Radio Science

Capturing the morphology of long-duration negative ionospheric disturbances using an empirical pattern recognition method

Authors


Abstract

[1] On the basis of an ionospheric definition of disturbed conditions independent of any causative mechanism, a feature-guided pattern recognition method reveals the dominant morphology of long-duration negative foF2 disturbances. A catalogue of negative disturbances lasting more than 24 hours is compiled from hourly foF2 data from 75 ionosonde stations and three solar cycles. Disturbances in each month and station are handled separately, and four local time intervals of disturbance commencement are considered. A median disturbance profile is produced only when a minimum occurrence probability holds. The time window under morphological investigation is selected such that nonsystematic features, precursor phenomena, and poststorm effects are not included in analysis. The disturbance patterns, first grouped according to major characteristic features and then fitted with simple mathematic functions, are described by a range of the normalized deviation of hourly foF2 to its corresponding monthly median and are provided to radio users along with their distribution in space and time. The present model is a nonconditional stand-alone model which may, in the event of an ionospheric disturbance at a certain location, predict its further development.

1. Introduction

[2] The prediction of the F2 layer critical frequency (foF2) is a challenging task for researchers, especially in the event of ionospheric disturbances, when large deviations of foF2 from the average quiet conditions grow, more or less, abruptly. Empirical models, established through statistical analysis of long measurement records, exhibit certain merits in the short-term forecasting of disturbed foF2 [Bilitza, 2002] and can be particularly successful if driven, apart from any storm time physical parameter function, by near real-time measurements too [Wilkinson, 1995]. On the other hand, their power is limited by the distribution of the selected data set, in terms of times and locations. Furthermore, and aiming toward better frequency management, empirical models are evaluated depending how well, compared to scaling errors, they capture the onset, amplitude and phase of ionospheric disturbances at each moment during the evolution of a storm.

[3] The International Reference Ionosphere (IRI) is an acclaimed by URSI and regularly updated empirical ionospheric model [Bilitza, 2001] which provides radio users with monthly averages for magnetically quiet conditions and corrected values for active conditions. Among efforts in modeling ionospheric response to disturbed magnetic conditions [e.g., Wrenn et al., 1987; Cander and Mihajlovic, 1998; Kutiev and Muhtarov, 2004], IRI has recently incorporated the STORM model of Araujo-Pradere et al. [2002]. Such models certainly introduce improvement over the former “climatological” IRI product; however they deal with only one out of several likely causes of ionospheric disturbances [Prölss, 1995] and usually ignore the regional nature of the ionosphere [Wilkinson, 1995].

[4] Another approach in the development of accurate storm models is the unconditional statistical study of the ionospheric medium in the first place, by employing a posteriori, feature-based pattern recognition methods in order to identify specific morphologies of ionospheric disturbances. In a second step, storm patterns of adequate frequency of occurrence at certain times and geographical theaters are linked to geophysical parameters, monitored in near real time, easing thus the short-term forecast. Following statistical analyses of day-to-day foF2 variability data, Kouris et al. [1998, 1999] have developed an ionospheric definition of disturbed foF2 conditions, based exclusively on their amplitude and duration. Kouris and Fotiadis [2002] validated the selection of appropriate amplitude/duration thresholds in defining disturbed ionospheric conditions with a study of a greater number of stations. Furthermore, aiming to enable detailed studies on storm morphology while avoiding at the same time grouping storms of different morphological features, Fotiadis et al. [2004] presented an analytical climatology of foF2 disturbances for 75 ionosonde stations and three solar cycles. In this latter work different classes of disturbances are identified according to their phase, local time onset and duration and their probability of occurrence is provided to radio users.

[5] Large-scale negative ionospheric disturbances have long been a popular research study [e.g., Matsushita, 1959] because they severely degrade and often disrupt HF communication systems. The purpose of this paper is to reveal and quantify the systematic and dominant morphological patterns of long-duration negative foF2 disturbances, so that they become available to radio users, allowing for their distribution in time and space and offering at the same time a simple prediction tool for their development once they occur at a single location [Kouris et al., 1998, 1999].

2. Disturbance Pattern Recognition Method

[6] In order to reveal the dominant patterns of long-duration negative foF2 disturbances and before the implementation of any method it should be ensured that the data under consideration (a) indeed represent the disturbed state and (b) have overall an adequate frequency of occurrence. Aiming to develop an ionospheric definition of disturbed conditions, independent of any potential causative mechanism, Kouris et al. [1998] defined a day as disturbed when the normalized deviation of hourly foF2 to its corresponding monthly median, hereafter called dfoF2, exceeds 0.30 in absolute value for at least 3 consecutive hours. Such deviations being greater (smaller) than the monthly median conditions are considered as positive (negative) disturbances. The disturbance ends when the absolute dfoF2 remains less than or equal to 0.20 for more than a 3-hour period [Kouris et al., 1999]. The selection of the 0.30 dfoF2 threshold ensures that an ionospheric disturbance is present since such deviations are normally observed for less than 10% of the monthly time [Kouris and Fotiadis, 2002]. Moreover, the minimum 3-hour duration criterion quite ascertains of a disturbed period by helping to avoid considering scaling errors as disturbed values. On the other hand, the selection of a more relaxed threshold for disturbance cease seems proper since for a midlatitude station dfoF2 lies within 0.20 for more than 85–90% in a month [Kouris and Fotiadis, 2002], while, again, a persistence period of more than 3 hours is supportive of the end of a disturbance.

