comparison of gps-tec measurements with nequick2 and …

Available online at www.sciencedirect.com

www.elsevier.com/locate/asr

ScienceDirect

Advances in Space Research 61 (2018) 1456–1475

Comparison of GPS-TEC measurements with NeQuick2 and IRImodel predictions in the low latitude East African region during

varying solar activity period (1998 and 2008–2015)

E. Mengistu a,⇑, B. Damtie b, M.B. Moldwin c, M. Nigussie b

aEthiopian Space Science and Technology Institute, Addis Ababa, EthiopiabWashera Geospace and Radar Science Laboratory, Bahir Dar University, Bahir Dar, Ethiopia

cDepartment of Climate and Space Sciences and Engineering, University of Michigan, Ann Arbor, MI, USA

Received 27 August 2017; received in revised form 5 January 2018; accepted 6 January 2018Available online 16 January 2018

Abstract

This paper examines the performances of NeQuick2, the latest available IRI-2016, IRI-2012 and IRI-2007 models in describing themonthly and seasonal mean total electron content (TEC) over the East African region. This is to gain insight into the success of thevarious model types and versions at characterizing the ionosphere within the equatorial ionization anomaly. TEC derived from five Glo-bal Positioning System (GPS) receivers installed at Addis Ababa (ADD, 5.33�N, 111.99�E Geog.), Asab (ASAB, 8.67�N, 116.44�EGeog.), Ambo (ABOO, 5.43�N, 111.05�E Geog.), Nairobi (RCMN, �4.48�N, 108.46�E Geog.) and Nazret (NAZR, 4.78�N, 112.43�E Geog.), are compared with the corresponding values computed using those models during varying solar activity period (1998 and2008–2015). We found that different models describe the equatorial and anomaly region ionosphere best depending on solar cycle, seasonand geomagnetic activity levels. Our results show that IRI-2016 is the best model (compared to others in terms of discrepancy range) inestimating the monthly mean GPS-TEC at NAZR, ADD and RCMN stations except at ADD during 2008 and 2012. It is also found thatIRI-2012 is the best model in estimating the monthly mean TEC at ABOO station in 2014. IRI show better agreement with observationsduring June solstice for all the years studied at ADD except in 2012 where NeQuick2 better performs. At NAZR, NeQuick2 better per-forms in estimating seasonal mean GPS-TEC during 2011, while IRI models are best during 2008–2009. Both NeQuick2 and IRI modelsunderestimate measured TEC for all the seasons at ADD in 2010 but overestimate at NAZR in 2009 and RCMN in 2008. The periodicvariations of experimental and modeled TEC have been compared with solar and geomagnetic indices at ABOO and ASAB in 2014 andresults indicate that the F10.7 and sunspot number as indices of solar activity seriously affects the TEC variations with periods of 16–32days followed by the geomagnetic activity on shorter timescales (roughly periods of less than 16 days). In this case, NeQuick2 derivedTEC shows better agreement with a long term period variations of GPS-TEC, while IRI-2016 and IRI-2007 show better agreement withobservations during short term periodic variations. This indicates that the dependence of NeQuick2 derived TEC on F10.7 is seasonal.Hence, we suggest that representation of geomagnetic activity indices is required for better performance over the low latitude region.� 2018 COSPAR. Published by Elsevier Ltd. All rights reserved.

Keywords: Ionosphere; Total electron content (TEC); GPS-TEC; NeQuick2; IRI model; Solar and geomagnetic activity

https://doi.org/10.1016/j.asr.2018.01.009

0273-1177/� 2018 COSPAR. Published by Elsevier Ltd. All rights reserved.

⇑ Corresponding author.E-mail addresses: [email protected] (E. Mengistu), bayliedam-

[email protected] (B. Damtie), [email protected] (M.B. Moldwin),[email protected] (M. Nigussie).

1. Introduction

The equatorial and low latitude regions of the iono-sphere have unique characteristics compared to mid andhigh latitudes due to the horizontal geomagnetic field lines


mailto:[email protected]






http://crossmark.crossref.org/dialog/?doi=10.1016/j.asr.2018.01.009&domain=pdf

E. Mengistu et al. / Advances in Space Research 61 (2018) 1456–1475 1457

and absorption of a larger fraction of incident solar energy(de Abreu et al., 2017 and references therein). The well-known equatorial ionosphere phenomena are the equato-rial ionization anomaly (EIA) or Appleton Anomaly,Equatorial Electrojet (EEJ) and equatorial spread-F(ESF) irregularities. Such processes are controlled by theneutral winds of the upper atmosphere which interact withthe conductive and magnetic regions of the ionosphere andproduced by dynamo effect, electric fields in the iono-spheric E and F-regions (de Abreu et al., 2017). Thedynamo electric fields are generated in the equatorial E-region by thermospheric winds. In the F-region, abovethe magnetic equator, the eastward electric field in combi-nation with the north-south magnetic field causes the iono-sphere plasma to drift upward with a vertical drift, E� B.The plasma lifted in this way then flow down the magneticfield lines to higher-latitude regions, forming two symmet-rical enhancements on each side of the equator through thecombined action of electrodynamic uplift, pressure gradi-ents (rp), and gravity (g) (Kelly, 2009). These joint effectsproduce a fountain like pattern of plasma motion called theequatorial fountain (Shunk and Nagy, 2009). The plasmathat slips down along the magnetic field lines will thenaccumulate and form large plasma densities at latitudesaround �15� to �20� of the dip equator. The resulting lat-itudinal structure in the ionization distribution, character-ized by the two low latitude ionization maximum densitywith a minimum centered at the dip equator is known asthe EIA (Abdu, 2005).

The total electron content (TEC) is an important param-eter widely used to study ionospheric characteristics. It isdefined as the total number of electrons present within anarea of one square meter cross section along the integratedpath from satellite to the receiver and is measured by TECu

(1TECu ¼ 1016 electrons=m2). Detailed knowledge of TECdistributions and variations is required to improve theoperation of space-based Earth observation, communica-tion and navigation systems especially for single frequencyGlobal Positioning Satellite (GPS) signal receivers. To thisend, a number of regional networks of GPS stations havebeen established, to provide near real-time forecasting ofionospheric TEC variations. Such a network has not beenbuilt over Africa, which hinders our ability to obtain a glo-bal understanding of the dynamics and structure of theionosphere (Akala et al., 2013; Yizengaw et al., 2012). Onthe other hand, the global distribution of TEC is possiblewith the ionospheric model data. Nowadays, several empir-ical models have been suggested which can estimate theTEC along with many other ionospheric parameters espe-cially in regions with scarcity of deployed ionospheric sens-ing instruments like Africa. The International ReferenceIonosphere (IRI) model is an international standard empir-ical model for the terrestrial ionosphere which is mostwidely used by the scientific community to compute thevariations of electron density, TEC and other parametersof the ionosphere. The IRI model is continuously updated

as new data become available, and in this study the latestIRI-2016 (Bilitza et al., 2017) model is used.

Since the IRI model mainly takes the data from the mid-latitude regions, therefore, estimation of TEC by the IRImodel in midlatitude is worthy compared to other latitudes(Kumar, 2016). The accuracy of the IRI model over equa-torial and low-latitude regions is important for its practicalapplication in satellite based communication and naviga-tion systems. The underestimation/overestimations in theIRI model TEC in equatorial and low-latitude regions havebeen highlighted in many studies (Kumar, 2016; Tariku,2015a,b; Chakraborty et al., 2014). For example, recently,Kumar (2016) investigated the performance of the IRI-2012 model during a deep solar minimum (2009) and amaximum year (2012) over the global equatorial regionand reported that the monthly and seasonal mean valueof the IRI-2012 model overestimates the observed GPS-TEC at all equatorial stations examined. The discrepancy(or over estimation) in the IRI-2012 model is found largerduring solar maximum (2012) than during solar minimum(2009). The discrepancy is maximum during the Decembersolstice and minimum during the March equinox. He alsoreported that the significant discrepancy in the IRI-2012model observed during the solar minimum year (2009)could be attributed to larger difference between F10.7 fluxand EUV flux (26–34 nm) during low solar activity period2007–2009 as compared to high solar activity period 2010–2012. The performance of IRI-2012 for estimating the ver-tical TEC variation over Ethiopian regions during the year2009–2011 has been carried out by Tariku (2015a). Accord-ing to his investigation, the modeled monthly and seasonalvTEC values are larger than the corresponding measuredvalues during the period of 2009–2010 when all modeloptions for the topside electron density are used. He alsonoted that the overestimation of vTEC values derived fromthe model decreases as the Sun transitions from very low tohigh solar activity. Recently there have been several otherstudies of TEC prediction performance of the IRI-2012model in other regions. For example, de Abreu et al.(2017) studied comparison of GPS-TEC measurementswith IRI-2012-TEC predictions in the Brazilian sector atPalmas (PAL) and Sao Jose dos Campos (SJC) duringsolar minimum (2009) and reported that the IRI-TECmodel gives the better estimates of the GPS-TEC duringthe nighttime from January to December months overthe studied region. They also reported that the modelTEC shows larger differences from the GPS-TEC in theafternoon hours and these discrepancies for all monthsare possibly due to the dynamics of EIA in consequenceof the equatorial fountain effect. Another comparativestudy on the morphological variations of the GPS-TECand its comparison with The IRI-2012 modeled TEC overBrazilian region during the periods of the years 2010–2013was studied by Venkatesh et al. (2014) which revealed thatthe performances of the model are better during low solaractivity periods compared to that during the high solaractivity years. Li et al. (2016) studied variation characteris-

1458 E. Mengistu et al. / Advances in Space Research 61 (2018) 1456–1475

tics of TEC and performance of the IRI-2012 model in pre-dicting TEC at the BJFS (Beijing Fang Shan station),China, (2009 and 2013). They reported that the IRI-2012model can reflect the climatic characteristics and solaractivity dependence of ionospheric TEC. They alsoreported that (i) the IRI-2012 model underestimated night-time TEC over both years, (ii) the periodic components ofTEC-IRI in low solar activity year (2009) exhibit goodagreement with that of TEC derived from Global Iono-sphere Map (TEC-GIM), and (iii) in high solar activityyear (2013), the differences between TEC-GIM and TEC-IRI are mainly due to periodic components and solaractivity.

Nigussie et al. (2013) studied the comparison of the IRI-2007 model with those from NeQuick2 over East Africanequatorial region and reported that the IRI-2007 modelgenerally overestimates the observed vTEC during thesolar minimum year. With respect to solar activity, theyreported that the performance of the model is better duringrising phase of solar activity period (1998) than lower solaractivity period (2007–2009). Using the data from ground-based GPS measurements during 2009–2011 over Kenyaregions (East Africa), Olwendo et al. (2012) noticed thatseasonal average of TEC estimated from the IRI-2007model are significantly higher than that GPS during theperiod. Furthermore, for solar activity phase 2009–2010,Olwendo et al. (2013) also reported the IRI-2007 modeloverestimation for all the seasons except during the Marchequinox.

