Real-time specification of HF propagation support based on a global assimilative model of the ionosphere



[1] An HF propagation program, HFNowcast, has been developed that takes advantage of the availability of a real-time global model of the ionosphere to specify in real time the range of frequencies that would be supported on a given HF communications circuit. HFNowcast determines the range of frequency support for a specified circuit, from the lowest usable F layer frequency (FLUF) to the highest (FMUF), together with the E layer ELUF and EMUF. The lack of real-time observations of HF absorption and HF noise at the receiver means that it is not possible to provide real-time specifications of the signal-to-noise ratio (SNR), the key parameter for HF communications. The global model (developed by Utah State University (USU)) includes particle E precipitation at high dip latitudes during the night, which often precludes the determination of foF2 and M(3000)F2 and thus the simple determination of propagation support in those regions. The FMUF is therefore derived by first calculating the vertical incidence ionogram at the control point(s) and then transforming that ionogram to the oblique incidence (OI) ionogram for the given circuit/hop length. HFNowcast has been validated favorably against several months of oblique propagation observations on midlatitude circuits, although its day-to-day variability does not match the observations very well, and it tends to underestimate the FMUFs. The systematic differences between the predictions and observations can be attributed directly to limitations of the current version of the USU model of the ionosphere.

1. Introduction

[2] HF propagation prediction programs such as VOACAP [Lane, 2001], ASAPS [ASAPS, 2005], and Rec-533 [ITU, 2005] provide monthly median predictions that are used to plan an HF communications link some months to years in advance. The planning is mostly in terms of the best frequency to use for a given circuit at a given time. The antenna and power requirements are usually determined (using the same programs) well before the transmitter and receiver sites are set up. The basic communications requirement is to provide a signal-to-noise ratio (SNR) that is large enough for the intended mode of operation.

[3] Monthly median prediction programs use worldwide monthly median maps of the key parameters foF2 and M(3000)F2 that were established using years of historical data at multiple ionosonde locations around the world. The best known maps are the URSI and International Radio Consultative Committee (CCIR) maps. ASAPS uses grid point maps that were developed independently by IPS Radio and Space Services. The purpose of the maps is to specify the values of foF2 and M(3000)F2 at arbitrary locations, time (time of day, month of year), and level of solar activity. The “prediction” aspect of the process arises when a prediction is made of the level of solar activity for the intended epoch of the required monthly median predictions.

[4] VOACAP and Rec-533 require the specification of the “sunspot number.” ASAPS requires the specification of an ionospheric index (called the T index) that is based on actual observations of the ionosphere rather than observations of the Sun. The sunspot number used in setting up the URSI/CCIR world maps is R12, the 12-month running average of the monthly sunspot number. (R12 is also used in empirical formulas to calculate foE and foF1.) To run the propagation programs, it is necessary to predict (or forecast, depending on the required lead time) the value of R12 (or T, but we shall generalize to R12).

[5] Except near and at solar minimum, there could be significant errors in the predicted value of R12. The scatter about the regression lines between median foF2 (say) and R12 also becomes important at higher levels of solar activity. In an attempt to minimize the effects of these errors, the prediction community has moved to lead times less than a month and to the use of an effective sunspot number, Reff. Reff is based on the most recent observations of foF2, which can now be obtained every 15 min or so. Reff is just the value of R12 that, when used in conjunction with the URSI/CCIR maps, will reproduce the observed values of foF2, usually in some least squares sense.

[6] The Air Force Weather Agency (AFWA) has implemented the Utah State University Global Assimilation of Ionospheric Measurements (USU-GAIM) [Scherliess et al., 2004; Schunk et al., 2004; Thompson et al., 2006]. Starting with a background model that is driven simply by real-time geophysical indices, USU-GAIM assimilates observations of such things as slant total electron content (TEC) made at ground-based GPS sites and thus generates an updated three-dimensional worldwide specification of the ionosphere from 92 to 1380 km. The current version of GAIM implemented at AFWA uses a Gauss-Markov technique to assimilate the observations and generates a new specification of the ionosphere every 15 min.