[7] Since the power of any statistical model lies on how long and representative the data set under consideration is, in the current investigation hourly dfoF2 values from 75 ionospheric stations and almost three solar cycles were calculated (exact periods/stations are fully reported by Fotiadis et al. [2004, Table I, pp. 1312–1313]; then a disturbance catalogue on hourly basis was compiled for each station, month and year, according to the previously mentioned definition [Kouris et al., 1998, 1999]. As large-scale negative disturbances are those considered with a minimum duration of 24 hours and, on the basis of the solar zenith angle at 300 km, these may commence at four local time (LT) windows: day (cos χ > 0.20), night (cos χ = 0) while dawn and dusk fall in between [Fotiadis et al., 2004]. Thus, in the present study, large-scale negative disturbances in each month, station and LT window were dealt with as separate cases/sets; that is, 12 × 75 × 4 sets of disturbances were initially considered. However, phenomena of such depth and duration are not present at all locations and seasons. In order to exclude any nonsystematic trends, a minimum frequency of occurrence from year to year was set to 0.15; that is, only sets of disturbances containing more than four disturbances in the full three solar cycle period were further considered for analysis. Table 1 reports ionospheric stations (41 out of 75) for which some sets of disturbances met the aforementioned combined selection criterion of disturbed state and minimum occurrence. The corrected magnetic latitude (CML) in Table 1 was calculated after the International Geomagnetic Reference Field (IGRF) model for the year 1986 and 300 km height.

Table 1. List of Stations With Significant Number of Long-Duration (>24 Hours) Negative foF2 Ionospheric Disturbances
StationCodeGeographicIGRF CML,a deg ϕm
Latitude, degLongitude, deg
  • a

    Corrected geomagnetic latitude (CML) is calculated with the International Geomagnetic Reference Field (IGRF) model for the year 1986 and 300 km height.

RomeROM41.812.535.7
PoitiersPOI46.60.342.3
LannionLAN48.8356.645.4
SloughSLO51.5359.448.4
JuliusruhJUL54.613.450.8
MoscowMOS55.537.351.0
UppsalaUPP59.817.656.3
LeningradLEN6030.755.9
ArkhangelskARK64.440.560.0
LyckseleLYC64.718.861.3
SodankylaSOD67.426.663.6
JohannesburgJOH−26.128.1−35.6
GrahamstownGRW−33.326.5−41.2
TaipeiTAI25121.517.5
WakkanaiWAK45.4141.738.1
IrkutskIRK52.510446.8
MagadanMAG6015153.2
YakutskYAK62129.655.5
VanimoVAN−2.7141.3−11.4
DarwinDAR−12.4130.9−22.2
TownsvilleTOW−19.3146.7−28.7
La ReunionLAR−21.155.9−30.9
NorfolkNOR−29168−35.9
BrisbaneBRI−27.5152.9−36.8
MundaringMUN−32116.2−44.5
CanberraCAN−35.3149−45.8
KerguellenKER−49.470.3−58.4
HobartHOB−42.9147.2−54.2
Campbell Isl.CLL−52.5169.2−60.2
Scott BaseSCO−77.9166.8−79.9
MauiMAU20.8203.521.3
Point ArguelloPNT34.6239.440.5
Wallops Isl.WAL37.8284.549.5
BoulderBOU40254.749.1
OttawaOTT45.4284.156.9
St. JohnsSTJ47.6307.355.2
GoosebayGOO53.3299.262.6
ChurchillCHU58.8265.869.5
Port StanleyPOR−51.7302.2−37.3
Argentine Isl.ARI−65.2295.7−49.8
Halley BayHAL−75.5333.4−61.5

[8] Now, for each set of disturbances, and assuming that LT hour may be a random variable within each LT window, the hourly dfoF2 of each disturbance were arranged with respect to their commencement local time hour within the corresponding LT window and a further time window 2 days before and 3 days after this point was examined. Then, for each storm time hour (independent variable) and set of disturbances the median, instead of the average, dfoF2 value was calculated forming thus a new disturbance pattern to be considered in the following three-step pattern recognition analysis.

