**2. Ionospheric models for space weather**

#### **2.1 Classification of ionospheric models**

The ionospheric models are classified as empirical and analytical models (datadriven) and theoretical and parameterized models (physics-based models).

Empirical models are developed from statistical analysis of long-time datasets or combine existing electron distribution models or statistical and numerical analysis of different TEC measurements at local, regional, and global observation and monitoring stations. They define electron density profiles based on geomagnetic/ geographical position, solar and geomagnetic activity to produce seasonal and monthly average or median values at specific times and locations. These models mainly reproduce long-term ionospheric response/climatological conditions which depend on the solar activity epoch (diurnal, seasonal, and solar cycle). Their dependence on actual measurements is a challenge in regions with limited and nonuniform distribution of ground and regional TEC measurement and monitoring stations. That's why global empirical models give more representative and accurate TEC estimation in the northern hemisphere and the mid-latitude regions compared to the southern hemisphere [5, 6].

Analytical models are a family of three models (NeQuick, COSTprof, and the ionosphere plasmasphere model) developed from the combination of improved DGR model [7] and diffusive equilibrium models for topside ionosphere and plasmasphere [8]; these apply Epstein functions and spherical harmonics to analytically describe the electron density profiles. The most important analytical model for space weather monitoring is the NeQuick model [9]; this has been adopted by European Geostationary Navigation Overlay System (EGNOS) project for real-time ionospheric corrections of the Galileo single frequency users. In the NeQuick 2 model [9], and the NeQuick model (Galileo Specific) [10], the topside and bottom side parameters have been formulated with modification in electron density, height and thickness parameters and computer package. These models have shown improved performance in TEC estimation during geomagnetic storms when compared with the IRI-2016 model [11]. In addition, the NeQuick 2 model underestimates the topside electron density when compared with IRI-Plas model due to plasmaspheric electron contribution between the 2000 to 20000 km [12], though it performed better than IRI-Plas in TEC estimation from 2013 to 2018 [13]. Generally, the NeQuick model correctly estimates the trend of GPS-TEC during quiet and storm days however the accuracy reduces during geomagnetic storms.

Theoretical models apply first principles to solve an interconnected set of equations including Boltzmann, energy, momentum, and continuity equations. They describe the state of ionospheric plasma along magnetic field lines in phase space

with boundary conditions and coupling mechanisms at the lower and upper limits. Magnetospheric, ionospheric, and thermospheric inputs parameters are provided by already developed empirical models for example MSIS-86 [14], HWM-91, or HWM-93 [15], Chiu and IRI empirical models. The accuracy of theoretical models depends on the accuracy, reliability, and representativeness of the empirical models used, the boundary conditions, the fundamental physical and chemical processes, assumptions taken, and the accuracy of numerical schemes. The biggest challenge with the theoretical models is the long time required for processing and computation of the model results which makes it hard to be adopted in the normal operational setting. However, physics-based models provide a more detailed study and explanation of physical and chemical processes about the behavior of ionospheric plasma and coupling mechanisms under different space weather conditions.

Parameterized models [16] were developed by US Air Force Research Laboratories and describe the climatological conditions of the ionosphere by producing electron density profiles, critical frequencies, and heights of E and F ionospheric regions from 90 to 25000 km and TEC. Parameterized models produce climatological conditions from a summarized output of theoretical models for varying geophysical conditions; this is more representative than average ionospheric conditions by empirical models. And hence the driving forces of geophysical conditions associated with winds and wave motion in the lower thermosphere to ionospheric density and TEC variation can be represented.