[7] This paper describes procedures developed at AFRL to provide propagation predictions based on the GAIM near real-time specifications of the ionosphere. The project is known as HFNowcast. A similar project was undertaken by QinetiQ [Angling et al., 2009] using the Electron Density Assimilative Model (EDAM) [Angling and Khattatov, 2006]. AFRL and QinetiQ maintained a close cooperation for this project.

[8] Section 2 describes the underlying approach and how it differs from those adopted for monthly median programs. Section 3 describes the functions of the various programs that together make up HFNowcast. In the absence of very time-consuming raytracing, the mode structure (1E, 1F, 2E, 2F mode) is defined basically by setting upper range limits for E and F layer hops, as described in section 4. Section 5 describes various validations of HFNowcast. At this stage in the progress of GAIM models, the errors in the forecast range of maximum usable frequencies (MUFs) are largely controlled by errors in the GAIM plasma frequency profiles. Section 6 presents brief conclusions.

2. Method of Approach

[9] As mentioned earlier, the most important parameter for HF communications is the SNR for each allocated frequency. The calculation of an SNR requires the calculation of the signal power (S) at the receiver and an observation or estimation of the HF radio noise power (N) at the receiver on the operating frequency. Only the most sophisticated HF systems actually measure the HF noise at the receive site in real time. Most monthly median predictions usually rely on a simple classification of the noise as being business, residential, rural, or remote.

[10] The S part of the SNR is a function of the transmitter power, the gains of the transmitting and receiving antennas, and the path loss along the circuit. In most situations, the largest components of the path loss are the inverse square spatial attenuation and the HF absorption. The inverse square loss can be derived by raytracing at multiple elevation angles through a suitably parameterized plasma frequency (electron density) profile that varies along the circuit. This is a tedious process, and it is not obvious that the current accuracy of the USU-GAIM profiles justifies that effort. Its utility should be addressed again when the “full physics” GAIM model is available from USU.

[11] HF absorption arises when the signals traverse the D and lower E regions during the day, these being the parts of the ionosphere that cause the absorption. Currently there are no observations of HF absorption being assimilated by GAIM, since GAIM does not model the D region. The fallback position is to use the absorption model given by Davies [1990], which is also used by ASAPS and Rec-533, but this is only a monthly median model that does not provide any day-to-day variability.

[12] The inability to provide real-time estimates of all of the components of the path loss and the HF noise at the receiver has led AFRL to concentrate on determining just the range of frequencies that would be supported on a given circuit at a given time. The actual circuit planning (choice of antennas and transmitter powers) is left to the monthly median programs such as ASAPS, VOACAP, and Rec-533. Once the circuit has been set up, the main task is to select the best frequency from the range of frequencies currently being supported, and this range can be provided by HFNowcast.

3. Programs Used to Determine the Frequency Support

[13] HFNowcast determines the instantaneous values of the lowest usable frequency (LUF) and maximum usable frequency (MUF) for a given circuit on the basis of the latest USU-GAIM worldwide specification of the ionosphere. The LUF and MUF are calculated separately for E and F propagation (ELUF, EMUF, FLUF, and FMUF). The E layer propagation is supported by the normal solar-controlled E layer and by a particle E layer at high latitudes. The particle E layer arises in GAIM via an empirical model of particle precipitation in the auroral ovals [Hardy et al., 1985]. Figure 1 shows the programs and data flow for the determination of propagation support.

Figure 1.

Programs and data flow for determination of propagation support.

[14] The USU-GAIM specification of the ionosphere is provided every 15 min. The USU grid points are generally too sparse for comfort, so the program Make_Finer_GAIM_Maps determines the profiles at a finer grid (with extra points halfway between the original grid points) by polynomial interpolation separately in latitude and longitude. The bilinear interpolation used by HFNowcast to determine the profile at a control point is then more reliable.

