Climate Change in the Thermosphere and Ionosphere From the Early Twentieth Century to Early Twenty‐First Century Simulated by the Whole Atmosphere Community Climate Model—eXtended

Motivated by numerous lower atmosphere climate model hindcast simulations, we performed simulations of the Earth's atmosphere from the surface up through the thermosphere‐ionosphere to reveal for the first time the century scale changes in the upper atmosphere from the 1920s through the 2010s using the Whole Atmosphere Community Climate Model—eXtended (WACCM‐X v. 2.1). We impose solar minimum conditions to get a clear indication of the effects of the long‐term forcing from greenhouse gas increases and changes of the Earth's magnetic field and to avoid the requirement for careful removal of the 11‐year solar cycle as in some previous studies using observations and models. These previous studies have shown greenhouse gas effects in the upper atmosphere but what has been missing is the time evolution with actual greenhouse gas increases throughout the last century, including the period of less than 5% increase prior to the space age and the transition to the over 25% increase in the latter half of the 20th century. Neutral temperature, density, and ionosphere changes are close to those reported in previous studies. Also, we find high correlation between the continuous carbon dioxide rate of change over this past century and that of temperature in the thermosphere and the ionosphere, attributed to the shorter adjustment time of the upper atmosphere to greenhouse gas changes relative to the longer time in the lower atmosphere. Consequently, WACCM‐X future scenario projections can provide valuable insight in the entire atmosphere of future greenhouse gas effects and mitigation efforts.

In the middle atmosphere, Santer et al. (2023) found that including both the warming of the troposphere and lower stratosphere and the cooling in the middle and upper stratosphere enhance the detectability of "fingerprints" on the climate system due to increases in carbon dioxide.Garcia et al. (2019) also found that simulations with the Whole Atmosphere Community Climate Model (WACCM) (Marsh et al., 2013), along with comparisons to the Sounding of the Atmosphere by Broadband Emission observations on the Thermosphere Ionosphere Mesosphere Energetics and Dynamics satellite, indicate that temperature trends in vertical regions up through the middle atmosphere into the lower thermosphere "remain an important signature of climate change, and they underscore the importance of global, continuous monitoring of this region of the atmosphere."Pearce (2023) discusses both the Santer et al. (2023) and Garcia et al. (2019) studies and in addition emphasizes the critical nature of future climate changes in the middle atmosphere.A summary of long-term trends in the mesosphere and lower thermosphere are discussed in studies by Beig et al. (2003) and more recently Laštovička (2021).They conclude the temperature trends are negative in the mesosphere and near zero in the mesopause region and are mainly due to CO 2 and CH 4 , as well as water vapor and ozone.Also, it should be noted that in this middle part of the atmosphere and extending into the upper atmosphere, the radiative relaxation time is on the order of days (Mlynczak et al., 2022), much shorter than the decadal timescale in the lower atmosphere.This means the response time to forcings like greenhouse gases is prompt in the upper atmosphere relative to the lag in the lower atmosphere.
In the upper regions of the atmosphere above about 100 km, various modeling and observation studies have examined the effects of greenhouse gas increases on the thermosphere and ionosphere.Laštovička (2021) also gives a summary of long-term trend studies in this upper region of the atmosphere.These studies include observations of the long-term neutral density (Emmert, 2015;Emmert et al., 2004Emmert et al., , 2008) ) and Incoherent Scatter Radar measurements (Zhang et al., 2011(Zhang et al., , 2016;;Zhang & Holt, 2013).The modeling studies typically involved two types of model realizations, some with early 20th century values of greenhouse gases compared to more recent values (Cnossen, 2014;Qian et al., 2008Qian et al., , 2013) ) and others with a doubling and halving of greenhouse gases (Akmaev & Fomichev, 1998;Liu et al., 2020;Roble & Dickinson, 1989).There have also been studies using continuous model simulations over the last half of the 20th century (Cnossen, 2020;Qian et al., 2006Qian et al., , 2021)).A promising machine learning technique has been applied to thermosphere density determination (Weng et al., 2020) and one unique study showed the changes in the thermosphere over the past ten thousand years using a model covering the upper mesosphere and thermosphere (Cai et al., 2023).This last study included a continuous upper atmosphere simulation of the last century but so far none of these studies have looked at whole atmosphere results continuously covering the entire past century.All these upper atmosphere observations and modeling studies make clear that in this upper region of the atmosphere, global natural variability on decadal to century time scales is dominated by solar cycle variations, with other sources of variability being relatively minor.Importantly, major sources of natural variability in the thermosphere and ionosphere differ significantly from those in the troposphere.Many historical earth system model simulations of the lower atmosphere 20th century climate have been performed and these models are subsequently used for future climate projections (Chen et al., 2021) but no continuous historical climate simulations have been made of the upper atmosphere climate over the past century.The models typically used for lower atmosphere climate only extend up into the stratosphere, for instance the Community Atmosphere Model (CAM) (Neale et al., 2013) or extend into the lower part of the thermosphere and ionosphere, for example, WACCM mentioned previously.There is clearly a need for simulation of climate change over the last century in the upper atmosphere.To address this, we examine whole atmosphere climate model simulations of the last century from the Whole Atmosphere Community Climate Model-eXtended (WACCM-X) with a focus on the effects of greenhouse gases on the upper atmosphere.Additionally, since this model self consistently simulates the Earth's atmosphere from the surface up through the thermosphere and ionosphere, we can see for the first time the effects of greenhouse gases at all these vertical levels over the past century in output from a single model.