The NeQuick2 model is a time dependent three-dimensional ionospheric electron density model now avail-able in its version 2 (Nava et al., 2008). By integrating theelectron density, NeQuick2 can also be used to obtain slantor vertical TEC. Using three months of data during lowsolar activity period (2007), Nigussie et al. (2012) testedthe validity of the model in the East African region fromGPS receivers located in the region (Nazret, Asab, BahirDar, Robe and Arba Minch stations) and reported thatthe performances of the model have increased substantiallywhen it is assisted by measurements than its standardusage. In other words, the ability of the model to reproducethe experimental TEC increases notably when the model isadapted by data ingestion from one station in comparisonwith the case when the model was used in a standard waybeing driven only by the daily solar flux. They also per-formed the validation of NeQuick2 TEC data ingestiontechnique against C/NOFS and EISCAT electron densitymeasurements during different days of the year in 2010(over NAZR, East Africa, and Tromso, Norway). Accord-ing to their report, the performance of the model afteradaptation shows considerable improvement in estimatingthe experimental data at both stations (Nigussie et al.,2016). Furthermore, by using ionosphere data collectedfrom four ground-based GPS receivers stationed in Ethio-pia and Nigeria and one Digisond stationed in Nigeria(during 1998, 2007, 2008 and 2009), Nigussie et al. (2013)tested the performances of NeQuick2 and IRI-2007 models

in describing the monthly median characteristics of theequatorial region ionosphere and reported that the perfor-mances of both models are better during a rising phase ofsolar activity period (1998) than low solar activity period(2007–2009). According to their finding, both models over-estimate the observed vTEC during low solar activity per-iod. They also reported that the modeled and experimentalNmF2 has shown good agreement; but the modeled andexperimental vTEC shows significant discrepancy mainlydue to inadequately computed ionospheric slab thicknessesusing IRI-2007 and NeQuick2. Moreover, by usingDigisondes, Ionosonde and GNSS data from stations inthe Republic of South Africa and sub-Saharan Africa,Elvidge (2014) presented the test scenario of the three maintypes of ionospheric models. According to their report, thedata assimilation models (modified NeQuick (mNeQuick),GPS Ionospheric Inversion (GPSII) and Electron DensityAssimilative Model (EDAM)) perform best followed bythe empirical model (IRI-2007) and then the physics basedmodels (Global Ionosphere Thermosphere Model (GITM)and Thermosphere Ionosphere Exosphere-General Circu-lation Model (TIE-GCM)).

However, to our knowledge, there is no simultaneouscomparative testing of GPS-TEC with NeQuick2, IRI-2016, IRI-2012 and IRI-2007 over the low latitude EastAfrican region. Furthermore, to our knowledge, the valida-tion of the IRI-2016 model has not been done over the EastAfrican region. Hence, the objective of the present study isto carry out the first extended comparative study of GPS-vTEC measurements with the NeQuick2, IRI-2016, IRI-2012, and IRI-2007 models prediction at the low latitudeEast African region. We study the vTEC prediction capa-bility of NeQuick2, IRI-2016, IRI-2012 and IRI-2007 mod-els at the low latitude East African region. This is done bycomparing the diurnal monthly and seasonal mean varia-tion of vTEC calculated from those models with the corre-sponding observations obtained from five ground-basedGPS receivers located in the region. Wavelet analysis isused to compare the periodic characteristics of daily meanvTEC from experimental and modeled outputs at Amboand Asab stations during 2014. For further comparison,we examined the wavelet coherence between the daily aver-aged variations of GPS and model derived vTEC at the twostations.

2. Data and method of analysis

The GPS-vTEC data observed at Addis Ababa station(hereafter referred to as ADD) during 1998 (January–August), 2008 (January–December), 2009 (January–Mayand December), 2010–2013 (January–December), 2014(January–October), 2015 (January–December), Asab sta-tion (hereafter referred to as ASAB) during 2014 (Jan-uary–December), Ambo station (hereafter referred to asASAB) during 2014 (January–December), Nairobi station(hereafter referred to as RCMN) during 2008 (January–December) and Nazret station (hereafter referred to as


NAZR) during 2008–2009 (January–December) and 2011(January–December) have been used for this study depend-ing on the availability of the GPS-TEC data. All the fiveGPS stations with their geographic and geomagnetic coor-dinates are listed in Table 1.

The GPS data at these stations were retrieved in stan-dard Receiver Independent Exchange (RINEX) formatand then processed by the GPS-TEC RINEX files process-ing software (version: 2.9.3) developed by Seemala (2014).To process the RINEX observation files, the program alsoneeds RINEX navigation files (obtained from website ftp://cddis.gsfc.nasa.gov/pub/gps/data/daily) for the observa-tion date to calculate elevation and azimuth angles of thesatellites (which are required for vertical TEC calculation).The differential code bias (DCB) files provided by the IGScode website (ftp://ftp.unibe.ch/aiub/CODE/) for satellitebiases are also required. The program uses them if theyare available, otherwise it calculates them. The programfirst calculates the slant TEC (sTEC) and then convertsthe sTEC to vTEC by assuming that the ionosphere is athin shell at an altitude of 350 km, via a mapping functiongiven by Mannucci et al. (1993) and Ma and Maruyama(2003). To eliminate multipath effects, different researchershave used observation data of above certain cutoff maskranging from 15� to 35� (Chu et al., 2005; Mushini et al.,2011; Meggs et al., 2006). Consequently, in this study, weused the averaged vTEC calculated by the software fromindividual PRNs above 20� elevation angles. Although,there are 1440 data points for a given day, we take onlythe vTEC value at one hour interval (in decimated, i.e., justpull out 1 per hour). Accordingly, we obtained 24 datapoints for a given day. The processed vTEC data weregrouped into monthly and seasonal sets. All the measure-ments (vTEC) within the hourly bin of a given month wereaveraged together to provide a single data point for eachuniversal time hour. Consequently, 24 data points wereobtained for each month of a given year.

The corresponding modeled vTEC were calculated onhourly basis using NeQuick2, IRI-2016, IRI-2012, andIRI-2007 models using the daily solar flux (F10.7) as a dri-ver of all models. From these hourly values, the averagedvalue of vTEC is calculated on hourly basis for a givenmonth. In the IRI model, initially it estimates the F-region peak parameters using URSI (International Unionof Radio Science) or CCIR (Consultative Committee forInternational Radio) coefficients. Then, it uses different for-mulations to derive the bottom-side and top-side verticalelectron density profiles. Thus, derived vertical electron

Table 1Geographic and geomagnetic coordinates of the GPS stations used for this st

Station Code Geographic co-ordin

Addis Ababa ADD (9.04�N, 38.77�E)Asab ASAB (13.06�N, 42.65�E)Ambo ABOO (8.99�N, 37.81�E)Nairobi RCMN (�1.22�N, 36.89�E)Nazret NAZR (8.57�N, 39.29�E)

density profile is integrated with respect to altitude to com-pute the TEC value. In the present study, the CCIR andNeQuick2 options for F-peak model and for the top-sideprofile estimation have been considered, respectively, forall IRI models. Further, the newly added and the defaultABT-2009 option for the bottom side thickness shapeparameter is considered in both IRI-2016 and IRI-2012.But, for IRI-2007, the default ‘B0 Table’ has been used.Moreover, the storm related models were set to on duringstorm days for all IRI models while they were set off forother days. Finally, the required vTEC by each IRI modelis obtained by integrating the electron density profile froman altitude of 90 km to 2000 km (upper boundary for IRImodels). However, vTEC by NeQuick2 is obtained by inte-grating the electron density profile from an altitude of 90km to the satellite height (�20,200 km). Lastly, themonthly difference (DvTECÞ between experimental vTECoand model derived vTECm values was calculated by:

DvTEC ¼ vTECm � vTECo: ð1ÞFor the seasonal grouping of the data, we used the asso-

ciated three months of data for each season: March equi-nox (February, March, and April), June solstice (May,June, and July), September equinox (August, September,and October), and December solstice (November, Decem-ber, and January). To study the seasonal discrepancy inthe NeQuick2, IRI-2016, IRI-2012 and IRI-2007 models,the percentage discrepancy in the modeled vTEC as com-pared to GPS-vTEC (vTECo) is computed using

%Dev ¼ vTECo � vTECm

vTECo

� �� 100%; ð2Þ

where vTECm represents the NeQuick2, IRI-2016, IRI-2012and IRI-2007 models. Further, we used the continuouswavelet transform (CWT) to identify the periodicities inmeasured GPS, NeQuick2, IRI-2016, IRI-2012, and IRI-2007 modeled vTEC variations. The CWT is a powerfulmathematical tool for non-stationary signals and givesinformation about the behavior of intermittency and peri-odic phenomena present in these signals (Torrence andCompo, 1998). Basically, the CWT is given:

Wxðs; bÞ ¼ 1

s

ZxðtÞw� t � b

s

� �dt; ð3Þ

where xðtÞ is the time-frequency domain, s and b are thescale and time, respectively, and w� is the conjugate com-plex of the mother wavelet w. Therefore, it is very impor-tant to correlate two time series in order to verify if these

ud.

ate (Lat., Long.) Geomagnetic co-ordinate (Lat., Long.)

(5.33�N, 111.99�E)(8.67�N, 116.44�E)(5.43�N, 111.05�E)(�4.48�N, 108.46�E)(4.78�N, 112.43�E)


time series have similar frequencies. Thus, the methodologyis to apply the wavelet coherence phase between two timeseries data (de Abreu et al., 2017 and the reference therein).Moreover, to study the effect of solar and geomagneticactivity on both the experimental and modeled vTEC(especially on their periodicities), the CWT analysis hasbeen applied. Here, the wavelet analysis part is carriedout by using the daily averaged vTEC obtained from obser-vation and models at ABOO and ASAB stations during2014 (high solar activity period).

3. Results and discussions

3.1. Comparison of the diurnal variations of monthly meanobserved and modeled vTEC

Fig. 1(a) shows the typical diurnal variations of monthlyaveraged measured and model derived vTEC from three

Fig. 1. (a) Left panel: Diurnal variation of the monthly averaged experimentmedium), RCMN station during 2008 (solar minimum), and ABOO station durof monthly mean variations of experimental and modeled vTEC discrepancie(LTÞ ¼ UTþ longitude=15.

different stations, namely; ADD (1998: solar medium),RCMN (2008: solar minimum), and ABOO (2014: solarmaximum). Here, note that the term solar medium is theoperational definition given for solar activity in betweensolar minimum and maximum (i.e., it can be ascendingor descending phase of solar activity period). We have ana-lyzed the whole year available data from each station overthe East African region. We found that the vTEC variationat all stations show typical diurnal characteristics such asvTEC minimum at predawn and continuing increase withlocal sunrise attaining a maximum around midday (atabout 1200 UT) followed by a decrease to a minimum dur-ing nighttime. This is the expected phenomena because asthe sun rises the ionization also increases which causesmore concentration of electrons near the F2 peak in theionosphere. Since TEC is directly related with maximumelectron density, it starts to increase and attains maximumvalue at the local noon time. As the Sun moves past zenith

al and models vTEC at ADD station during January–August 1998 (solaring 2014 (solar maximum), (b) Right panel: the corresponding contour plots at the three stations from top to bottom, respectively. Here, local time


and then sets, loss rates overcomes production rates, lead-ing to gradually decreasing TEC.

