Millstone Hill ISR observations of upper atmospheric long-term changes: Height dependency



[1] Ionospheric ion temperature is an excellent approximation to neutral temperature in the upper atmosphere, especially, for altitudes below 300 km. This analysis of long-term ionospheric ion temperature changes between 100 and 550 km at noon is based on a database of incoherent scatter radar observations spanning more than three solar cycles during 1968–2006 at Millstone Hill and provides direct evidence of long-term changes and their height dependency in the upper atmospheric temperature. A cooling trend at altitudes above 200 km and an apparent warming trend below 200 km are found. The cooling increases with height and shows variability with solar activity. The apparent warming varies with season and solar activity. It may result from the thermal subsidence caused by atmospheric contraction and pressure level change and from the ion temperature overestimation in the F1 region due to ion composition long-term changes. These long-term changes in ion temperature are accompanied by changes in electron density, being lower above the F2 peak and higher below the F2 peak. Electron temperature is accordingly enhanced. All these changes appear to be suggestive of a long-term greenhouse gas effect.

1. Introduction

[2] Greenhouse gases such as CO2 and CH4 are increasing in the lower atmosphere due to human activities. The effect of this on the upper atmosphere was first investigated in a theoretical modeling study by Roble and Dickinson [1989], suggesting a major greenhouse cooling in the thermosphere in response to increases in the CO2 and CH4 concentration. Such a cooling effect causes a global reduction in neutral densities including O, N2, O2 and in total neutral mass density at fixed altitude as the neutral temperature Tn decreases. If greenhouse gas concentrations are doubled, as predicted to happen by the end of the 21st century, Roble and Dickinson indicated that the decrease in thermospheric temperature will be as much as 50 K, and the decrease in thermospheric densities at a fixed height will be 40%–50%. The ionospheric consequences, as a result of thermal contraction, include a decrease in the F2 peak height, hmF2 [Rishbeth, 1990; Rishbeth and Roble, 1992], a decrease in topside ionospheric density and an increase in the F1 region ionospheric density, with less change in the maximum density NmF2. Some recent modeling studies using TIMEGCM and a more realistic CO2 concentration (single site measurements) have essentially confirmed prior model findings [Qian et al., 2006, 2008]; the greenhouse gas enhancement induced lower thermosphere changes and associated physics were further studied by Akmaev and Fomichev [1998].

[3] The above results, arising from theoretical considerations, imply that ionospheric parameters can be a sensitive indicator of greenhouse gas atmospheric effects. Many subsequent investigations of long-term ionospheric and thermospheric change have occurred and the area is a very active research topic. Most of these studies are based on long-term ionosonde observations which date to the early days of ionospheric research and are readily accessible. Bremer [1992] published the first such ionospheric long-term trend observational results. Recent studies include those by Ulich and Turunen [1997], Bremer [1998], Jarvis et al. [1998], Upadhyay and Mahajan [1998], Danilov and Mikhailov [1999], Mikhailov and Marin [2001], Xu et al. [2004], Yue et al. [2006], and Laštovička et al. [2006a, 2008]. Most of these studies concluded that hmF2 indicated a decreasing trend; other studies did not find a long-term trend or found a trend opposite to the sense predicted by Rishbeth and Roble [1992].

[4] Observations of thermospheric mass total density by satellites revealed a 2%–5% decrease per decade [Keating et al., 2000; Emmert et al., 2004, 2008]. These have been considered to be evidence of thermospheric cooling, supporting the Roble and Dickinson [1989] theoretical calculations of the greenhouse gas effect.

[5] However, the greenhouse gas effect may not be the sole reason for the observed secular changes in the ionospheric parameters (especially hmF2). Long-term changes in both solar and geomagnetic activity [Mikhailov and Marin, 2001; Solomon et al., 2010a], and secular variations of geomagnetic field orientation [Cnossen and Richmond, 2008; Yue et al., 2006] can impact trends in the upper atmosphere. These factors need to be carefully accounted for when pursuing upper atmospheric climate study.

[6] Progress has been made in identifying trends in the upper atmosphere in the past two decades, however, there are several outstanding issues yet to be fully addressed; any major ionospheric trends are probably associated with thermospheric trends in temperature, composition, and winds, and there are very few prior studies on trends in these parameters. Using incoherent scatter radar (ISR) observations for the upper atmospheric long-term trend study can provide an important insight into these issues, because of the following points, as pointed by Holt and Zhang [2008].