As mentioned above, decadal scale variability in the upper atmosphere is overwhelmed by the 11-year cycle of solar activity but here we are interested in the effects of greenhouse gases.To address this, we ran the WACCM-X simulations under a constant minimum phase of the solar cycle.This allows us to not only look at the changes between the early 20th century and the early 21st century, with different levels of greenhouse gases, but, for the first time, reveal the continuous decade to decade changes over the entire past century in the thermosphere and ionosphere.Our main objective is to examine the correlation between the changing greenhouse gas concentrations and the changing climate in the thermosphere and ionosphere over a time when the rate of change of these concentrations changes dramatically.We can thus begin to understand these continuous century scale changes at a level already achieved in the lower atmosphere.Furthermore, we can have more confidence in WACCM-X projections of 21st century future upper atmosphere climate, projections which are routine in the lower atmosphere.To study these upper atmosphere changes over the last century, we use WACCM-X results to examine the upper atmosphere climate through relevant atmospheric quantities in the thermosphere and ionosphere as they change decade-to-decade over 10 decades 1920s-2010s.The following sections describe the model, introduce the simulations performed, describe results of the simulations, and end with a discussion and conclusions from these results.

Model Description
WACCM-X is a self-consistent three-dimensional whole atmosphere global general circulation model extending from the surface to ∼600 km, with a dependence on solar cycle phase.The version 2 release of the model includes electrodynamics, ion transport, major and minor neutral species composition, electron and ion density and composition, and electron and ion temperature in addition to the previous neutral atmosphere and ionosphere representation (H.-L.Liu et al., 2010, 2018and J. Liu et al., 2018).These additions provide a more realistic simulation of thermosphere and ionosphere dynamics and energetics.The release of CESM version 2.1 includes version 2.1 of WACCM-X.This version of WACCM-X is based on the CAM version 4 and WACCM version 4 (Marsh et al., 2013;Neale et al., 2013).All three of these models are forced by observation-based CO 2 and methane (CH 4 ) at the surface lower boundary and by spectral solar irradiance, solar flux (F 10.7 ), geomagnetic activity (Kp) at the upper boundary.Of importance here is the calculation in the model of the upper atmosphere non-Local Thermodynamic Equilibrium radiative cooling rate of CO 2 , which is done using the radiative transfer algorithm described in Fomichev et al. (1993Fomichev et al. ( , 1998)).WACCM-X has a standard resolution of 1.9° × 2.5° in latitude and longitude and a 0.25 scale height resolution in the vertical above 1 hPa.There are two main modes of simulation using the WACCM-X (and WACCM) model(s), one with meteorological forcing in the lower atmosphere and the other with the model freely running without this constraint.For this study, we use this version 2.1 of the model with the standard resolution and the latter free-running mode to examine the changes from the surface through the thermosphere and ionosphere from the 1920s through the 2010s.

Time Slice Simulations
To examine changes for this period of 10 decades, we employ the same method as Solomon et al. (2018).In that study the changes between the periods of 1972-1976 and 2001-2005 were examined using solar minimum condition simulations from a model slightly older than this version of WACCM-X.They also did the same study under solar maximum conditions (Solomon et al., 2019) and found larger magnitude changes under solar minimum than under solar maximum conditions.In contrast to the two sets of simulations in the previous studies, here we perform 10 sets of simulations, one set per decade from 1921-1925 to 2011-2015 using an updated version of WACCM-X, version 2.1.Just as in these previous studies, we use the free running mode described previously.Also, based on their results, we choose solar minimum quiet geomagnetic conditions (F 10.7 = 70 and Kp = 0.3) to get larger magnitude changes over time.In addition, the choice of constant solar conditions in the above studies as well as in this study is based on the need to eliminate the large solar cycle variability, which dominates upper atmosphere long term change on century and decadal time scales.Also, as in the previous studies, the Earth's magnetic field varies over time as given by the International Geomagnetic Reference Frame.At the lower boundary, inputs over time of trace gases are based on the Chemistry-Climate Model Initiative standard (Eyring et al., 2013).More specifically, two of these trace gases are the greenhouse gases CO 2 and methane (CH 4 ) which are based on observations as discussed in Meinshausen et al. (2011).A data ocean and a nudged quasi-biennial oscillation are implemented based on observations.Simulations were started at the beginning of each decade and progressed for 6 years, with the first year to ensure equilibrium and the last 5 years of each decadal simulation to be used to represent that decade.We considered using the limited computing resources to do a single realization with annual values over these 10 decades, but this would not have accounted for the annual variability pointed out in Solomon et al. (2018).As in that study, the 5 years serve as a small ensemble for each decade, more like the ensemble mean method to address model variability in lower atmosphere simulations.Monthly mean output data from the model are used for each 5-year ensemble.Annual means are calculated for each year and then zonal means of those annual means.Subsequently, these 5-year zonal mean annual means are averaged to get one zonal mean annual mean for each of the 5-year periods.Finally, the mean is calculated from these zonal means using cosine (latitude) weighting, resulting in one global mean for the first half of each of the 10 decades for each variable output by the model.These global means along with the zonal means are used in this study to examine relevant atmospheric quantities.Compared to the much higher temporal frequency output of typical lower atmosphere simulations over the past century, these single global and zonal mean values for each decade can miss some of the variability on shorter time scales but these simulations are intended as a first step in the more thorough simulation of the thermosphere on century time scales.