However, close observation of diurnal variation invTEC shows that a clear discrepancy between the experi-mental and modeled vTEC is observed from month tomonth and station to station. For example, in 1998, allmodels underestimate the experimental vTEC at ADD sta-tion in the time span of 0000–0300 UT during January andFebruary. In 1998, the NeQuick2 and all IRI models over-estimate the experimental vTEC starting from 0300 UT(corresponding to sun rise hour) to about 0900 UT (corre-sponding to pre-noon hour) during January–March and0300–1000 UT during May–August except NeQuick2 dur-ing June (0300–0900 UT). Furthermore, this overestima-tion is observed during the time interval of 1600–2100UT in March 1500–2300 UT in April and July, and1200–2300 UT in May 1700–2300 UT in August. More-over, in June, the total discrepancy between the experimen-tal and all IRI models derived vTEC have been observedexcept at 0200 UT. A significant discrepancy among themodels themselves in June, July and August 1998 havebeen observed. For example, the discrepancy between theNeQuick2 and IRI models during June is very significantbetween about 0300–2300 UT. The NeQuick2 model showsbetter performance in estimating the experimental vTECduring June, while, all IRI models show better performancein the month of July. However, here, we noticed that allmodels show roughly an excellent monthly structure agree-ment in reproducing the experimental data over ADD sta-tion during January–May 1998. Starting from about 0800–0900 UT (corresponding to pre-noon hour) to 1500 UT(corresponding sunset hour), on the other hand, all theIRI and NeQuick2 models underestimate the observationduring the month of March. The NeQuick2 model alsoshowed an underestimation of vTEC, which is not preva-lent to all IRI models, during June in the time interval ofabout 1000 UT-1300 UT.

The mismodeling, which is calculated by using Eq. (1), isdisplayed in Fig. 1(b). Here, the color1 shows the size of thediurnal variation of the monthly averaged mismodelings.That is, deep yellow (overestimation) and deep blue (under-estimation) are being the worst agreement between theexperimental and modeled vTEC. Hence, Fig. 1(b) presentsthe exact difference in value between the measured andmodeled vTEC in all months of a given year at that partic-ular station. The DvTEC positive (negative) values indicatethat the models overestimates (underestimates) themeasured-vTEC. Keeping this in mind, a pronounced dis-crepancy (overestimation) between the experimental andsimulated vTEC is observed centered on 0600 UT (earlymorning) on the order of 20.93 TECu (by IRI-2007) atADD station during 1998. Such early morning variationsbetween the experimental and IRI-2007 modeled TEC val-

1 For interpretation of color in ‘Figs. 1 and 7, the reader is referred tothe web version of this article.

ues could be attributed to the fact that the shape of theelectron density profile is not well predicted by the IRImodel option (Akala et al., 2013; Ezquer et al., 1995). Gen-erally, high mismodelings of vTEC are observed in the timeintervals of 0500–0700 UT and 1300–1600 UT. These timeintervals include when double peaks of simulated vTEC areobserved (see the top panel of Fig. 1(a) and 1(b)). Theworst agreement (underestimation) of the models, as seenfrom the color bar, is observed in the order of ��9.96TECu (by IRI-2012) at ADD station during 1998. Gener-ally, our results show that the experimental vTEC showsa single peak whereas all the model derived vTEC showdouble peaks at ADD station in 1998. Such ionosphericstructure is also reported by Akala et al. (2013). Accordingto their report, the diurnal variations of the IRI-2007 withthe NeQuick topside option derived TEC show doublepeaks, while the corresponding GPS-TEC values show sin-gle peaks at both Lagos (Nigeria) and Pucallpa (Peru) sta-tions. However, ionospheric modeling studies havedemonstrated that TEC twin peaks or midday bite-outsat middle and lower latitudes can be created by a combinedeffect of E� B drift and altitude-dependent F region chem-ical loss processes (Pi et al., 1993). Modeling results alsoshow that considerable structuring in the local time varia-tion of the ionospheric ‘‘equatorial anomaly” can occurdue to the interplay of convection electric field penetrationand over shielding effects (Pi et al., 1993). The possiblecause of the midday bite-out ionospheric disturbances bythe meridional winds associated with traveling atmosphericdisturbances (TADs) is also addressed by Pi et al. (1993).The individual performance evaluations of those modelsresults show that they have different tendency of reproduc-ing measurements. For example, the NeQuick2 has shownminimum and maximum discrepancy with magnitude�8.90 TECu (in March) and 18.01 TECu (in May), respec-tively, at ADD station during the year 1998. Similarly, theobserved minimum and maximum discrepancy shown byIRI-2016, IRI-2012 and IRI-2007 models are (�8.46,17.93) TECu, (�9.96, 17.13) TECu and (�6.36, 20.93)TECu, respectively, at ADD station during 1998. Theseminimum and maximum discrepancies (by all models) havebeen observed during the month of March and May,respectively.

In 2008, at RCMN station, the experimental vTECshows a single peak and maximum value in the time inter-val of 1000–1400 UT. In contrast to ADD station (1998),the modeled diurnal variation of vTEC shows single peakcentered between 1100 and 1300 UT at RCMN station.In other words, all modeled vTEC shows similar patternwith the observation data. At the equatorial regions,Bilitza and Reinisch (2008) also explained that the steepestgradients, sharp peaks and deep valleys, and density crestsare on both sides of the equator (due to Fountain effect).As before, at RCMN station, the mismodeling calculatedby using Eq. (1) is shown in the Fig. 1(b). As seen from thisfigure, the worst overestimation and underestimation of theobservation data are observed in the time interval 0400–


2100 UT with order of magnitude 17.68 TECu in October(by NeQuick2) and �3.19 TECu in January (by IRI-2012),respectively. Similar investigation has been carried out atABOO station from January to December 2014 (high solaractivity period) and is displayed at the bottom panel ofFig. 1(a). It is clearly seen, from the plots, that the diurnalvariations of both measured and simulated vTEC shows aminimum around 0000–0300 UT (night time). However,they have different time and number of peaks. While thesimulated vTEC curves have double peaks structure withmaximum values between 0500 and 0700 UT (exceptNeQuick2 in August: only a single peak around 0900UT) and 1400–1600 UT, the observed diurnal variationof vTEC shows a single peak centered about between1000 and 1500 UT. This is in agreement with the work ofNigussie et al. (2013).

In 2014, all models have a common overestimation ofthe GPS-vTEC for the entire day in July at ABOO station.Moreover, the large discrepancy between NeQuick2 andIRI models in estimating the experimental vTEC isobserved during the month of June. An overestimationand underestimation of GPS-vTEC by NeQuick2 modelhas been observed for the time span of 0400–0700 UTand 0700–1700 UT during June at ABOO station. DuringAugust, large overestimation of the experimental vTECby NeQuick2 is observed for all hours in August. Similarly,overestimation of the GPS-vTEC by IRI models isobserved during August except for the time span of1100–1700 UT. During December, all models show nearlysimilar tendencies in estimating the GPS-vTEC at ABOOstation. In general, all models show a common perfor-mance (overestimation) around 0600 UT and 1700–2300UT except IRI models in February and NeQuick2 in Juneand November months of the year 2014 at ABOO station.An underestimation of GPS-vTEC by all models isobserved in the time span of 0800–1600 UT in January,February (except NeQuick2), March, April, October,November and December months of the year 2014 atABOO station. As before, the contour plot results dis-played in the bottom panel of Fig. 1(b) show the monthlyminimum and maximum discrepancies between the experi-mental and modeled vTEC with magnitude of �34.03TECu in May (by IRI-2016) and 25.94 TECu in May (byIRI-2007), respectively.

Fig. 2(a) displays hourly monthly averaged vTEC resultsobtained from the station ADD during 2008 (solar mini-mum), January–May and December of 2009 (deep solarminimum), 2010 (solar minimum), 2011 (ascending phaseof solar activity), and 2012 (solar maximum). The corre-sponding diurnal variations, computed using Eq. (1), aredisplayed in Fig. 2(b). A significant underestimation byall models is observed in the time span of 0500–1600 UTin January 0600–1500 UT in February 0600–1600 UT inMarch 0600–1500 UT in April except NeQuick2 (0700–1400 UT), 0700–1300 UT in May except NeQuick2(0700–1300 UT), 0700–1500 UT in July except IRI-2007(0700–1400 UT), 0700–1500 UT in August except

NeQuick2/IRI-2007 (0700–1300/1400 UT, respectively),0600–1500 UT in September except NeQuick2/IRI-2007(0600/0700–1200/13 UT, respectively), 0600–1600 UT inOctober except IRI-2007/NeQuick2 (0600/0700–1400/13UT, respectively), 0700–1400 UT in November exceptIRI-2007 (0700–1300 UT), 0600–1400 UT in Decemberexcept IRI-2007 (0600–1300 UT) of the year 2008 as shownin Fig. 2(a). The simulated vTEC shows a significant dis-crepancy (overestimation) from the experimental vTECmainly during the time intervals of 1800–2100 UT inMarch except IRI models, 0300–0700 UT and 1500–2300UT in April except IRI models, 0500–0600 UT and1300–1800 UT in May except IRI-2016 and IRI-2012,1300–1800 UT in June by IRI-2007, 1500–1600 UT in Julyby IRI-2007, 1400–1600 UT in August except IRI-2016 andIRI-2012, 1300/14–2300/1800 UT in September byNeQuick2 and IRI-2007, respectively, (0400–0600 UTand 1500–2300 UT by NeQuick2; 1500–2100 UT by IRI2007 and 1600–2100 UT by both IRI-2016 and IRI-2012)in October 1500–2100 UT in November and December ofthe year 2008 at ADD station (see Fig. 2(a)). As seen fromthe contour plot of the corresponding vTEC errors dis-played in the top left panel of Fig. 2(b), the difference ismainly observed between 0600 and 1600 UT. Moreover,as one can see from the color bar, the minimum and max-imum discrepancy between the modeled and experimentalvTEC is �19.6 TECu (by IRI-2012) and 8.66 TECu (byNeQuick2), respectively. Here, also, the individual modelsperformances are calculated by using Eq. (1). Accordingly,the maximum underestimation discrepancies by NeQuick2,IRI-2016, IRI-2012 and IRI-2007 are given by �15.57TECu, �19.4 TECu, �19.6 TECu and �19 TECu inFebruary, respectively, at ADD station during the year2008. Similarly, the maximum overestimation discrepancyby NeQuick2, IRI-2016, IRI-2012 and IRI-2007 are givenby 8.66 TECu in October, 6.90 TECu in November, 7.20TECu in November and 7.76 TECu in November,respectively.

