Evaluation of the IRI-2007 model options for the topside electron density
1. Introduction
The International Reference Ionosphere (IRI) is the de facto standard for ionospheric parameters and is widely used to specify ionospheric conditions. For many applications the Total Electron Content (TEC) is the parameter of interest because it determines the ionosphere's refractive and retarding effect on radio signals. TEC consists of a bottomside portion and a topside portion. The bottomside contributes only about 20% of the TEC. The dominant contribution comes form the topside and the topside electron density profile is therefore of utmost importance for many IRI users. A number of studies had noted discrepancies between the IRI-2001 model and measurements, especially in the upper topside. At issue is an overestimation of electron densities in the upper topside that increases with altitude reaching about a factor of 3 at 1000 km above the peak. The likely causes of this IRI artifact are the limited data base used to develop the original IRI model, primarily Alouette 1 topside sounder data with some AE-C, and DE-2 in situ data and typical profiles from the Jicamarca incoherent scatter radar, and a fitting process that is biased towards F-region densities because they are one or two orders of magnitude above densities in the upper topside. An improvement of the topside electron density model has been a high priority for IRI developers and users. Several groups investigated model corrections and new approaches to overcome this shortcoming of the IRI-2001 model. These efforts led to the introduction of two new options for the electron topside model in the latest version of the model, IRI-2007.
The first option is a correction factor for the IRI-2001 topside model that was developed by Bilitza based on over 150,000 topside profiles deduced from Alouette 1, 2, and ISIS 1, 2 topside soundings and describes variations with altitude, modified dip latitude, and local time. With this correction term the IRI model represents the topside sounder data quite well as shown in the examples in Fig. 1. The second option is the NeQuick topside model developed by Radicella and Leitinger and Coisson et al. based on ionosonde and topside sounder data. It is the most mature of the different proposals for the IRI topside that were described by Bilitza et al. The NeQuick topside model uses an Epstein-layer function with a height-dependent thickness parameter and in this way produces a smooth transition from an atomic oxygen ionosphere near the F-peak to a light ion ionosphere higher up. The model parameters were determined based on fitting this function to ISIS 1, 2 and Intercosmos 19 topside sounder profiles. We will call the first option IRI-2007-cor and the second option IRI-2007-NeQ in the remainder of the article.
In this study we use Alouette and ISIS topside sounder data to evaluate the two new options in relation to IRI-2001 and to assess which one of the two options shows better agreement with the data. We have also included the model of Triskova et al. (TTS model) that was developed based on Atmosphere Explorer C, D, and E data for low solar activities and Interkosmos 24 and 24 data for high solar activity.
2. Alouette and ISIS topside sounder data
The Alouette 1, 2 and ISIS 1, 2 satellites were ionospheric observatories designed and operated jointly by the USA and Canada. The launch dates and orbit characteristics for the four satellites are listed in Table 1. The primary instrument was the topside sounder for recording the electron density profile from the satellite down to the F-peak. But other instruments were included as well. ISIS-2, for example, carried a sweep- and a fixed-frequency ionosonde, a VLF receiver, energetic and soft particle detectors, an ion mass spectrometer, an electrostatic probe, a retarding potential analyzer, a beacon transmitter, a cosmic noise experiment, and two photometers. NASA support of the ISIS project was terminated on October 1, 1979. Partial operations were continued by the Canadian project team until March 9, 1984 and were then resumed in even more limited form by the Japanese Radio Research Laboratories (Kashima ground station) from August 1984 to January 24, 1990. These satellites have accumulated a large volume of data for the topside ionosphere. The inversion from the recorded ionogram to the topside electron density profile, however, involves a tedious and highly subjective manual process that could only be performed for a small percentage of the collected ionogram data. The National Space Science Data Center (NSSDC) has archived most of these data and has made them available on its ftp site. The different available data sets are listed in Table 1 and plotted in Fig. 1 showing their time and altitude coverage. The data stretch across almost two solar cycles and go up to 3500 km altitude. All except the second Alouette 1 data set were generated at the Communications Research Center (CRC) in Ottawa, Canada using manual scaling and the Jackson ionogram inversion program. The responsible group for the second Alouette 1 data set was the University of California Los Angeles (UCLA) Department of Meteorology using a data analysis process similar to the CRC.
Unfortunately, in many cases the profiles do not reach all the way down to the peak, because of the difficulties in scaling the trace in the cusp region near the peak. This is a problem of the inversion technique. Bent et al. tried to overcome this problem by using a model for the peak height and by assuming the lowest profile point density always at a constant ratio below the actual F2 peak density. We have used a different approach by fitting a Chapman layer to the lower part of the profile. The least-square fitting procedure finds the best Chapman parameters for representing the topside F-layer. Profiles are only considered if they had at least five profile points in the lower topside and if they could be fitted with a standard deviation of 1% or better. About 120,000 profiles remained after this fitting and selection process. The examples in Fig. 2a and b include the determined peak point and show that the procedure works reasonably well.
