produced by a positive feedback system between the ocean and atmosphere, which are never
in equilibrium, with the time scale being set by the spinup time of the ocean circulation gyres. So the existence of
decadal fluctuations in the ENSO timeline in Figure 1 is not
unexpected and can be viewed as a natural oscillation of
the system.
6. Development of a Gravity Wave Proxy
[24] Offermann et al. [2011] note that as of year 2011,
no conclusive results appear to be available on the longterm trend in gravity wave activity. They developed a
gravity wave (GW) proxy for their mesospheric temperature measurements in terms of the standard deviation about
the mean for a given time interval. Over 16 years,
they found an increase in gravity wave activity of 1.5%
per year for short-period waves of 3–10 min period.
Hoffmann et al. [2011] applied a similar technique to
their mesospheric wind measurements, first removing the
mean wind and tidal components, then computing variance,
focusing on waves with periods of 3 to 6 h. They also found
an increase in gravity wave activity over the time period
1990–2010. The data from their Figure 4 show a positive
trend of 2–6% per year (relative to the trend value in year
2000) for meridional and zonal wind fluctuations between
80 and 88 km altitude.
[25] We form a GW proxy for our thermospheric
temperatures in a manner similar to that used by
3902
OLIVER ET AL.: GLOBAL COOLING OF THE THERMOSPHERE
Hoffmann et al. [2011]. We parse the 1970–2012 Millstone
Hill temperature data into individual days, remove the mean
and tidal components for each day, and take the residual
fluctuation to be the gravity wave component. Let “hr” and
“ht” be, respectively, the numerical values of local time in
hours and height in kilometers. To each day’s set of Ti data,
we fit the formula
2(hr – p3 )
2(hr – p5 )
+ p4 cos
+
24
12
2(hr – p7 )
+ p8 (ht – 350).
p6 cos
8
Ti = p1 + p2 cos
(1)
[26] This formula includes the daily mean, 24 h, 12 h,
and 8 h tides, and a height gradient that is significant over
the 325–375 height span of our data. Parameters p1 –p8 are
determined by least squares fit. Subtracting this fit from
the data gives a set of residuals fit from which we calculate our GW proxy for that day. These residuals contain
a contribution GW from gravity waves and a contribution
noise from measurement error. We compute the gravity wave
contribution as
2
2
2
GW
= fit
– noise
.
(2)
Ideally, noise should be the mean of the Ti uncertainties, the
Ti provided by the database. For reasons discussed below,
we have adopted an alternate calculation for noise .
[27] We have already discussed the initial filtering of bad
Ti values from our database and the particularly damaging effects possible from bad estimates of Ti . The latter
effect bodes particular trouble for our GW proxy, for which
the data uncertainty becomes the “data.” One experiment of
1004 measurements covering 63 h in our database shows a
median Ti error bar of only 9 K while the median pointto-point scatter (absolute difference between neighboring Ti
values in the time series) is 236 K, occurring on a time scale
shorter than the Brunt-Väisälä buoyancy period of the fluid.
We cannot believe that these error bars truly represent data
uncertainty. It is possible for strong spurious signals such as
interference or clutter from other targets to contaminate the
ionospheric return and cause scatter in the deduced temperature while the strong signal may falsely yield a small error
estimate. The ratio between Ti and Ti point-to-point scatter
was seen to vary in a structured way throughout the timeline of years, the structure possibly correlated with changes
in data taking and analysis schemes. Several experiments in
the early 1980s show scatter many times the Ti values.
Most experiments since about year 2005 show scatter only
about 40% of the Ti . For purposes of our GW proxy, we
have chosen not to represent Ti uncertainty by the Ti given
in the database but to represent it by the point-to-point scatter in the temperature-minus-fit residuals noted above. Such
a choice has the advantages of basing data scatter on the
data themselves and of avoiding cases of poorly estimated
data uncertainty or any possible question of what the uncertainty estimate was intended to represent. Let ri , i = 1 : : : n,
be the n data-minus-fit residuals for a day’s data. The calculation si = ri – (ri–1 + ri+1 )/2 (end points omitted) may
be viewed as a measure of a data point’s deviation from
its immediate background. Application of this calculation to
a long sequence of normally distributed random numbers
having unity variance yields a sequence with a variance of
2
1.5. We therefore calculate noise
as the variance of s divided
by 1.5.
[28] We may question whether point-to-point
temperature-minus-fit differences are in part caused by the
very gravity waves we wish to detect. If we place our entire
database in chronological order, we find that the median
point-to-point time difference has the initially surprising
value of zero. This result means that most of our times
of measurement have multiple simultaneous altitudes of
measurement. Note that to the first order, we have removed
temperature height gradient from the residuals used to
compute the variance, so we actually usually have multiple
measurements of temperature at the same time. The happenstance is actually more favorable than having measurements
closely spaced in time. Not all times of measurements have
multiple heights of measurement in our chosen height range.
