Comparison of IRI _ PLAS and IRI _ 2012 model predictions with GPS _ TEC measurements in different latitude regions

The International Reference Ionosphere (IRI) is an empirical model for providing ionospheric parameters, including Total Electron Content (TEC), electron density, electron and ion temperature etc., in the altitude range from 50 km to 2000 km. Since the IRI model is limited up to 2000 km, IRI_PLAS model, plasmasphere extension of the IRI model, was proposed by the researchers. This paper investigates the TEC prediction performance of IRI_PLAS and IRI_2012 models by comparing GPS TEC data, in different latitude regions for magnetically active and quiet days. TEC data over 9 International GNSS Service (IGS) stations, located in different latitude regions, are used for the comparison. Evaluation of the diurnal results reveals good agreement with correlation coefficient >0.9 between GPS_TEC and empirical models for the quiet day irrespectively of the latitudinal data used. However, while the differences are not relatively large in most part of the active days, they reach high level, above 30 TECu, in some parts of the days.


Introduction
The ionosphere is a part of the Earth's atmospheric region where the enough ionized molecules and free electron density affect the propagation of radio frequency electromagnetic waves.Since they are much lighter than the free ions, free electrons mostly affect the propagation.Thus ionospheric total electron content (TEC) is of great importance for many study areas including space based observation systems, communication systems and space weather studies [Liu and Gao 2004].Ionospheric free electron density changes because of regular (diurnal period of earth, season etc.) and irregular (ionospheric and geomagnetic storms, traveling ionospheric disturbances (TIDs) etc.) variations [Spoelstra 1996].The widespread effect of the ionosphere on various areas and regular-irregular changes in the structure of the ionosphere have made the ionospheric studies popular subject in recent years [Adebiyi et al. 2016, Paul et al. 2016].In order to examine the structure and behavior of the ionosphere, continuous monitoring is crucial.For doing this, empirical models and GPS based TEC measurements are used.Empirical models can provide reliable simulation data for effective ionospheric study and forecasting [Akala et al. 2015].For this purpose, several empirical models like International Reference Ionosphere (IRI) [Bilitza 2001], Parameterized Ionospheric model (PIM) [Daniell et al. 1995], Parameterized, real time ionospheric specification model (PRISM) [Daniel and Brown 1995], Semi-Empirical Low Latitude Ionospheric Model (SLIM) [Anderson et al. 1987], Field Line Inter Hemispheric Plasma Model (FLIP) [Scali et al. 1997], Thermosphere-Ionosphere General Circulation Model (TIGCM) [Emery et al. 1996], Utah State University Global Assimilation of Ionospheric Measurements (USA_GAIM) [Scherliess et al. 2006], Bent model [Bent and Llewelly 1973], NeQuick model [Alcay et al. 2014], Sheffield University Plasmasphere Ionosphere Model (SUPIM) [Bailey et al. 1997] have been developed.Among these models, IRI which is regularly being improved and updated, is being widely used.Many studies have focused on the performance of IRI_2012 model [Leong et al. 2015, Tariku 2015, Rice and Sojka 2015, Kumar et al. 2014 and many others].Historical comparison of IRI and early ionograms are given in Rice and Sojka [2015].Tariku [2015] discussed the performance of the latest version of the IRI [IRI_2012] model for estimating the vertical total electron content (VTEC) variation over Ethiopian regions during the rising phase of solar cycle 24 (2009)(2010)(2011).According to the results, although the model overestimates VTEC values for most hours, it generally performs well in estimating diurnal VTEC values, particularly just after the midnight hours (0 UT-3 UT).
Since the IRI model specifies the ionosphere only up to 2000 km, it is necessary to extrapolate the ionosphere to the higher altitudes.For this reason, researchers proposed IRI_PLAS model, one of the possible candidate model for plasmasphere extension of IRI model [Gulyaeva and Bilitza 2012, Zakharenkova et al. 2015, Arikan et al 2015].Arikan et al. [2015] introduced a user friendly space weather service (URL-1), providing many output products.One of them is Total Electron Content in TECu including heights from 80 km to HPL (plasma-pause height, typically 20200 km).
Besides the empirical models, Global Positioning System (GPS) observations can be used to estimate GPS derived ionospheric TEC data (GPS_TEC).Because of the global coverage of the GPS, GPS_TEC is a good indicator of the geographical distribution of the ionization and proposed as an input for Ionosphere models [Misra and Enge 2006].Since the GPS satellites are located at the altitude of 20200 km, the amount of free electrons along the GPS ray path is composed mainly of Ionospheric Electron Content (IEC) and partly of Plasmaspheric Electron Content (PEC) [Balan et al. 2002, Cherniak et al. 2012, Karia et al. 2015, Akala et al. 2015].Zakharenkova et al. [2015] compared TEC values, computed using IRI_2012 and IRI_PLAS models with GPS based TEC data, derived from European mid-latitude GPS station Potsdam.According to the results, comparative data model analysis does not reveal good performance.Arikan et al. [2015] examined the performance of IRI_PLAS map for the magnetically active day by comparing Global Ionosphere Map (GIMs).It is observed that the global distribution of TEC is very different due to the effect of geomagnetic disturbances.
Unlike other studies, this study examines the TEC prediction performance of IRI_PLAS model in different part of the world.This paper is divided into four sections.Following the introduction, the next section provides brief overview of the GPS_TEC, IRI_2012 and IRI_PLAS models.Then, a detailed presentation of the results derived from measurements and empirical models are given.Finally, the last section summarizes key conclusions.