[9] The main goal of the feature-guided, empirical pattern recognition method was to capture the main phase disturbances' pattern and, in order to achieve this, precursor phenomena and post effects should be excluded from analysis. Thus a visual inspection of all profiles was carried out at first so as to select a more narrow time window before and after the disturbances' main phase which has a minimum duration of 24 hours. The rough rule used to ascertain visual inspection was to ensure that absolute dfoF2 values remain well within 0.20 (off the median) for at least 3 hours. An optimum time window for disturbance modeling was found to be within [−12, 41] hours of local storm time; however it should be noted here that the time window may depend also on the type of disturbance.

[10] The second step of this empirical method was to fit all disturbance profiles with a library of simple mathematical functions. In order to obtain reliable disturbance patterns it was assumed that a profile follows a distribution only when both the correlation coefficient was greater than 0.80 and at the same time the overall fitting standard error remains smaller than 4%. In case more than one functions satisfied the above combined criterion, then the simplest in terms of greater F statistic was selected, meaning fewer degrees of freedom and parameters to be calculated.

[11] After certain disturbance pattern types were extracted, the range of dfoF2 variation should be given for each type at each storm time hour of the [−12, 41] hours window. Therefore the minimum and the maximum dfoF2 value at each storm time hour was calculated for each pattern type, forming an “envelope” which offers radio users dynamic variability bounds [Wilkinson, 1995].

3. Morphology of Long-Duration Negative Ionospheric Disturbances

3.1. Pattern Description

[12] The visual inspection of the disturbance profiles verified the common belief [e.g., Matsushita, 1959; Campbell, 1996] that most large-scale negative disturbances follow in general the lognormal distribution. However, the fitting procedure resulted in an exponential pulse function (A pattern):

equation image

where only four parameters need to be specified. Tables 2a, 2b, and 2c provide users with all respective calculated function parameters, while Figure 1 illustrates the range of dfoF2 values (“envelope”) of observed storm patterns and the respective fitted functions. As seen from Figure 1, the A pattern disturbance maximizes about 6 hours from zero storm time while there can be either positive or negative prestorm dfoF2 deviations of some significance.

Figure 1.

Range of dfoF2 variation of observed storm patterns (circles) and the respective adjusted functions (curves) for long-duration (>24 hours) negative storms. One of the observed storm patterns is shown with diamonds. The y axis shows deviation from the foF2 monthly median (dfoF2), and the x axis shows local storm time hours after Kouris et al. [1998, 1999].

Table 2a. Limits of the Calculated Storm Patterns B and B*
 BB*
Upper LimitLower LimitUpper LimitLower Limit
a000.05−0.090.07−0.26
a010.05−0.090.02−0.20
b−0.27−0.40−0.23−0.41
c0.00−0.21--
d0−0.27−0.40−0.23−0.41
d1−0.16−0.16−0.13−0.11
e0−2.20−1.78−1.01−0.93
e10.070.050.020.02
Table 2b. Limits of the Calculated Storm Patterns Γ1 and Γ2
 Γ1Γ2
Upper LimitLower LimitUpper LimitLower Limit
a−0.15563−0.31256−0.17592−0.33170
b−0.09155−0.01133−0.04996−0.02261
c−0.04344−0.05705−0.07402−0.08652
d0.021830.016870.040580.05670
e0.00049−0.000340.00256−0.00514
f−0.00109−0.00094−0.00272−0.00450
g × 1040.7071.430.5118.29
h × 1051.581.334.709.16
i × 106−1.37−2.47−1.67−20.41
Table 2c. Limits of the Calculated Storm Patterns A and Γ3
 AΓ3
Upper LimitLower LimitUpper LimitLower Limit
a0.1500−0.2014−0.10826−0.33228
b−0.4577−0.33280.009880.09637
c−3.0446−1.7308−0.08003−0.06668
d15.168611.52510.027080.00354
e  0.00586−0.00120
f  −0.00461−0.00188
g  0.000250.00087
h  0.000200.00010
i × 105  −2.5473−4.3053
j × 106  −2.0471−1.3199
k × 107  4.30775.7113

[13] On the other hand, B pattern disturbances present no eventual peaks during their main phase and may be adequately described with the following step-like functions:

equation image

where, while they break out abruptly, recovery may progress in a more or less abrupt way. B pattern is quite shallow, reaching a depth of −0.40, and constant, remaining at −0.35 ± 0.05 for about 28 hours (Figure 1). Quite the same variation applies for B* pattern which seems to be a prolonged version of the aforementioned B pattern. B* pattern remains at maximum −0.40 for about 32 hours and recovers very slowly, requiring maybe more than 41 hours. On the other hand the prestorm activity is much more intense than that of the B pattern.

[14] The A pattern disturbances quite often present some secondary minima during their recovery phase requiring bimodal distribution functions, such as polynomials, to describe them (Γ patterns):

equation image

The following cases have been identified (Figure 1): (a) when the secondary peak during the recovery phase is less powerful than the peak of the main phase (Γ1 pattern) and (b) when the secondary peak reaches a greater depth than the main phase peak, resulting thus in a further deterioration (Γ2 pattern). In the Γ1 pattern the primary minimum of about −0.50 is reached at about 6 hours local storm time whereas the less deep secondary minimum appears around 24 hours. On the contrary, in the Γ2 pattern the first minimum emerges in less than 6 hours and the second more intense one (less than −0.50) again at about 24 hours local storm time. It should be noted here that the Γ2 pattern disturbance may last longer than 41 hours, causing thus far more deleterious effects on telecommunication systems.

