Trends in the F2 layer parameters at the end of the 1990s and the beginning of the 2000s



[1] The behavior of the critical frequency foF2 of the ionospheric F2 layer is considered for the period 1990–2010. Various available databanks of ionospheric vertical sounding data are described and compared. The analysis is performed in terms of comparison of the foF2 data for the period 1958–1980 to the data for 1990–2010. Two moments of the day (14:00 LT and the after-sunset moment) and two seasons (winter and summer) are considered for 12 stations for which the necessary data were available. The scatter of the foF2 values (in terms of the standard deviation, SD) relative to the dependence on the solar activity index F10.7 is considered. It is demonstrated that the values of SD for the period 1998–2010 are much higher than for the period 1958–1979. This increase in the SD includes two factors: real increase in the foF2 scatter and systematic decrease (negative trend) in foF2. This systematic decrease makes it possible to provide an independent evaluation of the trend in foF2: −0.03 MHz per year. Analysis of foF2 behavior for each of all 12 stations is performed. The obtained linear trends are negative for all stations, but the trend magnitude varies from one station to another. The trends for two moments of time are found of the same order of magnitude: −0.024 and −0.054 MHz per year for the summer and winter seasons, respectively. Some conclusions on the behavior of the hmF2 trends for the same period are presented. Possible causes of the changes in the trends in the F2 layer parameters are discussed.

1 Introduction

[2] The problem of long-term changes (trends) in ionospheric parameters has been discussed already for more than 15 years. We will not consider here even briefly dozens of papers dedicated to this problem, referring instead readers to the recent reviews [Danilov, 2012; Qian et al., 2011; Laštovička et al., 2012]. Note that only two factors in our opinion make important an attempt to find trends in parameters of the F2 layer especially for the period from the end of the previous to the beginning of this century. One is that the foF2 and hmF2 trends before 1990–1995 have been studied by several groups of authors. However, the obtained results were somewhat contradictory and had low statistical significance. The most complete study of ionospheric trends for that period at the global network of ionospheric stations was performed by Bremer [2001]. A collective experiment on revealing foF2 trends by various methods and various groups of researchers was performed by Laštovička et al. [2006]. In this experiment, the average value of the trend of −0.1 MHz per decade were obtained with some scatter. However, the cooling and contraction of the middle and upper atmosphere [Laštovička, 2009; Laštovička et al., 2012; Qian et al., 2011, 2012] develops quite intensely. The latter fact is confirmed by the data on satellite drag providing negative trends in the density [see Qian et al., 2011; Emmert et al., 2008; Emmert et al., 2012] and by the negative trends in the thermospheric temperature obtained from incoherent scatter data at Millstone Hill [Zhang et al., 2011; Zhang and Holt, 2012] and Saint Santin [Donaldson et al., 2010]. In this condition, one could expect that trends of the F2 layer parameters would increase in time and for the end of the previous and beginning of the current centuries would be much stronger than during the last two decades of the 20th century.

[3] The second factor is that the observed changes during the last decades in the density and temperature of the thermosphere exceed substantially the changes expected from theoretical models. Actually, the observed trends in ρ and T exceed what is predicted for the current increase in CO2. This problem was considered in detail by Qian et al. [2011] and Solomon et al. [2012]. In this situation, one can expect that trends of the F2 layer parameters also would be higher in the beginning of the new century than they were during the last decades of the previous century and were predicted by theoretical calculations by Rishbeth [1990].

[4] In this paper, we attempted to collect all the values of foF2 and hmF2 available in ionospheric data banks and to analyze the trends of these parameters from 1990 to the end of the observations available for the given station. The problem of data selection is rather complicated so we briefly consider it in the next section.

2 The Initial Data and Procedure

[5] While searching for long-term trends of various atmospheric and ionospheric parameters, the principal problem is finding rather long series of homogeneous reliable initial data. In the majority of papers dedicated to trends in foF2 and hmF2 and published before 2005, either the data of some particular station or the data for a group of stations from the data bank in the IWG format, which were collected by the Boulder Space Data Center and issued in 2000 in the form of a CD-ROM disk, were used. The disk contains data for several tens of stations during the period of their operation till 1999 if, certainly, they have not stopped operation earlier.

[6] For the aims of this work, the data on foF2 and hmF2 covering the first decade of the new century (desirably up to 2010) were required. Moreover, the initial data series should have begun in 1957–1958 (that is the case for the majority of stations). The latter requirement is related to the fact that in this paper (as well as in the previous publications of the authors), regression dependencies for the period 1957/1958–1979 (when according to our views there was no effect of long-term trends) were created to exclude the solar activity effect in the behavior of foF2 and hmF2.

[7] The detailed study showed that there are three sources of ionospheric data in which in principle one can find values of foF2 and hmF2 for the entire period from 1957/1958 to 2010. They are SPIDR (Space Physics Interactive Data Resource), the system of monthly medians (Rutherford Appleton Laboratory COST 251 VI Database Nov 96), and the data bank of Damboldt and Suessmann [2012]. Though the initial measurements at each ionospheric station are the same, the above three sources differ by amount of gaps in the data, by erroneous repetitions of the same values during many hours, and by some other parameters. So the principal aim of this paper was to use the above mentioned sources of foF2 and hmF2 data separately and to see if they lead to the same (at least qualitatively) trends in these parameters. Detailed discussion on the merits and defects of different sources of foF2 and hmF2 data can be found in Danilov and Konstantinova [2013a, 2013b].

[8] There are several formulae for recalculation of M3000 into hmF2. In the same way as in the previous publications [Danilov, 2011; Danilov and Vanina-Dart, 2010], a well known formula by Shimazaki [1955] was used here. We tried also for some situations the Bradley-Dudeney formula and found no differences to our trend results. However, the latter formula is not suitable for the whole analysis because it requires the data on the E layer, which we often do not have.

[9] As has been mentioned above, we used the period 1957/1958–1979 (below for the sake of brevity we will write 1958–1979, bearing in mind that the data for 1957, if they were available, were also included in the analysis) for creating the regression dependence of foF2 and hmF2 on the solar activity index F10.7. Doing that, we made two assumptions: First, there were no long-term changes (trends) in foF2 (hmF2) before 1980, and so the obtained dependencies provide a “pure” effect of the foF2 (hmF2) dependence on solar activity. Second, there is no ground to think that since 1980 the character of the foF2 (hmF2) dependence on solar activity has changed. In other words, all deviations of the foF2 (hmF2) values in the later years from the dependence of foF2 (hmF2) on F10.7 found for the “etalon” period 1958–1979 manifest long-term changes in foF2 (hmF2) (trends) not related to solar activity. The possible nature of these trends will be discussed below. Here we emphasize that all said above do not exclude a possibility of dependence of trends themselves on solar activity due to possible changes with solar activity in thermospheric parameters (temperature, density, composition, winds), which should govern the foF2 (hmF2) trends. For example, for the negative trends in the thermospheric density, such dependence on solar activity is detected; at low solar activity, the decrease in the density is stronger [Emmert et al., 2008, 2012].

[10] Examples of the foF2 dependence on the F10.7 index for the period 1958–1979 have been numerously presented in earlier publications. Here similar dependencies will be presented (see below Figures 1 and 2) in relation to the new approach to revealing trends described below. Now we emphasize that almost all the analysis will be performed in terms of the ΔfoF2 values, which are the differences between the observed at the given situation values of foF2 and its values for the corresponding F10.7 obtained in the “etalon” dependence for 1958–1979. Note especially the importance of years 1957 and 1958. In these years, the annual mean index F10.7 was ~230, and that makes it possible to create the regression dependence for a wide range of the F10.7 values from 60–70 to 230. For two stations analyzed in this paper (Grahamstown and Tashkent), however, the data on foF2 began later. In this case, the foF2 on F10.7 dependence covered a narrower interval of the F10.7 values up to ~190.