The Brunt-Väisälä buoyancy period for the mean Ti value
of 931 K for our database is 14 min. We find that 98.5%
of the point-to-point time differences in our database are
less than 14 min. We take this result to mean that our GW
proxy does not lose significant gravity wave information in
its calculation.
[29] We have parsed our database into 1232 days having
no data gap exceeding 12 h. Twelve hours of data is not long
enough to determine the mean and tidal components independently with high accuracy, but the short days yield such
large uncertainties in these determinations that they do not
adversely affect the yearly weighted averages that we will
show. Of our 1,756,521 data, 1,412,144 were included in
these 1232 days while the others were from day fragments
not usable or end pieces not needed. We also applied iterative Chauvenet editing to each individual day, eliminating
5% of the 1232 day database.
[30] As an endnote on our GW proxy development, we
will note two analysis modifications that we tested. We
included a 6 h tide in equation (1), and while we found
it to be statistically significant, it was weaker than the 8 h
tide and made an almost undetectable difference in the GW
proxy timeline that we show below. We also limited the time
gap allowed (in the ri–1 –ri+1 time difference discussed above)
in the GW proxy calculation. While this limitation eliminated increasing numbers of data points as the time limit was
reduced, it had little effect on the yearly average GW proxy
timeline that we will show at 30 min limit but fully eliminated all data from 1970 to 1975 at 15 min limit. Even at
a 3 min limit, the GW proxy timeline retained its long-term
pattern well and lost only one more year, 1976. From these
results we conclude that time sampling is not a major issue
for our GW proxy calculation.
[31] The day-by-day analysis gives a set of 1232 values
for the mean temperature, for the amplitudes and phases
of the 24 h, 12 h, and 8 h tides, and for the height gradient. Each set has variations with solar and magnetic activity
and with season. To separate these effects, we have fit
each set with a formula to model all effects. Let F be the
numerical value of the daily solar 10.7 cm flux intensity
in the usual units of 10–22 W m–2 s–1 , F be the centered
three-solar-rotation (81 day) average of F, Ap be the usual
daily magnetic activity index, d be the day number of the
year, and y be the year (including fractional part). Then
3903
dTi/dh (K/km)
GW Proxy (K2)
8−hour amp (K) 12−hour amp (K) 24−hour amp (K) Daily meanTi (K)
OLIVER ET AL.: GLOBAL COOLING OF THE THERMOSPHERE
1400
1200
1000
800
250
200
150
100
50
60
40
20
60
40
20
0
2500
2000
1500
1000
500
0
Year
1.5
1
0.5
0
1970
1975
1980
1985
1990
1995
2000
2005
2010
Year
Figure 2. Timelines of daily mean temperature, diurnal amplitude, semidiurnal amplitude, terdiurnal amplitude, a gravity wave proxy, and Ti height gradient computed from analysis of 1232 days
of observation.
we model each of our sets of analysis results according to
the expression:
q1 + q2 (y – 1991) + q3 (Ap – 14)+
q4 (F – 134) + q5 (F – 134)2 + q6 (F – F) + q7 (F – F)2 +
q8 cos(2d/365.2422) + q9 sin(2d/365.2422)+
q10 cos(2d/365.2422)(F – 134)+
q11 sin(2d/365.2422)(F – 134)+
(3)
q12 cos(4d/365.2422) + q13 sin(4d/365.2422)+
q14 cos(4d/365.2422)(F – 134)+
q15 sin(4d/365.2422)(F – 134),
to find the mean (q1 ), long-term linear trend (q2 ), linear
magnetic-activity dependence (q3 ), quadratic solar cycle
dependence (q4 –q5 ), quadratic solar rotation dependence
(q6 –q7 ), annual dependence (q8 –q11 ), and semiannual dependence (q12 –q15 ). The seasonal terms are allowed solar cycle
dependences in amplitude and phase. F = 134 is the mean F
value for our 1232 days of data while Ap = 14 is the mean
Ap value.