[15] Another disturbance morphology may be considered when a single main phase peak delays more than 6–8 hours and in specific more than 12 hours local storm time (Γ3 pattern) but sharply declines immediately after its peak. In this case however, after the completion of 24 hours of activity the storm effect may as well continue for more than 41 hours. Furthermore, a few hours before a Γ3 pattern disturbance breaks out a slight positive effect could be detected, while 12 hours before a stronger negative effect seems to be dominant.

3.2. Storm Pattern Space/Time Distribution and Model Accuracy

[16] In order to be potentially operational for radio users, the disturbance patterns identified above should be completed with the following information: (a) their space/time distribution and (b) their occurrence probability. Tables 3a, 3b, 3c, and 3d present the distribution of negative disturbance patterns with the month and location, depending on their LT window of commencement. On the other hand, although Fotiadis et al. [2004] have provided the latter for large-scale negative storms in general, probability should still be linked with the specific patterns found. Aiming however to ensure users of the model accuracy, Tables 4a, 4b, 4c, and 4d present, for each ionospheric station, month and LT storm commencement window, the corresponding mean absolute errors between the observed disturbances (in the period of available data) and the modeled patterns, in accordance to Tables 3a–3d. It should be noted that an absolute error is calculated (at each hour of the storm window) to the nearest dfoF2 model boundary only when a storm time dfoF2 deviation lies outside the respective envelope.

Table 3a. Distribution of Patterns of Long-Duration (>24 Hours) Negative foF2 Storms Commencing During Sunrise With the Location and Montha
IGRF CML, deg ϕmStationMonth
JanFebMarAprMayJunJulAugSepOctNovDec
  • a

    Patterns A, B, and Γ refer to Figure 1. A hyphen means not adjusted to the patterns of Figure 1.

American Longitude Zone
56.9OTT        Γ1   
49.5WAL  (B)A    Γ2A  
49.1BOU - A    A(B)  
40.5PNT   -A  (B)    
21.3MAU       Γ2    
−49.8ARI Γ1A       Γ2A
−61.5HAL Γ2          
 
European Longitude Zone
63.6SOD         B*Γ3 
61.3LYC         Γ3Γ3 
60.0ARK         Γ3  
56.3UPP    (B)    Γ3  
55.9LEN         Γ3  
51.0MOS    Γ2       
50.8JUL AB(B)     A  
48.4SLO  Γ2  Γ2  Γ3   
42.3POI   (B)        
35.7ROM   A     A  
 
Asian/Australian Longitude Zone
55.5YAK   Γ2     Γ3Γ3 
53.2MAG - Γ2Γ2   -AA 
46.8IRK     A      
−28.7TOW AA         
−36.8BRI         Γ1  
−44.5MUN  A     Γ2   
−45.8CAN  -A    -A  
−54.2HOB   Γ3    Γ3   
−58.4KER  Γ3     -   
−79.9SCO  Γ2Γ3        
Table 3b. Same as Table 3a but Commencing During the Day
IGRF CML, deg ϕmStationMonth
JanFebMarAprMayJunJulAugSepOctNovDec
American Longitude Zone
69.5CHU         Γ2  
56.9OTT  B*Γ3A   Γ3   
55.2STJ  Γ3         
49.5WAL   Γ3Γ2   B*   
49.1BOU  AB*    Γ2   
40.5PNT   Γ1        
21.3MAU   Γ3AΓ3 Γ3    
−37.3POR   Γ3        
−49.8ARI  Γ3      Γ3  
 
European Longitude Zone
61.3LYC   B*        
60.0ARK  Γ3Γ3        
56.3UPP   Γ3B*(B)      
55.9LEN  Γ3Γ3Γ3   Γ3Γ3  
51.0MOS  AΓ3A   A   
50.8JUL   Γ3        
48.4SLO  Γ3Γ3A  Γ3-   
45.4LAN  AΓ3    A   
42.3POI B*       Γ2  
35.7ROM   A        
−30.9LAR A        B* 
−41.2GRW A        B* 
 
Asian/Australian Longitude Zone
55.5YAK  AΓ3    Γ3   
53.2MAG  Γ3Γ3    Γ3B  
46.8IRK    Γ2       
17.5TAI Γ3     Γ3  Γ3 
−11.1VAN    Γ3   Γ3   
−22.2DARΓ3Γ3          
−28.7TOW         A  
−35.9NOR A       A  
−36.8BRI A          
−44.5MUN B*A      Γ3Γ3 
−45.8CAN         AΓ3 
−54.2HOB  AΓ3    -   
−58.4KER         B*  
−60.2CLL  B-        
Table 3c. Same as Table 3a but Commencing During Sunset
IGRF CML, deg ϕmStationMonth
JanFebMarAprMayJunJulAugSepOctNovDec
American Longitude Zone
62.6GOO         Γ2  
55.2STJ  - Γ2       
−49.8ARI  Γ2B*        
−61.5HAL Γ1          
 
