Development of an HF selection tool based on the Electron Density Assimilative Model near-real-time ionosphere



[1] The Electron Density Assimilative Model (EDAM) has been developed to provide real-time characterization of the ionosphere. A thin-client Web-based ionospheric situational awareness tool, which relies on data from EDAM, has been developed. The tool provides the high-frequency (HF) maximum usable frequency (MUF) and the optimum working frequency (FOT) on the basis of the current ionospheric electron density grid from EDAM. Two validations are reported. First, EDAM has been tested against a set of vertical ionosondes. The results show performance comparable to that of the Utah State University Global Assimilation of Ionospheric Measurements Gauss–Markov Kalman filter model. Second, MUFs estimated from EDAM were compared with those measured on an oblique path from New Zealand to Australia. Although EDAM demonstrates better performance than a monthly median model in some days, over the whole test period EDAM performed no better than the international reference ionosphere. This can be attributed to the lack of input GPS total electron content (TEC) data near the midpoint of the path and a probable TEC bias in the background model.

1. Introduction

[2] The Electron Density Assimilative Model (EDAM) has been developed to provide real-time characterization of the ionosphere. EDAM comprises a suite of programs that manage ionospheric data and assimilate them into a background ionospheric model. In order to demonstrate and assess the use of EDAM in an operational context, an ionospheric situational awareness tool (known as EDAM533) that relies on data from EDAM has been developed. The aims of this demonstration are twofold: first, to assess the operational benefit that may accrue from the use of an assimilative model; and second to investigate and implement the system architecture required to deliver a product derived from the assimilative model to a wide user community. This paper will provide a very brief overview of EDAM. It will then describe the development of EDAM533 and give details of the network architecture and the underlying maximum usable frequency (MUF) estimation algorithm. The user interface and two validation studies will also be described.

[3] For the demonstration system, the focus is on high-frequency (HF) communications users. This decision has been driven by a number of factors: HF propagation tools do not generally require integration into larger systems; HF operators are familiar with using propagation tools; and HF propagation provides a good test of the accuracy of the ionospheric models. The tool allows the user to point and click on a map to define transmitter and receiver locations, and the software returns the MUF and the optimum working frequency (FOT) based on the real-time ionospheric electron density grid from EDAM. The returned values are obtained by deriving foF2 and M(3000)F2 (see Table 1 for definitions) from the EDAM grid and then applying the MUF estimation algorithm from International Telecommunication Union Radiocommunication Sector (ITU-R) [2007]. A similar project, called HFNowcast [McNamara et al., 2009], is being undertaken by the U.S. Air Force Research Laboratory (AFRL) using electron density grids from the Utah State University Global Assimilation of Ionospheric Measurements (USU-GAIM) [Thompson et al., 2006]. McNamara et al. [2009] provide a good exposition of the rationale for providing MUF predictions rather than the more usual signal-to-noise ratio (SNR) predictions. In summary, in the absence of real-time noise and absorption measurements, it is not possible to provide real-time estimates of SNR.

Table 1. Ionospheric Parameters Used by REC533 to Estimate MUFa
foF2The F2 layer's critical frequency (the maximum o-ray frequency that is reflected at vertical incidence).
M(3000)F2The F2 layer's MUF factor for a 3000-km path. This relates the foF2 to the MUF for a 3000-km path thus: MUF(3000)F2 = foF2 × M(3000)F2

[4] A thin-client approach has been taken to provide EDAM533 to users. This approach requires only a Web browser on the client (users') computer. This is advantageous as there is a move toward allowing only Web services on some classified networks. The architecture uses a Web server (termed the HF server) to access the ionospheric data produced by EDAM. The client then makes requests to the HF server to run HF predictions; that is, the client passes transmitter and receiver information to the HF server and the server returns an MUF and FOT to the client.