[15] The nighttime profiles at high dip angles that correspond to an E layer generated by particle precipitation require identification and special treatment, because any interpolated profile could be nonphysical if the surrounding profiles are not all regular profiles or not all particle E profiles. The program Assign_ParticleE_Flags assigns each grid point a flag that indicates (by a series of checks) if the profile is a regular or particle E profile.

[16] The program HFNowcast reads the profiles from GAIM_fine_grid_profiles.dat, the particle E flags from GAIM_particleE_flags.dat, and the transmitter and receiver coordinates from circuit_details.dat. The frequency support is calculated on the basis of control points at the midpoint of the circuit for one-hop modes, for the quarter-points for two-hop modes, or for points 1500 km in from the transmitter and receiver for circuit lengths greater than ∼5000 km. If the four grid points surrounding a control point all correspond to a normal profile or to a particle E profile, the profile at the control point is calculated by bilinear interpolation. If the four grid points have different types of profiles, the profile for the closest grid point is taken as the control point profile. (This is an approximation, but no alternative suggests itself.)

[17] A significant part of the program logic (and computation time) is concerned with the particle E profiles, which must be identified as such. If there were no particle E profiles, the calculations of the supported frequency ranges could proceed by simple interpolation in a worldwide grid of foF2 and M(3000)F2 values. Values of M(3000)F2 derived from calculated particle E ionograms are invalid.

4. Propagation Geometry

[18] HFNowcast evaluates the LUF and MUF for both the E and F layers, for one-hop and two-hop propagation. (Sporadic E propagation is ignored, since Es is not included in GAIM.) The maximum hop lengths for the E and F modes are set by internally specified mirror reflection heights and minimum elevation angles. The highest-order mode considered is the two-hop mode. The longest circuit depends on the supplied mirror heights and is typically ∼5000 km. Longer circuits are analyzed using control points 1500 km in from each end. A minimum elevation angle can be set, to allow for a cutoff angle appropriate to tactical antennas. At present, this minimum elevation angle is applied only to 1F modes; if the elevation angle is too low, the 1F mode is replaced by a 2F mode.

4.1. Propagation Controlled by the Solar-Controlled E Layer

[19] The solar E layer sets the midlatitude and low-latitude EMUF and FLUF. The EMUF is calculated using the procedure given by Caruana [1995], which just multiplies foE by a circuit-length-dependent obliquity factor. The value of foE used is a monthly median value based on observations [Muggleton, 1975]. HFNowcast does not currently use the value of foE derived from the GAIM profiles because of the simple empirical nature of the E layer algorithms. The FLUF is determined by estimating the lowest frequency that would penetrate the underlying E layer (i.e., not be blanketed by the E layer). Again following Caruana [1995], the cutoff frequency is set equal to foE.secϕ, where ϕ is the angle of incidence of the ray for a 200 km reflection.

[20] The ELUF is actually set by D layer absorption. However, the D and E region densities are both strongly dependent on the solar zenith angle, so foE is used as a proxy for what the D layer is doing. HFNowcast follows Caruana [1995] and sets the ELUF to an empirical “absorption limiting frequency.” The ELUF increases with circuit length (up to the maximum hop length for E layer modes), is greatest in the middle of the day, and goes to zero at night. (In a more sophisticated model, the ELUF would be set equal to the lowest frequency that supplies the requisite SNR.)

4.2. Propagation Controlled by the F2 Layer or Particle E Layer

[21] HFNowcast works by determining the oblique ionogram for the specified circuit, with the nose frequency (where the low and high rays meet) defining the FMUF. The vertical incidence ionogram is first calculated from the profile and then transformed into the oblique ionogram for the specified circuit length using standard theory [Davies, 1990, section 6.3.4]. Because the vertical incidence ionogram is calculated from a given profile, the corresponding values of frequency, virtual height, and the real height at which reflection occurs (i.e., the ray apogee) are all known. The relationship between an oblique frequency fob and the equivalent vertical frequency, fv, that is reflected from the same real height hr, is given by Davies [1965, equation 4.11]:

equation image

where a is the radius of the Earth and Δ is the elevation angle of the transmitted ray. Taking each point on the vertical incidence ionogram in turn as fv gives the corresponding value of the oblique frequency fob on the oblique ionogram. The group path for each oblique frequency is given by Breit and Tuve's theorem generalized to a curved Earth/ionosphere [Davies, 1965, section]. This process thus builds up the oblique ionogram. The calculation of the oblique incidence ionogram also provides the elevation angle corresponding to the FMUF, which may be compared to the cutoff angle for a specified transmitting antenna. In practice, the VI ionogram is calculated only for frequencies greater than 0.7 foF2, since this is the part that transforms into the nose of the oblique incidence (OI) ionogram. The VI ionograms are actually ordinary-ray traces.

[22] Note that this VI-OI transformation also works for ionograms corresponding to particle E profiles. HFNowcast defines the propagation in this case to be E layer propagation, with the particle E layer blanketing the F2 layer (i.e., not letting signals penetrate up to the F layer, so FMUF = 0).

5. Validation of HFNowcast

[23] The initial validations of HFNowcast basically confirmed only that the algorithms had been implemented correctly in the Fortran code. These were followed by successful comparisons of the HFNowcast results with the Rec-533 estimates made by the QinetiQ program EDAM-533 [Angling et al., 2009] for the same ionosphere (in practice, the International Reference Ionosphere, 2007, available at HFNowcast was then compared with observations of the 1F MUF for a one-hop European circuit, Inskip (United Kingdom) to Rome, for December 2003, and the errors analyzed in terms of the errors in the GAIM plasma frequency profiles. HFNowcast has also been compared semiquantitatively with observations made on midaltitude circuits in the midlatitude United States and eastern Pacific during the HIDIVE campaign [Eccles et al., 2005]. Finally, HFNowcast has been compared with observations on the Irirangi (New Zealand) to Sydney (Australia) circuit for October 2006.

[24] An essential part of the validation process is validation of the USU-GAIM plasma frequency profiles. AFRL has spent a large effort over the past few years doing just this. One such worldwide validation has recently been published [McNamara et al., 2008]. For example, the average errors in MUF(3000)F2 (all ionosonde locations, all hours) were found to be 5 ± 5%, except for the few hours before dawn. The RMS error was ∼15% during the day and 30 ± 10% at night. The RMS errors in MUF(3000)F2 at most dip latitudes were 15 to 20%, but reached 40% near a dip latitude of 20° (the anomaly peaks).

[25] Another AFRL (unpublished, 2008) validation compared the GAIM profiles with those derived from autoscaled Digisonde ionograms. The study considered noon, sunset, midnight, and sunrise ionograms, September 2004, for Fairford, Grahamstown, Pt. Arguello, Puerto Rico, and Jicamarca. The GAIM specifications of foF2 for these five stations had an average error of 0.1 MHz and a standard deviation of 0.5 MHz. The specifications of hmF2 had an average overestimate of ∼14 km, with a standard deviation of ∼22 km. In terms of M(3000)F2, and assuming the validity of the Shimazaki [1955] formula, the +14 km corresponds to ΔM ∼ Δh/150, or ∼ −0.1, for M = 3.2. This error corresponds to an error of ∼3% in a calculated MUF. The 22 km corresponds to an error of ∼5%.

[26] The assimilation data provided to GAIM for the present study was restricted to the GPS TEC observations from the 95 locations indicated in Figure 2 (same for all months). GAIM will assimilate as well plasma frequency profiles derived from ionograms, but the general quality of the European ionograms was poor for December 2003 and February 2004, so the decision was made to assimilate only the GPS TEC data for the whole project. GAIM will also assimilate in situ DMSP (840 km) electron density observations, but these have minimal effect on the F2 layer subpeak.

Figure 2.

Locations of GPS ground sites that provided total electron content (TEC) observations.