Model Calculations 2.1.1. IRI_2012
The International Reference Ionosphere (IRI) is an international project recommended by the Committee on Space Research (COSPAR) and the Inter-national Union of Radio Science (URSI) to provide ionospheric parameters.The first version of IRI was released in 1978 and several steadily improved editions followed since then (1986, 1990, 1995, 2001, 2007, 2012 and 2016) (URL-2, Alcay et al, 2014).Although the fortran source code of 2016 is available (URL-2), the web interface has not been published yet in the preparation of this study.Therefore IRI_2012 model was used in this study.The IRI_2012 was released in 2013 and includes several important improvements and additions which is given in Bilitza [2015] in detail.The IRI model provides many parameters, including Total Electron Content (TEC) electron density, ion and electron temperature, ion composition for a given latitude, longitude, time date at altitudes ranging from 60 km to 2000 km.IRI based TEC data is derived by integrating the electron density profile from the lower boundary to the user specified upper boundary [Kumar et al. 2015].In this study, in order to compute IRI_2012 TEC values, a web interface from the IRI homepage (URL-3) was used.Relevant to this study, NeQuick and ABT-2009 options were used for topside electron density and bottomside thickness respectively.Since the CCIR option is recommended for continental areas and URSI option over the ocean areas [Aggarwal 2011], we used CCIR option for the F2 peak density calculations.

IRI_PLAS
Since the IRI model specifies the ionosphere only up to 2000 km, increasing number of the researchers focused on the information about the plasma conditions above the ionosphere in the plasmasphere.The international reference ionosphere extended to the plasmasphere, IRI_PLAS, [Gulyaeva et al. 2002] has been proposed an empirical model where the region of interest includes the plasmasphere up to 20200 km [Gulyaeva et al. 2011].A number of approaches have been proposed for extending IRI to the plasmasphere, including The Global Core Plasma Model (GCPM-2000), The Global Plasmasphere Ionosphere Density (GPID), The IMAGE/RPI plasmasphere model and The IZMIRAN plasmasphere model.For the details of such models, authors refer to Gulyaeva and Bilitza [2012].IONOLAB group presented a monitoring of space weather service that is available online at (URL-1) [Arikan et al. 2015].For a given date, time and location, this service provides many parameters, including IRI_PLAS TEC data.IONOLAB service provides IRI_PLAS data associated with the IZMIRAN plasmasphere model.IRI_PLAS source and executables are available at (URL-4).In this study, TEC values corre-sponding to IRI_PLAS model were estimated using IONOLAB service.