2. The Electron Density Assimilative Model

[5] The Electron Density Assimilative Model (EDAM) has been developed by QinetiQ [Angling and Cannon, 2004; Angling and Khattatov, 2006] to assimilate measurements into a background ionospheric model. This background model is provided by IRI2007 [Bilitza and Reinisch, 2008], and the majority of the input data are total electron content (TEC) measurements derived from International GPS Service (IGS) stations [Beutler et al., 1999]. Since the international reference ionosphere (IRI) does not include a plasmasphere, a simple exponential electron density profile, matched to the IRI scale height at 2000 km, is used above this height. The assimilation is based on a weighted, damped least mean squares estimation. This is a form of minimum variance optimal estimation (also referred to as Best Linear Unbiased Estimation (BLUE)) that provides an expression for an updated estimation of the state (known as the analysis) that is dependent on an initial estimate of the state (the background model) and the differences between the background model and the observations [Menke, 1989; Twomey, 1977]. The error covariance matrices of the background model and the observations are also used to control the relative contributions of the background and the observations to the analysis:

equation image
equation image

where xa is the analysis, xb is the background model, K is the weight matrix, y is the observation vector, B is the background error covariance matrix, and R is the error covariance matrix of the observations [Rodgers, 2000]. H is the nonlinear observation operator that relates the measurements to the state:

equation image

where ɛ is the observation error. The operator is nonlinear because, in EDAM, the background model is composed of the log of the ionospheric electron density. H is the Jacobian, whose elements are given by the partial differentials of the observation operator evaluated at the background model; that is,

equation image

[6] The use of a nonlinear observation operator requires that, at each assimilation step, the difference between the analysis and background model remains small; that is, it is only acceptable to make small corrections to the background model. Within EDAM, the differences between the analysis and the background model are checked at each assimilation step. If the difference of any element exceeds approximately 5% of the background model's value, the assimilation is rejected.

[7] The assimilation is conducted using a magnetic coordinate system that remains fixed in space with respect to the sun. An assimilation time step of 15 min has been used and the electron density differences between the voxels of the analysis and the background model are propagated from one time step to the next by assuming persistence combined with an exponential decay. The time constant for this decay is set at 4 h. Thus if the data feed is interrupted, the analysis will decay back to the background model.

3. EDAM533

3.1. User Requirements

[8] The user requirements for EDAM533 have been developed with assistance from the UK Defense HF Communications Service (DHFCS). Much of DHFCS's work is to provide frequencies to users on demand. Owing to the real-time nature of this activity, it is not necessarily well served by monthly median models. However, the real-time nature of an assimilative method can be used to provide better frequency assignments. The user requirements indicated that EDAM533 should have the following attributes:

[9] 1. It should have a simple map-based interface with the capability to define the locations of a transmitter and receiver. The map should include the capability to zoom into a particular region.

[10] 2. It should return the maximum usable frequency (MUF) and the optimum working frequency (FOT) based on the real-time ionospheric electron density grid.

[11] 3. It should allow up to 40 communications circuits to be defined for easy recall when required. The MUF and the FOT for these circuits should be continuously updated using the real-time ionosphere.

3.2. MUF and FOT Estimation

[12] EDAM533 provides users with an estimate of the current maximum usable frequency (MUF) between a transmitter and receiver based on the real-time ionospheric model produced by EDAM. This could be achieved by ray tracing through the EDAM electron density grid to synthesize an oblique ionogram and then taking the highest propagating frequency as the MUF. However, such an approach is computationally intensive and could potentially introduce unacceptable delays in providing MUF estimates to the user. Therefore, the approach taken in EDAM533 uses the MUF estimation algorithm as defined by REC533 of ITU-R [2007].

[13] In brief, REC533 estimates the MUF using monthly median values of foF2 and M(3000)F2 determined at the propagation path's control points (i.e., the midpoint for one hop paths, the quarter points for two hop paths, and so on). In EDAM533, the median values of foF2 and M(3000)F2 are replaced by real-time values estimated from vertical ionograms synthesized from the EDAM electron density profiles, and the MUF is then estimated. It should be noted that only the basic F region MUF estimation algorithm from REC533 has been implemented at this time; that is, no account is taken of E layer propagation or screening. These features may be added in a future upgrade. The optimum working frequency (FOT) is lower than the MUF and is calculated from the MUF using the technique defined by REC434 from ITU-R [1995].