[27] We do not address in this paper the question of whether or not the assimilation of the GPS TEC data makes the propagation predictions more accurate. This question was settled several years ago in early (unpublished, 2005, 2006) validations of GAIM by AFRL. The GAIM (actually the Gauss Markov Kalman Filter (GMKF)) model was found to be more accurate than its background model on virtually all occasions, for multiple ionospheric parameters and epochs.

5.1. Inskip-Rome 1F MUF

[28] Observations of the 1F MUF on the 1735 km Inskip (United Kingdom; 53.8°N, 357.2°E) to Rome (Italy; 41.8°N, 12.2°E) circuit have been provided to AFRL by QinetiQ (United Kingdom). USU GAIM was run with only GPS TEC data being assimilated. The quality of the ionograms and their autoscaling for relevant European ionosondes was considered too poor to be useful for assimilation. Two months have been considered, December 2003 and February 2004. We concentrate here on the December 2003 results.

[29] Since HFNowcast provides real-time specifications of the frequency support, it should track the day-to-day variability and also yield accurate monthly median values of the FMUF. Figure 3 compares the monthly median values of the FMUF (same as 1F MUF for this circuit) given by HFNowcast (HFN), ASAPS, and VOACAP. The median of the observed 1F MUF (or Maximum Observed Frequency (MOF)) is also plotted as the reference.

Figure 3.

Monthly median maximum usable frequencies (MUFs) for the Inskip to Rome circuit, December 2003, for different programs.

[30] The HFNowcast noon values (∼1200 UT) of the MUF are too low by ∼4 MHz (4/22 = 18%), while the ASAPS and VOACAP values are much closer to the observed values. The ASAPS and VOACAP values are in fact very similar for most of the 24 h. After sunset (∼1700 UT for this winter month), the ASAPS and VOACAP errors increase, while the HFNowcast errors decrease.

[31] Figure 3 shows that the daytime diurnal variation of the HFNowcast median FMUF is not smooth across 0900 UT. In fact, there is a discontinuity in the FMUF at 1000 UT. This discontinuity is in fact a consistent feature of the product of the GAIM values, MUF(3000) = foF2 × M(3000)F2, for each day, so it is not an artifact of HFNowcast.

[32] An indication that GAIM is tracking the day-to-day variability of the ionosphere is that the standard deviation of the HFNowcast errors is less than the standard deviation of the observations. Figure 4 shows that this situation holds for the HFNowcast FMUFs, although the improvement provided by GAIM is not particularly large. This is in contrast to the observation by Decker and McNamara [2007] that USU-GAIM successfully reproduced a portion of the Australian ionospheric weather in foF2 for March/April 2004.

Figure 4.

Diurnal variation of the standard deviation of the observed MUFs and of the errors in the GAIM Inskip-Rome MUFs, December 2003.

5.2. Accuracy of GAIM Profiles for the Inskip-Rome Midpoint

[33] The accuracy of the HFNowcast values of the 1F MUF relies mostly on the accuracy of the GAIM plasma frequency profiles, especially near the F2 peak. The errors inherent in the simple propagation models used in HFNowcast (specifically ignoring horizontal variations of the ionosphere in the reflection point) are currently smaller than the GAIM errors. This was shown by comparisons of HFNowcast with the QinetiQ program EDAM-533, which uses an accurate analytic raytracing approach that includes horizontal gradients. This section discusses the errors in the GAIM specifications of foF2 and M(3000)F2 at the locations of the Rome (RO041; Italy) and Chilton (RL052; United Kingdom) Digisondes and shows how they lead to the errors in the HFNowcast values of the MUF. The Chilton ionosonde is “near” the Inskip end of the circuit. (Dourbes is the closest Digisonde to the circuit midpoint, but the Dourbes ionograms were not suitable for analysis.) The Rome and Chilton hourly ionograms have been rescaled manually using SAO Explorer. The Chilton nighttime ionograms were of poor quality and affected by spread F, so even the manual scaling was unreliable. Only the daytime values of foF2 and M(3000)F2 are reliable (basically 0800 to 1600 UT/LT).

[34] Figures 567 through 8 show the diurnal variation of the observed and GAIM monthly median values of foF2 and M(3000)F2 for the two stations.