[7] 1. ISRs have the capability of directly monitoring the thermal status of the upper atmosphere, and radar observations of plasma temperatures and densities can be used to derive neutral temperature and composition [Oliver, 1979].

[8] 2. Ion temperature (Ti) variations contain a key feature for deriving long-term trends. Ti follows variations in the solar activity rather well (linear for most values of F10.7). This important feature makes it much easier to remove effects of the solar activity on long-term trends. This is in comparison with ionosonde measurements of Ne in the F2 layer (NmF2) which exhibit a somewhat more complicated response to solar activity [see Balan et al., 1994; Richards, 2001].

[9] 3. ISRs provide altitude profiles of a comprehensive set of ionospheric/thermospheric parameters. Examining physical consistency of results among multiple plasma parameters, which are closely related, we can verify variations of individual parameters and gain better understanding of physical processes.

[10] In an initial attempt to prove a direct measure of the upper atmospheric temperature trend in the Millstone Hill ISR data over 1978–2002, Zhang et al. [2005a] identified a negative Ti trend for most F2 region altitudes and seasons, but the confidence interval was rather large. Holt and Zhang [2008] showed a significant cooling trend in Ti at 375 km based on Millstone Hill ISR data for a very impressive 30 year period in 1978–2007. The rate of cooling for noontime at this attitude was roughly −4.2%/decade, and the confidence level of the trend value was carefully addressed. Recently, Donaldson et al. [2010] conducted an extended study of Ti trend with height and time dependency. Their ISR data was from observations made at St. Santin for a two-solar cycle period (1966–1987). Their results suggested a similar or even stronger cooling in the topside ionosphere and warming in the E-F region heights. They also indicated the local time dependency of the trend, being larger during the day than at night. It should be noted that this was a relatively small (short time span) data set for a period when the global warming signals in the ground/low atmospheric temperature just emerged. St. Santin is at a geomagnetic and geographic midlatitude while Millstone Hill has a very similar geographic latitude (42.6°N) but a higher magnetic latitude (∼ 52°N). In our study, we will use Millstone Hill data which spans roughly 40 years starting from the mid 1960s. This will allows us to address the trend in a real “long-term” sense and to reveal the up-to-date situation in the upper atmosphere as compared to what was shown in St. Santin data representing the very beginning phase of the global change.

[11] In this paper, we start with some brief introduction to the Millstone Hill ISR observation and database, then we describe the data processing method for trend determination. We will examine observations around noon when the best quality of data is expected, and focus on Ti results and present height dependency of the Ti trend. We will also address variability in the trend with season and solar activity, and discuss trends in electron temperature and electron density. Important information on the thermospheric temperature and composition can been also derived based on the same long-term Millstone Hill ISR data set and consideration of the energy equation for the ions. This is addressed in our accompany paper (J. M. Holt and S.-R. Zhang, Millstone Hill ISR observations of upper atmospheric long-term changes: Neutral parameters, submitted to Journal of Geophysical Research, 2011).

2. Measurements, Data, and Method

[12] Millstone Hill UHF incoherent scatter radar system operates with a zenith-directed 68 m diameter fixed parabolic antenna, which commenced operation in 1963, and a fully steerable 46 m antenna, which commenced operation in 1978. Radio signals are transmitted at 440 MHz by two 2.5 MW transmitters. Different experiment modes have different temporal and spatial resolutions depending on pulse length applied and integration time of signals. Long integration times provide high sensitivity but low time resolution, and large backscatter volumes provide good signal-to-noise ratios but comprised altitude resolution. The electron density and plasma temperatures are determined from the received power and spectrum.