A similar investigation was carried out in the deep solarminimum period of the year 2009 at ADD station and isdisplayed the middle panel of Fig. 2(a). Due to the scarcityof available data, we only analyzed data for the months ofJanuary–June of 2009. As seen from the results, large dis-agreement between the experimental and simulated vTECis observed during the months of April and May. The phys-ical cause for this difference is not known. However, for therest of the months, the diurnal variations of the observedand modeled vTEC show the usual pattern seen in 2008.In 2009, all models underestimate the experimental vTECduring the time intervals of 0000–0300 UT and 0600–1500 UT in January and February except NeQuick2 (over-estimation at 0600 UT in February), 0000–1700 UT inMarch except NeQuick2 and IRI-2007 (0000–0300 UTand 0600–1400 UT), 0800–1300 UT in December atADD. Similarly, overestimation of the experimental vTECis observed during the time interval of 1600–2300 UT inJanuary by IRI-2007, 1500–2300 UT in February and

Fig. 2a. Diurnal variation of the monthly averaged GPS and modeled vTEC at ADD during 2008 (solar minimum), January–May and December 2009(deep solar minimum), 2010 (solar minimum), 2011 (ascending phase of solar activity) and 2012 (solar maximum).


March by NeQuick2, and 1600–2300 UT in December atADD (see the middle panel of Fig. 2(a)). From the contourplot of the results displayed in the middle panel of Fig. 2(b), we observed that the minimum and maximum discrep-ancy are on the order of �27.08 TECu (by IRI-2016 inMay) and 13.59 TECu (by IRI-2007 in May), respectively.

The diurnal variation of the monthly mean experimentaland simulated vTEC obtained from the station ADD dur-ing the years 2010–2012 are displayed at the right top, leftand right bottom panels of Fig. 2(a), respectively. The dis-crepancies, calculated by using Eq. (1), are also displayedon the corresponding panels of Fig. 2(b). Olatunbosunand Ariyibi (2015) studied TEC variations at low-latitudestations within the EIA zone and reported that the maxi-mum value of experimental TEC was attained betweenthe hours of 1000–1400 UT at ADD station during 2012.Our present result is in agreement with this report providedthat the maximum time is 1300UT (in our result). Asbefore, the individual performance of each model is calcu-lated and results show that all models have a significantunderestimation during daytime hours between about0600–1800 UT at ADD station during the years 2010–2012 (see Fig. 2(a)). From the results displayed in the righttop panel of Fig. 2(b), we observe that the minimum andmaximum discrepancy with the measurements are �17.85TECu (by IRI-2016) and 12.92 TECu (by NeQuick2),respectively. The maximum underestimation carried out

by NeQuick2, IRI-2016, IRI-2012 and IRI-2007 modelsare given by �15.12 TECu, �17.15 TECu, �17.85 TECuin March and �16.57 TECu in February, respectively(see Fig. 2(b)). Again, the maximum overestimation ofthe experimental vTEC by NeQuick2, IRI-2016, IRI-2012and IRI-2007 models are given by 12.92 TECu in October,8.31 TECu in November, 9.11 TECu in November and8.99 TECu in October, respectively. In 2011, the minimumand maximum discrepancy values are �34.67 TECu (byIRI-2012) and 16.83 TECu (by NeQuick2), respectively(see left bottom panel of Fig. 2(b)). Similarly, in 2012,the discrepancy errors are given in the right bottom panelof Fig. 2(b) and results show that the minimum and maxi-mum error values are �30.18 TECu in March and 17.56TECu in August by NeQuick2, respectively. Furthermore,the maximum underestimation by IRI-2016, IRI-2012 andIRI-2007 models is �27.82 TECu in October, �23.88TECu in September and �27.12 TECu in October, respec-tively. Moreover, the maximum underestimation errors byIRI-2016, IRI-2012 and IRI-2007 models are 11.78 TECuin April, 11.88 TECu in April and 13.79 TECu in Novem-ber, respectively.

Based on the availability of GPS data, Fig. 3(a) displaysthe diurnal monthly mean experimental and its corre-sponding models vTEC at ADD station during the years2013–2015, from top to bottom panels, respectively. Thevariation of monthly mean vTEC discrepancy during the

Fig. 2b. The contour plot of the diurnal variations of the monthly averaged vTEC mismodelings at ADD during 2008–2012.


above mentioned years are as shown in the contour plots ofthe Fig. 3(b). During 2013–2015, both the modeled andexperimental vTEC showed a minimum around 0000–0300 UT. The GPS-vTEC has a single peak and reachesmaximum between 1000 and 1400 UT, which is agreementwith the report of Olatunbosun and Ariyibi (2015).According to the earlier study made by Appleton (1946),Martyn (1955) and Rastogi (1959), the occurrence of anoticeable trough supplemented by maximum TEC valuesin the pre-noon and the afternoon local time at equatorialregions is referred to as a noon bite-out, which is a charac-teristic feature at an equatorial station that falls in thetrough of the EIA. In our present study, such featuresare also observed in the diurnal variations of the modeledvTEC at ADD station with first peak at around 0600 UTand the second peak in between 1300 and 1400 UT atADD station. Habarulema et al. (2016) reported that anenhancement or launching of TIDs from the equator,which corresponds to increased DH (where DH / E� B

drift velocities during local daytime), is seen at around0600 UT over ADD on 9 March 2012. This indicates thatthe resulted modeled vTEC may be contaminated by the

effect of TIDs. In 2013, the NeQuick2 and IRI modelsshowed clear discrepancy during the entire daytime of eachmonth except April and November. During April andNovember, they show similar tendencies in estimating theobserved vTEC at ADD station. All models show betterperformance during nighttime than daytime hours, due tolarge discrepancies between simulated and experimentalvTEC observed during the daytime hours. The right toppanel of Fig. 4 shows the contour plot of the diurnal vari-ations of monthly averaged GPS-vTEC mismodelings byNeQuick2 and IRI models. As seen from this figure, thevTEC values for the entire years were overestimatedaround 0600 UT and in between 1500 and 1800 UT. In2013, the observed maximum underestimation and overes-timation, respectively, are �36.84 TECu in October and18.83 TECu in May by NeQuick2, �28.32 TECu inNovember and 11.45 TECu in May by IRI-2016, �28.59TECu in November and 11.96 TECu in December byIRI-2012 and �27.39 TECu in October and 16 TECu inNovember by IRI-2007 models.

In 2014, both simulated and experimental vTEC showsimilar diurnal structure with the results obtained at

Fig. 3. (a) Left panel: Diurnal variations of the monthly averaged experimental and simulated vTEC during 2013, 2014 (from January to October) and2015 at ADD station (from top to bottom, respectively). (b) Right panel: the corresponding contour plot of mean vTEC mismodelings by NeQuick2 andIRI models.


ABOO station (i.e., one is the mirror image of the other)except the magnitude difference (see the correspondingmonths displayed at the bottom panel of Fig. 2(a) andthe middle panel of Fig. 3(a)). This indicates that the iono-spheres over these two nearby stations are coupled, butunderstanding this needs further investigation. Fig. 3(b)shows that the maximum underestimation and overestima-tion are �30.35 TECu in March and 21.44 TECu in Augustby NeQuick2, �41.18 TECu in March and 15.04 TECu inMay by IRI-2016, �37.98 TECu in March and 16.94 TECuin May by IRI-2012 and, �36.18 TECu in March and20.88 TECu in May by IRI-2007 models, respectively. Gen-erally, tendencies to overestimate and underestimate themeasurements around 0600 UT and during daytime hours(0800–1500 UT), respectively, by all models have beenobserved at ADD station during the year 2015. The con-tour plots depicted in the bottom panel of Fig. 3(b) showedthe diurnal monthly averaged vTEC mismodelings during2015 at ADD station. Hence, the observed maximum

underestimation and overestimation of the GPS-vTECare �33.21 TECu in March and 12.21 TECu in Octoberby NeQuick2, �34.16 TECu in February and 10.51 TECuin May by IRI-2016, and �33.09 TECu in February and11.71 TECu in May by IRI-2012 and, �30.49 TECu inFebruary and 15.81 TECu in May by IRI-2007, respec-tively models.

In order to validate all models at NAZR station, asbefore, we examine the diurnal variations of monthly aver-aged experimental and simulated vTEC for the years 2008–2009 and 2011 and the results are displayed in the left col-umn of Fig. 4, from top to bottom, respectively. Tariku(2015b) studied patterns of GPS-TEC variation over low-latitude regions during the deep solar minimum (2008–2009) and solar maximum (2012–2013) phases. This studyused eight ground-based dual-frequency GPS receiversinstalled at different regions in Ethiopia. Accordingly,Tariku (2015b) reported that the diurnal variability ofvTEC showed maximum values nearly between 1000 and

Fig. 4. The same as Fig. 3, but for NAZR station during the period of the years 2008, 2009 and 2011.


1300 UT during both the low and high solar activityphases. Again, Tariku (2015a) studied TEC prediction per-formance of the IRI-2012 model over Ethiopia during therising phase of solar cycle 24 (2009–2011) and reported thatthe variability of the diurnal vTEC is minimal at predawnhours (near 0300 UT) and maximal between roughly 1000–1300 UT for both the experimental data and the model.Our present experimental result is in agreement with thisreport. As in the case of ADD station, the modeled vTECshows double peaks while the experimental vTEC shows asingle peak during the entire period of the study. Large dis-crepancies (overestimation) by all models are observed dur-ing early morning around 0600 UT (during the occurrenceof the first peak) and around 1300–1500 UT (during theoccurrence of the second peak) during the entire periodof the years 2008–2009 at NAZRT station. All modeledvTEC shows better agreement with the GPS-vTEC valuesjust after midnight (0000–0300 UT). Moreover, it is foundthat all models better estimates diurnal monthly meanvTEC values around 1100–1200 UT at NAZR station thanADD station during the same period of study, 2008. TheNeQuick2 and IRI models overestimate the measurementsin both before and after local noon hours with almost sim-

ilar performances in most months of the solar minimumyears (2008 and 2009). But, during the same months ofthe year 2011 at the NAZR station, we observe a differentscenario except that all models show nearly the same esti-mating performance. They show significant overestimationof GPS-vTEC during the time span of 0300–0900 UT inMay and June 1600–2300 UT in November and December(see Fig. 4). Large discrepancies (underestimation) betweenexperimental and models vTEC are observed during 0600–1600 UT in both November and December months of theyear 2011 at NAZR station.

During the years 2008–2009 and 2011, significant dis-crepancy among the models themselves is also found (seethe right column of Fig. 4). For example, the NeQuick2model shows a significant discrepancy with IRI modelsduring the time span of 0600–1500 UT in January 0500–2300 UT in February, April, September and October1600–2300 UT in March and August months of the year2008. Similarly, in 2009, a large discrepancy (overestima-tion) by NeQuick2 is observed during the months of Febru-ary–March and August–October. Significantunderestimation of the GPS-vTEC by all models isobserved from 0800 to 1300 UT in January 2009, as shown


from Fig. 4. Again, all IRI models underestimate the exper-imental vTEC in the time interval of 0800–1300 UT inFebruary and 0900–1300 UT in March (see Fig. 4). During2011, large discrepancies between NeQuick2 and IRI mod-eled vTEC are observed during the months of March,April, July, August and September. In other months ofthe year 2011, they mostly show similar performances inestimating the measurements. The right top panel ofFig. 4 shows the contour diagram of the diurnal variationof monthly mean GPS-vTEC mismodelings by NeQuick2and IRI models. In 2008, the calculated maximum underes-timation and overestimation values are �5.68 TECu inMarch and 9.73 TECu in October by NeQuick2, �7.42TECu in March and 6.30 TECu in November by IRI-2016, and �8.82 TECu in March and 6.60 TECu inNovember by IRI-2012 and, �6.72 TECu in March and8.63 TECu in April by IRI-2007, respectively. Similarefforts have been made during 2009 and 2011 at NAZRand results are given in the right middle and bottom panelsof Fig. 4.