It is worth noting that a data save and restoration effort at NASA's Goddard Space Flight Center (R. Benson, PI) was able to digitize a large number of the original analog telemetry tapes and use the Topside Ionogram Scaler with True Height Algorithm (TOPIST) program developed at the University of Massachusetts Lowell for the automated processing of these data. This effort will provide additional profiles for a follow-on evaluation study.
3. Comparing model and data -- topside profile shape
In IRI the electron density profile is normalized to the F2 peak density and height. Data-model discrepancies in the absolute electron density in the topside can therefore be attributed to two error sources: the IRI peak models or the IRI topside profile model. If we want to evaluate the reliability of the topside model, we first have to eliminate the influence of the IRI peak models. This is done by normalizing the IRI profile with the measured F-peak density and shifting it to the measured F-peak height. Fig. 2a and b shows the resulting profiles clearly illustrating the overestimation of the topside gradient in IRI-2001. For both examples the new IRI-2007 topside options reproduce the observed topside shape quite well. The TTS model was designed primarily as a representation of the absolute density values mostly above the F-peak and so does not include an option to normalize with the measured peak values. But we have included the model here also because it was proposed for the representation of the total ion density in conjunction with the IRI ion composition model that was developed by the same authors.
In Fig. 3a-c, we have plotted the model predictions versus the ISIS-2 topside sounder measurements for the three IRI models. The plot includes a total of 223,779 data points. If data and model agree well the data points should be all clustered around the central diagonal. The figure for IRI-2001 shows the noted problem at high altitudes where the model predictions become almost constant and exceed the sounder data. Values computed with the two new IRI-2007 topside options are plotted versus the sounder data in Fig. 3b and c. Both models overcome the problem at high altitudes and show good agreement with the data across the whole range covered by the sounder data. IRI-2007-cor seems to slightly overestimate the data while the IRI-2007-NeQ plot indicates an underestimation. To investigate this further we have plotted in Fig. 4a-c the data-model ratios at an altitude of 1000 km above the F2 peak. The IRI-2001 plot clearly shows the general overestimation reaching a factor of 10 at the magnetic equator and at high latitudes. With the IRI-2007 models the factor has been reduced to 2. In addition to the general overestimation the plots for IRI-2001 and IRI-2007-cor exhibit also a distinct latitudinal variation. The ratios are lowest (best agreement) at mid-latitudes and reach the largest values at the magnetic equator and at high latitudes. This again might be due to deficiencies in the least-square fitting process that was used to generate the IRI-2001 topside model. Insufficient weighting of the low topside densities could result in the latitudinal variations being dominated by the behavior at F-region heights which is quite different from the variation in the topside. The typical equatorial anomaly crests on both sides of the equator observed at F-region heights move closer towards the magnetic equator with increasing height until they merge into a single peak at about 1000 km above the peak. This characteristic behavior is well reproduced by the IRI-2007-cor model as can be seen in the latitude-height contour map shown in Fig. 5b. The old IRI-2001 model produces unrealistically steep profiles in the upper topside whereas the NeQuick option shows the typical double-hump structure of the EA extending up to 1500 km altitude and not merging to a single peak as expected.
4. Comparing model and data -- overall result
So far our comparisons were based primarily on the ISIS-2 data set because it provides the best coverage in altitude and latitude. The other 5 data sets in Table 1 produce similar results and similar plots to Fig. 3 and Fig. 4. To get an overall estimation of the data-model discrepancies we have computed the mean and standard deviation of the relative deviation between data and models for all six data sets separately. The results are listed in Table 2. The percentage deviation is computed as. In all cases we get a negative PD indicating that on average all 4 models overestimate the sounder data. As expected the largest discrepancies are found with the IRI-2001 and TTS model because IRI-2001 misrepresent the upper topside and TTS is not normalized to the F-peak. The two new IRI-2007 options produce significant improvements over the older model reducing the data-model difference by a factor of up to 10. For all six data sets best results are obtained with IRI-2007-NeQ showing at times a factor of 2 better representation than IRI-2007-cor. Taking all Alouette and ISIS data together we find that the 165% overestimation of the data with IRI-2001 is reduced to 46% with IRI-2007-cor and to 24% with IRI-2007-NeQ. The new topside models also significantly reduce the scatter of difference values as shown by the decrease of the standard deviation by a factor of 2-5.
6. Summary and conclusions
IRI-2007 introduces two new model options for the topside electron density, a correction of IRI-2001 and the NeQuick model. In our study we have evaluated these new options with six data sets of topside profiles deduced from Alouette 1, 2 and ISIS 1, 2 topside sounder measurements. The total number of profiles used for our study is 120,059. It is important to note that our evaluation is done after the influence of the F-peak models has been eliminated. This was done by normalizing the model profile with the measured peak density and height. Our main results are: -- The new options result in a significant improvement over IRI-2001. While IRI-2001 overestimated the data by an overall average of 165%, IRI-2007-cor reduced this number to 46% and IRI-2007-NeQ to 24%.