[13] Long-pulse data for the period 1964–1974 were originally presented as contours in a series of technical reports [e.g., Evans, 1969]. These are 210 diurnal variation contours either as monthly means or for individual days. They have been digitized recently and entered into Millstone Hill ISR database system (Madrigal). But data prior to 1968 were not included due to possible systematic errors arising from uncorrected instrumental effects. For the period 1970–1975, there are 87 high spatial resolution two-pulse data [Oliver and Glotfelty, 1996] which provide good E region density and temperature measurements as well as some F region coverage. The majority of the data, however, is from the 1976–2010 period when standardized data processing procedures have been applied. Previously we used this latter part of data (until 2002) to construct climatology models of the ionosphere [Holt et al., 2002; Zhang et al., 2005b], the ionospheric convection model [Zhang et al., 2007], and the ionospheric variability model [Zhang and Holt, 2008]. Data for the period 1976–2007 were also used in the trend study by Holt and Zhang [2008]. For our current paper, we concentrate on the zenith antenna observations through the end of 2006, and data since 2007 is not included to avoid complexity in data interpretation and processing because, as indicated by Emmert et al. [2010] and Solomon et al. [2010b], the years 2007–2009 were under extended solar minimum conditions where the neutral density was too low to be ascribed to long-term trend effects. Proper F10.7 terms may be needed to deal with data for these 2 years. In summary, both the two-pulse data and the data from 1976 onward were extensively used in various climatology studies; earlier data were mostly used in referred publications but not in systematical climatology studies yet.

[14] We take the same approach of data processing to derive the trends as described by Holt and Zhang [2008]. Data with F10.7 > 300 or ap > 80 are eliminated to minimize effects from extreme solar-geophysical conditions. Data for 1630–1730 UT (local noon) are first binned according to height. These height bins are 100–120, 120–140, 140–160, 160–180 and 180–200 km each with a 20 km interval for the E and low F region, and 200–250, 250–300, 300–350, 350–400, 400–450, 450–500, and 500–550 km each with a 50 km interval for the F region. Figure 1 shows the Ti data statistics. Figure 1 (top) shows Ti data counts in the log10 unit as a function of the float year with fraction representing the month. These data counts are the total number of data points in the entire altitude range concerned for a given month of the year. These are the data after a very preliminary filtering of obvious bad data. From 1975 to 1976, there were no Ti data included. Figure 1 (bottom) is the height distribution of the Ti data over the entire time period. The E-lower F region data are relatively few because of the small altitude bin size and of few E-low F region observations (especially after the two-pulse experiment years until the early 1980s).

Figure 1.

Data distribution as a function of (top) year/month and (bottom) altitude.

[15] For each height bin, medians of the data are computed for each month in the data subset, yielding ∼390 (for the F region) or 200–300 (for the E and lower F region) median-filtered points. Taking monthly median values serves three major purposes. (1)Most outliers are eliminated. (2) Overweighting of long-duration experiments is eliminated. (3) The need to correct for short-term autocorrelation (on the order of hours or days) in the data is eliminated [Weatherhead et al., 1998]. Months for which there are fewer than six points are eliminated, leaving ∼90% (for the F region) or 70%–80% (for the E and lower F region) of the monthly medians. This procedure helps to insure that outliers are reliably eliminated by the median filter.

[16] Least squares fits to Ti with terms of F10.7, ap and the year are then computed. The independent variables of the fits are time (year, including a frictional part corresponding to the day number of the experiment), F10.7 and ap. The F10.7 dependency shows a weak saturation effect at high values of F10.7 (see Figure 8). This is addressed by adding a quadratic term in the F10.7 dependence. Therefore, Ti = Tb + t × (yequation image) + f1 × (F10.7 − equation image) + f2 × (F10.7 − equation image)2 + a × (Apequation image), where y is the float year, equation image is the mean float year of the whole data set, F10.7 is the daily F10.7 cm flux, equation image is the mean F10.7 of the whole data set, Ap is the daily Ap index, and equation image is the mean Ap value of the whole data set. The background term Tb, long-term trend t, and F10.7 and ap term coefficients f1, f2 and a are obtained through least squares fits for any given bin. The relationship between the upper atmospheric temperature (especially the neutral temperature) and indices of solar F10.7 and magnetic Ap have been well explored, and similar equations have been widely used in empirical models [e.g., Hedin, 1987; Zhang et al., 2007; Donaldson et al., 2010].

