Trends in the ionospheric E and F regions over Europe

Continuous observations in the ionospheric E and F regions have been regularly carried out since the fifties of this century at many ionosonde stations. Using these data from 31 European stations long-term trends have been derived for different parameters of the ionospheric E layer (h′ E, foE), F1 layer (foF1) and F2 layer (hmF2,foF2). The detected trends in the E and F1 layers (lowering of the E region height h′E; increase of the peak electron densities of the E and F1 layers, foE and foF1) are in qualitative agreement with model predictions of an increasing atmospheric greenhouse effect. In the F2 region, however, the results are more complex. Whereas in the European region west of 30° E negative trends in hmF2 (peak height of the F2 layer) and in the peak electron density (foF2) have been found, in the eastern part of Europe (east of 30° E) positive trends dominate in both parameters. These marked longitudinal differences cannot be explained by an increasing greenhouse effect only, here probably dynamical effects in the F2 layer seem to play an essential role.

Abstract. Continuous observations in the ionospheric E and F regions have been regularly carried out since the ®fties of this century at many ionosonde stations. Using these data from 31 European stations long-term trends have been derived for dierent parameters of the ionospheric E layer (h¢ E, foE), F1 layer (foF1) and F2 layer (hmF2, foF2). The detected trends in the E and F1 layers (lowering of the E region height h¢E; increase of the peak electron densities of the E and F1 layers, foE and foF1) are in qualitative agreement with model predictions of an increasing atmospheric greenhouse eect. In the F2 region, however, the results are more complex. Whereas in the European region west of 30°E negative trends in hmF2 (peak height of the F2 layer) and in the peak electron density (foF2) have been found, in the eastern part of Europe (east of 30°E) positive trends dominate in both parameters. These marked longitudinal dierences cannot be explained by an increasing greenhouse eect only, here probably dynamical eects in the F2 layer seem to play an essential role.

Introduction
Based on the results by Roble and Dickinson (1989) with a globally averaged model of the coupled meso-, thermo-and ionosphere, Rishbeth (1990) predicted a lowering of E and F2 peak heights (hmE, hmF2) by about 2.5 km and 15±20 km if the atmospheric CO 2 and CH 4 content is doubled. The expected changes in the E and F2 peak electron densities should be small. In a more recent paper Rishbeth and Roble (1992) investi-gated the atmospheric greenhouse eect for December solstice conditions with the 3-dimensional TIGCM (thermosphere/ionosphere general circulation model) developed at the National Center for Atmospheric Research, Boulder. Assuming again a doubling of the greenhouse gases, a lowering of hmF2 by 10±20 km was derived whereas the electron density at the peak height of the F2 layer and above decreases (foF2 decreases up to 0.5 MHz) but below the F2 peak the electron density increases (including F1 and E regions).
Using the data of individual ionosonde stations Bremer (1992) with observations from Juliusruh and Poitiers and more recently Ulich and Turunen (1997a) with data from SodankylaÈ found a signi®cant lowering in hmF2 during the last 40 years. Recent investigations by Bremer (1996Bremer ( , 1997a and by Ulich and Turunen (1997b) gave, however, some hints that the trends in the F2 region are not so uniform as expected. Therefore, in this study the data of all available European ionosonde stations are considered to provide more reliable results about the thermospheric trends in dependence on latitude, longitude and altitude. In these analyses characteristic parameters of the F2 layer (hmF2, foF2), of the F1 layer (foF1) and of the E layer (h¢E, foE) have been used. The large amount of data necessary for these investigations were extracted from dierent CD-ROMs of the National Geophysical Data Center (NGDC), Boulder, Colorado, USA and from the WDC-C at the Rutherford Appleton Laboratory (RAL), Chilton, Didcot, UK.