Results
The first atmospheric quantity we discuss is temperature.Since the model covers the entire atmosphere from the surface to ∼600 km, we can look at temperature throughout the vertical range.Figure 1 shows the 5-year global mean temperature time series, one value for each decade, at four different atmospheric levels for the entire simulation period from the 1920s-2010s.Error bars are added for each decade based on the standard deviation of the 5 years that go into the mean value.Near the surface in Figure 1a, we see the temperature increase by nearly +0.95 degrees Kelvin (K) from the 1920s-2010s, which is very close to the observational record IPCC report value of +0.9 K.For the first two decades there is an increase of about +0.4 K. Then for the next three decades, from the 1940s-1970s, there is an ∼ −0.2 K decrease.After this, through the 2010s, there is an increase of over +0.7 K.In addition to the period of decrease around the 1940s-1970s, there is also a relatively slower increase in temperature between the 2000s and 2010s.These are "hiatus" periods mentioned previously which are seen in surface temperature observations.The error bars range from 0.1 to 0.3 K, changing significantly from decade to decade.Important caveats when comparing with lower atmosphere model results is that this simulation in contrast is not a fully coupled earth system simulation, for solar minimum, and only one realization per decade with a small five-member ensemble.In Figure 1b, we see the temperature time series for the upper stratosphere on the 1.25 Hectopascals (hPa) pressure surface near an altitude of 46 km where, unlike near the surface, the temperature has a decrease during this period of over −6 K.During the first four decades the decrease is slower, only changing by less than −1 K.After that, from the 1960s-1990s, the decrease is sharper, just under −4 K.At the end of the period, the decrease again slows down with just over a −1 K decrease from in the 2000s and 2010s.At this level, error bars are in the range of 0.1-0.4K, again varying from decade to decade.Figure 1c is this same temperature time series but for the upper mesosphere on the 0.0013 hPa pressure surface near 90 km in altitude.There is also an overall decrease at this level of ∼ −3.3 K.The decrease is consistent throughout the entire 10-decade period.Error bars range from 0.3 to 0.7 K.The time series for the thermosphere on the 2.3 × 10 −9 hPa pressure surface near 400 km altitude is shown in Figure 1d.A decrease is again seen at this level but larger, over ∼ −17 K.The first four decades have a noticeably slower decrease, ∼ −1 K, than the rest of the period.After the 1970s, the rate of decrease gets larger until the 2000s-2010s period, when there is again a slightly smaller decrease rate.Error bars range from 1 to 3 K.
Another important quantity to look at in the thermosphere is the neutral mass density.Figure 2a shows the same time series as in Figure 1 but for neutral density values at ∼400 km.Since the model has a native vertical pressure grid, the density for all decades is interpolated to the 1920s altitude grid and then the level with a global average altitude closest to 400 km is chosen.From the 1920s-2010s there is a decrease of just over 30% relative to the 1920s density.Just as in thermosphere temperature, the decrease is slightly smaller in the first four decades and less so, but still smaller, in the last decade compared to the middle decades of the time series.The error bars are on the order 0.01-0.04ng/m 3 or less than 0.5%.In addition to total neutral density, we can look at densities of constituents.Most have nearly the same change over time as the total density such as the atomic oxygen (a decrease of −7.4 × 10 6 cm −3 and error bar range of 1.0-2.0× 10 5 cm −3 ) and molecular nitrogen (a decrease of −1.3 × 10 5 cm −3 and error bar range of 1.0-2.0× 10 5 cm −3 ) shown in Figures 2b and 2c.Even though they both decrease over this period, the ratio of atomic oxygen to molecular nitrogen (O/N 2 ) at ∼400 km increases as seen  2d.This makes clear the decrease over time in atomic oxygen is relatively smaller than the decrease in molecular nitrogen.While the other constituents that make up the neutral density also decrease, one constituent that behaves differently is atomic hydrogen (shown in Figure 2e).Instead of decreasing, this constituent increases from the 1920s-2010s by 7.2 × 10 4 cm −3 or ∼15%.Again, the changes in the first four decades and the last two decades are slightly less than the four decades in between.The error bars of this increase are 2.0-8.0 × 10 3 cm −3 .
Turning to the ionosphere, the electron temperature five-year global mean time series on the 2.3 × 10 −9 hPa pressure surface near 400 km is shown in Figure 3a.The overall increase from the 1920s-2010s is over 22 K.All decades show an increase except the 1950s-1960s and 1980s-1990s, which both have a slight decrease smaller than −1 K. Error bars range from 8 to 17 K indicating a high annual variability relative to the change over the past century.The ion temperature, on the same pressure surface near an altitude of 400 km, decreases from the 1920s-2010s as shown in the 5-year global mean time series in Figure 3b.The total decrease for this time period is 19 K.This decrease is smaller for the first four decades than the decades that follow.The error bars are from 1.5 to 3.5 K. Figure 3c shows the time series of 5-year global mean electron density values from the 1920s-2010s just as neutral density at an altitude of ∼400 km.The electron density has a decrease from the 1920s-2010s of about −1.85 × 10 4 cm −3 or −16.6%.The decrease between the first five decades is slower than the decrease between the last five decades.Error bars are 1.5-3.0× 10 3 cm −3 .Figure 3d is the electron column density time series and shows a −0.57TECU decrease or about −10.2% with error bars of 0.1-0.15TECU.It is important to note this electron column density only includes the contribution to the total electron content (TEC) below the model top.