[17] In a trend study, the uncertainty of trend estimates is a very important concern. The standard deviation error bar can be used to express the uncertainty based on a Gaussian distribution assumption. An alternative approach is the bootstrap [Chernick, 1999; Efron and Tibshirani, 1993]. This relatively new and now widely accepted computer intensive technique makes no assumption about the distribution of the data. A large number of bootstrap data sets are generated from the actual data set (or from the residuals of the fit after removing terms of solar activity, magnetic activity dependency) by sampling with replacement. A least squares fit is computed for each of the bootstrap samples and the limits of the 95% bootstrap confidence interval are the points on the histogram of the resulting fit parameters such that 2.5% of the parameter values are less than and 2.5% greater than the limits. We use 10000 bootstrap samples in this paper.

3. Result and Discussion

[18] In this section, we present Ti trend results at various heights and provide some brief discussion on these results.

3.1. Height Dependency of the Trend

[19] The noontime ion temperature measurements are shown in Figure 2, where each panel corresponds to a height bin for which observed monthly median Ti values and a linear fit to them are given. These values are from observational data with terms of background, F10.7 and ap being removed. These are the terms in equation, Ti = Tb + t × (yequation image) + f1 × (F10.7 − equation image) + f2 × (F10.7 − equation image)2 + a × (Apequation image), which was mentioned in section 2. Very clear cooling trends can be seen in the upper F region (>200 km). They change with height, and are more significant at high altitudes. The derived cooling trend is on the order of −0.6 to −3.9 K/yr between 200 and 550 km. It is interesting to note that toward the ending years (2003–2006), Ti is consistently below the trend line, suggesting the strongest cooling in the recent years. At the beginning years (the late 1960s), however, Ti is below the trend line for higher altitudes (>400 km) and close to the level for the ending years; but at lowest heights (200–300 km), Ti is more or less on the trend line. It seems that the cooling trend emerged earlier at lower altitudes.

Figure 2.

Ti data and trend at various heights in the F region (>200 km). The data points are a result after removing terms of background, F10.7, and ap (see text). The solid line is a fit to the data points to give the long-term trend value (marked at the bottom left corner in each panel).

[20] The trend at 375 km was determined as −4.7 K/yr using the 1978–2007 data set over Millstone Hill in the work by Holt and Zhang [2008]; this rate is smaller here for the 1968–2006 period because of including additional data prior to 1978. Indeed for those early years (prior to 1978), the long-term change did not appear to be pronounced. The cooling trend started to emerge only at the breakpoint in the 1980s as shown in Figure 2 as well as in the work by Danilov [2008] and Donaldson et al. [2010], therefore inclusion of observations before this breakpoint can result in reduction of the derived cooling trend. The trend result is meaningful only for a specified period.

[21] In the E and lower F region, the situation is very different. As shown in Figure 3, ion temperature shows a warming trend at 180–200 km, then the warming becomes the largest around the F1 region (160–180 km) and weaker in the E region (100–120 km). The positive trend at the lowest height seems to be consistent with the Lidar observation shown in the work by She et al. [2009]. However, with the Pinatubo volcanic response accounted for, the Lidar trend is much smaller. At 140–180 km, our data show much larger scatter than in other areas, therefore the confidence limits of the trend estimate are wide. In the ISR data reduction, an ion composition model is used to determine plasma temperatures. The ion composition model for Millstone Hill ISR data has been assumed to be constant without any long-term changes, however, such constant ion composition may not be realistic given the fact of neutral composition long-term change. In fact, the Ti data based on a fixed ion composition model can overestimate the true ion temperature due to a molecular ion long-term increase, as implied in the electron density increase in the lower F region (or F1 region) (see discussions in section 4.2) and other ionosonde observations [Laštovička et al., 2006a]. This is mainly because that the dominant ion species in this region (and lower altitudes as well) are molecular ions. This overestimate would be most significant in the F1 region and impart an apparent warming in Ti. In the E region, the long-term decrease in the molecular ion ratio NO+/O2+ was reported by Danilov [2002], and can result in offset in the E region Ti measurement. See more discussions by Donaldson et al. [2010].

Figure 3.

Same as Figure 2 but for the E-lower F region altitudes (<200 km).

[22] On the other hand, as pointed out by Akmaev and Fomichev [1998] and [Akmaev et al., 2006], the long-term trend at a fixed altitude combines consequences of a true long-term temperature decrease at a fixed pressure and the subsidence of air cells as a result of the pressure level moving downward. The thermal contraction associated subsidence appears to enhance neutral temperature in the 100–200 km range, where there is a large positive temperature gradient with height. In fact, following the long-term temperature decrease, the hot air subsidence toward lower altitudes of low temperature and high pressure can cause an apparent warming. This warming is exactly shown in our data. Donaldson et al. [2010] have evaluated the true cooling and temperature change due to the pressure level change (subsidence). They found even in the region of 100–200 km, there exists a clear true cooling.