Experimental data and data analysis
Data from 31 European ionosonde stations (5°W ± 70°E, 35°N ±70°N) have been used for trend analyses. In Table 1 these stations are presented in dependence on latitude. The trend coecients given will be discussed later in Sects. 3 and 4.
In the trend analyses we used only monthly median values of the dierent ionosonde parameters. It is wellknown that these parameters depend markedly on solar and geomagnetic activity. To detect relatively small trends in the ionospheric parameters it is therefore necessary to remove these strong in¯uences, e. g. by a simple regression analysis. We calculated for the experimental data X (X hmF2, foF2, foF1,h¢E or foE) for each complete hour (0±23 UT) and each month the following regression equations th e f g epX 1 Here R is the solar sunspot number and Ap the planetary geomagnetic activity index. Instead of R also the solar¯ux data F 10.7 can be used to characterize the solar activity, the results of the trend analysis do not depend on this choice. Using the results of the regression analysis after Eq. (1) we calculated the deviations DX of the experimental hourly data X exp from these theoretical values X th D exp À th X 2 With these DX values linear trends can be estimated according to

D yearX 3
In this way it is possible to derive trend parameters b for each hour and each month. In this work the hourly and monthly DX values have normally been averaged to calculate more reliable yearly DX values and yearly trends after Eq (3). The test of the signi®cance of the trend parameter b can be made with Fisher's F parameter p r 2 x À 2a1 À r 2 X 4 Here r is the correlation coecient between DX and year after Eq. (3), and N is the number of years used in the analysis. The signi®cance levels to be exceeded by the F parameter after Eq. (4) can be found in Taubenheim (1969). Signi®cant trends with a con®dence level of more than 90% are maked in Table 1 or with solid symbols in Figs. 1±6.

F2 region
The peak height of the F2 layer (hmF2) and the electron density at this height (foF2) are used in the trend investigations of the F2 region. The value of hmF2 was derived from observed monthly M(3000)F2 values using the well-known formula of Shimazaki (1955) hmF2 1490aw3000F2 À 176X 5 As shown by Bremer (1992) the choice of this simple formula is not critical for the derived trends, the trends are nearly identical if the more accurate formula of Bilitza et al. (1979) is used for the estimation of hmF2.
In Fig. 1 the results of the trend analysis for hmF2 values are separately presented for European stations at longitudes less than 30°E in the upper part of this ®gure and at longitudes greater than 30°E in the lower part. In both cases the trends are ordered in dependence on latitude, shown in brackets below the name of the station. If the linear trends are signi®cant with more than 90% the yearly data DhmF2 are presented by solid dots, if the signi®cance level is smaller than 90% open circles have been used. The numerical values of the derived trend coecients can be found in Table 1. Whereas the trends do not show a dependence on latitude in the latitudinal belt investigated between 35°N and 70°N, there are marked dierences between the two longitudinal regions west and east of 30°E. In the part west of 30°E 11 of 15 trends are negative, from 9 signi®cant trends are even 8 negative. In the eastern part of Europe (east of 30°E) the results are quite opposite. Eleven of 13 trends including 7 of 8 signi®cant trends are positive. This marked dierence can also be seen in Fig. 2 where the hmF2 trend parameters are shown in relation to longitude. The two straight lines mark the median values of the individual trends in the two longitudinal zones: A0.18 km/y and 0.19 km\y. The corresponding mean values for both regions which slightly deviate from the median values shown in Fig. 2 are signi®cantly dierent with more than 99% as can easily be shown by a t-test (Taubenheim 1969).
The trends of foF2 are presented in Fig. 3 in a similar way as for hmF2 in Fig. 1. Also the foF2 trends do not markedly depend on latitude. But there seems to exist again a dierence between the two longitudinal regions west or east of 30°E. The median values of the foF2 trends are: A0.0024 MHz/y for the western part (stations in the upper part of Fig. 3) and 0.0015 MHz/ year for the eastern region (lower part of Fig. 3). The scatter of the values is however markedly higher than for the corresponding hmF2 trends presented already. Therefore, the concept that the mean trends of foF2 in both regions are dierent could be shown with only a low signi®cance level of about 84%. This behaviour can also be supported by the fact that in the western part 12 of 18 trends are negative (but from 6 signi®cant trends are 3 negative and 3 positive) whereas in the eastern part 8 of 13 trends are positive (here no signi®cant trends could be discerned!).
As a summary it can be stated that in the F2 layer above the European region in hmF2 and foF2 trends no marked dierences in dependence on latitude could be found but clear dierences between the trends in the longitudinal regions west and east of about 30°E. In the western part the trends are mainly negative and in the eastern part dominantly positive. The correlation coef-®cient between the hmF2 and foF2 trend values is due to the strong scatter of the individual trends (especially in the foF2 trends) only r 0.38. Nevertheless this value is after the t-test (Taubenheim, 1969) with 95% significance).