3.3. Graphical User Interface

[14] There are two Web pages accessible by users of EDAM533. The first is the HF prediction tool page (Figure 1). This is a map-based Web page, which allows the user to move transmitter and receiver flags around a map of the world and calculate the MUF and FOT associated with those locations. A warning is given if the foF2 and M(3000)F2 data derived from EDAM are, for whatever reason, out of date (i.e., older than 1 h). The second page is a stored circuits page (Figure 2). The circuit table allows the user to store up to 40 frequently used transmitter-receiver pairs and to get updated MUF and FOT values for each stored circuit every time they view the page. Furthermore, the MUF and FOT values are automatically updated every 15 min. Users only have access to their personal stored circuits and not those of other users. As with the HF prediction page, a warning is given if the EDAM data are out of date.

Figure 1.

Screen grab of HF prediction page of EDAM533.

Figure 2.

Screen grab of EDAM533 circuit table page.

3.4. Architecture

[15] The architecture of the demonstration system is based on the standard three-tier Web system. Tier one incorporates the client system, which accesses the EDAM533 Web site on a Web server (tier two, the HF server) via a browser application (using HTTP); behind the Web server, the third tier consists of supporting servers (e.g., an FTP server) that supply external data and services.

[16] EDAM receives external data (i.e., GPS TEC) that is assimilated to generate a global ionospheric model. The EDAM output is then processed to produce a data file containing foF2 and M(3000)F2 which are pushed to an FTP server as a file. This file is retrieved by the HF server and used, together with location information from the user (the HF client), to estimate MUF and FOT values. These values are then returned to the user and displayed in the Web browser. The user's client machine requires only a standard browser application with Javascript and cookies enabled in order to view the EDAM533 Web site. Figure 3 shows the main software components of the system, which are also described in Table 2.

Figure 3.

Schematic representation of EDAM533. IDL, Interactive Data Language. SOAP, Simple Object Access Protocol. PHP, Hypertext Preprocessor (scripting language). DB, database. WMS, Web Map Service. XEDAMDL, the EDAM data handling program.

Table 2. EDAM533 Software Componentsa
ComputerSoftware ComponentDescription
  • a

    SOAP, Simple Object Access Protocol. PHP, Hypertext Preprocessor (scripting language).

EDAM PCEDAMMain EDAM software, which obtains ionospheric data, assimilates it, and processes it to produce the data files required by EDAM533.
FTP serverFTP serverProvides a means to transfer data files from the EDAM PC to the HF server.
HF clientWeb browserAllows the user to access the EDAM533 Web site.
HF serverMUF estimation codeInteractive Data Language (IDL, code that implements the REC533 MUF estimation algorithm. It is accessed via the IDL-Java bridge.
 MUF estimation wrapper serviceA wrapper around the MUF estimation code, turning it into a Web service. This makes it easy to interface to the MUF estimation code, as the interface is well-defined in the Web Services Description Language (WSDL). Furthermore, the MUF estimation component can then be used via SOAP (i.e., calls so that it could form part of a service-orientated architecture (SOA) implementation. The wrapper makes it possible to update or replace the MUF estimation relatively easily without having to rewrite any code other than the wrapper. The methods exposed by the wrapper are used by the Web site PHP to get MUF and FOT values for individual or multiple sets of transmitter and receiver locations.
 EDAM file retrieval serviceA small IDL program that polls the FTP server for output files from EDAM. It checks for the most up-to-date file and then transfers it across onto the HF server for use by the MUF estimation code.
 Web site softwareComprises a set of dynamic Web pages written in PHP. These form the basis of the graphical user interface (GUI) that the user views via their Web browser.
 MapServerProvides maps to support the Web-based mapping interface (
 MySQL databaseStores circuit values and user authentication details (

4. Validation

4.1. Introduction

[17] Initial testing of EDAM533 focused on ensuring that the implementation of REC533 was correct and the system gave plausible results. Testing was conducted by comparing the EDAM533 results with those obtained from the standard Institute of Telecommunications Sciences (ITS) implementation of REC533 and with those obtained by direct ray tracing through IRI2007. Satisfactory comparisons have also been made between EDAM533 and the AFRL HFNowcast program [McNamara et al., 2009] when both have used identical ionospheric grids (IRI2007; i.e., no assimilation is used).