[23] In conclusion, the apparent warming throughout the E and lower F region can be caused by the pressure level change associated thermal subsidence, or as a sign of Ti overestimate due to ion composition long-term changes. The warming is unambiguous and fits well into the gradually decreasing cooling trend toward lower altitudes in the upper F region; the transition from cooling to warming as height decreases is smooth and gradual.

[24] A summary plot of the derived Ti trend is given in Figure 4, where both the rate of long-term change per year and the percentage change per decade are given as a function of height. The error bars are 95% bootstrap confidence levels. The average Ti for the entire data set at each altitude bin is calculated and used for the percentage change calculation. The trend are more cooling with increasing height from above 200 km, with a rate being up to −4 K/yr, or −3%/decade. Below 200 km, there is a clear warming trend, and the rate can be up to +3 K/yr, or +3%/decade. The error bars are large especially in the 150–200 km height range.

Figure 4.

The altitude profile of the Ti long-term trend at noontime. (left) The trend in the K/yr unit and (right) the same trend but in the percent/decade unit.

[25] Comparing our midday Ti trend for Millstone Hill with the diurnal average Ti trend for St. Santin [Donaldson et al., 2010], we see a very similar height dependency where the cooling increases with height in the F region and the apparent warming below 200 km with the greatest warming at 150–175 km. Our height profile of the trend is much smoother, and the cooling rate at K/yr is generally smaller.

[26] As noted at the beginning of this section, the cooling trend emerged around the early 1980s. The derived cooling trend should be more significant for the period 1980–2006 than for the period 1968–2006. The result for the 1980–2006 period is then calculated and shown in Figure 5. The overall height dependency of the long-term change is exactly the same as shown in Figure 4 where the earlier data set is included, but now the maximum cooling is above −5 K/yr or > −4%/decade, as compared to below −4 K/yr or −3%/decade. Data for below 150 km are probably insufficient and the corresponding result is not shown.

Figure 5.

Same as for Figure 4 except that the data used are for the 1980–2006 period.

3.2. Trend Variability With Season and Solar Activity

[27] We have selected two subsets of data corresponding to seasons of winter (November, December, January, and February) and summer (May, June, July, and August). These results are shown in Figure 6. There is a clear seasonal change in the average (background) Ti for altitudes above 200 km where Ti is high in summer than in winter. The seasonal change in the average (background) Ti is relatively weak below 200 km (see Figure 6, right) where Ti, close to neutral temperature Tn, increases rapidly with height. The long-term trend in Ti, however, hardly shows significant seasonal dependency in the upper F region above 200 km, where the seasonal change of trends (either in absolute trend value or percentage change) is well within the uncertainty range of the derived trend. Below 200 km the warming appears larger and at higher altitudes in winter than in summer. In summary, there is some seasonal variation in the background Ti but not quite in the Ti trend above 250 km; below 250 km there is some seasonal variation in the Ti trend but not in the background Ti.

Figure 6.

Altitude profiles of the Ti long-term trend at noontime for different seasons. (left) The trend in the K/yr unit and (right) the trend in the percent/decade unit.

[28] Ion temperature dependency on solar flux is very strong. Results from Emmert et al. [2004] indicated that the response of the long-term trend in total mass density for 400 km to solar activity is nonlinear, with the greatest density decrease (corresponding to neutral temperature cooling) at low solar activity, and the roughly equal decrease at median and high solar activities. Therefore we arranged data into two groups according to the daily 10.7 cm solar flux: F10.7 <110, 110≤F10.7≤200. Trend results for different solar activities are shown in Figure 7. Throughout the entire height range from 100 to 550 km, a lower solar activity corresponds to a stronger cooling. This stronger cooling, being consistent over the large height area, perhaps indicates some common driver which works differently at low and median/high solar activities. Qian et al. [2006] suggested different solar activity dependencies of CO2 and NO which cause long-term temperature changes in the lower atmosphere. Our results are consistent with their speculation.

Figure 7.