#### **2.2 TEC and electron density models**

Ionospheric models give average modeled TEC and electron density values in space and time, these may deviate from the actual measured values from ionosondes, radars, and GPS measurements by 20–40% during quiet times and can go high during disturbed and geomagnetic conditions. The most efficient models for space weather monitoring must be able to define electron density redistribution caused by both spatial and temporal ionospheric variations. The earliest primary objective behind the development of ionospheric models was to estimate correct ionospheric delay for singlefrequency GPS users however, several modifications and improvements with longtime modeled data have enhanced the model's capacity for space weather monitoring and prediction. Their performance in monitoring space weather-induced electron density variations depends on several factors like latitude, altitude, and space weather conditions: hence the development of various regional and altitude-specific models. Global ionospheric models combine several regional and altitude-specific models to give TEC averages globally however their challenge is limited performance on extreme space weather conditions and induced ionospheric plasma irregularities discussed later in Section 3.

Electron density profiles in models are described by ionospheric characteristics, the most important for TEC calculation are peak density (NmF2), critical frequency (foF2), and peak height (hmF2) at the F2 layer where there is maximum electron density. The CCIR and ITU-R maps are standard models for ionospheric parameters set by the International Radio for Consultative Committee (CCIR) and the International Telecommunication Union (ITU). Currently, CCIR maps provide peak parameters for most empirical and theoretical ionospheric models for prediction.

The earlier ionospheric parabolic layer models [17–20] utilized parabolic functions to describe electron density profiles; however, these were challenged by the discontinuities between F1 and E regions, this was later improved by Dudeney [21] by employing trigonometric functions to represent the E-valley (**Figure 2b**).

*Ionospheric Electron Density and Electron Content Models for Space Weather Monitoring DOI: http://dx.doi.org/10.5772/intechopen.103079*

Chapman layer models utilize the Chapman functions to estimate the rate of photoionization with consideration of one neutral gas species and ignoring transport processes (**Figure 2a**). The neutral density at any height *N h*( ) (Eq. (1)) is determined by the shape of the layer with scale height (*H*) and peak height (*hm*).

$$N(h) = N(h\_m) \exp\left(\mathbf{k}\left[\mathbf{1} - \frac{h - h\_m}{H} - \exp\left(-\frac{h - h\_m}{H}\right)\right]\right) \tag{1}$$

The constant 0.5 1 ≤ ≤ *k* has been derived empirically in comparison with TEC; this has redefined the height parameters of the Chapman functions representing E, F1, and F2 layers. Distinctively, the α -Chapman layer occurs at *k* = 0.5 while the β-Chapman layer at *k* = 1 [22].

Refer to **Table 1** for details of empirical and semi-empirical electron density and TEC models.

#### *2.2.1 The international reference ionosphere (IRI) model*

This is the only standard ionospheric empirical model that takes data from all available data sources. The IRI density profile is based on a combination of global and regional models for D, E, F1 & F2 ionospheric layers with their characteristic parameters merged by mathematical functions. The IRI working group composed of the International Scientific Community, the Committee on Space Research (COSPAR), and the International Union of Radio Science (URSI) updates the model yearly with new data from a network of data sources to validate and improve the model performance. In addition to TEC and electron density, the model describes electron and ion temperatures, and ion composition from an altitude of 50–2000 km at any given time

#### **Figure 2.**

*The representation of electron density models, (a) is a Chapman layer model modified after [22] and (b) is a parabolic layer model modified after [20]. NmE, NmF1, and NmF2 represent peak densities of E, F1, and F2 layers respectively, and hmE, hmF1, and hmF2 represent peak heights of E, F1, and F2 respectively.*


**Table 1.**

*A table of empirical and semi-empirical ionospheric models for TEC and electron density.*

and location. The input parameters to the model represent solar activity, ionospheric index, and magnetic activity and over the years, the model has been updated to the current IRI-2016 [25].