European Longitude Zone
56.3UPP   Γ2   Γ1Γ2   
55.9LEN   Γ2 A Γ1    
51.0MOS   B*Γ1 Γ1     
50.8JUL - Γ2Γ2Γ1Γ1 Γ2   
48.4SLO Γ2 Γ2Γ2Γ1Γ1 Γ2-  
45.4LAN Γ2       -  
42.3POI        Γ1   
 
Asian/Australian Longitude Zone
55.5YAK   Γ1Γ2Γ1  -   
53.2MAG   Γ2   Γ2    
−54.2HOB  Γ1         
−60.2CLL  BΓ2        
−79.9SCO  Γ2 A       
Table 3d. Same as Table 3a but Commencing During the Night
IGRF CML, deg ϕmStationMonth
JanFebMarAprMayJunJulAugSepOctNovDec
American Longitude Zone
69.5CHU A          
56.9OTT -B*Γ2B   BΓ1  
55.2STJ  B*     Γ2   
49.5WAL Γ2ABBΓ2Γ2Γ2Γ2   
49.1BOU AΓ1ABA B*AΓ1Γ1 
40.5PNT   AΓ3 Γ1AA   
21.3MAU    Γ3       
−37.3POR         AB 
−49.8ARI Γ2Γ2B      Γ2 
 
European Longitude Zone
63.6SOD           Γ2
56.3UPP         Γ2  
55.9LEN        Γ2Γ3  
51.0MOS  Γ2Γ2    Γ2 Γ3 
50.8JUL B* Γ3    Γ2   
48.4SLO BA(B)Γ2   Γ2(B)A 
45.4LAN AABΓ2   B B 
42.3POI B*AAB   Γ1B  
−35.6JOH Γ3          
−41.2GRW Γ1          
 
Asian/Australian Longitude Zone
55.5YAK  Γ2Γ3    Γ2Γ3  
53.2MAG AAΓ3    Γ2Γ2  
46.8IRK   A    (B)   
38.1WAK    A   Γ2Γ1  
−28.7TOW          A 
−35.9NOR B A      A 
−44.5MUN B* Γ1     Γ1A 
−45.8CANΓ2BAΓ1Γ2   AAA 
−54.2HOB Γ2BB   B Γ1Γ2Γ2
Table 4a. Distribution of Mean Absolute Errors Between All Observed Long-Duration Negative foF2 Disturbances Commencing During Sunrise and the Respective Patterns of Table 3aa
IGRF CML, deg ϕmStationMonth
JanFebMarAprMayJunJulAugSepOctNovDec
  • a

    From top to bottom: stations of the American, European, and Asian/Australian longitude zone. Patterns A, B, and Γ refer to Figure 1. A hyphen means not adjusted to the patterns of Figure 1.

American Longitude Zone
56.9OTT        .04   
49.5WAL  .10.02    .050  
49.1BOU - .03    0.07  
40.5PNT   -.01  .06    
21.3MAU       .10    
−49.8ARI .040       .04.02
−61.5HAL .04          
 
European Longitude Zone
63.6SOD         .05.05 
61.3LYC         .04.05 
60.0ARK         .03  
56.3UPP    .05    .04  
55.9LEN         .04  
51.0MOS    .08       
50.8JUL .03.08.15     .08  
48.4SLO  .07  .02  .08   
42.3POI   .07        
35.7ROM   .01     .02  
 
Asian/Australian Longitude Zone
55.5YAK   .06     .02.04 
53.2MAG - .05.03   -.05.07 
46.8IRK     .01      
−28.7TOW 0.03         
−36.8BRI         .03  
−44.5MUN  .03     .04   
−45.8CAN  -0    -.02  
−54.2HOB   .03    .03   
−58.4KER  .02     -   
−79.9SCO  .09.05        
Table 4b. Same as Table 4a But Commencing During the Day With Respect to the Patterns of Table 3b
IGRF CML, deg ϕmStationMonth
JanFebMarAprMayJunJulAugSepOctNovDec
American Longitude Zone
69.5CHU         .07  
56.9OTT  .05.02.03   .03   
55.2STJ  .03         
49.5WAL   .02.02   .03   
49.1BOU  0.02    .06   
40.5PNT   .08        
21.3MAU   .02.020 .10    
−37.3POR   .01        
−49.8ARI  .07      .03  
 