Figure 1.

Dependence of foF2(SS + 2) on the F10.7 index at (a) Wallops and (b) Tashkent stations for two time intervals.

Figure 2.

Dependence of foF2(SS + 2) on the F10.7 index at (a) Boulder and (b) Juliusruh stations for two time intervals.

[11] In the same way as in the previous papers, values of foF2 for two seasons (January–February and June–July) and two moments of the day were considered. The first moment corresponds to 1400 LT, whereas the second one, (SS + 2), corresponds to the moment of 2 h after sunset. The reason for choosing these moments has been numerously discussed earlier, and briefly it is that in the daytime we expect stronger influence of aeronomical parameters on foF2 trends, whereas 1–2 h after sunset the influence of dynamical processes on foF2 should be the strongest. Calculating values of foF2 and hmF2 for each situation indicated above, we used only the days with Ap ≤ 30 to avoid periods with high magnetic activity.

[12] Because of the difficulties in finding initial data, we performed the procedure of searching for trends for each of four situations (1400 LT, winter; 1400 LT, summer; SS + 2, winter; and SS + 2, summer; for the sake of briefness, we will designate them as 14JF, 14JJ, SSJF, and SSJJ) using independently several sources. It could be only SPIDR, only Damboldt and Suessmann, only medians, or a combination of these data banks to the IWG format data for the period before 1999. Each combination of situation and source we will, for the sake of brevity, call position.

[13] The principal method of dealing with the hmF2 data was the same as for the foF2 data. Analytical description of the curve in figures of Figure 5 type (similar to Figures 1 and 2 for foF2) was used as a model dependence of hmF2 on F10.7 for the given position. The model values were subtracted from the experimental values of hmF2 for the later period, and the obtained value ΔhmF2 = hmF2(obs) − hmF2(mod) was used for further analysis. The results of this analysis will be described in section 6.

3 Analysis of the Scatter in foF2 Data

[14] After analyzing a large number of stations in four sources described above, 12 stations were selected for which it was possible to perform a search for foF2 trends for the period after 1990 for all four situations and, as a minimum, for two (and desirably for all three to four) sources. As a result, 180 positions were considered. Analysis of linear trends found for these positions will be considered in detail in the next section.

[15] Now we consider the results of a statistical analysis of the scatter in the initial data. Danilov [2009] and Danilov and Vanina-Dart [2010] showed that after approximately 1980, an increase in the initial data scatter (the standard deviation, SD, was calculated) in hmF2, and the ratio foF2(ss + 2)/foF2(14) was observed. It was assumed that this increase in the data scatter is caused by the increase in instability of the system of horizontal winds in the thermosphere (which influence both hmF2 and foF2(ss + 2)) due to the cooling and contraction of the upper atmosphere (see above).

[16] For the comparison of the degree of scatter in the initial data on foF2 during the earlier (“etalon”) and analyzed periods, we also use here the standard deviation (SD) of the observed values of foF2 from the regression dependence of foF2 on F10.7 for two periods: 1958–1979 and 1998–2010. The latter period is taken as the approximately 11 year cycle adjacent to the end of the analyzed period (2010). Since the table for all positions would be very cumbersome, we present as an example in Table 1 values for four stations with various situations and various sources (19 positions). It should be emphasized that in this part of the study no smoothing of foF2 was performed and “pure” values of foF2 for each time of the day averaged over two corresponding months (January–February, or June–July) were considered.

Table 1. Standard Deviations SD and Values of Δ (See Text) for Four Stations
StationsSituationSourceΔ (MHz)SD(1) (MHz)SD(2) (MHz)
TashkentSSJJIWG + medians−0.730.180.82
TashkentSSJJIWG + SPIDR−0.860.180.99
TashkentSSJJIWG + Damboldt−0.710.180.8
Moscow14JJIWG + SPIDR−
Moscow14JJIWG + Damboldt−
SloughSSJJIWG + SPIDR−0.360.170.46
SloughSSJJIWG + medians−0.320.170.53
SloughSSJJIWG + Damboldt−
TashkentSSJFIWG + medians0.360.410.8
TashkentSSJFIWG + SPIDR0.390.410.8
TashkentSSJFIWG + Damboldt0.410.410.82

[17] One can see in Table 1 that in all presented cases the scatter (SD) of the initial data for the “etalon” period is substantially less than for the period 1998–2010. Such picture is observed for all 127 considered positions except for seven.

[18] Now we consider in detail the SD values for two periods. For the first period, the situation is simple: small values of SD show that the scatter of data is small and the experimental points are located close to the regression curve. The situation for the SD values for the second period is more complicated. The increase in SD as compared to the first period in this case is related to two effects. The first is the increase in the real scatter of foF2 due to, probably, the same causes as the increase in the scatter in hmF2 and foF2(SS + 2)/foF2(14) analyzed in earlier papers (see above). In other words, if we drew a curve of the foF2 on F10.7 dependence for 1998–2010, we would obtain higher scatter of the points relative to this curve than the scatter of points relative to the “etalon” regression curve (1958–1979). The second is the high value of SD for the second interval, which includes also the foF2 trend, because in the presence of a trend, points of the later period would more and more move from the “etalon” regression curve even if there is no scatter of these points relative to their own dependence on F10.7.

[19] All said above is illustrated well by Figure 1. Figure 1a shows the picture for Wallops station. One can see that the scatter of points (data for the 1958–1978 period) relative to the solid (“etalon”) curve is small (SD(1) = 0.19). Crosses show the values of foF2 for the corresponding value of F10.7 but for the period 1998–2010. One can distinctly see that, first, crosses lie substantially lower than the “etalon” regression curve. That manifests the existence of a systematic downward shift, i.e., negative trend. Second, the scatter of the foF2 relative to the curve drawn through crosses (dashed curve) is 0.33, which is substantially larger than the scatter of the earlier points relative to the solid curve. The summated effect of the trend and increase in the scatter gives for the crosses a value SD(2) = 1.02 relative to the “etalon” regression curve. Absolutely the same picture is observed also for Tashkent (Figure 1b): a systematic shift of crosses (1998–2010) relative to the “etalon” curve (i.e., negative trend) and their stronger scatter.

[20] Table 2 shows values of the SD(2)/SD(1) ratio averaged over all sources for each situation at seven stations. One can see that the regularity mentioned above that the SD(2) for later period is substantially higher than the SD(1) value for the earlier period is fulfilled for all stations and all situations except for two moments at Rome station. However, at averaging over all moments, the SD(2)/SD(1) value is obtained more than unity for all stations. One can see from the lower line of Table 2 that the scatter of the averaged values of SD(2)/SD(1) for each station is relatively small. Averaging these values over all stations, we obtain SD(2)/SD(1) = 2.1 ± 0.54. In other words, on the average for all stations, the scatter of the foF2 values in the period 1998–2010 is approximately higher by a factor of 2 than in the “etalon” period.

Table 2. Ratios SD(2)/SD(1) for Various Stations and Situations

[21] In Table 2, one more fact draws attention. A tendency is seen that for the same season, the SD(2)/SD(1) value is higher for the SS moments than for 1400 LT. This is true for all pairs except three out of 14 available. The most probable explanation of this fact is that the foF2 values in the after-sunset period are more sensitive to dynamical processes (see above) and so demonstrate stronger variability than the foF2 values in the daytime.