[32] We wish to show our data sets for daily mean, tidal
amplitudes, GW proxy, and Ti height gradient binned into
years. We first subtract the short-term variations of the model
(magnetic activity, solar rotation, and seasonal dependences)
in order to show the long-term trend and solar cycle dependences. The solar rotation and seasonal effects are largely
averaged out when we bin results into years, but all years
do not necessarily have balanced solar rotation and seasonal
contributions. Additionally, we have subtracted the solar
cycle dependence for the GW proxy and height gradient
plots. That dependence is very small for the GW proxy; it is
expectedly sizable for the height gradient but not an item of
interest in our discussions. Figure 2 shows those daily mean,
tidal amplitude, GW proxy, and height gradient results. The
straight line on each plot is its long-term trend (q1 and q2
from the fit). Let us first discuss the daily mean and diurnal
harmonics. The daily mean not only shows the large solar
3904
OLIVER ET AL.: GLOBAL COOLING OF THE THERMOSPHERE
cycle effect but also shows a continual cooling over time
of 2.1 K/yr. By comparison, the global analysis shown in
Figure 1 gave a rate of 2.0 K/yr and a trendline temperature
only 2 K different at the middle of the time span. Figure 2
would also show the decadal oscillations evident in Figure 1
if the solar cyle effect were removed. Akmaev [2012], noting that these incoherent scatter temperatures decline much
faster than greenhouse gas increases can explain and much
faster than is consistent with the decline observed in thermospheric densities, urged community reanalysis of the radar
data. Our day-by-day analysis may qualify as that reanalysis. The long-term cooling is evident and strong. The
24 h amplitude also shows strong solar cycle modulation
and so may be strongly related to the solar source. The 12 h
and 8 h amplitudes show weak solar modulation. We ask
if this implies that these two components are controlled by
tides propagating from lower altitudes under weak solar control. The diurnal harmonics show outlier points even in the
yearly averages. The temperature results for years 1982 and
1983 are known to be faulty owing to the failure of the
raw-data analysis program to account properly for spectral
aliasing [M. J. Buonsanto, 1997, personal communication].
The year 2003 saw only four experiments run due to the
ongoing development and testing of a new data acquisition
system. As some of these experiments reflect an intermediate
stage of radar data processing, they require experimentspecific evaluation and calibration. In addition, two of these
periods experienced magnetic Ap indices exceeding 100.
Year 2003 is an outlier on the diurnal-harmonic plots. It
is such an outlier on the GW proxy plot that it is missing
off scale.
[33] The GW proxy is shown in the fifth panel of Figure 2.
We would wish to see some correspondence between this
GW trend and the temperature trend shown in Figure 1,
including an onset in the late 1970s, a constant progression
thereafter, and decadal fluctuations. By covering points to
the right, we see no clear long-term trend until at least the
late 1970s. We also fitted an alternate trendline to this GW
proxy timeline, one for which the trend was constrained to be
nil until a breakpoint year and thereafter began its constant
rate of rise. The breakpoint year so determined was 1974.4˙
6 years, perhaps confirming the existence of a change in
slope in the decade of the 1970s. The GW proxy values for
1973–1977 are unexpectedly negative. We can only suggest
that excessive noise was subtracted in equation (2) because
the noise was not of the random character expected and
the factor 1.5 used to compute the noise variance was too
small. It may be that the entire GW proxy trendline should
be shifted upward slightly on the plot. The fitted trendline clearly shows an increase in the GW proxy continuing
to the current day. It shows a slope of 4.7%/yr (for year
2000, chosen to match reference year chosen for the corresponding calculation with the Hoffmann et al. [2011] data).
This compares with the 1.5%/yr reported by Offermann
et al. [2011] and the 2–6%/yr computed from the data of
Hoffmann et al. [2011] for mesospheric temperature and
winds. There are also decadal fluctuations in the GW proxy.
The major decrease of 1995–2001 mirrors a period of temperature increase above the trendline shown in Figure 1.
Other such correspondences can be seen in later and earlier years. While the GW-temperature anticorrelation is not
fully realized everywhere, there is enough correspondence
in general and during particular time periods, we believe, to
suggest cause and effect.
[34] The sixth panel of Figure 2 shows the long-term trend
of the Ti height gradient at 350 km altitude. The scatter
in these yearly gradient results is likely due to the variety
of operating modes used through the years and the resulting variety of altitudes used in the gradient calculation.
While the daily mean temperature has decreased by less than
10% over the observing time span, the temperature gradient has decreased by more than 50%. This result may give
some evidence that heat transfer from high to low altitude is
increasing with time, as one might expect from a condition
of increasing gravity wave cooling.