European Longitude Zone
61.3LYC   .02        
60.0ARK  .04.02        
56.3UPP   .01.03.05      
55.9LEN  .05.01.01   .03.02  
51.0MOS  .07.02.02   .03   
50.8JUL   .07        
48.4SLO  .03.02.03  .05-   
45.4LAN  0.02    .02   
42.3POI .03       .04  
35.7ROM   .02        
−30.9LAR .04        .01 
−41.2GRW .03        .02 
 
Asian/Australian Longitude Zone
55.5YAK  .04.02    .05   
53.2MAG  0.02    .03.07  
46.8IRK    .06       
17.5TAI .04     .07  .03 
−11.1VAN    .03   .04   
−22.2DAR.030          
−28.7TOW         .01  
−35.9NOR 0       .03  
−36.8BRI 0          
−44.5MUN .02.03      .03.01 
−45.8CAN         0.05 
−54.2HOB  .04.05    -   
−58.4KER         .04  
−60.2CLL  .05-        
Table 4c. Same as Table 4a But Commencing During Sunset With Respect to the Patterns of Table 3c
IGRF CML, deg ϕmStationMonth
JanFebMarAprMayJunJulAugSepOctNovDec
American Longitude Zone
62.6GOO         .08  
55.2STJ  - .04       
−49.8ARI  .03.06        
−61.5HAL .08          
 
European Longitude Zone
56.3UPP   .03   .08.07   
55.9LEN   .05 0 .09    
51.0MOS   .05.08 .05     
50.8JUL - .04.03.04.05 .07   
48.4SLO .06 .060.05.05 .05-  
45.4LAN .07       -  
42.3POI        .07   
 
Asian/Australian Longitude Zone
55.5YAK   .09.03.06  -   
53.2MAG   .03   .03    
−54.2HOB  .07         
−60.2CLL  .06.05        
−79.9SCO  .07 .05       
Table 4d. Same as Table 4a But Commencing During the Night With Respect to the Patterns of Table 3d
IGRF CML, deg ϕmStationMonth
JanFebMarAprMayJunJulAugSepOctNovDec
American Longitude Zone
69.5CHU .04          
56.9OTT -.05.03.02   .05.07  
55.2STJ  .03     .02   
49.5WAL .06.04.06.05.02.04.03.05   
49.1BOU .08.08.02.04.09 .02.03.08.06 
40.5PNT   .02.03 .07.02.04   
21.3MAU    .07       
−37.3POR         .02.07 
−49.8ARI .04.05.07      .04 
 
European Longitude Zone
63.6SOD           .07
56.3UPP         .05  
55.9LEN        .05.06  
51.0MOS  .04.05    .04 .05 
50.8JUL .10 .06    .06   
48.4SLO .07.04.07.07   .05.04.02 
45.4LAN .04.03.06.05   .04 .05 
42.3POI .04.03.02.05   .05.03  
−35.6JOH .04          
−41.2GRW .06          
 
Asian/Australian Longitude Zone
55.5YAK  .06.03    .03.04  
53.2MAG .05.04.03    .03.08  
46.8IRK   .02    .03   
38.1WAK    .03   .02.05  
−28.7TOW          .03 
−35.9NOR .10 .06      .02 
−44.5MUN .04 .08     .06.02 
−45.8CAN.08.04.05.05.04   .03.02.04 
−54.2HOB .04.05.06   .05 .04.05.02

[17] Before stepping into the analysis of the space/time disturbance pattern distribution of Tables 3a–3d it should be mentioned that (B) in Tables 3a–3d marks a shorter step function, for more than 3 hours, compared to the B pattern (Figure 1). Furthermore, as seen in Tables 3a–3d, an overall equinoctial bias of occurrence of large-scale negative foF2 disturbances is evident, possibly related to the known “Russell-McPherron effect” due to the IMF orientation with respect to the geomagnetic field [Russell and McPherron, 1973].

[18] Tables 3a and 3b show a quite similar disturbance pattern distribution for disturbances commencing at sunrise and day LT hours respectively. For corrected magnetic latitudes – CML greater than 54° (44°) the Γ3 pattern is dominant in autumn (spring) months when disturbances break out during sunrise (the day). Below these CML boundaries dominant seem the A pattern during sunrise hours in the midlatitudes and the Γ3 pattern during daytime hours, especially closer to the equatorial zone (CML < 22.5°). The overall strong Γ3 pattern appearance during the day means that its single peak is rather retarded (for about 12 hours) and appears at late night or presunrise hours.

[19] The distribution of disturbance patterns breaking out at sunset is pretty characteristic: almost only the Γ1 and Γ2 patterns may be present. For CML between 45° and 65°, the Γ2 patterns seem to prevail at the equinox months close to summer while in the summer months Γ1 patterns appear more often meaning that moving to summer the second maxima of the disturbance becomes much less intense. On the other hand, the disturbance patterns' distribution commencing at night is much more complicated. The Γ2 pattern disturbances may be observed only at locations for which CML is greater than about 49° from May (November) to September (February) for the Northern (Southern) hemisphere. Overall from its space/time distribution, the Γ2 pattern may be characterized as the “equinox near local summer” pattern. Finally, Table 5 summarizes the above findings on the dominant disturbance patterns at specific latitude zones and seasons, stressing also longitude zones and specific stations which form some exception, presenting thus a more local character.