In general, as a common test scenario, our monthlyresults show that all models (most frequently) either over-estimate or underestimate the experimental vTEC at all sta-tions located at the low latitude East African region duringour period of study. This is in agreement with the previ-ously reported studies (de Abreu et al., 2017; Nigussieet al., 2012). Furthermore, according to our present results,both the NeQuick2 and IRI models show a significant per-formance difference in estimating the measurements fromstation to station. For example, our results show that theexperimental vTEC at all stations show a single peak dur-ing our study periods. But, the modeled vTEC show differ-ent trends at the two low latitude stations of the EastAfrican region. At RCMN station, which is located atlow latitudes of the southern hemisphere, the modeledvTEC shows a single peak (maxima) during solar minimum(2008). However, at the remaining stations (ADD, NAZRand ABOO stations, which are located in the northernhemisphere), the modeled vTEC shows double maxima inmost months of the study periods. Basically, the diurnaldouble maxima (observed by modeled vTEC) manifestsas twin peaks in the diurnal trend of the electron densityor vTEC during the day at that station. But this is notobserved in the observations. This signature was firstobserved by Kohl et al. (1968) as a bite-out event, i.e., asudden depletion of peak electron density or TEC followedby a recovery. Therefore, our present results indicate thatthere is uncertainty in the driving forces of those modelsusing solar (solar flux indices, F10.7) and geomagneticactivity as inputs. Furthermore, the effects of those drivingforces are also different from hemisphere to hemisphere onthe low latitude of the East African region.

Guo et al. (2015) studied the temporal-spatial variationof Global GPS derived TEC during 1999–2013 andreported that the effect of solar activity on TEC is strongerin low latitudes than in mid-high latitudes, and stronger inthe southern hemisphere than in the northern hemisphere.

Furthermore, according to their report, the effect in lowlatitudes in the northern hemisphere is stronger than thatin low latitudes in the southern hemisphere. Moreover,according to reports presented by Martyn (1955) andRao (1966), the pre-noon peak in the diurnal variation atequatorial latitudes is influenced by horizontal winds inaddition to production and loss processes, while the after-noon peak is determined by vertical E� B electrodynami-cal drifts and diffusion along the magnetic field lines.Habarulema et al. (2016) also explained that the decreaseof DH , which has a linear relationship with E � B driftvelocities during local daytime, after sunrise is because ofthe disturbance dynamo electric field contribution to low-latitude ionospheric drifts. Accordingly, the different per-formances observed by those models over the two low lat-itude hemispheres may be due to the combined effect ofsolar and geomagnetic activity. Apart from this, accordingto reports made by different scholars, the discrepancybetween modeled and experimental vTEC is shown mainlydue to inadequately computed ionospheric slab thickness(or poor estimation of the hmF2 and foF2) (Chakrabortyet al., 2014; Nigussie et al., 2013).

3.2. Seasonal performances of the NeQuick2, IRI-2016, IRI-

2012 and IRI-2007 models

Percentage deviation in vTEC estimated by NeQuick2,IRI-2016, IRI-2012 and IRI-2007 models at ADD (2008,2010–2013 and 2015), NAZR (2008, 2009 and 2011),RCMN (2008) and ABOO (2014) are shown in Figs. 5and 6. The values were computed using Eq. (2) and resultsshow that the magnitude of percentage deviation in the IRImodeled vTEC is found maximum during March equinoxand December solstice whereas minimum during June sol-stice and September equinox at ADD station during 2008,2011, 2013 and 2015. During 2010 and 2012, the IRI mod-eled vTEC show maximum and minimum deviations dur-ing equinoctial and solstice seasons, respectively, at ADDstation (see Fig. 5). Similarly, the magnitude of percentagedeviation (by NeQuick2 model) is found maximum duringMarch equinox and December solstice whereas minimumduring June solstice and September equinox at ADD sta-tion during 2008 and 2010–2012 (see Fig. 5). From this fig-ure, the NeQuick2 modeled vTEC shows maximumdeviation during September equinox and December solsticewhile minimum deviation is observed during March equi-nox and June solstice in 2013. However, the observed trendby NeQuick2 model at ADD station (see Fig. 5) during2013 is opposite with 2015. The IRI-2007 model showsthe best performance in estimating the measured vTEC atADD station during solstice seasons of the year 2008 withmaximum discrepancy of 3.42% in June solstice and12.87% in December solstice (see Fig. 5). The NeQuick2model also shows the best performances in the other twoseasons with maximum discrepancy of 10.8% duringMarch equinox and �6.12% during September equinox in2008 (see Fig. 5). Its performance during June solstice is

Fig. 5. Seasonal mean variation of percent discrepancy in NeQuick2 and IRI-models over ADD station during 2008, 2010–2013 and 2015.

Fig. 6. Seasonal mean variation of percent discrepancy in NeQuick2 and IRI-models at NAZR (2008–2009 and 2011), ABOO (2014) and RCMN (2008)stations.



also good because it has shown insignificant discrepancywith the IRI-2007 model (about 1.1%). This implies thatthe NeQuick2 is the most preferable model at ADD stationduring 2008 (solar minimum). However, its performance atNAZR and RCMN stations during the same period (2008)is very poor (see Fig. 6). Again, from Fig. 5, it is seen thatthe NeQuick2 and IRI-2007 are the best models in estimat-ing the Equinoctial and Solstice seasons, respectively, atADD station during 2010 (solar minimum). The magnitudeof the discrepancy observed by the IRI-2007 model duringJune solstice and December solstice are 0.5% and 2.21%,respectively, while the NeQuick2 modeled vTEC shows14.91% in March equinox and 0.94% in September equinoxas shown in Fig. 5.

In 2011, the IRI-2016 model shows better agreementwith observation during June solstice with about �0.12%deviation whereas the NeQuick2 model shows the best per-formances in the other three seasons with maximum devia-tion of 16.93%, �2.09% and 11.59%, respectively, duringMarch equinox, September equinox and December solstice(see Fig. 5). In 2012, the NeQuick2 is the best model forJune solstice with 0.93% and September equinox with�8.92% deviations, while the IRI-2007 is the best modelin March equinox with 6.79% deviation and December sol-stice with 1.28% deviation as shown in Fig. 5. In 2013, thebest models are the IRI-2007 and IRI-2016 models at ADDstation (see Fig. 5). The IRI-2007 shows the best perfor-mance during both March and September equinoxes andDecember solstice with deviation of 5.81%, �0.07% and6.8%, respectively. Similarly, the IRI-2016 shows the bestperformance during June solstice with maximum discrep-ancy of 1.72%. In 2015, the IRI-2007 model best performsin all seasons (except June solstice) with error of magnitude15.94% in March equinox, 4.84% in September equinoxand 8.8% in December solstice at ADD station. Similarly,the IRI-2012 model shows best performance in June sol-stice with maximum discrepancy of �0.46% at ADD sta-tion during 2015.

In 2008, we find three models at NAZR, work best dur-ing different seasons: namely, IRI-2007 in March equinoxwith 1.47% deviation, IRI-2012 in June solstice andDecember solstice with �14.11% and �6.02% deviations,respectively, and IRI-2016 in September equinox with max-imum discrepancy of �0.04% (see Fig. 6). In 2009, the IRI-2012 model best performs in all seasons with error of about�4.7% in March equinox, �13.81% in June solstice,�4.99% in September equinox and �15% in December sol-stice at NAZR station. In 2011, the NeQuick2 is the bestmodel during March equinox and June and December sol-stices with maximum discrepancy values of 1.44%, �4.02%and �0.75%, respectively, at NAZR station. From Fig. 6,we also found that the IRI-2007 has better performanceduring September equinox with maximum discrepancy of0.51% at NAZR station during 2011. Similar investigationat ABOO station show that the IRI-2007 and IRI-2016show better performance during March and Septemberequinox with maximum error of 6.22% and �1.16%,

respectively, during the year 2014 (see Fig. 6). TheNeQuick2 and IRI-2012 models show better performanceduring June and December solstice with maximum discrep-ancy of about �12.26% and 0.49%, respectively, at ABOOstation. At this station, the other models also show betteragreement with observation during December solstice (seeFig. 6). A similar investigation carried out at RCMN sta-tion reveals that the IRI-2012 model shows better perfor-mance during March equinox and December solstice withmaximum discrepancy of �6.96% and �22.86%, respec-tively, during 2008. Compared to IRI models, NeQuick2simulations is unable to correctly predict the measuredvTEC for all the seasons at RCMN station (see Fig. 6).

We also evaluate the seasonal mean differences of eachmodel as compared to measured vTEC during varyingsolar activity periods. For example, the observed seasonaldiscrepancy in the IRI-2012 model is larger during the year2011 than 2008 except in June solstice, which indicates thatthe IRI-2012 model performs better for the three seasonsduring solar minimum (2008) than ascending phase of solaractivity (2011) at NAZR station (see Fig. 6). But, this holdstrue only during March equinox and December solstice ofthe same solar activity years (2008 and 2011) at ADD. Thereverse is true for the other two seasons at ADD stationduring the year 2008 and 2011. Of course, the observed dis-crepancy during March equinox in 2008 and 2011 at ADDis almost equal (the difference is about 0.08%) as shown inFig. 5. As can be seen from Fig. 5, all IRI models show bet-ter estimating performance during solar maximum (2012)than solar minimum (2008) except in September equinox.Therefore, our present result at NAZR is in agreement withthe results recently presented by Kumar (2016) andVenkatesh et al. (2014). But, our results at ADD stationis partly in contradiction with the results reported by theabove two scholars. Moreover, our present finding atNAZR station is partly in contradiction to the results pre-sented by Tariku (2015a) that found the overestimation ofvTEC values derived from the IRI-2012 model decreases asthe Sun transitions from very low to high solar activity.However, our result at ADD station during 2008, 2012(expect in September equinox), 2013 and 2015 supportsthe work of Tariku (2015a) (see Fig. 5).