[22] Now we return to Table 1. There is a column designated as Δ. It shows an averaged difference between crosses and “etalon” curve in figures of the type of Figures 1 and 2, in other words, how strongly the foF2 values for the period 1998–2010 differ on the average from the corresponding (i.e., taken for the same F10.7) values for the period 1958–1979. Actually, it is an equivalent to the trend that we are searching for in this paper but calculated not per 1 year (as will be done in the next section) but as a difference between two periods distanced in time. One can see in this column that in three cases out of four presented in Table 1, the Δ value is negative. This means that from the period 1958–1979 to the period 1998–2010, the foF2 value decreased, though differently at different stations. The fourth case is shown in Table 1 deliberately as an illustration of the exception from the rule. Out of the total number of 127 considered positions, a positive value of Δ was obtained only in nine. Also, only in one case shown in Table 1 is this value positive for all sources for the given situation (SSJJ). The rest of the four cases are scattered over various stations and situations. A cause of such positive value of Δ symbolizing a positive trend is not yet clear. Probably it is some sort of fluctuation related to an error in the data. At any case, the overwhelming majority of the calculated values of Δ point out to a negative trend in foF2 after 1979.

[23] This fact is important for our analysis in two aspects. First, at a strong scatter in foF2 (high values of SD(2)), it is not always possible to create a statistically significant dependence of ΔfoF2 on time in order to obtain a significant linear trend (this question will be considered in the next section). Using the Δ value, we can, at last, know the sign of the trends between the two periods (1958–1979 and 1998–2010) even if the scatter of particular values of foF2 does not allow us to calculate a linear trend.

[24] The second aspect due to which it is important to have an estimate of the change in foF2 by the above-indicated method is the following. In order to obtain a detailed picture of changes in ΔfoF2 in time, one has, due to high values of SD(2), to apply 11 year smoothing as is done in the next section. In the above described method, no smoothing is applied, and the sign and amplitude of the changes in foF2 are derived directly from the initial data.

4 Linear Trends in the Critical Frequency

[25] As has been already mentioned above, the strong scatter of foF2 values in later years of the considered interval makes it difficult to obtain a pronounced picture of the changes in ΔfoF2 with time for calculation of a linear trend. In the same way as in the majority of our earlier publications, a running smoothing with the 11 year window was applied. As usual, the obtained smoother value was referred to the middle of the considered interval. We took the ΔfoF2 values from 1985 up to the end of the data series available for the given position (data series ending before 2006 were not considered). So we were obtaining the time behavior of the smoothed values of ΔfoF2 from 1990 to (in the best case) 2005. In the worst case, the smoothed points ended in 2001. Thus, considering figures in this sections (as well as below in section 6), one should bear in mind that, though points end in the first half of the first decade of the new century, they really manifest the behavior of the critical frequency till the last years of the decade.

[26] Several available sources for each situation at each station were analyzed. Twelve stations with several sources satisfying the requirements described in section 3 were chosen. On the total, 180 positions were obtained. Out of these 180 positions, in only 22 cases was no statistically significant negative trend found: the trend was absent or even had a positive sign.

[27] Due to obvious reasons, it is impossible to present in the paper the table with all 180 positions. Table 3 shows results of calculations for three stations with different sources (13 situations). It follows from Table 3 that in two cases (Boulder and Wallops) the trends obtained for various sources are relatively close to each other and make it possible to obtain for a given situation an average trend with relatively small SD. In the case of Rome, the situation is different. The trends obtained using different sources vary from −0.019 to −0.057, which makes it possible by formal averaging to obtain the average value of the trend equal to −0.041 but with a large SD.

Table 3. Examples of Trends in foF2 in MHz Per Year for Various Positions at Three Stations
Station and SourceBeginning (Year)End (Year)R2ΔfoF2 (MHz)Trend (MHz/yr)Average (MHz/yr)
Rome, SSJJ
IWG + SPIDR199420050.93−0.63−0.057−0.041
IWG + Damboldt199420020.89−0.2−0.025 
Medians + SPIDR199420050.94−0.58−0.053 
Boulder, 14JF
IWG + SPIDR199020030.76−0.59−0.045−0.046
Wallops, SSJF
IWG + SPIDR199020040.91−0.77−0.055−0.064
IWG + medians199020020.92−0.63−0.053 

[28] In the data for Rome, two facts draw attention. First, the lowest, strongly different from the others, value of the trend is obtained for the source “medians.” This is a manifestation of the fact which we mentioned above describing the sources: medians in some cases could give results differing from the results for other sources. Second, it is distinctly seen that three sources for which the coefficient of determination R2 is high (0.93–0.94) give values of the trend very close to each other from −0.052 to −0.057, which are different from the trends for two other sources (including medians) with lower R2.

[29] Finally, the most important difference of the data for Rome from the data for Boulder and Wallops is that at the two latter stations in the situations presented in Table 3, the trend begins directly in 1990, whereas for Rome for all sources it begins only in 1994. Examples of these two types of behavior are presented in Figure 3 (beginning of the trend in 1990) and Figure 4 (beginning of the trend in 1994).

Figure 3.

Trends for various sources for stations (a) Rome (situation SSJF) and (b) Boulder (situation 14JF). Different marks correspond to different data sources (shown at the bottom of each panel). Lines show a linear approximation of each set of data. Numerals show the coefficient of determination R2.

Figure 4.

Trends for various sources for stations (a) Juliusruh (situation 14JJ) and (b) Moscow (situation 14JF). Different marks correspond to different data sources (shown at the bottom of each panel). Lines show a linear approximation of each set of data. Numerals show the coefficient of determination R2.

[30] In the examples presented in Figure 3, the slopes of the approximating lines are close, and so all sources give close values of the negative trend in foF2. However, it is not the case for all stations and situations. Rome station presents an example in Table 2, which was discussed above. In spite of the smoothing, the data scatter sometimes stays rather high (for example, for the source IWG + SPIDR for Boulder station), but the coefficient of determination R2 = 0.76 at the number of points N = 14 makes us, according to Fisher criterion, consider the obtained trend reliable with a statistical significance above 95%.

[31] In figures of the Figure 4 type, the pronounced trend begins as a rule in 1994 (in some cases, in 1995). A small increase (jump) in ΔfoF2 of 0.1–0.3 MHz as a rule precedes this date. We can say nothing on the possible causes of this jump. The point of the beginning of the trend can be in the region of both small negative (Figure 4a) and positive (Figure 4b) values of ΔfoF2. In the former case, the initial values of ΔfoF2 are slightly below the corresponding values ΔfoF2 in the period 1958–1979. This looks natural, because a small negative trend (for example of the order of 0.01 MHz per year found in the collective experiment of Laštovička et al. [2006]) should have led to a small negative value in ΔfoF2 near 1990. The presence of small positive values of ΔfoF2 in 1990–1994 is not completely understandable. It could be (in the same way as positive values of Δ in Table 3) caused by problems with equipment. It is easy to assume that during several decades at ionospheric stations, changes in the equipment or processing methods could occur, which was not compensated completely, and led to higher values of foF2 than the ones measured earlier.

[32] Anyway, all four panels in Figures 3 and 4 give a negative trend for the period after 1994. The final list of the obtained linear trends in foF2 is presented in Table 4. The trend values averaged over all the sources available for each situation are presented. The corresponding columns show the beginning and end of the linear trend. One can see that, although the scatter of trend values for different stations is high, one can obtain an average value for each situation with acceptable SD (the second line from the bottom).