Based on the results in Figures 1-3, we can quantitatively evaluate the relative changes over these 10 decades by looking at a different aspect of the data and that is vertical profiles of the global mean accumulating change when dividing the period up into the first four decades, adding on the middle three decades, and then the last three decades.This is shown for neutral temperature in pressure coordinates in Figure 4a.Near the surface, there is a little less than a 1 K cumulative increase for the 1920s-2010s period, with nearly half from the last few decades and the smallest accumulated increase in the three decades prior.There is a cumulative decrease in temperature at all levels above the troposphere.In the stratosphere at just under 50 km in altitude, there is a maximum accumulated decrease of −6.5 K for all 10 decades.In the first four decades, the cumulative decrease is only −0.8 K, −3 K from the next three decades, and another −2.5 K from the last three decades.There is a minimum in the cumulative temperature decrease just above 80 km in the mesosphere of about −3 K.The first four decades make up −0.8 K of this change and the next three decades with only a −0.5 K of this decrease.The final three decades have the remaining −1.7 K change.In the thermosphere, we can see in Figure 4a that the total cumulative decrease is just over −17 K.In the first four decades the accumulated decrease is only about −2.5 K, in the next three decades −5.5 K, and in the last three decades just over −9 K.For neutral density on altitude, Figure 4b shows a −25% cumulative decrease around an altitude of 400 km for the 1920s-2010s period.Of this change only just under −4% occurred in the 1920s-1950s.The next three decades had just over a −9% cumulative decrease and the last three decades have an accumulated decrease of −12%.Other neutral constituent number densities such as atomic oxygen, molecular oxygen, and nitrogen (not shown) have changes over these 10 decades very similar to neutral mass density but atomic hydrogen number density changes differently.Figure 4c shows the atomic hydrogen mass density cumulative changes for altitudes above 100 km for the three different periods described above during the 1920s-2010s.Below 120 km, atomic hydrogen decreases by a maximum of just under −4% for the entire 1920s-2010s period.Above 120 km up to 170 km, the atomic hydrogen increases through the 1980s by as much as +9% and then, from the last three decades, the cumulative change is near zero or less positive and even negative below 120 km.Above 170 km, the atomic hydrogen increases throughout the entire 10 decades by a maximum of +15%, with +3% in the first four decades, +7% in the next three decades, and +5% in the last three decades.Figure 4d shows the same representation for CO 2 mixing ratio on altitude, illustrating the change becoming larger when progressing through the 20th and early 21st centuries.It is clear in the latter part of this period, instead of being vertically well mixed, the CO 2 mixing ratio has a gradient in the stratosphere.This is likely due to the difference in the adjustment of the turbulent troposphere and the more stable stratosphere to the ever-larger rate of increase of CO 2 .
These same changes can be examined in the ionosphere for the 10 decades.Figure 5a shows the electron density global mean vertical profiles on altitude above 100 km from the 1920s-2010s separated into the first four decades and two subsequent three-decade periods.There is a different behavior in the lower ionosphere than in the upper ionosphere with the transition around 210 km.Below this altitude, the electron density increases by about +7% from 1920s to 2010s.Around +1% of this cumulative increase occurred in the first four decades, about +2% in the next three decades, and +4% in the last three decades.Above 210 km, the electron density at 400 km decreases by −17.5% from the 1920s-2010s.About −3% of this cumulative change is in the first three decades, just under −6% in the next three decades, and just under −9% in the last three decades.For electron temperature on pressure, Figure 5b, in the lower ionosphere, below around 150 km, there is a cumulative decrease with a maximum of −18 K.This decrease is only −2 K for the first four decades, −7 K in the next three decades, and −9 K in the last three decades.Above 150 km there is a cumulative increase of electron temperature for the 1920s-2010s with a maximum of 35 K.Of this increase, only +6 K is in the first four decades, +10 K in the next three decades, and +19 K in the last three decades.Figure 5c shows the corresponding ion temperature on pressure changes above Up to this point, we have described annual mean results, but we also want to get an idea of the seasonal and latitudinal changes in the thermosphere and ionosphere over these 10 decades.To do this we examine zonal mean time series at equinox and solstice at a given vertical level.Figures 6a and 6e show time series for all latitudes on the 2.84 × 10 −8 hPa pressure level near an altitude of 306 km for zonal mean neutral temperature changes from the 1920s-2010s for the months of March and June.There is latitudinal dependence in the temperature changes in both March and June.In the month of March, the temperature changes are larger in the northern hemisphere low-to mid-latitudes and smaller in the southern hemisphere and the northern hemisphere high-latitudes.At all latitudes, most of the March cumulative temperature change occurs after about 1970 and tends to be more inconsistent before then.In June, the temperature changes from mid-to high latitudes in the summer hemisphere (northern hemisphere) are systematically smaller than those in the rest of the latitudes.As in March, most of the cumulative change in June also occurs after about 1970, and even after about 1980 in the high summer (northern) hemisphere latitudes.The largest cumulative changes are ∼ −18 K at 50°N latitude for March and ∼−20 K at 30°N latitude for June.Figure 6b is the same as Figure 6a but for neutral density on an altitude level near 395 km.The cumulative negative change of neutral density at all latitudes in March is nearly the same, −0.2 g/cm 3 , except near the south polar region where the cumulative change is slightly less, around −0.18 g/cm 3 .For June, in Figure 6f, the neutral density cumulative change is least near the north pole, ∼ −0.17 g/cm 3 .This cumulative change is more negative when moving further south, reaching a maximum of −0.10.1029/2023JD039397 9 of 16 density.There are some similarities to electron density with larger cumulative change in the northern hemisphere polar region for both months.In addition, the rate of cumulative change is positive in some latitude regions for limited time periods.The results for all these variables for September equinox and December solstice (not shown) are very similar to these March and June results.