The latest IRI-2016 model showed better performanceduring solar minimum (2008) than solar medium (2011)except in June solstice at NAZR station (see Fig. 6). How-ever, the reverse is true for IRI-2007 model except inMarch equinox. Generally, comparing among IRI models,we found that IRI-2007 shows poor performance at NAZRduring solar minimum (2008–2009) except in March equi-nox in 2008. The NeQuick2 model showed better estimat-ing performance during solar medium (2011) than solarminimum (2008–2009) at NAZR, which is in agreementwith the report of Nigussie et al. (2013). However, theNeQuick2 showed a better performance during solar min-imum (2008) for both March equinox and June solsticethan solar medium (2011) at ADD station; the reverse istrue for the other two seasons during both 2008 and 2011


(see Fig. 5). Generally, our results show that the ionosphereover the two nearby stations (with a distance of less than100 km apart) exhibit different features. In other words,the ionosphere over ADD station may be contaminatedby other factors possibly due to the orientation of the geo-magnetic field lines and dynamics of neutral wind which inturn produces the EEJ currents, leading to the electrondensity variability in the equatorial and low latitude station(like ADD station). Patel et al. (2017) studied GPS-TECvariation during low to high solar activity period (2010–2014) under the Northern Crest of Indian EIA region. Inthat study, they reported the diurnal and seasonal variationof GPS-TEC, dependence of GPS-TEC with solar activity,geomagnetic condition and EEJ strength. Accordingly,they reported that there is a positive correlation observedbetween GPS-TEC and EEJ strength. Therefore, we sug-gest that the different performance observed by those mod-els in these two nearby stations (ADD and NAZR stations)may be due to the effect of the EEJ and its variability withtime which cannot be captured by the present models. Theway data are binned, averaged and parameterized whiledeveloping those models may also be the other factor.Lastly, we want to state that the collected data from bothRCMN and ABOO stations (during this study) were notenough to generalize the best model (with respect to solaractivity) at these stations, and hence needs further investi-gation by collecting more data from the stations, whichincludes from solar minimum to solar maximum periods.

3.3. Comparison of the periodicities in the measured and

modeled vTEC variations

In order to make comparison of the periodic character-istics of GPS and modeled vTEC, wavelet analysis has beencarried out. The wavelet analysis distribution of daily meanof GPS, NeQuick2, IRI-2016, IRI-2012 and IRI-2007derived vTEC are presented in Fig. 7 from top to bottom,respectively, at ABOO (first column) and ASAB (secondcolumn) stations during 2014. The wavelet analysis distri-butions of daily mean values of both solar-and geomag-netic indices are also displayed in the third column ofFig. 7. The white bands correspond to the cone of influenceobtained using the method proposed by Torrence andCompo (1998), where any oscillation under this cone doesnot have any statistical significance. The oscillations pre-sented here have different characteristics at both stationsbetween GPS and modeled vTEC except in some cases.

The top panel of the left column of Fig. 7 presents theoscillations of GPS-vTEC with periods of 2–4 daysobserved in February (days 40–59) (i.e., in between thetwo broken white vertical lines) and periods around 16–32 days observed between about June and August (days125–225) at ABOO station. The same oscillations wereobserved in February and between about the end of Juneand August at ASAB station (top panel of the middle col-umn of Fig. 7). This may be an indicator for the two iono-spheric regions in these periods that they are roughly

coupled (not necessarily) via unknown mechanisms. But,it needs further investigation in order to come up with asolid conclusion (probably, a coupling mechanism betweenthem) because a further detail about the coupling issue isbeyond the scope of this study. The IRI-2012 modeledvTEC shows a completely different scenario from that ofthe experimental vTEC at both stations during 2014(fourth left and middle columns of Fig. 7). Generally, ourresult shows a large discrepancy between the GPS-vTECand IRI modeled vTEC particularly in the periods of 16–32 days at both stations during June–August (days 125–225) (third to fifth left and middle columns of Fig. 7). Incontrast, the wavelet analysis results (NeQuick2 derivedvTEC) show better agreement with measurements in theperiods of 16–32 days at both stations (second left and mid-dle columns of Fig. 7). However, in contrast to IRI-2016and IRI-2007, the NeQuick2 modeled vTEC shows theabsence of oscillations with lower periods between 2 and4 days during February at both stations. According to asuggestion by de Abreu et al. (2017), the oscillation withperiods of between 2 and 4 days is possibly associated withthe propagation of planetary waves. The analyses were per-formed according to the yellow color of the power spec-trum which characterizes the highest energy level andsolid curved black line (95% significant level).

However, it is important to study the mechanismsresponsible for the discrepancy between observations andmodeled data as well as discrepancies among the modelsperformances themselves. Hence, to better understand thediscrepancy between the observations and simulated vTECin the day-to-day ionospheric variations and uncertaintyon the driving forces, again, the wavelet analysis has beencarried out. It is clear that the Earth’s ionosphere is createdby the ionizing effect of solar radiations (measured inexpression of solar indices) and energetic particles. Conse-quently, solar activity has the most important effect on theionospheric electron density or TEC. There is also evidencethat the ionospheric response depends on the geomagneticactivity level (Patel et al., 2017; Wang et al., 2008;Olatunbosun and Ariyibi, 2015). Therefore, studies of theeffects of solar and geomagnetic activity on vTEC are veryimportant. Accordingly, to investigate the effect of solaractivity on the ionosphere over the low latitude East Afri-can region, we evaluated the daily averaged value of solarradio flux (F10.7) and sunspot number (SSN): the Solarflux is estimated by the intensity of solar radio emissionat the reciprocal of the period of 2800 MHz or the 10.7cm flux commonly measured in solar flux unit (1 s.f.u. =1022 W m�2 Hz�1). Similarly, the daily mean values of themagnetic indices like Dst, Ap, Kp and AE are used to indi-cate geomagnetic activity since the ionosphere is linked tothe Earth’s magnetic field. The Dst index basically mea-sures the intensity of the ring current which increases dur-ing magnetic storms. The Kp-index indicates the level ofdisturbance of the magnetic field caused by a geomagneticstorms and is derived using an average of the horizontalfield intensity H (or D element if it is more disturbed than

Fig. 7. CWT power spectra analysis distribution of 2014 daily averaged GPS and modeled vTEC at ABOO (left column) and ASAB (middle column). Inboth columns, the GPS, NeQuick2, IRI-2016, IRI-2012 and IRI-2007 models derived vTEC are shown from top to bottom, respectively. The right (lastcolumn) column shows the CWT power spectra analysis distribution of daily averaged values of different solar and geomagnetic indices during 2014. Here,the vertical axis represents the daily period and the horizontal axis represents the day of the year.


H) observations from a network of about 12 geomagneticobservatories distributed around the world between geo-magnetic latitudes 48� and 63�. It is a global measure ofthe magnetic deviations from the regular daily variationduring a 3-h period (Schunk and Nagy, 2000). The Kpindex is quite often used in modeling but its logarithmicnature makes it inappropriate if daily values are needed.The Ap (daily equivalent planetary index) is the daily aver-age of a 3 h interval value that is directly derived from theKp index (Davies et al., 1990). Substorms are generallymonitored by AE which is defined by AE = AU-AL, whereAU and AL are eastward and westward electrojet indicesrespectively, and are measured in nT. The AU and ALindices are calculated from the upper and lower envelopesof the H component, respectively. The AE (auroral electro-jet) index is customarily used to determine onset (when AEincreases) and recovery phases (when AE decreases) of sub-storm activity.

Accordingly, as one can see from the third column ofFig. 7 (third–sixth panels), the oscillations of all geomag-netic indices with periods of 2–4 days has been observedin February during 2014 (solar maximum). This is in agree-ment with the experimental vTEC except AE to someextent. In contrast to this, both solar indices (first and sec-ond panels of the third column of Fig. 7) show the absenceof oscillations with lower periods between 2 and 4 days inFebruary which shows a total discrepancy between theGPS-vTEC and the two solar indices for this period ana-lyzed. Furthermore, the SSN shows a unique phenomenonwhich is not observer in both the vTEC and geomagneticindices (see between the second broken white and brokenred lines of the third column of Fig. 7). These results indi-cate that a geomagnetic activity has a measurable effect onthe low latitude of the equatorial vTEC.

Using dual frequency GPS receiver at low latitude sta-tions of Ile-Ife (7.52�N, 4.28�E), Addis Ababa (9.04�N,38.77�E) and Bangalore (13.03�E, 77.57�E), all locatedwithin 0–15�N of the equatorial anomaly region, the mea-surement of ionospheric TEC for 2012 has been carried outby Olatunbosun and Ariyibi (2015) and they reported thatgeomagnetic storms enhance TEC variations at the threestations. Therefore, our present result is in agreement withthe report recently presented by Patel et al. (2017),Olatunbosun and Ariyibi (2015) and Wang et al. (2008).In comparison to the previously mentioned models, a sig-nificant performance in showing this phenomenon can beseen only by IRI-2016 and IRI-2007 models. However,the oscillations of both solar indices with periods of 16–32 days show a significant agreement with both the exper-imental and NeQuick2 derived vTEC at both stations. Asdiscussed in the introduction, the performances ofNeQuick2 model in estimating the experimental data hasbeen tested at both the low latitude of the East Africanregion and high latitude (Tromsø in Norway) (Nigussieet al., 2012, 2016). During that study, the performancesof NeQuick2 at both regions was investigated by assistingthe model with measurements. They first computed themodel input (effective ionization level, Az) when the mod-eled sTEC best fits the measured sTEC by single GPS recei-ver (reference station). They, then, quantify theperformances of the model in reproducing the measuredTEC by running the model at different locations, whereGPS stations are available (or where there is an experimen-tal TEC from other sources as well), using the same Azlevel that calculated from a single station as a driver ofthe model. Finally, they reported that the ability of themodel to reproduce the experimental TEC increases nota-bly when the model is adapted by data ingestion from

Fig. 8. Wavelet coherence between daily averaged variations of GPS andmodeled vTEC at ABOO station from January to December of 2014. Thevertical axis represents the daily period and the horizontal axis represents


one station in comparison with the case when the modelwas used in a standard way being driven only by the dailysolar flux (F10.7). And this performance holds true up tocertain distance (�620 km) (Nigussie et al., 2012) awayfrom the reference station. Even though, they have shownthat the capability of NeQuick2 (in describing the EastAfrican region of the ionosphere) can be improved sub-stantially by data ingestion, still a discrepancy betweenthe experimental and modeled data remains a challenge.Though they have the same unit, the calculated Az valuesare also different from that of F10.7. Furthermore, our pre-sent results show a total discrepancy between the GPS-vTEC and solar activity indices during oscillations withperiods of 2–4 days as in the case of NeQuick2 derivedvTEC during 2014. This indicates that the observed dis-crepancy between the GPS-vTEC and NeQuick2 modeledvTEC may be attributed to the non-inclusion of the effectsof geomagnetic activity as a driver of the model next tosolar flux (F10.7). In other words, it is important to includethe impacts of the EEJ which controls the EIA and E � Bdrifts in the model which could probably be very much use-ful for improving the NeQuick2 model predictions particu-larly in the equatorial and low-latitude regions.