7. Discussion of the Thermosphere/Gravity
Wave/ENSO Connection
[35] Let us now review Figures 1 and 2 within the climate context of the times and under the assumption of a
gravity wave teleconnection between behavior at the surface
and behavior in the thermosphere. Our working hypothesis is that winds over the wave-corrugated ocean created
the atmospheric gravity waves, which propagated to the
thermosphere and cooled the air there. According to this
hypothesis, the gravity wave flux and thermospheric temperature would have been stable before the climate regime shift
of 1976–1977. The climate regime shift would create a step
increase in ocean forcing, and the oceans would respond by
launching Rossby waves westward from eastern boundaries,
corrugating the ocean as they propagated. The time scale for
the oceans to fill with Rossby waves is a few decades, so
the corrugation would be overspreading the oceans even to
the present day. The increasingly corrugated ocean would
launch increasing fluxes of gravity waves into the atmosphere. This scenario would make the thermospheric cooling
a long, transient process that would end when the Rossby
waves have fully covered the oceans. The temperature timeline seen in Figure 1 and the gravity wave timeline seen in
Figure 2 meet these expectations in terms of long-term trend
and perhaps in the identification of period of onset, though
variability in the data may make the identification of an onset
more subjective.
[36] The fluctuations of thermospheric temperature about
its trendline are anticorrelated with both the decadal fluctuations in ENSO activity and fluctuations in the gravity
wave proxy. Presumably, this would mean that high ENSO
activity corresponds to either increased ocean corrugation or
increased wind speeds over existing corrugation. As wind
speed is part of the ENSO MEI index, it provides a natural
explanation. Concerning the ocean-corrugation explanation,
we have already noted the statement of Latif and Barnett
[1994] that the decadal-scale period is set by the spin-up time
of ocean circulation. Perhaps we should not discount the
possibility that such forcing in circulation may also launch
Rossby waves from forcing boundaries.
[37] But if thermospheric temperature fluctuations are
anticorrelated with ENSO decadal activity, then why is the
long-term temperature trend positively correlated with the
post-regime-shift long-term “relaxation” in ENSO activity?
We can only speculate here that of two competing factors
for gravity wave generation, the increasing Rossby wave
corrugation of the oceans and the decreasing winds speeds
3905
OLIVER ET AL.: GLOBAL COOLING OF THE THERMOSPHERE
indicated by the ENSO MEI index, the former must have
been the more important.
[38] We are aware of the body of literature on the filtering
of gravity waves by winds and recognize that a long-term
trend in gravity waves may be caused by a long-term trend in
the winds (and vice versa). Indeed, there are reports of longterm trends in the winds in the mesosphere [e.g., Hoffmann
et al., 2011; Ratnum et al., 2013] and in the thermosphere
[e.g., Brum et al., 2012] that may bear upon the question of
the long-term trend in gravity waves. We do not discount this
wind trend explanation for our gravity wave observations,
but owing to the observed correlations between behavior
at sea surface level and behavior in the thermosphere, we
have chosen to explore the possibility that it is an ocean
source and not the propagation path winds that are responsible for the long-term changes seen in the thermosphere. Our
prior discussion on ocean Rossby waves at least points to
a physical mechanism that may initiate cause and effect on
the time scale observed. Additionally, Yiğit and Medvedev
[2010] have noted the importance of the faster gravity wave
harmonics as a particularly important source of waves reaching the thermosphere. Such fast waves would be less likely
to undergo wind filtering in the mesosphere.
8. Discrepancy Between Thermospheric
Temperature and Density Declines
[39] The review of Qian et al. [2011] notes that the thermospheric cooling of 2 K per decade expected from the
current rate of CO2 increase is not far from being compatible with the rate of density decrease (mainly atomic oxygen
O) indicated by satellite drag measurements. Those density measurements must therefore seem highly incompatible
with their quoted 47 K per decade cooling measured by the
radars. The expected relation between density and temperature is based upon assumptions of diffusive equilibrium
and fixed conditions at some lower boundary of the thermosphere. Our normal expectation is that when energy is
lost from the thermosphere, the reduced upward pressure
gradient force no longer opposes the gravitational force as
well, and the atmosphere readjusts by contracting. Gravity wave cooling, however, does not lose energy from the
system but merely distributes it downward, so there must
be expansion at lower altitudes while there is contraction
at higher altitudes. It is certainly true, of course, that the
transported energy, distributed now among more molecules,
causes a smaller temperature rise at lower altitude than
the temperature fall induced at higher altitude. Changes
at the lower boundary would be critical to this densityversus-temperature expectation. We ask here if an increase
in gravity wave activity might change the O density at the
lower boundary by changing the level of the turbopause. A
lower turbopause height would release the light O from the
mixed air scale height at a lower altitude, resulting in an
increase in O density at higher altitudes. We may estimate
the lowering needed. If exospheric temperature should drop
from 1000 K to 900 K, O density at 350 km should drop
28.4%, according to the tables of the Jacchia [1977] model.