We also choose another way to look at these seasonal characteristics of the changes over the 10 decades.Figure 7a shows a global map on a pressure surface of 2.84 × 10 −8 hPa near an altitude of 306 km of the neutral temperature changes from the 1920s-2010s for the month of March.The changes range from just under a −11 K decrease in the south tropical Atlantic Ocean and eastern Southern Ocean areas to about a −23 K decrease over eastern Canada and the north tropical Atlantic Ocean region.Figure 7b is the same as Figure 7a with the same color scale but for September.The temperature decrease is overall larger ranging from −14 K particularly in the eastern Southern Ocean to −27 K also over eastern Canada.The same map representation for December is shown in Figure 7c with changes ranging from −10 K in the south tropical Atlantic Ocean region to −23 K over eastern Canada.In Figure 7d the same the map results for June are shown with decreases from −13 K over western Russia to −25 K over eastern Canada.

Discussion
Since this is the first continuous simulation of the whole atmosphere over a ten-decade period, it is interesting to compare with previous shorter period studies, studies examining differences due to increasing levels of greenhouse gases, and studies using differences many decades apart.We will examine temperature first and the study most like ours is Solomon et al. (2018), which, as mentioned earlier, used a similar WACCM-X model as we use here but for a shorter period from the 1970-2000s.They reported global mean changes of +0.2 K/decade near the surface, −1.1 K/decade near the stratopause, −0.7 K/decade near the mesopause, and −2.8 K/decade at 400 km.The changes from our results for this same 1970s-2000s period are +0.1 K/decade, −0.7 K/decade, −0.5 K/ decade, and −1.9 K/decade, respectively.For the entire period from the 1920s-2010s, we find +0.1 K/decade, −0.7 K/decade, −0.34 K/decade, and −1.89 K/decade, respectively.What is clear is that including the decades prior to the 1970s causes lower cooling per decade in levels above the surface relative to the cooling per decade for just the period from the 1970s onward.Cnossen (2020) used a different version of the WACCM-X model for  Other studies examining a doubling of CO 2 include Akmaev and Fomichev (1998) using the Spectral mesosphere/ lower thermosphere model and Roble and Dickinson (1989) using a global mean model, show a thermosphere temperature change of −40 to −50 K and −50 K, respectively.It should be noted that for the ∼−6K cooling in the upper stratosphere, lower mesosphere region, ozone is a significant factor.Results of a couple of thorough studies of climate change in this region of the atmosphere including the role of ozone can be found in Santer et al. (2023) and Garcia et al. (2019).For the stratopause to mesopause region, Garcia et al. (2019)  There have been previous studies that have looked at the spatial distribution of the changes in thermosphere temperature and ionosphere.One of those is Liu et al. (2020) where they show the change of zonal mean neutral temperature with a doubling of CO 2 at altitudes from 50 to 400 km in June.They see the largest changes in the lower latitudes of both the southern and northern hemispheres and at even higher latitudes in the northern hemisphere.This agrees with Figure 6e at 300 km for the northern hemisphere during the entire ten-decade period but only for the period beginning in the 1970s in the southern hemisphere.Both studies also show an increase in cooling in the polar regions (70 o -90°N/S latitude) of both hemispheres, more so in the southern polar region.Again, this is the case in the southern hemisphere for only the 1970s onward but for the entire ten-decade period in the northern hemisphere.Zhou et al. (2023) and Yue et al. (2022) found this difference between the northern and southern hemispheres to be related to the larger movement of the magnetic pole in the north relative to the south, inducing a hemispherical asymmetric change in Joule heating rates.
We also look at neutral mass density changes over this ten-decade period and compare those also with previous studies.Observations discussed in Keating et al. (2000), Emmert et al. (2008), andSaunders et al. (2011) find −5%, −5.5% ( ± 1.4%), and −7% per decade neutral density changes under low solar activity during the latter part of the 20th century.Solomon et al. (2015) found under solar minimum conditions a −4.9% decrease of neutral density using the Thermosphere-Ionosphere-Electrodynamics General Circulation Model (TIE-GCM).The previously mentioned Solomon et al. (2018) modeling study found a change of −3.9%/decade at 400 km for 1970s-2000s period.We find a change of −5%/decade for this same period, which is very close to observation studies, particularly the first and second, and larger than the modeling study.Cnossen (2020) found −2.4%/ decade change at 400 km for the 1950 to 2015 period.We find a change −3%/decade from the 1950s-2010s.The reason for this difference was discussed in Cnossen (2020) as follows.A larger density change occurs during solar minimum conditions and a smaller density change during solar maximum conditions due to the larger effect of nitric oxide cooling as solar activity increases (Qian et al., 2006).Since Cnossen (2020) used results from simulations for all solar activity levels before removing the solar cycle, the density change is lower compared to the results here under just solar minimum conditions.Liu et al. (2020) found a −40% to −60% change with a doubling of CO 2 .We find a change of −2.9% per decade for the entire period and −5.0% for the 1970s-2000s period.For a doubling of CO 2 , as mentioned above, we have a 30.4% increase of CO 2 in this study as shown in Figure 8a.We find a change of density at 200 km of −17%, which is equivalent to a −59.275% change for a doubling of CO 2 , near the larger end of the decrease in the Liu et al. (2020) analysis.