GPS_TEC
A dual frequency GPS receiver can provide code and phase observations.The geometry free linear combination of GPS signals (ionospheric observable) is classically used for ionospheric investigation which is generated by subtracting simultaneous pseudorange or carrier phase observations [Nohutcu et al. 2010].Using such observations, slant TEC values are generated for each satellite and then they are converted VTEC values using single layer model and mapping function as explained in Arikan et al. [2003], Schaer [1999].In this study, GPS TEC values over the IGS stations were estimated using ionolabtecv1.0,downloaded from the IONOLAB service.Using this program, STEC values are converted into VTEC values using the mapping function, provided in Arikan et al. [2003] and Nayir et al. [2007], for every position of the satellite with a 30s time resolution.Then, in order to obtain an accurate and robust estimate of the TEC values in the zenith direction of the GPS receiver combining the computed VTEC data from all available satellites, Reg-Est (Regularized Estimation of TEC) algorithm is applied as discussed in Arikan et al. [2003Arikan et al. [ , 2004] ] and Nayir et al. [2007].

Results and Discussion
The TEC values from the GPS, IRI_2012 and IRI_ PLAS models over the 9 IGS stations (Figure 1), located in different regions, were computed and compared.
In addition, Table 1 illustrates the details of the stations including geographical coordinates, receiver and antenna information, etc.
Although the performance of IRI models is limited for the active days since they provide monthly average of the parameters, in order to determine the level of this deficiency, both active (18.03.2015, 20.12.2015, 21.12.2015) and quiet days (02.06.2015, 27.09.2015, 27.10.2015)data and the GPS_TEC data generally reach highest level, respectively.TEC values corresponding to northern hemisphere mid-latitude stations, BARH, MADR, DAEJ are provided in Figures 7, 8 and 9.During the quiet days, IRI_2012 TEC data are generally closer to the GPS_TEC data except 02.06.2015.Particularly during post-noon hours discrepancies between IRI_ PLAS TEC and GPS_TEC data reach largest level.On the active days, although GPS_TEC and models derived TEC data generally show similar trend for the particular hours, differences between them are at the high level.For instance, GPS_TEC data corresponding to BARH station on 20.12.2015 during 17-20 UT, reaching maximum 40 TECu level due to the increasing activity level as depicted in Kp values (Figure 3).However, while such influence can be observed slightly over MADR stations, any increase is not detected over DAEJ during this time period.Besides, although both GPS and empirical models derived TEC data exhibit similar trend over BARH and MADR on 18.03.2015,TEC differences corresponding to DAEJ station during post mid-night periods are very high.During such period, while GPS_TEC data are 15 TECu level, maximum values of IRI_2012 and IRI_PLAS TEC are 43 and 38 TECu respectively.
Similar to the northern hemisphere stations, TEC data corresponding to quiet days in the southern hemisphere high latitude stations (PARC, HRAO and YARR), exhibit similar behavior (Figures 10,11        12).Moreover, while IRI_PLAS TEC data are generally consistent to the GPS TEC data on 27.10.2015for PARC and HRAO stations, IRI_2012 TEC data are closer to the GPS_TEC data on 27.09.2015for DAEJ and HRAO stations.Meanwhile except the northern hemisphere stations, the effect of the activity is apparent clearly in terms of TEC increase over the southern hemisphere stations.GPS_TEC data overestimate the models based TEC data on 18.03.2015during post-midnight period over PARC stations and on 20.12.2015 over PARC and YARR stations during post-sunset and post-midnight period, respectively.