Modeling the topside profile of the electron density has been an existing challenge in the IRI model. This IRI topside has been formulated with a correction factor and the NeQuick model was adopted as a default topside model up to approximately 2000 km [6]. However, this is still not comparable to the altitude approximation of

#### *Ionospheric Electron Density and Electron Content Models for Space Weather Monitoring DOI: http://dx.doi.org/10.5772/intechopen.103079*

GPS-TEC at about 20200 km (geostationary orbit altitude). In the latest update of the IRI model (IRI-2016) [25], the representation of the topside ion densities during very low and high solar activity has been improved and the computer program updated to turn IRI into a real-time model for space weather forecasting. The extension of the IRI model to the IRI plasmasphere model (IRI-Plas) [32] has been proposed to improve the representation of the plasmaspheric electron contribution. However, model results have shown overestimation when compared with actual measurements from GPS and ionosondes on varying space weather conditions [33, 34]. This, therefore, suggests that the IRI-Plas extension needs further improvement and update to correctly estimate the plasmaspheric electron content with response to dynamic space weather. Despite the regular updates and improvements, limited data contribution to yearly IRI updates from a limited number of ionospheric monitoring stations in African region (**Figure 1**) in addition to the equatorial anomaly irregularities is the most cause of significant bias between IRI modeled results and experimental GPS-TEC [35]. NeQuick being a 3-dimensional and with IRI models are the only recommended models for calculation of TEC by Recommendation ITU-R 218/3.

#### *2.2.2 Comparison of IRI modeled results with satellite measurements*

The recent expansion of satellite missions with GPS receivers onboard for precise orbit determination and radio occultation has improved the understanding of the topside ionosphere. This has been important for continuous validation and improvement of empirical models to accurately predict TEC and electron density variations for different space weather phenomena [36, 37]. The comparison of satellite measurements with models is based on error assessment using root mean square, absolute and relative values, and other statistical parameters. Previous studies found a consistent underestimation of TEC recorded by a dual-frequency altimeter on a TOPEX satellite at an altitude of 1336 km by the IRI model mostly occurring during high solar activity, at the equatorial anomaly, and the high latitude regions [36, 38]. The significant error comes from the systematic technique of TEC measurement from the altimeter and the absence of experimental territory results over ocean surfaces for IRI validation [39]. The orbit altitude over which TOPEX TEC is approximated is 1336 km which is less than 2000 km for IRI, this is also a possible cause of discrepancy between the two TEC values.

Further inconsistencies between IRI modeled and SWARM Langmuir probe electron density values are observed by Singh et al. [40] and Pignalberi et al. [37]. SWARM is a constellation of 3 satellites each with a GPS receiver, A and C fly at 460 km altitude and B at an altitude of 540 km. The in-situ electron density and TEC measurements give a more detailed description of the topside ionosphere by simultaneous measurements at different altitudes. This has been used in modeling and improving the topside formulation for the validation of empirical models [41]. In addition, the comparison of IRI modeled values with the satellite data depend on the selected input model parameters. Arıkan, et al. [42] compared the IRI-2016 F2 Layer Model Parameters with Ionosonde Measurements. He noted that the "ON" storm model improves the performance of the IRI during severe storm times. In addition, the CCIR option for peak F2 electron density produced more accurate modeled results in the low latitudes and the southern hemisphere while the URSI option is recommended for the northern hemisphere and mid-latitudes. Also, the IRI model with the NeQuick topside improved its performance when compared with the TOPEX data [43].

## *2.2.3 The coupled thermosphere ionosphere plasmasphere electrodynamics model (CTIPe)*

CTIPe model [44] combines the global thermosphere model [45], high latitude ionosphere model [46], low and mid-latitude ionosphere/plasmasphere model [47] and an extension of the electrodynamic circulation [48]. The model inputs include F10.7 averaged over 41 days for ionospheric, heating, and dissociation processes, high latitude electric potential patterns generated by 1 minute average from ACE satellite, solar wind density, solar wind speed, interplanetary magnetic field (IMF) in the XYZ planes, tilt, and total magnetic field provided by the Space Weather Prediction Centre (SWPC) real-time databases; these are coupled with the Weimer model to produce particle and auroral precipitation patterns [49]. The model runs 30 minutes ahead of real-time with a 10-minute interval of updated results and has been adopted at the SWPC by NOAA to study the thermosphere-ionosphere conditions for space weather nowcasting and forecasting purposes.