As was discussed above, the changes in H density over these 10 decades as shown in Figures 2e and 4c are different than the changes in other atmospheric densities.The reason for the increase rather than decrease is that H density in the thermosphere is determined not only by changes in CO 2 but also by changes in methane (CH 4 ) (Nossal et al., 2016).This is related to the idea that a main source of H in the upper atmosphere is from CH 4 and other hydrogen containing intermediary species in the middle atmosphere such as water vapor.From Figure 8b, we can see that in this simulation CH 4 mixing ratio has increased by +80% during the 1920s-2010s which resulted in an 15.25% increase of H density in the thermosphere during this same period (Figure 4c).In addition, this rate of change of CH 4 corresponds well with the rate of change of H density with a slower rate in early 20th century and faster rate afterward.Toward the end of the period examined in this study, there is an indication that this H density rate of increase slowed, which corresponds with a slowing of the rate of increase of CH 4 as shown in Figures 2e and 4c 2016) study, they ran climate simulations with a one-dimensional version of the NCAR Thermosphere-Ionosphere-Mesosphere-Electrodynamics General Circulation Model (TIME-GCM).They found that higher thermospheric hydrogen density is correlated with lower temperature due to CO 2 cooling of the upper atmosphere as well as with increases in CH 4 .
In the ionosphere, there have been several model studies of the long-term changes.Cnossen (2020) looked at TEC results from a modeling study using a different version of WACCM-X for the 1950 to 2015 period and found a global mean change of −0.19 TECU/decade ( ± 0.03 TECU).In Figure 3d is the time series of electron column density from the model.It should be noted this is only the column total up to the model top so does not include the entire column of TEC.From Figure 3d, the change from the 1950s-2010s is −0.096TECU/decade and −0.063 TECU/decade for the 1920s-2010s.There are a few possible explanations why the TEC rate of change of the current study is lower than that of the Cnossen (2020).First, results here are under solar minimum conditions instead of a realistic varying solar cycle.Second, as previously mentioned, the two studies were done with different versions of the WACCM-X model.In addition, and likely a more minor effect, this study ends in the early 2010s instead of 2015.We can also look at the electron column density changes over time with latitude for March (Figure 6d) and June (Figure 6h).Depending on latitude, a range of change values from −0.045 TECU/decade to 0.2 TECU/decade are apparent.This makes clear a global mean would give a lower change than Cnossen ( 2020), but we are including the three decades before 1950.A closer look at Figures 6d and 6h leads us to conclude that this is reasonable since there are latitudes where the electron column density increases in the period leading up to 1950.Zhou et al. (2023) and Yue et al. (2022) found that a larger vertical drift in the equatorial region because of decreased geomagnetic fields will result in less electron density around the equator and more in the EIA region which would explain the low latitude electron density changes in Figure 6.Zhang et al. (2011) analyzed nearly four decades of incoherent scatter radar observations and found at noontime a change in electron density of +0.45%/decade at 150 km and between −0.05% and −0.35%/decade at higher altitudes.In Figure 5a, at 150 km the change we see is +0.67%/decade and decreasing to as low as −1.94%/decade near the top of the model.These discrepancies between observations and the model results could be due to the time-of-day difference of noontime versus evening and single geographic point versus global mean, the latter of which can result in significant differences particularly due long-term changes of the Earth's magnetic field at this Millstone Hill location.The transition level from the positive to negative changes in electron density in Figure 5 of 210 km is near the same level as that of Zhang et al. (2011) but lower than that from Zhou et al. (2022) where they find the transition at 280 km.This could be due to their study only analyzing data from 30°North to 30°South in latitude, which is the equatorial and low latitude region and usually has a higher F2 peak compared to the global mean domain of our analysis.Qian et al. (2008) found this transition to occur around 260 km when using TIE-GCM global means and a doubling of CO 2 , but that study did not use a whole atmosphere model.Further in-depth investigation is needed to reveal the sources of these inconsistencies.Zhang et al. (2011) also analyzed observations of electron temperature and found changes of ∼ −5 K/decade at 150 km and, above that altitude, a range of +10 to +45 K/decade with a change of ∼+25 K/decade at most altitude points.Figure 5b shows the electron temperature changes from the 1920s-2010s reach a minimum of −2.1 K/decade below 150 km and a maximum of +3.9 K/decade above this altitude.The reason for these differences, particularly large at higher altitudes could be the same as those mentioned above for electron density.What is clear is the changes from the 1920s-1980s in electron density and electron temperature are only half the changes for the entire period from the 1920s-2010s.In Solomon et al. (2018), they found a decrease in ion temperature of −2.7 K/decade at height of the peak ion density from the 1970s-2000s.Here we find as shown in Figure 3b and a decrease of −3.7 K/decade for this time range near this altitude.The reason for this difference appears to be due to the two different model versions being used and, more specifically, the steeper gradient in ion temperature changes near the top of the model.Observations of ion temperature in a similar period as the Solomon et al. (2018) are mentioned in Zhang and Holt (2013) and Zhang et al. (2016) where they found a gross decrease of −4 K/decade, very similar to our −3.7 K/decade value.But they also found noontime decreases at a few locations on the order of −10 to −30 K/decade, much higher than in our simulations.