[18] Two further tests are presented here: first, a validation of EDAM itself with respect to data from vertical ionosondes and, second, comparisons between EDAM533 results and MUFs measured using an oblique sounder.

4.2. Validation of EDAM

[19] As part of an ongoing collaboration with AFRL, the test scenario described by McNamara et al. [2008] has been replicated to test the performance of EDAM. In this test a network of GPS receivers provided input data to the assimilative model, and the model output is compared to data from 21 vertical ionosondes. The list of GPS stations used by AFRL and the data files containing checked foF2 and M(3000)F2 data were provided by AFRL to ensure that identical conditions were maintained for the test. Figure 4 shows the locations of the input GPS stations (Figure 4, dots) and validation ionosonde locations (Figure 4, triangles). Full details of the ionosondes can be found in Table 1 of McNamara et al. [2008]. The test was conducted for the month of September 2006.

Figure 4.

Map showing locations of input GPS stations (dots) and test ionosonde stations (triangles).

[20] After assimilation, foF2 and M(3000)F2 can be estimated (by synthesizing the vertical ionogram) from the EDAM electron density grid at the ionosondes' locations. These values can be compared with the measured values to assess the performance of the model. Figure 5 shows the observed and EDAM-estimated hourly median foF2 for Boulder (Colorado) for September 2006. Such results for all stations can be accumulated to provide an overall view of the results. Figures 6 and 7 show the estimated monthly median values of foF2 and M(3000)F2 plotted against the observed values for all 21 ionosondes.

Figure 5.

Observed and estimated hourly median values of foF2 for Boulder, Colorado (September 2006).

Figure 6.

Scatterplot of EDAM median values of foF2 against the observed medians (September 2006, 21 ionosondes).

Figure 7.

Scatterplot of EDAM median values of M(3000)F2 against the observed medians (September 2006, 21 ionosondes).

[21] The mean and RMS errors in foF2 and M(3000)F2 have also been plotted as a function of the absolute value of the geomagnetic latitude (Figures 8–11). As expected, owing to the greater variability of the equatorial ionosphere, the lower latitude stations tend to exhibit the highest RMS errors.

Figure 8.

Mean foF2 errors for day and night, versus the absolute value (abs) of the magnetic latitude.

Figure 9.

RMS foF2 errors for day and night, versus abs(magnetic latitude).

Figure 10.

Mean M(3000)F2 errors for day and night, versus abs(magnetic latitude).

Figure 11.

RMS M(3000)F2 errors for day and night, versus abs(magnetic latitude).

[22] Figures 57 are directly comparable to Figures 4, 5, and 10 of McNamara et al. [2008], respectively; Figures 811 are directly comparable to Figures 6, 7, 11, and 12 of McNamara et al. [2008], respectively. It can be seen that the performance of EDAM is comparable to that of the USU GAIM model used by McNamara et al. [2008].

4.3. Comparison With Oblique Soundings

[23] A previous study has compared measured MUFs between the UK and Rome with estimated MUFs from EDAM [Angling and Khattatov, 2006]. In this paper, EDAM estimates of MUF are compared with values measured on a ∼2250-km path from Irirangi (New Zealand, 39.3°S, 175.4°E) to Sydney (Australia, 33.9°S, 151.2°E) during October 2006. Data are available for 1–3 October and 19–31 October and have been scaled by an automatic process [Hutchinson, 2007]. Figure 12 shows the nearby GPS stations used in the assimilation (labeled with their four-character designations) and the path from Irirangi to Sydney.

Figure 12.

Map showing the assimilated GPS stations and oblique ionosonde test path between Irirangi (New Zealand) and Sydney (Australia).