The model simulations have reproduced ionospheric thermosphere conditions of TEC and electron density during quiet and storm periods [50, 51]. The discrepancies observed when the simulated results are compared with experimental data result from non-representation of ionospheric complex electrodynamic processes associated with plasma transport and prompt penetration electric fields (PPEF) [51] and delayed ionospheric response to forcing parameters related to space weather conditions [50, 52]. From CTIPe simulations, Vaishnav et al. [53] revealed that the ionospheric delay to solar flux changes increases with reduced eddy diffusion. This effect is more observable in the mid and low-latitude regions due to stronger solar activity and EUV acting as the forcing parameters for ionization processes. Therefore, in addition to not reliable results and uneasy accessibility for model runs with longer processing times, the estimation of correct and accurate forcing parameters also significantly affects the CTIPe modeled electron density and TEC results.

## **3. Special features of ionospheric variability for space weather**

Ionospheric models monitor the quite long-term ionospheric (climatological) conditions occurring repeatedly and solely dependent on the sun's activity. These represent reference ionospheric conditions of electron density and TEC of the upper atmosphere influenced by diurnal, seasonal, latitudinal, local time, and solar cycle. Ionospheric storms and associated effects disturb quite ionospheric conditions with sudden changes in electron density and TEC, that is negative and positive storms [54]. The variability of the lower ionospheric (D, E & F1) regions is unique and complex to understand since its variability is influenced by forcing from the lower atmospheric and tidal processes in addition to solar activity. The latitude location adversely affects the distribution of electron density and TEC with changes in thermospheric composition and neutral winds induced by direct solar wind coupling at the high latitudes and the equatorial anomaly at the low latitudes.

#### **3.1 Ionospheric storms**

The strength of an ionospheric storm can be defined by changes in TEC and electron density while geomagnetic indices define the intensity of the geomagnetic storm. However, there is not necessarily a strong positive correlation between TEC

#### *Ionospheric Electron Density and Electron Content Models for Space Weather Monitoring DOI: http://dx.doi.org/10.5772/intechopen.103079*

and geomagnetic activity indices. Positive storms are more observed during the main phase while negative storms occur regularly during recovery; this is observable at all latitudes [55]. Latitudinal location determines ionospheric storm response effects (**Figure 3b**) with the penetration of magnetospheric electric fields from high to low latitudes; this causes wind perturbations and thermospheric composition changes associated with *ExB* drift and plasma transport processes. The result is positive storms observed at mid and low latitudes while the negative storms at the geomagnetic equator at storm onset [54, 56].

Physics-based models have shown tremendous progress in modeling ionospheric storm response effects, the Coupled Magnetosphere Ionosphere Thermosphere Model (CMIT) simulated the electron density and TEC for three storms showing positive and negative storms and thermospheric composition changes [56], (CMIT is a physics-based model coupling Lyon-Fedder-Mobarry global magnetosphere MHD code [57] with Thermosphere-Ionosphere- Electrodynamics General Circulation Model (TIE-GCM) [48], CTIPe simulations on storm day (20th November 2003) produced storm onset features, TEC and neutral density changes [58], GGCM-CTIM produced localized effects of auroral precipitation and electric fields for strong Joule heating, thermospheric upwelling, and neutral density enhancement on storm day (August 24, 2005) [59] (GGCM is Open Global General Circulation Model which calculates, 3-D global magnetosphere and 2-D high-latitude ionosphere solving resistive MHD equations with solar wind input). A more coupled model RCM-CTIPe which considers the high latitude field-aligned currents describes the electrodynamic coupling to reproduce the behavior of storm time electric field [60], Rice Convection Model (RCM)–CTIPe Model is a self-consistency that describes the electrodynamic coupling between magnetosphere, thermosphere, ionosphere and plasmasphere