If Hair is the scale height of the air below the turbopause and
HO is the scale height of O above the turbopause, then lowering the turbopause height by the distance ht will increase
the O density at the original turbopause height by the factor
exp(ht /Hair )/ exp(ht /HO ). Setting this expression equal to
1/(1–0.284) and adopting physical quantities appropriate for
the turbopause height level (a temperature of 200 K, gravitational acceleration of 9.5 m/s2 , mean molecular weight
of air of 29 atomic mass units), we find ht = 4.5 km. A
lowering of the turbopause level by about 4 km can completely offset the thermospheric density decline at 350 km
that would be caused by the 80–90 K thermospheric temperature decline portrayed in Figures 1 and 2. Would one
expect an increase in gravity wave activity to lower the
turbopause? One might expect the greater wave activity to
create more turbulence and so raise the turbopause height.
On the other hand, the gravity waves are thought to transfer
heat from higher to lower regions of the thermosphere and
so may actually increase the positive temperature gradient in
the low thermosphere, creating an atmosphere more stable
against the development of turbulence. Walterscheid [1981]
estimates that gravity waves cause a net heating below 110
km and cooling above. Yiğit and Medvedev [2009] say that
the net effect is cooling above the turbopause. If the region
of the turbopause is heated and the positive temperature gradient there is strengthened, stability against turbulence is
increased, and the boundary between turbulent and nonturbulent air may move to a lower altitude. Note that the Ti
height gradient results shown in the sixth panel of Figure 2
seem to indicate that increasing amounts of heat are being
transferred from higher to lower altitudes with increasing
time, though this result applies near 350 km altitude and not
at turbopause level. Confirmation that the height of the turbopause has experienced a long-term decline comes from
mass spectrometer measurements made aboard rockets at
midlatitudes during 1966–1992 [Pokhunkov et al., 2009].
These results show the turbopause to have decreased by 3.64
km over 19.4 years, a rate of decrease exceeding that needed
to explain the long-term radar-temperature/satellite-density
relation. If it should be true that gravity waves act both to
cool the thermosphere and lower the turbopause, then the
offsetting influences of these two effects would combine to
constrict the thermospheric density response to the wave
action. Note that feedback effects should be at work in these
long-term changes in gravity waves and turbopause height.
Yiğit and Medvedev [2010] note that the gravity wave cooling occurs lower in the thermosphere at solar maximum,
when the higher density dissipates the waves at lower altitude, and higher in the thermosphere when density is lower.
The long-term increase in gravity wave flux should steadily
cool the thermosphere, decrease density, allow waves to
propagate to higher altitudes, and so accentuate the cooling
in a positive feedback effect. A lowering of the turbopause
should cause O density and hence total density to rise, leading to wave dissipation and cooling at a lower altitude.
Model calculations could quantify these effects.
9. Concluding Remarks
[40] We have analyzed Ti data near 350 km altitude
over the time span 1966–2012 to seek explanations for
three outstanding questions concerning the long-term cooling observed in the upper thermosphere since the late 1970s:
(1) Why is the cooling so much larger than expected, (2)
why has the cooling lasted so long, and (3) why is the thermospheric density response to the cooling so small? We
3906
OLIVER ET AL.: GLOBAL COOLING OF THE THERMOSPHERE
have speculated that gravity waves may have caused this
cooling, based on recent simulations that show that gravity waves are expected to cool the upper thermosphere by
an amount comparable to the long-term cooling observed.
A gravity wave proxy formed from the nontidal fluctuations in Ti showed a positive long-term trend throughout its
timeline, consistent with the increasing cooling observed.
Fluctuations of Ti about its long-term trendline were seen
to anticorrelate both with fluctuations in the gravity wave
proxy and with decadal fluctuations in El Niño–Southern
Oscillation (ENSO) activity. The time scales of the longterm trend and the decadal fluctuations are characteristic
of the ocean, not the atmosphere. We have suggested that
the following scenario may explain these behaviors: (a) the
climate regime shift of 1976–1977 launched slow Rossby
waves across the oceans, waves which continue to propagate to this day; (b) winds over this increasingly corrugated
ocean have launched increasing fluxes of gravity waves into
the atmosphere; (c) these increasing fluxes of gravity waves
have propagated to the thermosphere to produce increasing
amounts of cooling; and (d) the decadal increases in winds
associated with the decadal increases in ENSO activity have
produced decadal increases in gravity wave fluxes, which, in
turn, have caused decadal cooling in the thermosphere. The
strong thermospheric cooling seen would be expected to produce thermospheric density declines much larger than those
observed via satellite drag if the density at the base of the
thermosphere were unchanging. We have noted that a lowering of the turbopause by about 4 km would raise atomic
oxygen densities and completely compensate for that large
expected density decline at 350 km altitude and referenced
evidence of that lowering. We have asked if the heat pumped
by gravity waves from the upper thermosphere to the lower
thermosphere may augment the positive temperature gradient in the turbopause region, thereby making it more stable
against the development of turbulence, and hence lowering the turbopause itself. Our proposed answers to our three
posed questions are that the magnitude of the cooling is due
to the agency of atmospheric gravity waves, that the length
of the cooling is due to agency of ocean Rossby waves, and
that the small density response is due to the lowering of
the turbopause.