[24] Figures 13–15 show the measured MUFs (solid lines), estimated MUFs from IRI2007 (dotted lines), and estimated MUFs from EDAM (dashed lines) for 1–3, 19–25, and 25–31 October 2006, respectively. A number of spuriously high MUFs can be seen, especially in the 19–25 October period, which are due to incorrect autoscaling. These have been removed before statistics are calculated for the data.

Figure 13.

Observed, IRI2007, and EDAM MUFs for 1–3 October 2006.

Figure 14.

Observed, IRI2007, and EDAM MUFs for 19–25 October 2006.

Figure 15.

Observed, IRI2007, and EDAM MUFs for 25–31 October 2006.

[25] While EDAM shows significant benefit for some days (i.e., 30 October), on other days the EDAM results can be comparable or even degraded in comparison to IRI. The mean, standard deviation, and RMS of the MUF errors for each period are given in Table 3. The results are divided into day and night (0600–1759 and 1800–0559 local time at the path midpoint). The performance of EDAM is similar to that reported by Angling and Khattatov [2006]. However, in this case, over the whole of the test period, EDAM performs no better than IRI. This can partly be attributed to the fact that the test path is over the ocean and there are no data inputs near the midpoint. However, it can also be seen that EDAM exhibits a consistently negative MUF bias. In the absence of a vertical sounder near to the midpoint of the path it is difficult to determine the reason for this. However, one possible cause is that the background model may exhibit positive TEC bias while being unbiased in foF2. This is plausible since IRI is expected to perform well (at least in a median sense) in terms of foF2, but is much less well constrained in terms of TEC. Thus, when TEC measurements are assimilated, the EDAM TEC is pulled down in relation to the IRI TEC; consequently, the foF2 and MUF are also reduced.

Table 3. Mean MUF Error, RMS MUF Error, and Standard Deviation of MUF Errors for EDAM and IRI
  Day (0600–1759 LT)Night (1800–0559LT)
Mean (MHz)Standard Deviation (MHz)RMS (MHz)Mean (MHz)Standard Deviation (MHz)RMS (MHz)
EDAM1–3 Oct−−
19–25 Oct−−
26–31 Oct−−
All data−−
IRI20071–3 Oct0.
19–25 Oct0.51.71.8−
26–31 Oct0.
All data0.

5. Conclusions

[26] An HF selection tool (EDAM533) has been implemented. The MUF predictions generated by EDAM533 are based on foF2 and M(3000)F2 values derived from the EDAM real-time electron density grids. These parameters are used to estimate the MUF by means of the ITU-R REC533 algorithm. EDAM533 has been implemented as a thin client; the client PC requires only a standard Web browser to access the system. This design has been chosen to ease any transition from unclassified operation to usage on classified networks where Web services will be preferred in future.

[27] A previous study [Angling and Khattatov, 2006] has shown that MUFs derived from EDAM can be more accurate than those generated by a median model. In this work, the performance of EDAM has been assessed by comparing estimated values of foF2 and M(3000)F2 with observed values from vertical ionosondes. By comparison to the results reported by McNamara et al. [2008], it can be seen that EDAM performance is comparable to that of the USU GAIM.

[28] Additional testing has been conducted against MUFs observed on a path between New Zealand and Australia. Although EDAM provides better estimates of MUF on some days, over the course of the whole test period (one month) EDAM provides little improvement compared to IRI. This disappointing result can be ascribed to the lack of input data near the midpoint of the path in question and a probable TEC bias in the background model. This demonstrates the need for better background models to support the development of assimilative models.


[29] EDAM has been developed under funding from the United Kingdom Ministry of Defense Science and Technology program. IGS data were obtained from the SOPAC Data Centre. Differential code biases were obtained from the Centre for Orbit Determination in Europe. The scaled vertical ionosonde data were obtained from the IPS Radio and Space Services Web site ( and from the University of Massachusetts Lowell DIDBase ( [Reinisch et al., 2004] and provided to QinetiQ by AFRL as part of a continuing program of cooperation. The Irirangi-Sydney oblique ionosonde data were provided by IPS Radio and Space Services and the New Zealand Defense Force.