One common theme in the results and comparisons with previous studies described here is the correspondence of changes in thermosphere variables with changes of greenhouse gases.This is obvious when noticing that as the rate of increase of CO 2 changes more rapidly in the latter decades of the 1920s-2010s as shown in Figure 8a, the rates of decrease of neutral temperature (Figure 1d) and neutral density (Figure 2a) also change more rapidly.This can also be seen more clearly by comparing the vertical profile changes of CO 2 in Figure 4d with the changes in Figure 4a for neutral temperature.This can also be seen in some of the ionosphere variables like the electron density and electron and ion temperatures in Figure 5.By comparing Figure 8a with Figure 1, we can estimate the climate sensitivity in our simulations in the form of change of temperature for a doubling of CO 2 .Since in this study we have a 30.4% increase in CO 2 from the 1920s-2010s in Figure 8a, from Figure 1 this works out to +3.125 K near the surface, −19.737K in the upper stratosphere, −10.855K in the upper mesosphere, and −55.921K in the thermosphere.The reported climate sensitivity for this version of the model from previous lower atmosphere simulations is +5.3K (Gettelman et al., 2019) but paleoclimate studies like Zhu et al. (2021) have found this estimate to be too high.A couple of caveats are that our simulations are not using a fully coupled climate model, are only for five-member decadal ensembles, and for fixed solar minimum conditions.It is important to note when comparing long term changes due to greenhouse gases throughout vertical regions of the atmosphere that the physical processes involved are very different and that we have simplified the upper atmosphere and lower atmosphere in this study.Given that, when looking at the correlation of thermosphere temperature with CO 2 mixing ratio in Figure 8c and CH 4 mixing ratio in Figure 8d, the correlation is very high with correlation coefficients of 0.995 and 0.95, respectively.The reason for the lower CH 4 correlation coefficient relative to CO 2 is the smaller change of CH 4 in the last few decades in Figure 8d.It is clear the CH 4 rate of increase in Figure 8b has slowed more than the temperature rate of decrease in Figure 1d toward the end of the time series.
It is important to note that the Earth's magnetic field changes significantly over these 10 decades and has been shown to have a sizable geographically local effect on changes in the thermosphere and ionosphere (Cnossen, 2020).This can be seen in Figures 6 and 7 for equinox and solstice months over the Atlantic sector.This effect of the magnetic field was further examined in a study by Qian et al. (2021) for the period from the 1960s-2010s.They found, when looking at global mean changes, such as in this study, the varying geographically local magnetic field effects result in both positive and negative changes and offset, resulting in negligible contribution in a global mean sense.Since the Qian et al. (2021) study examined simulations from the same version of the WACCM-X model as this study, the global mean changes discussed here, especially in the thermosphere, can be attributed almost entirely to greenhouse gas increases with the caveat as pointed out in Zhou et al. (2021) that these magnetic field changes can redistribute CO 2 in the upper atmosphere.
From everything we have shown thus far, it is clear the WACCM-X model in this specific simplified configuration can give a reasonable simulation of the past century of climate changes, especially in the upper atmosphere.With this capability moving forward, there is added confidence in similar simplified projections, as well as fully coupled transient projections into the next century.This has been done in a couple of recent studies, one by Brown et al. (2021) where they also did decadal WACCM-X simulations but for each decade of the 21st century focusing on future satellite drag predictions.In another by Cnossen (2022), they examined the period 2015-2070 using a WACCM-X model configuration like that used by the lower atmosphere CAM and WACCM models when doing a middle of the road scenario projection for the Intergovernmental Panel on Climate Change (IPCC) assessments.An obvious next step would be to choose other IPCC type configurations and do entire 21st century WACCM-X projections to predict how the upper atmosphere will change under various IPCC scenarios with different specified levels of greenhouse gases along surface global warming mitigation efforts over the next century.

Conclusions
Examination of the connection of greenhouse gas increases to changes in the Earth's atmosphere have been ongoing throughout much of the 20th century and early 21st century.Models like CAM and WACCM, which are the building blocks of WACCM-X, have simulated changes during this time for hindcast verification with observations of these models' capability to accurately simulate the Earth's changing climate from the surface up through the mesosphere.Using the ability of WACCM-X to self-consistently simulate the entire atmosphere using historical concentrations of greenhouse gases, while keeping solar conditions constant at low solar activity, we can now represent changes in the atmosphere up to around 600 km spanning the decades from the 1920-2010s.In this paper, we show for the first time the decade-to-decade continuous changes of upper atmosphere variables for 10 decades from the early 20th century to the early 21st century, revealing the consistent impact greenhouse gases have had on upper atmosphere climate.Upper and middle atmosphere cooling, as well as lower atmosphere heating are revealed by a single model consistent with fundamental climate change theory.Just as the lower atmosphere 20th century model simulations, we verify our results using comparisons with observations and with previous modeling study results.With straightforward simplification of the WACCM-X model by limiting solar activity to the minimum phase of the solar cycle with a quiet sun and using global means to eliminate the effect of the changing magnetic field, we can examine quantities in the thermosphere as they are affected almost purely by greenhouse gas changes.While the effect on long term thermosphere and ionosphere changes due to the Earth's magnetic field can be large regionally, Qian et al. (2021) concluded this magnetic field evolution has a negligible effect on global mean long-term changes in the thermosphere and ionosphere.With just over a 30% increase of the major greenhouse gas, CO 2 , during the 10 decades of this study, we find globally, near 400 km in altitude, the neutral temperature decreases by −1.7 K/decade and the neutral mass density decreases by −2.9%/decade.The O and N 2 densities also decrease but the O/N 2 ratio increases.In contrast to these other densities, H density near 400 km increases by +1.7%/decade in large part due to a +81.6% increase in CH 4 .In the ionosphere, we find a change in electron temperature of +22K, ion temperature of −19K, and electron density of −16.6%, all at 400 km.There is also a −0.57TECU (−0.063TECU/decade) or about −10.2% decrease in electron column density during this 1920s-2010s period.It should be noted again that the electron column density simulated by the model only includes the column contribution below the model top.Nevertheless, these results compare well with previous studies which made use of observations and various flavors of atmospheric models.