Altitude profiles of the Ti long-term trend at noontime for different solar activity levels. (left) The trend in the K/yr unit and (right) the trend in the percent/decade unit.

4. Some Further Discussions

4.1. Ti Dependency on Solar Activity

[29] The most important step to deriving long-term Ti trends is to remove dominant variations with solar activity and season. Comparing average Ti profiles for seasonal and solar activity variability in Figures 6 and 7, we can see that the solar activity dependency is much stronger than the seasonal one. In fact, the dependency is height-dependent as demonstrated in Figure 8 for altitudes above 200 km and Figure 9 for altitudes below 200 km. Those data points are observational data after removing terms (mentioned in the equation in section 3.1) of Tb (background), yequation image (long-term trend) and Apequation image (magnetic activity); a least squares fitting to these data points using a function with linear and parabolic terms, f1 × (F10.7 − equation image) + f2 × (F10.7 − equation image)2, is given by solid lines. The strong linear relationship between Ti and F10.7 is pronounced for most altitudes and most F10.7 values, although the saturation, a prominent feature of the electron density response to F10.7 increase, is also visible in the Ti response to F10.7. The linearity between Ti and F10.7 largely holds true for approximately F10.7 ≤ 160 where the majority of data stays. The linear slope f1 for the Ti versus F10.7 dependency (marked at the bottom right corner of each panel) is highest at the 250–400 km height range. The slope at the topside decreases monotonically with increasing height. At the bottomside, it decreases monotonically with decreasing height. At those lowest heights, Ti hardly varies and is nearly independent of F10.7. The ions are largely heated by electrons through Coulomb collisions, and the electron density peaks in the range of 250–300 km during the day and increases with solar activity. The ions are cooled by the neutrals through ion-neutral collisions, and the solar activity dependency of neutral temperature is pronounced and well defined. These features in electron density and neutral temperature can be used to understand solar activity dependency of the ion temperature.

Figure 8.

Solar activity dependency of Ti for different height bins above 200 km. The data points are after removing from original observational Ti values terms of background, long-term trend, and ap. The solid line is a fit to the data points with a function of linear and parabolic terms, f1 × (F10.7 − equation image) + f2 × (F10.7 − equation image)2. The line slope is given at the bottom right corner of each panel.

Figure 9.

Same as Figure 8 but for height bins below 200 km.

4.2. Trends in Other Parameters

[30] Long-term changes in neutral temperature are reflected in ion temperature. This is in particular true for altitudes below 300 km. As a consequence of neutral temperature change, ionospheric electron density change can be expected, because a lower neutral temperature gives rise to a lower balance height which is determined by the equilibrium between photochemical loss and diffusion, and therefore reduces the F2 peak height [Rishbeth et al., 1978]. The neutral composition change induced by long-term cooling will also modify the O/N2 and then alter electron density in the ionosphere. Here we use electron density measurements from the same ISR data set to demonstrate corresponding ionospheric changes. The trend is derived using the same approach as for Ti.

[31] Figure 10 shows altitude profiles of the long-term trend in Ne and the average Ne. It can be seen that Ne increases at heights below 200 km (except for the lowest height where there seems to be a outlier). Between 200 and 300 km where the F2 peak height lies, the trend in Ne is nearly zero. At higher altitudes, a decreasing trend in Ne is very clear, and the decrease grows with height until at 400 km. This height profile of the trend in Ne agrees well with expectation of neutral temperature cooling caused ionosphere and thermosphere change [Rishbeth and Roble, 1992; Qian et al., 2008], as well as most ionosonde results for the ionospheric peaks from the E region to the F region [Laštovička et al., 2006a, 2006b, 2008]. It should be noted that most prior studies were using ionosonde observations from the 1950s to the 1990s period [Laštovička et al., 2006b]. Considering the breakpoint mentioned in section 3.1, we may expect those prior trend values are quantitatively small; in fact, another exercise we performed indicates that using the same ISR data but for the 1995–2006 period, Ne trends in the upper F region are nearly doubled as compared to those shown in Figure 10 for the 1968–2006 period.

Figure 10.

The altitude profile of the Ne long-term trend at noontime. Ne is in the log10 (m−3) unit. (left) The trend in the log10 (m−3)/yr unit and (right) the same trend in the percent/decade unit.