F1 region
The peak electron density of the F1 layer, foF1, has been used for a trend analysis at F1 region heights near 170± 180 km. Unfortunately only values of 12 stations were available, 4 stations for the region east of 30°E and 8 stations in the western part of Europe. In Fig. 4 the trend results are therefore presented only in their dependence on latitude and not subdivided into dierent longitudinal regions. The trends are in most cases positive (9 of 12 trends including 4 of 5 signi®cant trends) with a median value 0.001 MHz/year. In agreement with the results in the F2 region in the F1 layer the trends do not markedly depend on the latitude, but in contrast to the F2 layer in the F1 layer no marked dierences could be found also for dierent longitudinal regions. At longitudes west of 30°E 6 of 8 stations have positive trends, and at longitudes east of 30°E 3 of 4 trends are positive.

E region
The E region re¯ection height h¢E and the peak electron density foE have been used for trend analyses at E region heights (110±120 km). In Fig. 5 the results for h¢E are presented. Unfortunately only observations at 10 stations are available, and all of them are from the western part of Europe. From most of measurements negative trends were derived (8 of 10 including 5 of 6 signi®cant trends) with a median value A0.07 km/year. Also here the trends do not show a marked dependence on latitude.
In Fig. 6 the foE trends for 25 European stations are shown, separated again for two longitudinal regions west (upper part) and east of 30°E (lower part). The trends are relatively uniform, 22 of 25 trends (including 10 of 11 signi®cant trends) are positive. The median trend value is 0.0013 MHz/year. Marked longitudinal and latitudinal dependencies could not be found in the European E region investigated.