Table 5. Summary of the Dominant Disturbance Patterns in Different CML Zones and Months
 IGRF CML, deg ϕmMonthsPatternSuccessZone/Stations of Exceptiona
  • a

    Reports theaters of trend exception.

Sunrise22.5°–54°allA15/24Europe
 >54°Sep–Nov: NH, Mar–May: SHΓ311/13America
Day<22.5°allΓ310/11-
 >44°Mar–Apr: NH, Sep–Oct: SHΓ318/27-
Sunset45°–65°Jun–Aug: NH, Dec, Jan–Feb: SHΓ19/11-
 45°–65°Feb–May, Sep–Nov: NH, Aug–Nov, Mar–May: SHΓ218/25-
Night>49°May–Sep: NH, Nov–Feb: SHΓ216/22Boulder (49.1°ϕm)

4. Discussion

[20] It is hereby of practical interest to identify the specific merits and drawbacks of different foF2 storm time empirical models. Thus, along with the present product, the following discussion involves the recently incorporated in IRI STORM model of Araujo-Pradere et al. [2002] and the one of Kutiev and Muhtarov [2004]. The main difference between the present approach and the aforementioned models is that they are conditional on previous geomagnetic activity since they are driven by linear filters incorporating a weighted, accumulative effect of past ap (or Kp) values respectively. As a result they predict the ionospheric response to systematically increased geomagnetic activity. However, neither the aforementioned indices are fully representative for geomagnetic activity at all latitudes, nor they may be considered as the only causes for perturbed ionospheric conditions. Furthermore, ionosphere presents a regional character which is not accounted for in such models. On the other hand, it should be stressed that the main advantage of these models is that these geomagnetic drivers (storm indices) may be quite predicted. However, for example the STORM model in its development used the final validated ap index and not the available real-time index, but this is surely only a small source of error. The approach presented in this paper focuses solely on the morphology of long-duration negative foF2 disturbances and, while not currently linked with a storm index, still directly provides radio users with operational disturbance patterns and their distribution in time and space, depending on season and local time of commencement; thus they may act accordingly.

[21] The STORM model exhibits a 33% improvement over a monthly median model, capturing more than half of the increase in variability on storm conditions, being a significant advance over climatology [Araujo-Pradere and Fuller-Rowell, 2002]. However, smaller ionospheric variations of the order of ±10–15%, though a likely ionospheric response to increased geomagnetic activity levels, are not reproduced since they are not considered linked with the geomagnetic activity history [Araujo-Pradere and Fuller-Rowell, 2002; Kutiev and Muhtarov, 2004]. This outcome is also due to the aforementioned conditional modeling. On the contrary, the approach hereby presented reveals patterns of ionospheric foF2 storms defined only by their power, while in order to cope with smaller day-to-day variations, Fotiadis and Kouris [2006] have introduced a foF2 variability model with latitude, bearing a mean error of about ±0.04, which may likely act as complementary to any foF2 storm model.

[22] Regarding the prediction success for different seasons and latitudes, the STORM model may not apply for the equatorial latitudes, that is, within ±20° of the magnetic equator, and seems not so successful in predicting ionospheric response in winter and near equinox months [Araujo-Pradere and Fuller-Rowell, 2002] where positive disturbances prevail. On the other hand, the proposed model identifies dominant storm patterns at certain latitudes depending on their frequency of occurrence. For example, since long-duration negative foF2 disturbances are observed in equatorial latitudes with frequency of less than 5 in 33 years, the present approach does not deal with them. Moreover, disturbances of positive phase are not a subject of investigation here but their patterns could be calculated in a similar future analysis.

[23] Finally, it is of special interest to compare the feature-capture capability between STORM and the proposed model. A closer look at the STORM validation paper [Araujo-Pradere and Fuller-Rowell, 2002] shows that there is a delay of simulated foF2 data up to 6–12 hours to the observed ones at the main phase of the storm, toward reaching the maximum storm depth. Furthermore, at certain storm times there seems to be a difference of more than 20% at the maximum foF2 deviation from the median between the simulated and the actually observed values. Sudden and deep phase transitions are not captured by STORM as well as recurrent type foF2 disturbances. This may be the result of a filter, built only for midlatitudes, producing a quite smooth response while the dynamic range of foF2 variations is large.

[24] Because of the lack of a full validation test for the proposed model, an example of application is shown in Figure 2. First, two foF2 disturbances have been identified [Kouris et al. 1998, 1999] for a few stations during September and October 1998, a period other than the years of the modeled patterns (1964–1995). Then at each station, the disturbances are aligned to the proposed patterns of Figure 1 depending on the month and the LT window of storm commencement as reported in Tables 3a–3d. The agreement between observed and modeled data is quite evident for all three stations for the storm of 19 October 1998 (Figure 2). On the other hand the proposed dfoF2 envelopes for the September 1998 storm are successful at the European stations hence not at the American zone station (Boulder, Colorado). However, even in this miss case, it seems that the Γ2 pattern could also fit the Boulder case and possibly a mechanism keeping track of the real-time ionospheric changes could make a correction to the best pattern in future model formulation.