[41] In our scenario, the climate regime shift of 1976–
1977 “shook” the oceans, and the oceans and overlying
atmosphere broke out in waves, as fluids are prone to do.
The ocean effect is a long-term transient response to forcing
of unknown origin. The atmospheric effect is a steady state
response to ocean forcing. This scenario may have validity,
but it is unproved. The ocean Rossby wave source is purely
speculative. Those Rossby waves, if they should exist, cannot be detected amongst the larger ocean eddies with current
observational capabilities. (We ask if the very smallness of
these ocean waves may be essential in that it leads to atmospheric gravity waves that are able to propagate to the upper
thermosphere without growing to their breaking amplitude
and dissipating at lower altitude.) For the existence of
increasing amounts of gravity waves in the thermosphere,
however, we have the evidence from our data of the increasing amounts of nontidal fluctuations in temperature. If the
theorists and modelers [e.g., Walterscheid, 1981; Yiğit and
Medvedev, 2009] are correct, this increasing gravity wave
flux should increasingly cool the thermosphere.
[42] These speculations need broad vetting by additional
studies based on other data sets and of applicable theory.
[43] We believe that the magnitude of the thermospheric
temperature decline has been so great that the CO2 theory
of its cooling cannot be maintained. It is certainly possible,
however, that increased greenhouse warming at surface level
has initiated the train of events that has led to the cooling by
others means.
[44] Acknowledgments. This work used data accessed from the the
Millstone Hill Madrigal Database. The Millstone Hill incoherent scatter radar is supported by the U.S. National Science Foundation Upper
Atmospheric Facilities Program under a cooperative agreement with the
Massachusetts Institute of Technology. This work was supported in part
through NSF grants AGS-0925893 to Boston University and AGS-1042569
to the Massachusetts Institute of Technology. We thank John Holt and
Bill Rideout for indispensable help in interpreting and using the Madrigal
Database; Dudley Chelton, Richard Walterscheid, and Erdal Yiğit for critically informative discussions; and former Boston University undergraduate
students Tyler Wellman, Jessica Donaldson, and Patrick Walsh for their
work in developing our analysis tools.
[45] Robert Lysak thanks the reviewers for their assistance in evaluating
this paper.
References
Akmaev, R. A. (2012), On estimation and attribution of long-term temperature trends in the thermosphere, J. Geophys. Res., 117, A09321,
doi:10.1029/2012JA018058.
Brum, C. G. M., C. A. Tepley, J. T. Fentzke, E. Robles, P. T. dos
Santos, and S. A. Gonzalez (2012), Long-term changes in the thermospheric neutral winds over Arecibo: Climatology based on over three
decades of Fabry-Perot observations, J. Geophys. Res., 117, A00H14,
doi:10.1029/2011JA016458.
Chelton, D. B., and M. G. Schlax (1996), Global observations of oceanic
Rossby waves, Science, 272, 234–238.
Chelton, D. B., M. G. Schlax, R. M. Samuelson, and R. A. de Szoeke
(2007), Global observations of large oceanic eddies, Geophys. Res. Lett.,
34, L15606, doi:10.1029/2007GL030812.
Chelton, D. B., M. G. Schlax, and R. M. Samuelson (2011),
Global observations of nonlinear mesoscale eddies, Prog. Oceanogr., 91,
167–216.
Dickson, R. R., and J. Namias (1976), North American influences on the
circulation and climate of the North Atlantic sector, Mon. Weather Rev.,
104, 1255–1265.
Djuth, F. T., L. D. Zhang, D. J. Livneh, I. Seker, S. M. Smith, M. P. Sulzer,
J. D. Matthews, and R. L. Walterscheid (2010), Arecibo’s thermospheric
gravity waves and the case for an ocean source, J. Geophys. Res., 115,
A08305, doi:10.1029/2009JA014799.
Donaldson, J. K., T. J. Wellman, and W. L. Oliver (2010), Long-term change
in thermospheric temperature above Saint Santin, J. Geophys. Res., 115,
A11305, doi:10.1029/2010JA015346.