Discussion
As the ionospheric plasma parameters are markedly controlled by the solar activity it is necessary to remove this part carefully if derivation of long-term trends are required which are normally more than one order of magnitude smaller than the solar induced variations. In this study a linear dependence of the plasma parameters on solar as well as geomagnetic activity is adopted to describe the solar in¯uence. This model is very similar to a model used by Kane (1992) to get a better description of the hysteresis eect in foF2 observations. Kane (1992) Fig. 1. Yearly trends of DhmF2 after elimination of the solar and geomagnetic in¯uences for European stations at longitudes west of 30°E (upper part) and at longitudes east of 30°E (lower part). Signi®cant trends (>90%) are marked by solid dots only used the solar 10.7 cm radio¯ux (F10.7) instead of the solar sunspot number R but also the geomagnetic Ap index. After Kouris et al. (1994) foF2 is slightly better correlated with solar activity if a quadratic relation is assumed (f (R, R 2 ) a + bR + cR 2 ). This eect can be seen in Fig. 7 where the monthly mean noon values (12 LT) of foF2, hmF2, foF1, foE, and h¢E from observations at Juliusruh between 1957 and 1996 are presented showing their dependence on solar activity. The data for January (crosses) and July (solid circles) are shown together with the linearly (complete lines) and quadratically ®tted curves (dashed). The dierences between the linear and quadratic approximations are often very small but sometimes remarkable (e.g. hmF2 in July). The quality of the dierent approximations of the solar-induced variations can be tested by correlation coecients between the observed data with the dierent model results of these data. In Fig. 8 these correlation coecients are presented for foF2 and hmF2 data from Juliusruh for January and July dependent on local time. All correlation coecients are highly signi®cant, also for the parameters foF1 and foE (not shown here), although the correlation of h¢E with solar activity is markedly smaller. The dierences between the correlation coecients for the same ionospheric parameter are very small. In most cases the correlation with the linear f (R) model (open circles in Fig. 8) is slightly smaller and the best correlations have been found sometimes for the f (R, Ap) values (crosses) and sometimes for the f (R, R 2 ) values (solid circles). In Fig. 9 the trends in foF2 and hmF2 for Juliusruh are presented using dierent models to remove the solar-induced variation. Here all data (all 24 h of all 12 months) have been used as in Figs. 1±6. The general agreement of the three dierent curves of the DfoF2 and DhmF2 values in Fig. 9 is very good, indicating that the derived trends do not depend greatly on the model used to eliminate the solar-induced variation. In detail small dierences may occur in the individual data, the general behaviour, however, is little in¯uenced by the choice of the model. In particular the f (R, R2) and the f (R, Ap) models give very similar results. Independent of the model used the removal of the solar-induced variation requires a long data series. Therefore in this study nearly all data series investigated are longer than 20 y with only very few exceptions, as can be seen in Figs. 1±6.
The model calculations for an atmospheric greenhouse eect are normally carried out for a doubling of the greenhouse gases CO 2 and CH 4 . To compare these theoretical results with our experimental trend data it is necessary to take into account the real changes of these gases during the last 40 y, corresponding to the main interval of ionosonde observations which has been analysed here. After the recent IPCC report (Houghton et al., 1996) and Brasseur and de Rudder (1987) the atmospheric content of these gases has increased during the last 40 y for CO 2 by about 15%, CH 4 by about 44%, N 2 O by about 7% and for the CFCs and HCFCs partly by essentially higher rates (yearly increases between 0.5%). Taking into account the dierent amounts of these greenhouse gases, for the total direct radiative forcing, after Houghton et al. (1996) an eective increase of the greenhouse gases of about 20% can be assumed for the last 40 y.
After the model calculations of Rishbeth (1990) and Rishbeth and Roble (1992) the F2 peak height hmF2 should be lowered by about 10±20 km if the greenhouse gases are doubled, corresponding to a lowering of about 2±4 km for the last 40 y. The median experimental trends derived in this study however, give a lowering of 7.2 km (at longitudes west of 30°E) or an increase of 7.6 km (at longitudes east of 30°E) for the last 40 y. Whereas the experimental lowering of hmF2 in the western part of Europe is only slightly stronger than the theoretically expected result, in the eastern part even the direction between predicted and observed trend values is contrary. Such a behavior of regional dierences in the hmF2 trends has also been found by Ulich and Turunen (1997b) using data from ionosonde stations all around the world. It is dicult to decide if technical changes of the equipment or changes in the evaluating algorithm could be responsible for the considerable dierences especially for the unexpected positive trends in the eastern part of Europe. Therefore, for three stations of this region the hmF2 trends are shown in Fig. 10 once again. Indeed there seem to be some discontinuities in the trends (marked by dotted vertical lines) which could be caused by technical changes, but nevertheless there remain large parts of the observations with positive trends. Therefore, the trends of individual stations may be partly in¯uenced by technical changes, but it seems to be impossible to argue that the positive trends in the eastern part of Europe and in other regions of the world as found by Ulich and Turunen (1997) are only caused by arti®cial changes.