Habarulema and Ssessanga (2017) studied adaptation ofa climatological model (IRI-2012) with the use of TECdata derived from the Global Navigation Satellite System(GNSS), and most importantly its subsequent validationwith both radio occultation from Constellation ObservingSystem for Meteorology, Ionosphere and Climate (COS-MIC) and ionosonde data. By adjusting the solar activityindices used within the standard IRI-2012 model with theaim of minimizing error differences between IRI TECand GNSS TEC, the adjusted indices are used as driversof the IRI-2012 model on a regional scale and results forelectron density (Ne) profiles, maximum height of the F2

layer (hmF2), TEC, and critical frequency of the F2 layer(foF2) generated. Validation was done by direct compar-ison with ionosonde and COSMIC-derived data parame-ters. By averaging results over low, equatorial, and mid-latitude regions, they reported that the adapted IRI-2012model shows an improvement of about 18% in estimatingTEC during the storm period of 9 March 2012. Further-more, they reported that the Ne profiles from the adaptedmodel accurately approximates ionosonde Ne profiles espe-cially below 300 km altitude but underestimates the iono-sonde NmF2 and hence foF2 in mid-latitude regions.Moreover, they also reported that in most cases both stan-dard and adapted models match COSMIC data for topsideelectron density representation. As mentioned previously,such better performance of the Ionospheric model (afteradaptation) in the equatorial region is also reported byNigussie et al. (2016, 2012). This indicates that theobserved discrepancy between the GPS-vTEC and IRImodeled vTEC is due to inadequately computed indices(particularly, F10.7) used as drivers of the model on aregional scale.

In order to study the agreement between GPS observedand modeled vTEC, we applied the wavelet coherencebetween the daily averaged variations of GPS-vTEC andsimulated vTEC during the year 2014 at ABOO and ASABstations and results are presented in Figs. 8 and 9, respec-tively. The arrows indicate the phase difference betweenboth time series where right arrows indicate series are inphase, left arrows indicate series are out of phase (180�),and vertical arrows indicate the second time series lagsthe first by 90�. According to Carey et al. (2013) the waveletcoherency allows identification of scales and times whenthe time series are experiencing oscillations at a similar fre-quency, i.e., both the time series are in effect coupled. Thus,as shown in both Figs. 8 and 9 by NeQuick2, most of thehigh energy regions of the wavelet coherence spectrum existon the scale of 16–32 days (similar to that observed inFig. 9) and pass the significance testing, which means thatthe 27-day period of the sun’s rotation influences the peri-odic variation of vTEC (Kutiev et al., 2012). This is inagreement with the report presented by Guo et al. (2015).The solar activity, including the sunspot with the strongmagnetic field, the solar flare, and the bright eruption,can generate more ultraviolet and X-rays, which are thegoverning source of ionization. The solar activity can alsogenerate the high-speed solar wind to influence the Earth’smagnetosphere to result in molecular dissociation (Guoet al., 2015 and references therein). The sunspots are facingEarth every 27 days because of the sun’s rotation, so theradiation has a cycle of 27 days to make the TEC variationwith the period of 27 days. The IRI-2016 and IRI-2007models have shown better estimating performance duringthe periods roughly between 4 and 8 days in the monthof February and this is in agreement with the results dis-played on Fig. 7. We do not discuss comparison resultsof the IRI models themselves and leave to future work with

the day of the year.

Fig. 9. The same as Fig. 8, but, for ASAB station.


an extended case study. This study shows that in the equa-torial and low latitude regions during 2014 large differencesare observed between the GPS and IRI modeled vTEC par-ticularly in the periods of 16–32 days as compared toNeQuick2 derived vTEC at both stations (see Figs. 8 and9). Hence, further studies are needed to fully understandthe discrepancies between experimental and simulatedvTEC values as well as among the models themselves byincorporating additional sources of experimental data. Thiswill include observations from solar minimum to solarmaximum periods.

Here also, besides the driving mechanisms of iono-spheric variability described above, there are other unpre-dictable phenomena related to solar activity that departfrom the usual state of the ionosphere. This state is com-monly called the irregular (or non-repeatable) variationof the ionosphere. The effect of Traveling Ionospheric Dis-turbances is one example (Habarulema et al., 2016; Seeber,2003). Jonah et al. (2015) using GPS-TEC data at equato-rial and low latitude regions in the South American sectorfor 2001 and 2009, observed oscillations with periodsaround 1–5 and 8–10 days. They suggest that these are con-tributions of tides and planetary waves from the loweratmosphere in the TEC variability. Hence, from our pre-sent results, the observed oscillations with periods of 2–4days may also be contributions of TIDs or planetary wavesfrom the lower atmosphere in the vTEC variability, butneeds further investigations.

4. Conclusions

In this paper, we report the capability of the NeQuick2,IRI-2016, IRI-2012 and IRI-2007 models in predictingdiurnal monthly and seasonal mean variation of GPS-TEC over the low latitude East African region during vary-ing solar activity periods (1998 and 2008–2015). Studies ofthis kind are both important and difficult because of the

presence of the EIA, irregularities, and the relatively sparsemeasurements available in this region. The main conclu-sions of the study are as follows:

1. There is no outstanding model performing equally wellunder all conditions (season, solar and geomagneticactivity) and locations.

2. Simulated TEC at all stations (located in the northernhemisphere) show diurnal double maxima which mani-fests as twin peaks in the diurnal trend of the Ne orTEC; a unique feature not seen at RCMN (a stationlocated at the southern hemisphere or barely beneathequator).

3. The observed diurnal monthly discrepancy is foundmaximum in the time when double peaks of modeledTEC are simulated for all stations (except at RCMN).

4. Seasonal variations in measured TEC show semiannualperiodicity with maxima in March equinox and Decem-ber solstice, while minima are observed in June solsticeand September equinox over ADD (2008, 2011, 2013and 2015), ABOO (2014), NAZR (2008) and RCMN(2008) stations. This trend is also seen from the IRImodeled TEC at ADD (2008, 2010, 2012 and 2013),RCMN (2008) and NAZR (2008–2009). Similarly,NeQuick2 modeled TEC shows this trend at ADD dur-ing 2013 and 2015.

5. Another seasonal trend in measured TEC has beennoted over ADD (2010 and 2012) and NAZR (2009) sta-tions; the largest diurnal peaks in measured TEC tend tooccur in equinoctial seasons, while smallest during sol-stice seasons. NeQuick2 modeled TEC also shows thistrend at ADD (2008, 2010 and 2012), ABOO (2014),RCMN (2008) and NAZR (2008–2009).

6. In 2011, at NAZR, both measured and modeled TECshow maximum during September equinox and Decem-ber solstice while minimum are observed during Marchequinox and June solstice, with largest value observedin September equinox and lowest in June solstice (bymeasured and NeQuick2), and March equinox (by IRImodels).

7. The seasonal trend observed in measured TEC equallyexplains the semiannual variability noted on the sea-sonal variations in magnitude of percentage deviationsby IRI models over ADD. Similarly, the magnitude ofpercentage deviation by NeQuick2 shows semiannualperiodicity with maxima in March equinox and Decem-ber solstice, while minima are observed in June solsticeand September equinox over ADD during 2008 and2010–2012. However, the discrepancy is found maxi-mum during September equinox and December solsticewhereas minimum during March equinox and June sol-stice in 2013; the reverse is seen in 2015.

8. Comparison among the IRI models yields that the IRI-2007 is the best model in estimating the measured sea-sonal mean TEC at ADD (except in June solstice during2011–2013 and 2015) for all solar activity phases. The


IRI-2016 and IRI-2012 models, respectively, show betterperformance in June solstice during 2011–2013 and 2015at ADD station.

9. Compared to IRI models, NeQuick2 shows better per-formance in reproducing GPS-TEC during all seasonsof the years 2008 and 2011 (except in June solstice) atADD. Similarly, NeQuick2 modeled TEC shows betteragreement with observations during all seasons exceptin September equinox at NAZR in 2011. In contrast,NeQuick2 shows a very poor performance at RCMNin 2008. NeQuick2 shows better performance duringMarch equinox and June solstice while IRI-2016 andIRI-2012 models show better prediction during Septem-ber equinox and December solstice at ABOO in 2014.

Competing interests

The authors declare that they have no competinginterests.

Acknowledgements

The latest IRI-2016, IRI-2012, and IRI-2007 modelsdata were obtained using the web based version availableat https://omniweb.gsfc.nasa.gov/vitmo/iri2016_vitmo.html,https://omniweb.gsfc.nasa.gov/vitmo/iri2012_vitmo.html and https://omniweb.gsfc.nasa.gov/vitmo/iri_vitmo.html, respectively. The GPS data have been downloadedfrom the website www.unavco.org. The solar and geomag-netic data have been obtained from the National Aeronau-tics and Space Administration (NASA) by the websitehttp://omniweb.gsfc.nasa.gov/form/dx1.html. Cross wave-let and wavelet coherence toolbox for MATLAB were pro-vided by A. Grinsted (2002–2014): obtained from thewebsite http://www.glaciology.net/wavelet-coherence. Thework of M.B. Moldwin was partially supported by USNSF Grant AGS/450512.

The work of Melessew Nigussie was partially supportedby the Air Force Office of Scientific Research and AirForce Material Command USAF under award FA9550-16-1-0070.

References

Abdu, M.A., 2005. Equatorial ionosphere thermosphere system: electro-dynamics and irregularities. Adv. Space Res. 35 (5), 771–787.

Akala, A.O., Seemala, G.K., Doherty, P.H., Valladares, C.E., Carrano, C.S., Espinoza, J., Oluyo, S., 2013. Comparison of equatorial GPS-TECobservations over an African station and an American station duringthe minimum and ascending phases of solar cycle 24. Ann. Geophys.31, 2085–2096.

Appleton, E.V., 1946. Two anomalies in ionosphere. Nature 157, 691–693.Bilitza, D., Reinisch, B.W., 2008. International reference ionosphere 2007

improvements and new parameters. Adv. Space Res. 42 (4), 599–609.Bilitza, D., Altadill, D., Truhlik, V., Shubin, V., Galkin, I., Reinisch, B.,

Huang, X., 2017. International reference Ionosphere 2016: fromionospheric climate to real-time weather predictions. Space Weather15, 418–429. https://doi.org/10.1002/2016SW001593.

Carey, S.K., Tetzlaff, D., Buttle, J., Laudon, H., McDonnell, J., McGuire,K., Seibert, J., Soulsby, C., Shanley, J., 2013. Use of color maps andwavelet coherence to discern seasonal and inter annual climateinfluences on stream flow variability in northern catchments. WaterResour. Res. 49, 1–14. https://doi.org/10.1002/wrcr.20469.

Chakraborty, M., Kumar, S., de Kumar, B., Guha, A., 2014. Latitudinalcharacteristics of GPS derived ionospheric TEC: a comparative studywith IRI 2012 model. Ann. Geophys. 57 (5), A0539. https://doi.org/10.4401/ag-6438.

Chu, F.D., Liu, J.Y., Takahashi, H., 2005. The climatology of ionosphericplasma bubbles and irregularities over Brazil. Ann. Geophys. 23,379384.

Davies, K., 1990. Ionosphere Radio, Peter Peregrinus Ltd, London.10.1049/PBEW031E.

de Abreu, A.J., Martin, I.M., de Jesus, R., Bolzan, M.J.A., Venkatesh, K.,Fagundes, P.R., Alvesi, M.A., Gende, M., Kumar, S., 2009. Compar-ison of GPS-TEC measurements with IRI2012-TEC predictions in theBrazilian sector during the unusual solar minimum 2009. Ann.Geophys. 60 (3), A0331. https://doi.org/10.4401/ag-7300.