Emmert, J. T., and J. M. Picone (2011), Statistical uncertainty of 1967–2005
thermospheric density trends derived from orbital drag, J. Geophys. Res.,
116, A00H09, doi:10.1029/2010JA016382.
Fritts, D. C., and T. S. Lund (2011), Gravity wave influences in the thermosphere and ionosphere: Observations and recent modeling, in Aeronomy
of the Earth’s Atmosphere and Ionosphere, IAGA Spec. Sopron Book
Ser., vol. 2, pp. 109–130, Springer, New York, doi:10.1007/978-94-0070326-1_8.
Graham, N. E. (1994), Decadal-scale climate variability in the tropical
and North Pacific during the 1970s and 1980s: Observations and model
results, Clim. Dyn., 10, 135–162.
Graham, N. E. (1995), Simulation of recent global temperature trends,
Science, 267, 666–671.
Hertzog, A., G. Boccara, R. A. Vincent, F. Vial, and P. Cocquerez
(2008), Estimation of gravity wave momentum flux and phase
speeds from quasi-Lagrangian stratospheric balloon flights. Part II:
Results from the Vorcore campaign in Antarctica, J. Atmos. Sci., 65,
3056–3070.
Hidalgo, H. G., and J. A. Dracup (2003), ENSO and PDO effects on
hydroclimatic variations of the Upper Colorado River Basin, J. Hydrometeorol., 4, 5–23.
Hoffmann, P., M. Rapp, W. Singer, and D. Keuer (2011), Trends of mesospheric gravity waves at northern middle latitudes during summer, J.
Geophys. Res., 116, D00P08, doi:10.1029/2011JD015717.
3907
OLIVER ET AL.: GLOBAL COOLING OF THE THERMOSPHERE
Holt, J. M., and S. R. Zhang (2008), Long-term temperature trends in
the ionosphere above Millstone Hill, Geophys. Res. Lett., 35, L05813,
doi:10.1029/2007GL031148.
Jacchia, L. G., (1977), Thermospheric temperature, density, and composition: New models, SAO Spec. Rep. 375, Smithson. Astrophys. Observ.,
Cambridge, Mass.
Latif, M., and T. P. Barnett (1994), Causes of decadal climate variability over the North Pacific and North America, Science, 266,
634–637.
Latif, M., and T. P. Barnett (1996), Decadal climate variability over the
North Pacific and North America: Dynamics and predictability, J. Clim.,
9, 2407–2423.
Laštovička, J. (2012), On the role of ozone in long-term trends in the
upper atmosphere-ionosphere system, Ann. Geophys., 30, 811–816,
doi:10.5194/angeo-30-811-2012.
Lognonne, P., R. Garcia, F. Crespon, G. Occhipinti, A. Kherani, and J.
Artru-Lambin (2006a), Seismic waves in the ionosphere, Europhysics
News, 57(4), 11–15.
Lognonne, P., J. Artru, R. Garcia, F. Crespon, V. Ducic, E. Jeansou, G.
Occhipinti, J. Helbert, G. Moreaux, and P.-M. Godet (2006b), Groundbased GPS imaging of ionospheric post-seismic signal, Planet. Space
Sci., 54, 528–540.
Manabe, S., and R. Wetherald (1967), Thermal equilibrium of the atmosphere with a given distribution of relative humidity, J. Atmos. Sci., 24,
241–259.
Mantua, N. J., S. R. Hare, Y. Zhang, J. M. Wallace, and R. C. Francis
(1997), A Pacific interdecadal climate oscillation with impacts on salmon
production, Bull. Am. Meteorol. Soc., 78, 1069–1079.
Maximenko, N. A., O. V. Melnichenko, P. P. Niiler, and H. Sasaki (2008),
Stationary mesoscale jet-like features in the ocean, Geophys. Res. Lett.,
35, L08603, doi:10.1029/GL033267.
Merzylakov, E. G., Ch. Jacobi, Yu. I. Portnyagin, and T. V. Solovjoya
(2009), Structural changes in trend parameters of the MLT winds based
on wind measurements at Obninsk (55N, 37E) and Collm (52N, 15E), J.
Atmos. Sol. Terr. Phys., 71, 1547–1557.
Minobe, S. (1997), A 50–70 year climatic oscillation over the North Pacific
and North America, Geophys. Res. Lett., 24, 683–686.
Namias, J. (1952), The annual course of month-to-month persistence in
climate anomalies, Bull. Am. Meteorol. Soc., 33, 279–285.
Namias, J. (1969), Seasonal interactions between the North Pacific Ocean
and the atmosphere during the 1960s, Mon. Weather Rev., 97, 173–192.