#### **Figure 3.**

*Comparison of GPS-TEC at different latitudes and different storm and quiet days in 2015. Figure a and b shows the TEC variation for the 4 stations during the quietest and disturbed day of each month in 2015 respectively. The unmarked line is for ffmj (50.091°, 8.665°) mid-latitude, the square marked line is mas1 (27.764°, −15.633°) low latitude, the star marked line is for nklg (0.354°, 9.672°) the equatorial anomaly station and the line marked by 1 is nyal (78.930°, 11.865°) a high latitude station. The stations were selected within the same longitudinal band between 8° to 15° to avoid local time effects. The quietest and most disturbed days were got from the international list of quiet and disturbed days at the WDC for geomagnetism, Kyoto (http://wdc.kugi.kyoto-u.ac.jp/qddays/ index.html). The RINEX files were got from CDDIS (https://cddis.nasa.gov/Data\_and\_Derived\_Products/ GNSS/daily\_30second\_data.html) and processed using GOPI software (https://seemala.blogspot.com/).*

systems. It couples three models together, that is CTIPe [61], RCM [62], and TIE-GCM [48]. From this point of view, the self-consistency theoretical models coupling the magnetosphere, plasmasphere, ionosphere, and thermosphere systems are more dependable for reproducing storm ionospheric effects in electron density and TEC variations. However, empirical studies of storm time modeling depend on predicting the behavior of the foF2 parameter during storm periods. The storm time ionospheric correction model developed by Araujo-Pradere et al. [63] was incorporated in IRI-2000 [64]. However, the percentage deviation between experimental and IRI-2007 predicted foF2 was 100% (high) when evaluated at low latitudes [65]. Therefore, it's difficult to represent the complex dynamic processes in the data-driven models with numerical and statistical methods.

### **3.2 Latitudinal features**

#### *3.2.1 Electron density and TEC at high latitudes*

The plasma and density redistribution at high latitudes are density depletion zones (**Figure 3**), aurora ovals, and polar cap patches. These features are due to energetic particle precipitation, open magnetic field lines, direct IMF, solar wind coupling, and induced polar cap convection [66]. TDIM, a physics-based mid-latitude model [67] was modified to include high latitude ionospheric effects of particle precipitation, electric field convection, plasma transport processes, electron density, and electron temperatures [68]. The IRAP plasmasphere-ionosphere model (IPIM), a 3-D multifluid mathematical model for interhemispheric studies [69] showed that electron density depletion results from a decrease of neutral atomic oxygen concentration and <sup>2</sup> *O N*/ ratio; these results also closely match with the ionosonde and EISCAT radar data [70]. Theoretical models generally use statistic inputs that define precipitation, convection, and conductance rates to reproduce the complex mechanisms occurring in the auroral regions. The recommendation by the COSPAR Scientific Assembly [71] was implemented in 2013 where auroral boundaries and storm time model for aurora regions were integrated into the IRI model [72]. This has improved the performance of the IRI model in predicting the electron density and TEC over the high latitude regions [73].

#### *3.2.2 Low latitude effects on TEC and electron density*

The main features of interest in the low latitudes are the equatorial ionization anomaly (EIA) and the equatorial plasma fountain (EPF), these impact TEC and electron density variation in the F region. A recent review [74] clears the misinterpretation regarding the development and formation of the two related low latitude phenomena. The EIA removes plasma from the magnetic equator by the upward *ExB* drift to form a trough at the magnetic equator and ionization crests on either side of magnetic latitudes about ±15°-30°; this varies from quiet to disturbed conditions (**Figure 3**). This phenomenon is controlled by the equatorial electrojet (EEJ) which causes the fountain effect during quiet times. The EPF determines the resultant direction of plasma flow because of field-aligned plasma diffusion and field perpendicular *ExB* plasma drift. During disturbed days, the EIA becomes strong because of storm time equatorward wind and eastward (westward) PPEF during the main (recovery) phase associated with <sup>2</sup> *O N*/ composition changes. Using multi-instrument and long-term monitoring observations, the strength and asymmetry of the