[32] Electron temperature is another plasma thermal parameter. Different from ion temperature which can be very close to Tn, electron temperature for the F region is not mainly associated with Tn but is much more variable. It has some clear seasonal changes, but its dependency on solar activity is complicated. As noted by Zhang et al. [2004], an enhanced solar EUV flux gives rise to more photoelectrons which in turn elevate plasma temperatures by heating processes (∼Ne); meanwhile, the increased electron density due to the enhanced EUV flux leads to an enhanced electron cooling rate through Coulomb collisions (∼Ne2), which may lead to a lower Te. The actual response of Te to a change in solar EUV is the result of, in addition to effects of heat conduction at high altitudes, these two competing processes, and depends much on the level of background Ne. Te trends, derived using the same ISR data set and the same method as for Ti, appear to be positive in the height range we are considering (Figure 11). The larger range of confidence limits is due to the difficulty of successfully removing solar activity dependency discussed above. The positive trend for above 200 km is perhaps related to the strong anticorrelation between Te and Ne for a given energy input: when Ne decreases (as shown in Figure 10), Te increases. At lower altitudes of below 200 km, the thermal coupling between Te, Ti and Tn are much stronger that the apparent warming shown in Ti occurs in Te in a very similar way.

Figure 11.

The altitude profile of the Te long-term trend at noontime. (left) The trend in the K/yr unit and (right) the trend in the percent/decade unit.

5. Summary

[33] This paper reports the long-term change in the midday ionospheric ion temperature over the 100–550 km height range, as measured by an incoherent scatter radar at Millstone Hill during a four solar cycle period between 1968 and 2006. Ionospheric ion temperature is an excellent approximation to neutral temperature in the upper atmosphere, especially, for altitudes below 300 km. A cooling trend at altitudes above 200 km and an apparent warming trend below 200 km are found. The cooling increases with height and shows variability with solar activity. The apparent warming varies with season and solar activity. It may result from the thermal subsidence caused by atmospheric contraction and pressure level change, and from the ion temperature overestimation in the F1 region during the radar data reduction due to ion composition long-term changes. These long-term changes in ion temperature are accompanied by changes in electron density, being lower above the F2 peak and higher below the F2 peak. Electron temperature is accordingly enhanced. All these changes appear to be in agreement with theoretical expectation of thermospheric and ionospheric responses to greenhouse gas changes.

[34] The cooling at higher altitudes and warming at lower altitudes are qualitatively well determined, however, the precise rate of long-term change depends on the exact time period considered since it seems that the cooling emerged “only” since the early 1980s. The earlier from this breakpoint the data set begins, the smaller the derived trend is. In fact, the data set for the 1980–2006 period gives a maximum cooling of above −5 K/yr or −4%/decade, and a −1.9 to −3.8%/decade cooling between 250 and 350 km (where Ti is close to the exospheric temperature), while the data set for the 1968–2006 period gives a maximum cooling of less than −4 K/yr or −3%/decade, and a −1% to −2% cooling between 250 and 350 km. Similarly, Ne trends in the upper F region for the 1995–2006 period are nearly doubled as compared to those for the 1968–2006 period.

[35] Our investigation has been focused on observations at noon. Much smaller cooling trend has been noted for St. Santin at midnight [Donaldson et al., 2010]. In a future paper, we will examine data for other local times. Other than sites at Millstone Hill and St. Santin, ISR observations have been made at Sondrestrom, Greenland for more than 2 solar cycles; in Poker Flat, Alaska area ISR observations were made in the 1970s and have now resumed since 2007. These data sets may be used to detect long-term trends at high latitudes as a major extension to the work described here.


[36] We thank William Rideout and members of the Haystack Observatory Atmospheric Sciences Group for assembling and maintaining the Madrigal Database. W. L. Oliver was responsible for verifying the two-pulse data and maintaining it for many years prior to its inclusion in the Madrigal Database. The Millstone Hill incoherent scatter radar is supported by the U.S. National Science Foundation (NSF) Upper Atmosphere Facilities Program under a cooperative Agreement between NSF and Massachusetts Institute of Technology. Work by James Kurdzo was supported by the NSF REU program at MIT Haystack Observatory in 2008.

[37] Robert Lysak thanks Alexey Danilov and another reviewer for their assistance in evaluating this paper.