Similar dierences to those for hmF2 trends were also detected in the foF2 trends in Fig. 3. In the western part of Europe we detected a mean foF2 decrease during the last 40 y of about 0.1 MHz which is in relatively good agreement with the results of Rishbeth and Roble (1992) who predicted a lowering up to about 0.5 MHz for a doubling of the greenhouse gases. But in the eastern part of Europe an increase of about 0.06 MHz was detected in contrast to the predicted reduction of Rishbeth and Roble (1992). In general the dierences between the foF2 trends in the western and eastern part of Europe qualitatively agree with the dierences in the hmF2 trends, however the dierences in the hmF2 trends seem to be more pronounced. The reason for these longitudinal dierences are not yet known, but they are probably caused by the complex dynamics of the F2 region. The hmF2 and foF2 maps derived with the TIGCM by Rishbeth and Roble (1992) contain structures which could be connected with dierent regional trends of these F2 region parameters. However detailed experimental and theoretical investigations are necessary in the future to reach a better understanding of trends in the F2 region dependent on latitude and longitude.
At lower heights (F1 and E regions) where dynamical processes are not so dominant as in the F2 layer latitudinal as well as longitudinal dierences could not be detected. Using the median foF1 trend of 0.001 MHz/ year a mean foF1 increase of 0.04 MHz during the last 40 y was derived from the experimental results. Rishbeth and Roble (1992) derived, with their model calculations, an increase in the electron density near the peak height of the F1 layer between 5% and 50%, with a median value In the E region two parameters have been used for trend investigations, h¢E and foE (see Figs. 5 and 6). With the median trend value of -0.07 km/y we get a lowering of the E region by about 2.8 km during the last 40 y which is in the same order as predicted by Rishbeth (1990) for a doubling of the greenhouse gases. Therefore, the experimental value is higher than the predicted value by about a factor of 5. Using the mean foE trend of 0.0013 MHz/year a foE increase of about 0.05 MHz was observed during the last 40 y. After the model calculations of Rishbeth and Roble (1992) an increase of the electron density at E region heights of about 5% (corresponding to an increase in foE by about 2.5%) is predicted for a doubling of the greenhouse gases. For typical foE values of 2±3 MHz an foE increase of about 0.05±0.08 MHz can be expected (or 0.01±0.02 MHz for the last 40 y). Also for this parameter the mean experimental trend value is slightly higher than the prediction. The observed foE increase is also in qualitative agreement with trends of the ion ratio [NO + ]/ [O 2 + ] obtained from mass spectrometer measurements by Danilov and Smirnova (1997 1957± 1996) and dierent model values assuming a linear (open circles) or quadratic dependence on solar sunspot number R (solid circles) and for a twofold regression with R and geomagnetic Ap values (crosses) in dependence on local time LT Based on observations at 31 European ionosonde stations over the last 40 y long-term trends have been derived at dierent ionospheric parameters of the E, F1 and F2 regions. In spite of marked dierences between the results derived at single stations some common results could be found: 1. Technical changes of equipment or the evaluation algorithms used may partly in¯uence the trend results derived for individual stations. However the comparison of the results of many dierent stations allows the derivation of representative (median) trends for the European region (5°W A70°E ; 35°N -70°N). 2. The trends in the E and F1 layers with lowering of the E region height, h¢E, and increases of the peak electron densities, foE and foF1, do not depend on latitude and longitude and are in qualitative agreement with modelling results of the atmospheric greenhouse eect. The experimental mean trends are, however, stronger by a factor of 2..5 compared with the theoretical predictions for the E region and smaller by a factor of 2 with F1 region predictions. 3. The trends in the F2 region (hmF2 and foF2) do not depend on latitude but show marked dierences for the longitudinal regions west or east of 30°E.
Whereas at longitudes west of 30°E the trends in hmF2 as well as foF2 are in general negative as expected after model calculations for an atmospheric greenhouse eect, the trends in the European region east of 30°E are in general positive. These longitudinal dierences are more pronounced in the hmF2 data but could also be detected in the foF2 data. The cause of these regional dierences which were also found by Ulich and Turunen (1997b) in world-wide hmF2 data is unknown at present and needs further investigations.