Table 4. Trends in foF2 for 12 Stations and Four Situations and the Results of Averaging of the Trends in Each Column (the Third and Fourth Lines From the Bottom)a
StationSummer SS + 2Summer 1400 LTWinter SS + 2Winter 1400 LT
  1. aThe two bottom lines show the results of averaging if we put the trends obtained for Hobart and Townsville stations according to months but not the seasons (see text).
Average trend and SD−0.031−0.028−0.046−0.044
Average trend and SD−0.027−0.021−0.050−0.054

[33] However, a curious fact draws attention. The trends values for the Southern Hemisphere stations (Hobart, Grahamstown, and Townsville) are put into the table according to the real season (that is, the data for January–February and June–July are put into summer and winter columns, respectively). But these data do not fit their columns. For example, the values of −0.011 and −0.006 for Townsville station look impertinent in winter columns where the majority of other stations give higher negative trends. These values look more pertinent in the summer columns where according to other stations negative trends are lower.

[34] If we “forget” about winter and summer and put the trends obtained for Hobart and Townsville stations according to months but not the seasons, the statistics becomes even more favorable (the bottom section of Table 4). The exact cause of the detected effect is obscure, but one can assume that changes in the system of thermospheric winds are tied to seasons of the Northern Hemisphere and so similar effects in both hemispheres could be observed in the same months but not in the same seasons.

[35] In both bottom sections of Table 4, a seasonal effect (in the lowest line, it is better provided statistically) is clearly seen: both for 1400 LT and for the SS + 2 moment, the negative trends are higher in winter than in summer.

5 Trends in hmF2

[36] In the same way as for foF2, two approaches were used to derive hmF2 trends. Examples of figures similar to Figures 1 and 2 for foF2 are shown in Figure 5. In the first approach, the values of ΔhmF2 obtained for the period after 1985 till the end of the data available were smoothed, and the dependence of the obtained values on time was drawn. Points in the graphs (Figures 6-9) were approximated by a line the slope of which provided a linear trend in hmF2.

Figure 5.

Dependence of hmF2 on the F10.7 index at stations (a) Juliusruh, (b) Slough, (c) Wallops, and (d) Tashkent for various situations. Years refer to the points of the etalon period.

Figure 6.

Changes in the smoothed values of ΔhmF2 in time for various stations and situations: (a) Juliusruh, SSJF; (b) Slough, SSJF; (c) Wallops, 14JF; (d) Slough, 14JF. Different marks correspond to different data sources (shown at the bottom of each panel). Lines show a linear approximation of each set of data. Numerals show the coefficient of determination R2.

Figure 7.

Changes in the smoothed values of ΔhmF2 in time for various stations and situations: (a) Point Arguello, 14JJ; (b) Tashkent, SSJJ; (c) Tomsk, 14JF; (d) Tomsk, SSJF. Different marks correspond to different data sources (shown at the bottom of each panel). Lines show a linear approximation of each set of data. Numerals show the coefficient of determination R2.

Figure 8.

Changes in the smoothed values of ΔhmF2 in time for various stations and situations: (a) Juliusruh, 14JF; (b) Moscow, SSJF; (c) Ashkhabad, 14JF; (d) Grahamstown, SSJF. Different marks correspond to different data sources (shown at the bottom of each panel). Lines show a linear approximation of each set of data. Numerals show the coefficient of determination R2.

Figure 9.

Changes in the smoothed values of ΔhmF2 in time for various stations and situations: (a) Moscow, 14JF; (b) Moscow, SSJJ; (c) Slough, 14JJ; (d) Tashkent, SSJF. Different marks correspond to different data sources (shown at the bottom of each panel). Lines show a linear approximation of each set of data. Numerals show the coefficient of determination R2.

[37] In the second approach, an arithmetic average of the ΔhmF2 values (without any smoothing) was calculated for the period 2000–2010. If the initial data series ended earlier, the averaging was performed from 2000 to the end of the series. The obtained value characterized the mean change in hmF2 from the period before 1980 to the period after 2000. Being divided by the interval in 30 years (conventionally: 1975–2005), it gave a value of the yearly trend in hmF2 averaged over the period after 1980. It is worth emphasizing that in this case no smoothing was needed and it was possible to estimate the mean value of the trend even without drawing ΔhmF2 dependence for the period after 2000.

[38] As a result of consideration of more than 20 ionospheric stations, 10 stations were selected for which it was possible to estimate trends in hmF2 using the two methods described above.

[39] The results of estimation of trends using the first method for these 10 stations are shown in Table 6. In the same way as for foF2, estimates of trends for the same station and same situation were performed using the data of different sources of data described above in section 2 or combinations of such data (for example, the data in the IWG format before 1999 and the SPIDR data for 2000–2010). The degree of agreement of the results (trends in hmF2) obtained made it possible to estimate the reliability of the derived trends.

[40] One can see in Table 5 that for some stations (Ashkhabad, Grahamstown, Hobart, and Point Arguello) the data useful for application of the methods described were only in the SPIDR system. For other stations, there was a possibility to use for each situation various sets of initial data. Table 5 shows the station name, source of the initial data, period for which trend was obtained, the coefficient of determination R2, and the trend value k (per year), which is the slope of the approximating line in Figures 6-9 (see below).

Table 5. Trends in hmF2 for 10 Stations for Various Situations and Various Data Sources
Station and SituationSourceBeginning (Year)End (Year)R2k (km/yr)
14 JFSPIDR199020050.91−1.3
14JFIWG + SPIDR199020050.93−1.7
14JJIWG + SPIDR199520050.92−5.6
SSJFIWG + SPIDR199020050.91−1.3
SSJJIWG + SPIDR199020050.89−1.3
14 JFSPIDR199520050.95−1.3
14JFIWG + SPIDR199420050.96−1.4
14JJIWG + SPIDR199020050.72−1.8
SSJFIWG + SPIDR199020050.83−0.6
SSJJIWG + SPIDR199020050.62−0.8
14 JFSPIDR199520010.99−4.5
14JFIWG + SPIDR199520010.98−4.4
14JJIWG + SPIDR199020050.5−0.7
SSJFIWG + SPIDR199520010.97−4.0
SSJJIWG + SPIDR199420050.96−2.5
14 JFSPIDR199020010.83−2.2
14JJDamboldt  Positive 
SSJFIWG + SPIDR199020010.72−1.7
SSJJIWG + SPIDR199020010.74−1.4
14 JFSPIDR199020050.94−1.2
14JFmedians + IWG + SPIDR199020050.96−2.6
14JFmedians + SPIDR199020050.97−2.8
14JJmedians + IWG + SPIDR199320050.89−3.5
14JJmedians + SPIDR199020050.99−7.1
SSJFmedians + IWG + SPIDR199020020.90−4.7
SSJFmedians + SPIDR199020050.90−1.4
SSJJmedians + IWG + SPIDR199020030.95−1.9
SSJJmedians + SPIDR199020030.91−3.7
Point Arguello
SSJFSPIDRNo trend   
SSJJSPIDR1990 0.20−0.5

[41] The principal result of the analysis by the first method is that out of 72 positions presented in Table 5, the trend was not pronounced or was distinctly positive only for three positions. In other cases, the trend was negative and statistically significant according to Fisher criterion. This result allows us to make the main conclusion: a negative trend in hmF2 is typical for the period considered. Three deviations from the general picture could have an occasional character.