Figure 2.

Range of dfoF2 variation of observed storm patterns (circles) and the respective adjusted functions (curves) according to the distribution Tables 3a–3d for the storms of 24 September 1998 and 19 October 1998 (marked with diamonds) for three ionospheric stations. The y axis shows deviation from the foF2 monthly median (dfoF2), and the x axis shows local storm time hours after Kouris et al. [1998, 1999]; below zero local storm time, the UT of the ionospheric disturbance commencement is shown.

[25] Moreover, one might argue in favor of the ability of the present model to capture the features mentioned above since the study is delimited to short time windows of about 54 hours. The purpose behind the selection of a time window of this size was to capture systematic trends such as primary and secondary maxima and different slopes in storm recovery phase. Therefore the key factor in studying disturbances in a greater time window should be the systematic observation of certain features. For example, in Table 6 the hours of delay of an A pattern storm after a positive ionospheric sudden storm commencement are reported. It is evident that in most cases this effect is absent, compared to randomly selected stations of Tables 3a–3d, and, when present, the hours of delay may vary significantly from 4 to 6 hours up to 14 hours. That suggests that this storm feature should not be considered as prominent for the A pattern long-duration negative foF2 disturbances.

Table 6. Hours of Delay of A Pattern Long-Duration Negative foF2 Storms After a Positive Ionospheric Sudden Storm Commencement (ISSC) Featurea
IGRF CML, deg ϕmStationMonth
FebMarAprMayJunJulAugSepOctNov
  • a

    A hyphen marks absence of ISSC.

Sunrise Commencement
53.2MAG        3-
49.1BOU  -    24 
−44.5MUN 10        
 
Day Commencement
51.0MOS 4 -   -  
−44.5MUN -        
−54.2HOB -        
 
Night Commencement
53.2MAG--        
49.5WAL -        
49.1BOU- 14 -  6  
48.4SLO 9       6
45.4LAN-14        
42.3POI --       
−45.8CAN 5     --5

[26] On the contrary, let us now examine the ionospheric behavior before and after a long-duration negative foF2 disturbance with onset at sunset hours. Tables 7a and 7b report, in accordance to some stations of Tables 3a–3d, the storm times with foF2 deviations less than 20% off the monthly median before and after a disturbance at sunset. It is evident that in all cases such decreases in foF2 are detected both before (about 19 hours) and after (about 53 hours) the onset of a sunset negative storm. Therefore such systematic trends may be considered as associated with large-scale negative storms commencing at sunset.

Table 7a. Local Storm Time With Variability Being Less Than 20% Off the foF2 Monthly Median, Before the Commencement of Long-Duration Negative Disturbances at Sunset Hoursa
IGRF CML, deg ϕmStationMonth
MarAprMayJunJulAugSep
56.3UPP      −19
55.5YAK −19−20    
53.2MAG −14/−21     
51.0MOS  −13    
50.8JUL  −19    
−60.2CLL−19      
Table 7b. Same as Table 7a But After the Commencement of Long-Duration Negative Disturbances at Sunset Hoursa
IGRF CML, deg ϕmStationMonth
MarAprMayJunJulAugSep
56.3UPP      49/55
55.5YAK 5251    
53.2MAG 50/55     
51.0MOS  53    
50.8JUL  52    
−60.2CLL53      

[27] In summary, the calculated disturbance patterns hereby reported offer a prediction on the likely further development of an established ionospheric storm [Kouris et al., 1998, 1999], usually a product of a previous enhanced geomagnetic activity. A future connection of specific storm patterns with a suitable storm index would enhance the operational power and application capability of the present model which at its present form is still a reliable empirical, nonconditional stand-alone disturbance model for radio users.

5. Conclusions

[28] The main goal of this paper is to reveal the morphology of large-scale negative foF2 disturbances. Special attention is paid to exclude nonsystematic trends amid sampling effects as well as prestorm and poststorm phenomena not linked with specific disturbance patterns. The former is achieved by setting a minimum occurrence probability requirement and the latter by adapting the local-time window under investigation. Each disturbance pattern is presented by a variation envelope, both from observed data and from fitted functions, thus allowing radio users for fine tuning adjustments. The distribution of the disturbance patterns found is provided for each month and station separately depending on the local time of storm commencement. The present model provides radio users with the most probable expected disturbance patterns in the event of a negative ionospheric disturbance at a certain location, after enhanced geomagnetic activity.

Acknowledgments

[29] The authors are grateful to the editor, associate editor, and referees for their invaluable comments in improving this paper.