Since the days of equinoxes 18.03.2015and 27.09.2015refer to the equivalent conditions at all stations, similar trend is observed.However, due to the influence of the geomagnetic conditions, GPS_ TEC values show sudden increase and decrease,  12).However, at other southern hemisphere stations (PARC, HRAO), while all three results show similar trend, differences between GPS_TEC and IRI_PLAS TEC are above 8 TECu in some part of the day.In addition, particularly for the HRAO, IRI_2012 results are mostly close to the GPS_TEC data.Comparing to the PARC and HRAO, differences of mid-latitude stations are at lower level.Days 20.12.2015 and 21.12.2015refer to the same magnetic storm with the 1st day (20.12.2015) during the main phase of the storm and the 2nd day (21.12.2015) at the recovering phase of the storm.The influence of the storm on 20.12.2015 is apparent at all stations however at different level.Particularly, the influences of the storm are observed just after midnight hours (0 UT -3 UT) and after 12 UT.Such condition is more clear at GPS_TEC data corresponding to BARH and PARC after 15 UT and YARR during morning hours.In addition, during the recovering phase of the storm (21.12.2015), the influence of the storm on TEC values are not clear except PARC and HRAO.Particularly, influence of the increasing level of the storm after 12 UT on 20.12.2015 and its decreasing level during 0 UT -3 UT on 21.12.2015 is observed clearly at PARC station (Figure 10).
In addition, one of the quiet days (02.06.2015) refers to summer conditions in the northern hemisphere and winter conditions in the southern hemisphere.It is interesting to note that GPS_TEC data overestimate the model based data at all northern hemisphere stations.Particularly in some part of the WHIT, differences are above 8 TECu.However GPS_TEC data is below the models based data at the southern hemisphere stations.Such results prove the significant effect of the seasonal conditions on TEC data which is not able to reflected properly by the models.
In order to facilitate the comparison and obtain more quantitative estimation, some statistical values corresponding to differences between GPS_TEC and empirical models and also between empirical models, including minimum, maximum, range, mean and Std values are provided in Tables 2-10.According to the values given in tables, the superiority of the IRI_PLAS model is not apparent, comparing to the IRI_2012 model for the quiet days.For some quiet days and stations, while IRI_PLAS TEC data are closer to the GPS_TEC data, for others IRI_2012 TEC data show good agreement with GPS_TEC data.On the active days, although maximum differ-ences between GPS_TEC and empirical models are generally less than 10 TECu, they reach high values in some time periods due to the increase of the TEC data caused by the geomagnetic activity observed from the GPS measurements (Tables 5, 8 and 10).The most intense difference is observed between GPS_TEC and empirical models TEC on 18.03.2015for PARC station located at southern hemisphere high latitude.
In order to further examine the consistency and find out how the TEC estimated from empirical models are correlated with GPS_TEC, the correla- tion coefficients between such data sets were computed using the following equations: (1) (2) (3) Where GPS i is respective GPS_TEC data,GPS i is their mean, IRI_2012 i and IRI_PLAS i are respective IRI_2012 and IRI_PLAS TEC data, (IRI_2012 i ) and IRI_PLAS i are their mean values.Subscripts "i" denotes numerical position in the data.The correlation coefficients between GPS_TEC data and empirical models derived TEC data are depicted in Table 11.It is noticed that the correlation coefficients between GPS_TEC and empirical models exhibit high agreement which are generally above 0.90 for the quiet days at all station except TIXI.Although correlation coefficients are above 0.97 for TIXI on 27. 09.2015 and 27.10.2015, there is no correlation between such data sets on 02.06.2015.Although TEC differences between GPS_TEC and empirical models are not at the large level, less than 8 TECu, the reason of such correlation is due to the trends of data sets which are not consistent (Figure 6).While GPS_TEC data have decreasing trend during post noon hours, models derived TEC data have decreasing trend until 17 UT and then they have increasing trend.As expected, correlation coefficients between GPS_TEC and empirical models derived TEC data for the active days are at different level due to the activity level and its effects on TEC values which are detected by GPS based observations.On the other hand, there are good correlation coefficients between IRI_PLAS and IRI_2012 models, about 0.99 level, at all stations for both geomagnetic conditions.