Ezquer, R.G., Ortiz de Adler, N., Radicella, S.M., Mosert de Gonzalez,M., Manzano, J.R., 1995. IRI and BPM total electron contentpredictions for Tucuman and BPM total electron content predictionsfor Tucuman. Adv. Space Res. 15, 121–124.

Elvidge MSci., S., 2014. On The Use of Multi-model Ensemble Techniquesfor IONOSPHERIC and Thermospheric Characterisation 2014University of Birmingham UK (PhD thesis).

Guo, J., Li, W., Liu, X., Kong, Q., Zhao, C., Guo, B., 2015. Temporal-spatial variation of global GPS derived total electron content, 1999–2013. PLoS ONE 10 (7), e0133378. https://doi.org/10.1371/journal.pone.0133378.

Habarulema, J.B., Katamzi, Z.T., Yizengaw, E., Yamazaki, Y., Seemala,G., 2016. Simultaneous storm time equatorward and poleward large-scale TIDs on a global scale. Geophys. Res. Lett. 43, 6678–6686.https://doi.org/10.1002/2016GL069740.

Habarulema, J.B., Ssessanga, N., 2017. Adapting a climatology model toimprove estimation of ionosphere parameters and subsequent valida-tion with radio occultation and ionosonde data. Space Weather 15, 84–98. https://doi.org/10.1002/2016SW001549.

Jonah, O.F., de Paula, E.R., Muella, M.T.A.H., Dutra, S.L.G., Kherani,E.A., Negreti, P.M.S., Otsuka, Y., 2015. TEC variation during highand low solar activities over South American sector. J. Atmos. Terr.Phys. 135, 2–35.

Kelly, M.C., 2009. The Earth’s Ionosphere Plasma Physics and Electro-dynamics, second ed. Elsevier Inc., ISBN 13: 978-0-12-088425-4.

Kohl, H., King, J., Eccles, D., 1968. Some effects of neutral winds on theionospheric F-layer. J. Atmos. Terr. Phys. 30 (10), 1733–1744.

Kumar, S., 2016. Performance of IRI-2012 model during a deep solarminimum and a maximum year over global equatorial regions. J.Geophys. Res. Space Phys. 121. https://doi.org/10.1002/2015JA022269.

Kutiev, I., Otsuka, Y., Pancheva, D., Heelis, R., 2012. Response of low-latitude ionosphere to medium-term changes of solar and geomagneticactivity. J. Geophys. Res. 117, A08330.

Li, S., Li, Li, Peng, J., 2016. Variability of ionospheric TEC and theperformance of the IRI-2012 model at the BJFS station, China. ActaGeophys. 64(5), 1970–1987. 10.1515/acgeo-2016-0075.

Ma, G., Maruyama, T., 2003. Derivation of TEC and estimation ofinstrumental biases from GEONET in Japan. Ann. Geophys. 21,2083–2093. https://doi.org/10.5194/angeo-21-2083-2003.

Mannucci, A.J., Wilson, B.D., Edwards, C.D. A new method formonitoring the Earth’s ionosphere total electron content using theGPS global network. In: Proceedings of ION GPS-93, Institute ofNavigation, 1993, pp. 1323–1332.

Martyn, D.F., 1955. Theory of height and ionization density changes atthe maximum of a chapman-like region, taking account of ionproduction, decay, diffusion and total drift. In: Proceedings CambridgeConference, Physical Society, London, pp. 254–259.

https://omniweb.gsfc.nasa.gov/vitmo/iri2016_vitmo.html,https://omniweb.gsfc.nasa.gov/vitmo/iri2012_vitmo.html



https://omniweb.gsfc.nasa.gov/vitmo/iri_vitmo.html

https://omniweb.gsfc.nasa.gov/vitmo/iri_vitmo.html

http://www.unavco.org

http://omniweb.gsfc.nasa.gov/form/dx1.html

http://www.glaciology.net/wavelet-coherence

http://refhub.elsevier.com/S0273-1177(18)30032-2/h0005










https://doi.org/10.1002/2016SW001593

https://doi.org/10.1002/wrcr.20469

https://doi.org/10.4401/ag-6438










https://doi.org/10.1371/journal.pone.0133378

https://doi.org/10.1371/journal.pone.0133378

https://doi.org/10.1002/2016GL069740

https://doi.org/10.1002/2016SW001549









https://doi.org/10.1002/2015JA022269

https://doi.org/10.1002/2015JA022269




https://doi.org/10.5194/angeo-21-2083-2003


Meggs, R.W., Mitchell, C.N., Smith, A.M., 2006. An Investigation intothe relationship between ionospheric scintillation and loss of lock inGNSS receivers. In: Characterising the Ionosphere, Meeting Proceed-ings RTO-MP-IST-056, Paper 5, Neuilly-sur-Seine, France: RTO, pp.51 510.

Mushini, S.C., Jayachandran, P.T., Langley, R.B., MacDougall, J.W.,Pokhotelov, D., 2011. Improved amplitude and phase scintillationindices derived from wavelet detrended high latitude GPS data. GPSSolut. 16 (3), 363373.

Nava, B., Coısson, P., Radicella, S.M., 2008. A new version of theNeQuick ionosphere electron density model. J. Atmos. Sol. Terr. 70,1856–1862.

Nigussie, M., Radicella, S.M., Damtie, B., Nava, B., Yizengaw, E.,Ciraolo, L., 2012. TEC ingestion into NeQuick2 to model the EastAfrican equatorial ionosphere. Radio Sci. 47, RS5002. https://doi.org/10.1029/2012RS004981.

Nigussie, M., Radicella, S.M., Damtie, B., Nava, B., Yizengaw, E.,Groves, K., 2013. Validation of the NeQuick2 and IRI-2007 models inEast African equatorial region. J. Atmos. Sol. Terr. Phys. 102, 26–33.https://doi.org/10.1016/j.jastp.2013.04.016.

Nigussie, M., Radicella, S.M., Damtie, B., Yizengaw, E., Nava, B.,Roininen, L., 2016. Validation of NeQuick2 TEC data ingestiontechnique against C/NOFS and EISCAT electron density measure-ments. Radio Sci. 51. https://doi.org/10.1002/2015RS005930.

Olatunbosun, L.G., Ariyibi, E.A., 2015. Studies of total electron contentvariations at low-latitude stations within the equatorial ionizationanomaly zone. IOSR J. Appl. Phys. (IOSR-JAP) 7(5), 12–24. e-ISSN:2278-4861 (Ver. I).

Olwendo, O.J., Baki, P., Mito, C., Doherty, P., 2012. Characterization ofionospheric GPS total electron content (GPS-TEC) in low latitudezone over the Kenyan region during a very low solar activity phase. J.Atmos. Sol. Terr. Phys. 84–85, 52–61.

Olwendo, O.J., Baki, P., Cilliers, P.J., Mito, C., Doherty, P., 2013.Comparison of GPS TEC variations with IRI- 2007 TEC prediction atequatorial latitudes during a low solar activity (2009–2011) phase overthe Kenyan region. J. Adv. Space Res. 52 (10), 1770–1779. https://doi.org/10.1016/j.asr.2012.08.001.

Patel, N.C., Karia, S.P., Pathak, K.N., 2014. GPS-TEC variation duringlow to high solar activity period (2010–2014) under the Northern Crestof Indian equatorial ionization anomaly region. Positioning 8, 13–35.https://doi.org/10.4236/pos.2017.82002.

Pi, X., Mendillo, M., Fox, M.W., Anderson, D.N., 1993. Diurnal doublemaxima patterns in the F region ionosphere: Substorm-related aspects.

J. Geophys. Res. 98 (A8), 13677–13691. https://doi.org/10.1029/93JA00502.

Rao, B.C.N., 1966. Control of equatorial Spread-F by the F layer height.J. Atmos. Terr. Phys. 28, 1207–1217.

Rastogi, R.G., 1959. The diurnal development of the anomalous equa-torial belt in the F2 region of the ionosphere. J. Geophys. Res. 64, 727–732.

Schunk, R.W., Nagy, A.F., 2000. Ionospheres, Physics, Plasma Physicsand Chemistry. Cambridge, Atmospheric and Space Science Series.Cambridge University Press, New York.

Shunk, R.W., Nagy, A.F., 2009. Ionospheres, Physics, Plasma Physics,and Chemistry. Cambridge Atmospheric and Space Science Series.Cambridge University Press, New York.

Seeber, G., 2003. Satelite Geodesy. de Gruyter, Berlin, Germany.Seemala, G.K., 2014. GPS-TEC Analysis Application, Indian Institute of

Geomagnetism (IIG). Navi Mumbai, India.Tariku, Y.A., 2015a. TEC prediction performance of the IRI-2012 model

over Ethiopia during the rising phase of solar cycle 24 (2009–2011).Earth Planets Space 67, 35. https://doi.org/10.1186/s40623-015-0206-2.

Tariku, Y.A., 2015b. Patterns of GPS-TEC variation over low-latituderegions (African sector) during the deep solar minimum (2008 to 2009)and solar maximum (2012 to 2013) phases. Earth Planets Space 67, 35.https://doi.org/10.1186/s40623-015-0206-2.

Torrence, C., Compo, G.P.A., 1998. Practical guide to wavelet analysis.Bull. Am. Meteorol. Soc. 79 (1), 61–78.

Venkatesh, K., Fagundes, P.R., Seemala, G.K., de Jesus, R., de Abreu, A.J., Pillat, V.G., 2014. On the performance of the IRI-2012 andNeQuick2 models during the increasing phase of the unusual 24thsolar cycle in the Brazilian equatorial and low-latitude sectors. J.Geophys. Res. Space Phys. 119, 5087–5105. https://doi.org/10.1002/2014JA019960.

Wang, X., Sun, Q., Eastes, R., Reinisch, B., Valladares, C.E., 2008. Short-term relationship of total electron content with geomagnetic activity inequatorial regions. J. Geophys. Res. 113, A11308. https://doi.org/10.1029/2007JA012960.

Yizengaw, E., Zesta, E., Moldwin, M.B., Damtie, B., Mebrahtu, A.,Valladares, C.E., Pfaff, R.F., 2012. Longitudinal differences ofionospheric vertical density distribution and equatorial electrodynam-ics. J. Geophys. Res. 117, A07312. https://doi.org/10.1029/2011JA017454.








https://doi.org/10.1029/2012RS004981

https://doi.org/10.1029/2012RS004981

https://doi.org/10.1016/j.jastp.2013.04.016

https://doi.org/10.1002/2015RS005930







https://doi.org/10.4236/pos.2017.82002

https://doi.org/10.1029/93JA00502

https://doi.org/10.1029/93JA00502









https://doi.org/10.1186/s40623-015-0206-2

https://doi.org/10.1186/s40623-015-0206-2

https://doi.org/10.1186/s40623-015-0206-2



https://doi.org/10.1002/2014JA019960

https://doi.org/10.1002/2014JA019960

https://doi.org/10.1029/2007JA012960

https://doi.org/10.1029/2007JA012960

https://doi.org/10.1029/2011JA017454

https://doi.org/10.1029/2011JA017454

comparison of gps-tec measurements with nequick2 and …

Documents