Namias, J., X. Yuan, and D. R. Cayan (1988), Persistence of North Pacific
sea surface temperature and atmospheric flow patterns, J. Clim., 1,
682–703.
Offermann, D., J. Wintel, C. Kalicinsky, P. Knieling, and R. Koopmann
(2011), Long-term development of short-period gravity waves in middle
Europe, J. Geophys. Res., 116, D00P07, doi:10.1029/2010JD015544.
Oliver, W. L., Y. Otsuka, M. Sato, T. Takami, and S. Fukao (1997), A climatology of F region gravity wave propagation over the middle and upper
atmosphere radar, J. Geophys. Res., 102, 14,499–14,512.
Pokhunkov, A. A., V. V. Rybin, and G. F. Tulinov (2009), Quantitative characteristics of long-term changes in parameters of the upper
atmosphere of the Earth over the 1966–1992 period, Cosmic Res., 47,
480–490.
Qian, L., J. Laštovička, R. G. Roble, and S. C. Solomon (2011), Progress in
observations and simulations of global change in the upper atmosphere,
J. Geophys. Res., 116, A00H03, doi:10.1029/2010JA016317.
Ratnum, M. V., G. K. Kumar, N. V. Rao, B. V. K. Murthy, J. Laštovička, and
L. Qian (2013), Evidence of long-term change in zonal wind in the tropical lower mesosphere: Observations and model simulations, Geophys.
Res. Lett., 40, 397–401, doi:10.1002/grl.50158.
Roble, R. G., and R. E. Dickinson (1989), How will changes in carbon
dioxide and methane modify the mean structure of the mesosphere and
thermosphere?, Geophys. Res. Lett., 16, 1441–1444.
She, C. Y., W. Thiel, and D. A. Kreuger (1998), Observed episodic warming
at 86 and 100 km between 1990 and 1997: Effects of Mount Pinatubo
eruption, Geophys. Res. Lett., 25, 497–500.
Schlax, M. G., and D. B. Chelton (2008), The influence of mesoscale eddies
on the detection of quasi-zonal jets in the ocean, Geophys. Res. Lett., 35,
L24602, doi:10.1029/GL035998.
Timmermann, A., J. Oberhuber, A. Bacher, M. Esch, M. Latif, and E.
Roeckner (1999), Increased El Niño frequency in a climate model forced
by future greenhouse warming, Nature, 398, 694–697.
Trenberth, K. E. (1990), Recent observed interdecadal climate changes in
the Northern Hemisphere, Bull. Am. Meteorol. Soc., 71, 988–993.
Trenberth, K. E., and T. J. Hoar (1996), The 1990–1995 El Niño–Southern
Oscillation event: Longest on record, Geophys. Res. Lett., 23, 57–60.
Trenberth, K. E., and J. W. Hurrell (1994), Decadal atmosphere-ocean
variations in the Pacific, Clim. Dyn., 9, 303–319.
Trenberth, K. E., and D. P. Stepaniak (2001), Indices of El Niño evolution,
J. Clim., 14, 1697–1701.
Vadas, S. L., D. C. Fritts, and M. J. Alexander (2003), Mechanism for the
generation of secondary waves in wave breaking regions, J. Atmos. Sci.,
60, 194–215.
Walsh, P. L., and W. L. Oliver (2011), Is thermospheric long-term cooling
due to CO2 or O3 ? Ann. Geophys., 29, 1779–1782, doi:10.5194/angeo29-1779-2011.
Walterscheid, R. L. (1981), Dynamical cooling induced by dissipating
internal gravity waves, Geophys. Res. Lett., 8, 1235–1238.
Wolter, K., and M. S. Timlin (1993), Monitoring ENSO in COADS with
a seasonally adjusted principal component index, in Proceedings of the
17th Climate Diagnostics Workshop, pp. 52–57, NOAA, Washington,
D. C.
Wolter, K., and M. S. Timlin (1998), Measuring the strength of ENSO
events: How does 1997/98 rank?, Weather, 53, 315–324.
Wolter, K., and M. S. Timlin (2011), El Niño/Southern Oscillation behavior since 1871 as diagnosed in an extended multivariate ENSO index
(MEI.ext), Int. J. Climatol., 31, 1074–1087.
Yiğit, E., and A. S. Medvedev (2009), Heating and cooling of the thermosphere by internal gravity waves, Geophys, Res. Lett., 36, L14807,
doi:10.1029/2009GL038507.
Yiğit, E., and A. S. Medvedev (2010), Internal gravity waves in the thermosphere during low and high solar activity: Simulation study, J. Geophys.
Res., 115, A00G02, doi:10.1029/2009JA015106.
3908