Not only do we reveal for the first time these 1920s-2010s continuous upper atmosphere changes, but we also find that the correlation coefficients of greenhouse gases CO 2 and CH 4 with upper atmosphere temperature during these 10 decades are very large, 0.995 and 0.95, respectively.This high correlation confirms, when the solar cycle variability of solar irradiance and geomagnetic disturbances are not factors, that there is an undeniable relationship continuously over the entire past century of greenhouse gas increases and long-term upper atmospheric changes.In the upper atmosphere, the radiative relaxation time is on the order of days, so the response of the thermosphere is expected to be prompt from forcings such as from CO 2 and CH 4 .This is exemplified by the model as the greenhouse gas changes are about 5% of the total and thermospheric changes are smaller in the early decades of the 1920s-2010s and these changes are larger in the later decades, when greenhouse gases increase by over 25%.It is also clear from Figure 1 this correlation in the model is smaller when going down in altitude through the mesosphere, stratosphere, and troposphere.This is expected since the solar cycle variability is less prominent and the radiative relaxation time scales of greenhouse gases affecting the atmosphere are longer due to the mechanisms and processes being very different and becoming more complex in these lower regions.Importantly, this high correlation is another affirmation that the thermosphere is an important region of the atmosphere for observing, modeling, and predicting climate change.So, with a model that produces a reasonable simulation of the thermosphere and ionosphere over the past century, we can be confident when using WACCM-X for future projections of upper atmosphere changes over the century ahead and for modeling the effects on the atmosphere of ongoing and future climate mitigation efforts.Zhou, and two anonymous Reviewers for their time examining the manuscript and providing invaluable and insightful ideas and comments.We also thank the model developers of the CESM, CAM, WACCM, and WACCM-X models.WACCM-X simulations were performed using computational resources at the NCAR-Wyoming Supercomputing Center (doi:10.5065/D6RX99HX).The National Science Foundation National Center for Atmospheric Research is a major facility sponsored by the National Science Foundation under Cooperative Agreement No. 1852977.This work was supported by NSF Grant AGS-2050072 in cooperation with the University of Wisconsin-Madison and Embry Riddle University.

Figure 1 .
Figure 1.1920-2010s temperature time series 5-year global mean decadal values for four vertical levels near, (a) the Earth's surface (b) 46 km in the stratosphere, (c) 90 km in the mesosphere, and (d) 400 km in the thermosphere.For reference, the model lower boundary carbon dioxide and methane concentration time series are included in Figure 8.
3 g/cm 3 near the south pole.In both March and June, after about 1950, the rate of this cumulative change gets quicker until about the year 2000.The cumulative change in the period prior to 1950 contributes a relatively small amount to the total cumulative change in both March and June at all latitudes.The same representations as neutral density but for electron density in March and June are shown in Figures 6c and 6g.The largest negative cumulative changes of electron density are in the southern hemisphere low-latitude region of about −0.6 × 10 5 cm −3 or about −20% relative to the 1920s values in March and around −0.3 × 10 5 cm −3 or about 15% less than the 1920s values in June.Again, for both months, the cumulative change at all latitudes prior to 1950 is relatively smaller than afterward.The decreases become larger after the 1970s at all latitudes.Figures 6d and 6h are the same but for electron column

Figure 6 .
Figure 6.Five-year zonal mean decadal value differences relative to the 1920s at March equinox (top) and June solstice (bottom) for neutral temperature on the 2.84 × 10 −8 hPa pressure surface (a) at ∼295 km and (e) at ∼285 km, neutral density (b) at ∼377 km and (f) at ∼395 km, electron density (c) at ∼377 km and (g) at ∼395 km, and (d) and (h) electron column density.
and Mlynczak et al. (2022) found a cooling ranging from −0.5 to −0.8 K/decade for the years 2002-2018.Our simulations show a similar cooling of about −0.5 to −0.7 K/decade for the 1970s-2000s and −0.34 to −0.7 K/decade for the 1920s-2010s.The lower mesopause change values for the 1920s-2010s is expected as previously discussed due to the lower rate of greenhouse gas increases in the earlier decades but interestingly the values are the same in the stratopause region for the two time periods.The latter requires further investigation.
compared to Figure 8b.In the previously mentioned Nossal et al. (