Table 6. Averaged Values of Trends in hmF2 for 10 Stations for Daytime and After-Sunset Conditions, Summer, and Winter
Stationk (km/yr)SD (km)
Point Arguello2.22.58
1400 LT2.51.73
SS + 21.91.05

[42] The absolute value of the obtained negative values of k varies in Table 5 significantly: from 0.46 (Point Arguello, SSJJ) to 7.1 (Wallops, 14JJ). If one averages formally all 69 values presented in Table 5, one would obtain a value of k = −2.14 km per year with the standard deviation SD = 1.42 km per year, which could be considered as an average global trend in hmF2 in the period after 1990. However, such averaging assumes similar change (the same trend) in hmF2 at all stations. It is obvious from very general considerations that the changes in the dynamical regime of the thermosphere occurring due to the cooling and contraction of the middle and upper atmosphere and influencing changes in the F2 layer height could and should occur differently in different regions of the globe. The model calculations [Qian et al., 2011] give the same result: in different geographic regions, trends in hmF2 of different magnitude and even different sign could be observed.

[43] Before coming to values of k obtained for different stations, we consider examples of ΔhmF2 dependencies on time, which served as a basis for deriving the k values. These examples are shown in Figures 6-9.

[44] Figure 6 presents an example of the most “favorable” cases where the initial data till 2010 (and therefore the smoothed data till 2005) were available and a well-pronounced decrease in ΔhmF2 was beginning directly in 2000. One can see that in all four panels, a negative trend is well pronounced and statistically significant (high values R2 ~ 0.8–0.97). In three cases (Figures 6a, 6b, and 6d), a good agreement between the two used sets of data is observed. In Figure 6c, there is a discrepancy between the data of only SPIDR and combinations of the SPIDR data to the IWG data and medians: the time behavior of ΔhmF2 according to two latter groups is almost the same. Most probably, the difference in the time behavior according to the SPIDR data is related to the fact that the SPIDR data for Wallops station begin only in 1967, so the reliability of creation of the hmF2(F10.7) dependence for the period before 1980 (see above section 3) is lower than in the case of using medians for the period 1957–1967.

[45] Figure 7 shows an example of cases when the initial data ended before 2010 (mainly in 2006) and so the smoothed points end in 2001. One can see that in Figures 7c and 7d, a fairly satisfactory agreement between various sets of data giving values of k close to each other (see Table 5) is observed. A substantial difference between the ΔhmF2 behavior according to the IGW and SPIDR data is seen in Figure 7b. The cause of this difference is not clear, but the preference apparently should be given to the SPIDR data.

[46] It is not in all cases that a pronounced decrease in ΔhmF2 began directly from 1990. Approximately in a half of cases, ΔhmF2 began to decrease in 1992–1995 (see Figures 8 and 9). Before that, either the absence of changes in ΔhmF2 or even its small increase (Figures 8c and 8d) was seen. It is worth noting that, as one can distinctly see in Figure 8b, different sets of the initial data give a similar picture: the absence of changes till 1993 and a well-pronounced statistically significant decrease with almost the same slope k. A similar picture is seen also in Figures 8a, 9a, and 9d. In the situation SSJF at Tashkent station (Figure 9d), the time behavior of ΔhmF2 for the SPIDR data differs from that for the IWG data. The cause of the difference is not clear.

[47] One can see in Figures 6-9 that for the majority of the analyzed time dependencies of ΔhmF2, the coefficient of determination R2 is above 0.80 and in many cases above 0.9. That provides a high statistical significance of the obtained slope k according to Fisher criterion. The same picture is observed also almost for all other dependencies of ΔhmF2 on time not shown in Figures 6-9.

[48] Note one more important fact related to Figures 6-9. The analyzed values of ΔhmF2 lie either in the region of negative values (Figures 6a, 6c, 7a, 7b, 8a, 8d, and 9c), or begin in the first years in the region of positive values and then transit into the region of negative values (Figures 6b, 6d, 7c, 7d, and 8c), or lie completely within the region of positive values (Figures 8b, 9a, 9b, and 9d).

[49] It is evident that the above-indicated difference is caused by a different prehistory of changes in hmF2 before 1990. In the first case, after the “etalon” period 1957–1980, the negative trend in hmF2 already existed and led to negative values of ΔhmF2 in 1990. In two other cases, trends in hmF2 before 1990 were positive, and that led to positive values of ΔhmF2 in 1990. However, the point is important that, independently of the fact whether analyzed ΔhmF2 lie in the regions of negative or positive values, they in the majority of cases (69 out of 72) show a negative trend in the period in question. This fact shows that around 1990, there occurred a substantial change in the processes governing changes in the hmF2 height. Probably, to changes in the regime of the thermospheric horizontal winds (they influence substantially the value of hmF2 via the vertical drift), the effect of general “contraction” of the thermosphere, that is, the decrease in the height of constant density levels (including those where the F2 layer maximum is formed), was added. This led in some cases to a substitution of a positive trend by a negative one and thus to dominating of negative trends in the period after 1990.

[50] Now we return to the problem of averaging of the obtained values of k. It has been already mentioned above that averaging of all values presented in Table 5 gives k = −2.14 ± 1.42 km/yr. The SD value is large enough (1.42 km/yr), but one has to remember that this value of k is obtained by averaging of a large number of points (69), so its statistical significance is high enough.

[51] Analyzing Table 5, one can see that in statistical aspect, the situation is different for different stations. For example, for Point Arguello station, there are only three estimates that are strongly different from each other. At the same time, for Tomsk station, there are 10 values of k that differ between themselves much less than for many other stations.

[52] We performed a series of formal averaging of k values shown in Table 6. First of all, we averaged values of k for each station over all four situations and sets of initial data. The results are shown in the 10 top rows of Table 6. The SD values are also shown there. The data of Table 6 confirm the statement mentioned above that the reliability of the hmF2 trends obtained for different stations is different. The statistically most reliable trend is obtained for Tomsk and Hobart, the least reliable trend is seen for Point Arguello. For Slough station, if one takes as erroneous two values of −5.8 and −5.6 km/yr obtained for the 14JJ situation and differing strongly from k for other situations, we obtain statistically significant value of the trend equal to 1.4 km/yr with a small value SD = 0.2 km/yr.

[53] Now we come to the second method of estimation of the hmF2 trend. First of all, we come back to Figure 5. We remind that points and approximating solid line refer to the period before 1979. We think that at that time there was still no pronounced ionospheric trends, and so the obtained dependence could be taken as a sort of an “etalon.”

[54] Crosses in Figure 5 show values for the same situations and values of F10.7 but for the 11 year period from 2000 to 2010 (if the data available end earlier, the crosses from 2000 up to the end of the data are shown). One can see that in all examples shown in Figure 5, crosses lie systematically below the approximating line. This very fact is an illustration of the presence of a negative trend according to the second method.

[55] We emphasize that at such comparison no assumptions are made except one: the dependence of the ionizing radiation of the Sun on the solar activity index F10.7 stays unchanged. There is no ground for the opposite assumption except for the years of the deep minimum of solar activity (2008 and 2009). The difference between the lines in Figure 5 and each of the crosses provides the value of ΔhmF2, which we smooth and show in Figures 6-9 as a function of time.

[56] Figure 5 provides only examples with a well-pronounced effect of lower values of hmF2 in the later period. Values of ΔhmF2 averaged over the period after 2000 for each station and situation are shown in Table 7. One can see that in the overwhelming majority of cases (56 situations out of 68) the averaged values of ΔhmF2 are negative. That means that between the period 1958–1979 and the first decade of our century the hmF2 value was decreasing. The rest of the 12 cases correspond to a positive trend in hmF2 before the beginning of the analyzed period. However, as we have noted above, positive values of ΔhmF2 do not yet mean the absence of a negative trend in the considered period. For some situations, for example, for Moscow station (see Figures 9a and 9b) the smoothed values of ΔhmF2 lie in the region of positive values (this fact showing that between the period before 1980 and the analyzed period the hmF2 value was increasing); however, after 1993–1995, a well-pronounced negative trend is observed.