#### *Ionospheric Electron Density and Electron Content Models for Space Weather Monitoring DOI: http://dx.doi.org/10.5772/intechopen.103079*

anomaly, time of formation, and crest to trough ratio parameters display seasonal and diurnal variability from quiet to active solar activity and geomagnetic storms [75, 76] and references therein. Theoretical models are superior in representing the equatorial anomaly features; SUPIM [77] incorporates *ExB* drift and neutral wind velocities to produce the plasma fountain (reverse effect) during upward drift (downward drift) and through this, the EIA structures have been modeled during both quiet and active solar periods [74], Bhuyan et al. [78] reproduced electron density and temperature distribution with reasonable accuracy and the EIA phenomena around the Indian equatorial and low latitude regions using theoretical simulation. Several data-driven models for the low latitude have been developed [79, 80] and their results compared with IRI and actual measurements, this is important for continuous validation and improvement of the performance of global and regional ionospheric models.

Equatorial plasma bubbles (EPB) are plasma density fluctuations occurring when the bottom side F layer is raised above the upper limit which is dependent on solar activity, the major drivers are *ExB* plasma drift caused by pre-reversal enhancement in addition to Rayleigh–Taylor instability, gravity waves, neutral winds, and electric field fluctuations [74]. Recent modeling efforts include data-driven and numerical simulation models [81, 82], these are important to relate the nonlinear occurrence of EPBs not easily studied through simulation by theoretical models. Most numerical models cannot predict the day-to-day variability of EPBs; therefore, a need to add real-time assimilation methods of data and integrate the combined models into the global ionosphere and thermosphere electron density models for the prediction of such phenomena.

#### **3.3 Plasma irregularities at low altitudes**

#### *3.3.1 Traveling ionospheric disturbances (TIDs)*

The dynamic change in thermal neutral density within the mesosphere-lower thermosphere (MLT) altitudes depends strongly on low atmospheric forcing associated with internal gravity wave propagation rather than solar and geomagnetic activity [83, 84]. TIDs are simply gravity waves (GWs) signatures resulting from the interaction of GWs with ionospheric plasma traveling at ionospheric altitudes. Traveling atmospheric disturbances (TADs) also manifest with modification of neutral and electron density in the low altitude ionosphere. These are initiated by fluctuating aurora electrojet and or particle precipitation from enhanced heating during storm periods and their intensity varies with the solar cycle and geomagnetic activity [84]. The traveling disturbances exhibit a very complex nature of waves mixed with plasma instabilities associated with propagation characteristics and corresponding electron densities which requires classification based on wavelengths, magnitude, direction, seasonal and diurnal factors. The long-time monitoring and observations by ground and space-based instruments have supported the development and validation of empirical models [85] and semi-empirical models [86, 87] to predict spatial and temporal dynamics of TID parameters at the MLT. However, much work needs to be done regarding the associated density variations to integrate them into the global TEC models for space weather forecasting.

#### *3.3.2 Sporadic E layers (Es)*

The irregular and dynamic layers of compressed and enhanced ionization form below the F region and descend through the E-F valley region as observed by ionosonde and ISR. They are associated with electron density fluctuations generated from the combined effects of sheer wind mechanisms and electric field structures of the earth's magnetic field. While there is no significant correlation with geomagnetic activity, it's important to statistically understand their occurrence mechanisms and origin to formulate correct inputs for empirical prediction models. The Ground-totopside model of Atmosphere and Ionosphere for Aeronomy (GAIA) has been used for numerical simulation and prediction of Es events using the horizontal wind model (HWM14) [88]. In addition to more recent efforts for numerical and datadriven models, the statistic trend of the critical frequency of Es layer (foEs) has been studied; this has shown diurnal, seasonal, geographical dependence with spatial and temporal variations [88, 89] and references therein.