Conclusion
In the present study, we compared TEC values from GPS measurements with corresponding TEC data from the IRI_PLAS and IRI_2012 models over the 9 IGS stations, located three different latitude regions.Comparison was applied for three geomagnetic active and quiet days.Although the drawbacks of empirical models are well known for the active days since they give monthly average values, in order to determine the magnitude of the differences, active days were considered in this study.According to the results, while maximum differences between GPS_ TEC and empirical models based TEC data are gen- erally less than 10 TECu level, in some part of the active day differences reach above this level particularly in the northern hemisphere mid-latitude and southern hemisphere high latitude stations.In addition maximum differences are observed over the stations located at southern hemisphere high latitude.However for the quiet day, GPS_TEC and empirical models derived TEC data are consistent and correlation coefficient between them are generally over 0.90.It is interesting to note that although IRI_PLAS model includes the plasmasphere part and better TEC prediction performance is expected comparing to the IRI_2012 model, the superiority of the IRI_PLAS model is not clear in the results.While IRI_PLAS model reveals good performance in TEC representation and mimics the GPS_TEC data at some stations, for others IRI_2012 TEC data are much closer to the observed GPS_TEC.Such conditions are not depend on the latitudinal location of the stations.In addition, the results corresponding to 02.06.2015 exhibit the influence of the seasonal conditions which cause the (GPS_TEC) -(Models_TEC) values positive in the northern hemisphere and negative in the southern hemisphere stations.

Figure 1 .
Figure 1.Location of the IGS stations used in the experiment.

Figure 2 .
Figure 2. Kp and Dst indice values for the quiet days.

Figure 3 .
Figure 3. Kp and Dst indice values for the active day.

Figure 4 .
Figure 4. TEC comparison over the northern hemisphere high-latitude station WHIT.

Figure 5 .
Figure 5. TEC comparison over the northern hemisphere high-latitude station KIRU.

Figure 6 .
Figure 6.TEC comparison over the northern hemisphere high-latitude station TIXI. and

Figure 7 .
Figure 7. TEC comparison over the northern hemisphere mid-latitude station BARH.

Figure 8 .
Figure 8. TEC comparison over the northern hemisphere mid-latitude station MADR.

Figure 9 .
Figure 9. TEC comparison over the northern hemisphere mid-latitude station DAEJ.

Figure 10 .
Figure 10.TEC comparison over the southern hemisphere high-latitude station PARC.

Figure 11 .
Figure 11.TEC comparison over the southern hemisphere high-latitude station HRAO.

Figure 12 .
Figure 12.TEC comparison over the southern hemisphere high-latitude station YARR.

Table 1 .
Details of the stations used in the experiment.

Table 2 .
Some statistical values corresponding to the differences of TEC values for WHIT.

Table 3 .
Some statistical values corresponding to the differences of TEC values for KIRU.
both TEC values and trend.Similar condition is derived at one of the southern hemisphere high latitude station (YARR), located at the eastern part of the region (Figure

Table 4 .
Some statistical values corresponding to the differences of TEC values for TIXI.

Table 5 .
Some statistical values corresponding to the differences of TEC values for BARH.

Table 6 .
Some statistical values corresponding to the differences of TEC values for MADR.

Table 7 .
Some statistical values corresponding to the differences of TEC values for DAEJ.

Table 8 .
Some statistical values corresponding to the differences of TEC values for PARC.

Table 9 .
Some statistical values corresponding to the differences of TEC values for HRAO.

Table 10 .
Some statistical values corresponding to the differences of TEC values for YARR.

Table 11 .
Correlation coefficients between GPS_TEC and empirical models based TEC values.