Table 7. Average Values of ΔhmF2 Between the Period 1958–1980 and the Period Shown in Table for Various Stations
StationSourcePeriodΔhmF2 (km)
IWG + SPIDR2000–2010−15.6−60.8−16.8−21.3
IWG + SPIDR2000–2010−18.2−40.5−9.2−14.3
IWG + SPIDR2000–2006+7.7+22.4+9.2−5.0
medians + IWG + SPIDR2000–2010−35.3−74.1−52.8−24.9
medians + SPIDR2000–2010−35.8−111.6−17.1−39.6
Point ArguelloIWG + SPIDR2000–2008−29.8−84.7+10.7−10.3

[57] The scatter of the ΔhmF2 values in Table 7 is rather high: from −118.6 km to +51.3 km. However, these values present extreme ones. Formal averaging of the values in Table 7 gives −27.8 ± 24.9 km (56 cases) for negative values and +15.6 ± 12.7 km (16 cases) for positive values. One can see that negative values of ΔhmF2 prevail, demonstrating that after 1980 a decrease in the F2 layer height (negative trend in hmF2) occurred more often. It is impossible to obtain estimates of the trends themselves from the above-presented data due to two reasons. First, the statistical significance of the obtained averaged values of ΔhmF2 is small because of the scatter of data for different stations and different situations. Second, the time interval to which the above-indicated values of ΔhmF2 correspond is not exactly determined. One can state only that between the period 1957–1979 and the period after 2000, the ΔhmF2 value on the average changed by this or that factor. If very conventionally we accept that the Δt interval on the average was 30 years, the above-indicated values of ΔhmF2 correspond to a negative trend of about 1 km per year and positive trend of about 0.5 km per year. The value of negative trend obtained in such a way does not contradict the trend values obtained for different stations by the first method and shown in Table 6.

[58] We emphasize once more that the second method is not aimed at obtaining any exact estimates of the hmF2 trends. It is aimed at clarifying the question how the hmF2 value was changing between the “etalon” interval 1958–1979 and the analyzed interval after 2000. At the same time, this method has two advantages: visual character (see Figure 5) and the absence of any intermediate procedures, because the values of hmF2 observed at the same values of F10.7 are directly compared between themselves.

6 The Dependence of foF2 on hmF2

[59] Bearing in mind the changes which according to the data of the previous sections occur in the critical frequency and height of the F2 layer, it is worth to consider possible changes of the relation between foF2 and hmF2. We selected 10 stations for which the needed initial data were found in the same way as indicated above in this paper. We considered the same four situations: 14JJ, 14JF, SSJJ, and SSJF as in the analysis of foF2 and hmF2 trends. Since (as was described in detail by Danilov and Konstantinova [2013a, 2013b] and Danilov [2012]) various sources of the initial data could give results differing from each other, we tried (if there was a possibility) to use several sources for each situation.

[60] We drew the dependencies of foF2 on hmF2 for each situation and all available sources for two periods: 1957–1979 and 1980–2010. If for the given station there were no data till 2010, we used the data up to the end of the series available but not earlier than 2006. The series of data ending before 2006 were not analyzed. We characterized the obtained dependencies of foF2 on hmF2 by two parameters: the coefficient of determination R2 (making it possible to estimate the statistical significance of the obtained dependence using Fisher criterion) and the standard deviation squared SD2.

[61] In the period 1957–1979, the foF2 dependence on hmF2 was well pronounced, had approximately linear character, and was characterized by high (above 0.70, and most often 0.90–0.99) values of R2 (see Figures 10-13). The cause of such dependence is well known from the physics of the F2 layer. The increase in hmF2 means raising the F2 layer into the region of lower recombination (lower values of [N2]/[O]) and, respectively, increase in NmF2 (foF2) in the layer maximum. Positive phase of an ionospheric storm at middle latitudes is the most visual example of such process. The storm-induced circulation generated by the heating at polar latitudes is equatorward and lifts the F2 layer along the geomagnetic field lines increasing hmF2. In that case, an increase in foF2 occurs, forming the positive phase of an ionospheric storm.

Figure 10.

Dependence of foF2 on hmF2 at Ashkhabad station for two situations (14JF and 14JJ) for the “etalon” (full line, full circles) and later periods (dashed line, crosses).

Figure 11.

Dependence of foF2 on hmF2 at Boulder station for two situations (14JF and SSJJ) for the “etalon” (full line, full circles) and later periods (dashed line, crosses).

Figure 12.

Dependence of foF2 on hmF2 at Moscow station for two sources and the SSJF situation for the “etalon” (full line, full circles) and later (dashed line, crosses) periods.

Figure 13.

Dependence of foF2 on hmF2 at Tashkent station for two situations (14JF and 14JJ) for the “etalon” (full line, full circles) and later periods (dashed line, crosses).

[62] Figures 10-13 show examples of the foF2 dependencies on hmF2 for the two periods indicated above for various stations and situations. In all figures, points and solid line correspond to the “etalon” period 1957–1979 (characterized by standard deviation SD(1)), whereas crosses and dashed curve correspond to the later period (after 1980). It is distinctly seen that the foF2(hmF2) dependence is much worse pronounced for the later period than for the earlier one. The crosses systematically provide a stronger scatter relative to the “etalon” foF2(hmF2) dependence (solid curve), this fact being manifested in much higher values of the standard deviation SD(3). If we draw the separate approximation via crosses (which is done in the form of a dashed line), even then the scatter of the foF2 values relative to this line (values of SD(2)) would be substantial. The systematic shift of crosses relative to points (see, for example, Figure 10 for Ashkhabad) increases the SD(3) value as compared to the SD(2) value.

[63] For different stations, the character of the dependence of foF2 on hmF2 changes from the earlier period to the later one slightly differently. In the case of Ashkhabad station (Figure 10), we see a systematic shift of the dependence for the later period. In the case of Boulder station (Figure 11), we have a conservation of the dependence character but a substantial (see the corresponding values of SD) increase in the scatter. For Moscow station (Figure 12), we have a very strong increase in the scatter and a substantial change in the slope of the new dependence of foF2 on hmF2 as compared to the “etalon” one. A substantial increase in the standard deviation and decrease in the coefficient of determination in the later period are typical features of all the presented examples.

[64] We have analyzed 128 cases (station plus situation plus source). We considered the SD(3)2/SD(1)2 ratio where SD(3) and SD(1) are the standard deviations for the later and earlier periods, respectively. In only four of them was the SD(3)2/SD(1)2 ratio less than unity. In the rest of the cases, the scatter of the data in the later period was stronger than in the earlier one. Because of obvious reasons, we cannot present in the paper a table for all considered cases, so we present in Table 8 examples for Tashkent (one situation, but different sources) and Moscow (one source but different situations) stations. Table 9 shows an example for Slough and Tomsk stations for the 14JF situations and different sources. These examples show that, although for particular situations and sources the SD(3)2/SD(1)2 values differ slightly, all of them have the same order of magnitude and manifest an increase in the foF2 scatter in later period.