### **4. Deep learning and neural networks**

Artificial Neural Networks (ANN) have become dependable methods for modeling ionospheric variability to predict space weather and associated effects. This is because they can solve the high and complex nonlinearity involved in the magnetosphere-ionosphere-thermosphere coupling processes without the knowledge of physical, chemical, and coupling processes and be used for forecasting. The ANNs are data-driven models with massive parallel distributed processing units (neurons) orderly arranged in layers that learn nonlinear patterns (inputs) aided by a learning algorithm with logical adjustment of weights and bias. The input parameters are trained with varying numbers of hidden layers and neurons, training algorithms, and training functions till they have generalized well to a required output as TEC or electron density, refer to [90, 91] for details in computation techniques in neural networks.

Generally, the input parameters are selected based on their effect on the ionospheric variation, that is spatial and temporal variation represented by geographical positions, date and time, geomagnetic activity represented by Dst, Kp, Ap, and SYM/H indices, solar activity represented by F10.7 index, sunspot number, and proton flux and EUV flux. While CMEs and solar flares enhance the geomagnetic activity causing geomagnetic storms, the intensity of the storm activity depends on the state of the interplanetary field described by solar wind parameters [92]. Several studies have also revealed that there is a positive correlation between TEC and solar wind parameters mostly during storm days [93]. Therefore, other input parameters considered include IMF in the XYZ plane, total magnetic field, solar wind speed, temperature, density, and electric field. The specific inputs are selected for specific NN models based on the purpose and or type of the model (regional or single station model, long time or short time prediction model, storm time models). Recently, the addition of IRI's parameters related to TEC and electron density variation like peak density (NmF2), critical frequency (foF2), and peak height (hmF2) at F2 has proved to improve the capacity of NN models to learn long term trends of solar cycle variations [94, 95].

Neural networks have been applied for other special cases of modeling for example prediction of topside electron content using SWARM data [96], topside ionosphere, and plasmasphere model for electron density and TEC [97], bottom side electron density profile model over Grahamstown, South Africa [98]. With the progressive development of computation techniques and the availability of extensive datasets for space weather and ionospheric parameters, researchers are now able to develop

*Ionospheric Electron Density and Electron Content Models for Space Weather Monitoring DOI: http://dx.doi.org/10.5772/intechopen.103079*

more reliable and accurate NN models that can capture most spatial and temporal TEC and electron density variations. And for this reason, NN models are flexible and have become reference points to validate and improve the existing global and regional empirical models. However, the development of such comprehensive NN models requires very large datasets which may not be available in certain areas and at certain times mostly in African equatorial regions, strong knowledge of computational techniques, and long times for computational processing.

## **5. Summary and conclusion**

We reviewed the development of ionospheric models and discussed several improvements and updates made. Throughout the chapter, we emphasized the need for continuous validation of ionospheric models with updated datasets. Theoretical modeling has improved from 1-D to 3-D representing the M-I-T coupling, physical and chemical processes occurring during special space weather conditions. This is important in explaining and understanding the energy transfer mechanisms from the solar wind to the earth's atmosphere during storms, the effect of wind and thermospheric composition on ionospheric TEC and electron density variations, and the effect of forcing from the lower atmosphere. Empirical models have been improved to globally predict the climatological condition of TEC and electron density like IRI. The wide network of ground monitoring and observation stations with space-based measurements has provided extended databases for space weather and ionospheric parameters. This has supported the development of reliable and accurate neural network models for prediction. The role of International and national Organizations like COSPAR, URSI, ITU, CCIR, and research and education institutions has been emphasized in supporting and maintaining ground stations for ionospheric observation. These measurements have become backbones for continuous validation and improvement of ionospheric models. However, it's important to note that some regions still have a sparse and non-uniform distribution of ground monitoring stations which is a challenge for developing representative data-driven models.