Table 8. Examples of the Obtained Values of R2 and SD2 for Tashkent and Moscow Stations
Source (foF2-hmF2)R2SD(1)2SD(2)2SD(3)SD(3)2/SD(1)2
14JJ SPIDR-SPIDR0.810.391.171.213.10
14JJ IWG + Dam-SPIDR0.800.381.361.423.74
14JJ SPIDR-Dam0.810.401.081.403.50
14JJ IWG + Dam-Dam0.810.411.241.563.78
14JF SPIDR-SPIDR0.950.381.643.308.68
14JJ SPIDR-SPIDR0.71v0.390.702.92
SSJF SPIDR-SPIDR0.790.672.594.586.84
SSJJ SPIDR-SPIDR0.870.100.330.545.4
Table 9. Example of the Obtained Values for Slough and Tomsk Stations
Source (foF2-hmF2)R(1)2SD(1)2SD(2)2SD(3)2SD(3)2/SD(1)2
14JF SPIDR-SP0.930.420.760.992.36
14JF IWG + m-SP0.910.470.891.042.21
14JF SP-IWG + SP0.910.580.891.202.07
14JF IWG + m-i + SP0.900.580.861.292.22
14JF IWG + Da-SP0.920.480.900.982.04
14JF IWG + D-i + SP0.900.580.871.202.07
14JF SPIDR-SP0.930.511.111.262.47
14JF IWG + D-SP0.930.471.181.372.91
14JF SP-Damb0.930.471.151.222.60
14JF IWG + D-Dam0.930.461.161.393.02

[65] Table 10 shows the SD(3)2/SD(1)2 values for all 10 stations averaged over situations and sources for two seasons. One can see that for all 10 stations and both seasons (except for Hobart in summer), SD(3)2/SD(1)2 > 1. Table 10 shows also that on the whole the effect of the increase in the foF2 scatter relative to hmF2 is stronger in winter than in summer.

Table 10. Mean Values of SD(3)2/SD(1)2 for Two Seasons

7 Discussion

[66] One of the principal conclusions of this paper is that the degree of scatter of the foF2 values (determined by the standard deviation SD) in the period 1998–2010 is found much (by a factor of 1.5–2.5, see Table 2) higher than in the “etalon” period 1958–1979. As far as there is no ground to think that the ionospheric observations at the global network are conducted less accurately, the obtained conclusion manifests an increase in the critical frequency variability at the same conditions of solar activity. In principle, such increase in the variability could occur due to two causes. One is an increase in the variability of aeronomical parameters governing the electron concentration in the F2 layer maximum in the conditions of photochemical equilibrium. They are, first of all, the temperature and composition of the thermospheric gas. The second cause could be an increase in variability of the system of horizontal winds in the thermosphere, which via the vertical drift and changes in the height of the layer maximum hmF2 influences NmF2. Both processes could and should occur as a result of the cooling and contraction of the middle and upper atmosphere. Currently, it is difficult to say which one out of them influences stronger the increase in SD(2)/SD(1) (see Table 2). However, the fact noted while describing Table 2 on stronger increase in SD(2)/SD(1) for the moment 2 h after sunset (SS + 2) when the influence of dynamical processes should be the strongest, makes it possible to assume that the main role in the increase in foF2 scatter is played mainly by the instability of thermospheric dynamics inevitably changing in the process of cooling and contraction of the upper atmosphere.

[67] The analysis of the total change in foF2 from the period 1958–1979 to the period 1998–2010 (the Δ value in Table 1) allows us to state that during the time between the two above-indicated periods there occurred a decrease in the critical frequency, and the value of this decrease can be 0.5–1.0 MHz. On the basis of these values, one cannot obtain a linear trend in the foF2 per year because they refer to the interval between two periods, but a confirmation of the very fact of a negative trend in the critical frequency is very important. One can conventionally estimate the trend assuming that the Δ value is −0.60 MHz and refers to a period of 20 years. In that case, the negative trend in foF2 would be −0.03 MHz per year. This value is exactly in the middle between the averaged linear trends over all stations for summer (−0.024 MHz per year) and winter (−0.052 MHz per year) shown in Table 4.

[68] Describing Table 4 in section 5, we have actually already discussed the results of calculation of linear trends in foF2. The scatter of the obtained values from one station to another is strong enough but not too strong to prevent calculating an average value over all stations for each situation with reasonable values of SD (see the bottom lines in Table 4).

[69] The first result concerning the obtained trends worth discussing is the seasonal dependence of the trends. For both considered moments (1400 LT and SS + 2), the value of the negative trend is approximately higher by a factor of 2 in winter than in summer. No substantial difference between the foF2 trends for the 1400 LT and SS + 2 moments is found. A detailed analysis of these two facts still should be done, but preliminarily one can state that the derived trends in foF2 are caused not only (and not mainly) by a decrease in the temperature but by changes in other parameters (composition and density of the thermospheric gas, system of horizontal winds), which are changed in the process of cooling and contraction of the upper atmosphere.

[70] A very important candidate to the cause of negative trends in foF2 is the possible decrease in the atomic oxygen concentration. Such a decrease is considered [Solomon et al., 2012] as a probable source of stronger effects in cooling of the thermosphere than is expected from numerical simulations for real values of CO2 increase. The decrease in [O] could be caused by stronger eddy diffusion at the turbopause level leading to stronger downward transport of O atoms into its chemical loss [Emmert et al., 2012]. There are indications that an increase in the eddy diffusion efficiency could really exist, but the problem requires a special detailed consideration.

[71] The mean (averaged over four situations and the data sets available for each station) values of hmF2 trends shown in Table 7, though have no high statistical significance, nevertheless, make it possible to draw some conclusions. First of all, the average trends for all stations are negative. That allows us to state that in the period considered there occurred a systematic decrease in the F2 layer height caused by the general cooling and contraction of the upper atmosphere. The average values of the trends for different stations do not strongly differ between themselves and make it possible to state a global character of the obtained decrease in hmF2 at the edge of the centuries.

[72] Using all the data on k presented in Table 5, we attempted to find a difference between the trends in hmF2 for daytime and nighttime conditions (1400 LT and SS + 2) and for summer and winter. The results of the corresponding formal averaging are shown in the four bottom lines of Table 6. One can see that no statistically significant difference in the hmF2 trends between the winter and summer seasons is found. As for two moments of local time considered in the paper, there the difference is more distinctly pronounced and statistically significant: the mean negative trend in the daytime (1400 LT) is higher than at the moment 2 h after sunset. However, the SD values are high enough so the above conclusion should be considered as preliminary.

[73] If for the sake of discussion we accept that the hmF2 trend in the daytime is actually slightly higher than 2 h after sunset when the influence of dynamical processes on hmF2 is maximal, we come to an inevitable conclusion that in the considered period, long-term changes in the F2 layer height are caused not only by changes in the system of horizontal winds (that should have strongly influenced the hmF2 trends in the SS + 2 period) but also by changes in photochemical parameters caused by cooling and contraction of the thermosphere, that is, by downward shift of the density levels at which the F2 layer maximum is formed (that should be manifested mainly in the daytime).

[74] In the opposite case, the values of k for the SS + 2 moment should be substantially larger in magnitude than trends for the daytime. The conclusion on changes in photochemical parameters was obtained above in this section analyzing trends in foF2 and also by Danilov and Konstantinova [2012] on the basis of analysis of foF2(hmF2) dependence for the periods before and after 1980.

[75] As for the seasonal difference between k values for hmF2, one should note that all four anomalously high values of k (k < −5 km/year) in Table 5 fall in the summer period. That determines the obtained small difference in k for summer and winter in Table 6. It is difficult to say whether this difference manifests a real seasonal difference.

[76] An important conclusion of the paper is the fact that before the analyzed period at some stations an increase in hmF2 as compared to the “etalon” period 1957–1980 occurred. This conclusion agrees with the results of Danilov and Vanina-Dart [2010] and Bremer [1998] according to which at various stations both positive and negative trends in hmF2 were obtained before 1990. However, the results of this paper show that, beginning from the mid-1990s at all analyzed stations, a negative trend in hmF2 dominates.

[77] The problem of comparison of the obtained results to results of other authors is the most difficult. In the vast majority of papers on F2 layer trends (see references in Danilov [2012], Bremer [1998], and Bremer et al. [2012]), the data before the mid-1990s were considered. Several relatively recent papers on F2 layer trends for particular stations give trends of the same order of magnitude as obtained in this paper. Rojas and Milla [2012] obtained negative trends in hmF2 from 1.63 to 2.35 km/yr depending on local time on the basis of Jicamarca digisonde observations. Khaitov et al. [2012] obtained a mean negative trend in foF2 k = −0.01 MHz/yr on the basis of observations at Tomsk station. A negative trend in foF2 of −0.015 MHz/yr was reported by Gnabahou1 et al. [2012] for the Ouagadougou station.

[78] The collective experiment on determination of the foF2 trends headed by J. Laštovička [Laštovička et al., 2006] gave for Juliusruh station trends between 0.01 and 0.02 MHz/yr, whereas the averaging of the Juliusruh line in Table 4 of this paper gives 0.03 MHz/yr.

[79] The most severe conflict of our results is with the recent paper by Bremer et al. [2012], where no statistically significant trends in foF2 were obtained. A detailed consideration of the conflict needs a separate detailed work because quite different approaches and methods of trend derivation were used. As for hmF2, negative trends were obtained by Bremer et al. [2012], but a detailed comparison also requires a special work. It is out of the scope of this paper.

[80] On the whole, the statistics of the studies described in section 7 looks in the following way:

[81] The data of 10 ionospheric stations are used.

[82] Number of the analyzed cases N = 128.

[83] Number of cases with SD(3)2/SD(1)2 < 1 is 4.

[84] The average value over all 128 cases is SD(3)2/SD(1)2 = 3.2 ± 2.1

[85] The average SD(3)2/SD(1)2 value for winter = 3.9 ± 2.5 (N = 64)

[86] The average SD(3)2/SD(1)2 value for summer = 2.6 ± 1.4 (N = 64)

[87] The average SD(3)2/SD(1)2 value for 1400LT is 4.2 ± 2.8 (N = 34) for winter and 2.6 ± 1.2 (N = 30) for summer.

[88] For the SS + 2 moment, the seasonal differences are slightly smaller: 3.5 ± 1.9 and 2.7 ± 1.6, respectively.

[89] Although one can see in Tables 8-10 that some scatter of the SD values is seen for various stations and various situations, the average values of SD(3)2/SD(1)2 are statistically significant and make it possible to draw some conclusions.

[90] The main conclusion of section 7 is that the dependence of the critical frequency foF2 of the F2 layer on its height hmF2, which is well pronounced (with high values of R2 and small SD) for the “etalon” period 1957–1979, is distorted substantially in the period after 1980. In our opinion, it is a direct proof of the fact that a change occurs in photochemical parameters of the thermosphere responsible for formation of the F2 layer: temperature, effective recombination coefficient, and composition. If there occurred no such change and the vertical distribution of these parameters under a fixed level of solar activity stayed unchanged, the shift of the F2 layer upward or downward under the influence of vertical drift would bring the layer maximum into the same aeronomical conditions as before. In that case, the same values of foF2 would correspond to the hmF2 values. The analysis performed in this paper shows that it is not the case. The rather distinct relation between hmF2 and foF2 observed during the “etalon” period is either broken completely or has much worse statistical characteristics in the later period. It can occur only if the vertical distribution of photochemical parameters of the thermosphere has changed. Such change should inevitably occur under the observed by other methods (satellite drag, incoherent scatter [see Laštovička, 2009; Laštovička et al., 2012; Danilov, 2012]) intensification of the cooling and contraction of the upper atmosphere.

[91] The obtained stronger effect in winter than in summer (see Table 10 and the summated statistics above) points out the fact that the temperature can hardly play the main role in the occurring process. The character of the temperature dependence of the main ion-molecular reaction O+ + N2 governing the effective recombination coefficient in the F2 layer is such that temperature change should have been manifested stronger in summer time than in winter. Changes in the neutral composition of the thermospheric gas could cause the changes considered in this paper. With the increase in cooling and contraction of the upper atmosphere, both the oxygen atoms concentration and the concentration of molecular compounds should decrease at a fixed height. The electron concentration is proportional to [O]n (n = 0.7–0.85) [Mikhailov et al., 1995], and so simultaneous decrease in both [O] and [N2] with the [O]/[N2] ratio kept nearly unchanged still would lead to a decrease in Ne. The analysis of the trends in foF2 described above in this paper leads to the same conclusion. Possible decrease in the atomic oxygen concentration, which is considered as one of the causes of the cooling increase in the thermosphere [Solomon et al., 2012] and which can be a result of the intensification of the eddy diffusion around the turbopause, also is able to contribute to the effect of the foF2 decrease with development of cooling and contraction of the upper atmosphere. Emmert et al. [2012] noted that such intensification (if it exists) would explain their results on CO2 measurements.

[92] As a by-product of this study, it is worth noting that the utilization of the considered databases on foF2 and hmF2 shows that the most consistent results are obtained using the SRIDR database, especially when there are data for the entire period 1958–2010. A combination of the SPIDER data with the IWG data (when there are no SPIDER data for the earlier years) also provides reliable enough results. The Damboldt database and medians in our opinion are less suitable for trend studies.

8 Conclusions

[93] Two methods were used to derive trends in the F2 layer parameters foF2 and hmF2 for the period 1990–2010. Several sources of the initial data were used. For foF2, negative trends were obtained for the vast majority of 72 positions (station + LT moment + source) by the linear method. The second method provided a visual proof that there was a decrease in foF2 from the “etalon” period 1958–1979 to the considered period after 1990. It is also found that the scatter of the foF2 data relative to the foF2 dependence on solar activity (the F10.7 index) is higher in the period considered than in the “etalon” period.

[94] For hmF2, the picture of the trend behavior with time is more complicated. For some stations, an increase in hmF2 as compared to the “etalon” period is found. However, for the majority of cases, linear trends in hmF2 are negative after 1990–1994.

[95] The results presented in section 7 suggest an explanation of the fact (numerously mentioned in various publications, see above) that after approximately 1980, long-term changes in various parameters in the ionosphere and thermosphere began to occur. The explanation is based on the analysis of the relation between the main parameters of the F2 layer (foF2 and hmF2) in the “etalon” period before 1980 and the analyzed period after 1980. The results of the analysis show that the relation between foF2 and hmF2, which was distinctly pronounced in the “etalon” period (this fact is clearly understood from the point of view of the F2 layer physics) is broken (becomes less pronounced and less statistically significant) in the later period.

[96] The above-described event could be observed only if there is a change in the vertical distribution of the photochemical parameters governing the NmF2 (foF2) values in the F2 layer maximum. Such change should inevitably occur at cooling and contraction of the upper atmosphere.

[97] Thus, the results obtained in this paper could be qualitatively understood in the frames of the concept of cooling and contraction of the upper atmosphere [Laštovička et al., 2008, 2012; Laštovička, 2009]. As a by-product of this investigation, some (rather moderate) differences have been found between the results based on different international databases.


[98] The work was supported by the Russian Foundation for Basic Research (project 11-05-00102-а).