Figure 5.

Diurnal variation of the observed and GAIM monthly median values of foF2 at Rome (RO041), December 2003.

Figure 6.

Diurnal variation of the observed and GAIM monthly median values of foF2 at Chilton (RL052), December 2003.

Figure 7.

Diurnal variation of the observed and GAIM monthly median values of M(3000)F2 at Rome (RO041), December 2003.

Figure 8.

Diurnal variation of the observed and GAIM monthly median values of M(3000)F2 at Chilton (RL052), December 2003.

[35] Concentrating on the noon values of foF2 and M(3000)F2, the GAIM values of MUF(3000)F2 = foF2 × M(3000)F2 are too low by ∼6% for Rome and ∼29% for Chilton. The ∼18% underestimate in the HFNowcast 1F MUF is therefore not surprising.

[36] The February 2004 results for this circuit also showed a significant underestimate of the 1F MUF during the middle of the day, along with a limited tracking of the day-to-day variability of the ionosphere. The major difference between the 2 months was that the HFNowcast, ASAPS, and VOACAP results were very similar during the day, indicating that the smaller ASAPS and VOACAP errors found for December 2003 were fortuitous. The GAIM profiles for February 2004 have not been investigated in detail.

5.3. Comparison With HIDIVE Results

[37] The HIDIVE campaign [Eccles et al., 2005] recorded HF signal strengths on circuits from WWV (Fort Collins, Colorado) and WWVH (Hawaii) to various locations in the western United States. Although these observations provide the signal strength, in contrast to the more limited LUF/MUF given by HFNowcast, they can provide useful validations of the HFNowcast values of the FLUF as well as the FMUF. Semiquantitative comparisons have been made for selected days between HFNowcast and HIDIVE for December 2003 and September 2004.

[38] Figure 9 shows a sample HIDIVE plot for reception of WWV and WWVH signals at Bear Lake Observatory (BLO, Utah; 41.9°N, 111.4°W; circuit lengths 550 and 3990 km) for 1 December 2003. (Subsequent plots on the HIDIVE web site give the results for days 2 to 7.)

Figure 9.

HIDIVE observations of the signal strength of signals from WWV (top curve, green) and WWVH (second from the top, blue) recorded at Bear Lake Observatory, 1 December 2003. (Screen capture from Web site).

[39] Superimposed on the HIDIVE curves as thick colored horizontal bars are the intervals for which the HFNowcast MUF exceeded 5 MHz, and the HFNowcast FLUF was less the 5 MHz (which define the range of frequency support for the 1F mode). These bars are placed at convenient y coordinates, since HFNowcast does not provide signal strengths.

[40] Considering first the WWVH (blue) data, it can be seen that the HFNowcast support between 0500 and 1400 UT (nighttime) is consistent with the observations. The observed signal strength falls to zero (nothing observed) during the middle of the day because of HF absorption. The observed signal strength on the WWV circuit (green) also drops dramatically (by over 35 db) during the day. In both cases, the HFNowcast value of the FLUF exceeded 5 MHz, so no propagation was predicted. The observed WWV propagation lasted through the night (depending on the required signal strength), whereas HFNowcast predicted no propagation between 0500 and 1400 UT, as the FMUF fell below 5 MHz.

[41] Similar comparisons of the HFNowcast MUFs and the HIDIVE signal strength plots for other days confirm the general validity of the HFNowcast ranges of frequency support for circuits from WWV and WWVH to Bear Lake Observatory and Klamath Falls (Oregon, 42.2°N, 112.8°W). As with the Inskip-Rome circuit, HFNowcast tends to underestimate the FMUF, and its day-to-day variability of the predicted propagation does not match the observed variability very well. On the basis of our experience with GAIM [see, e.g., McNamara et al., 2008], it seems likely that the GAIM profiles had errors similar to those illustrated earlier for Rome and Chilton. Some of the HFNowcast errors that occur as the HF absorption changes after dawn and before dusk can be attributed to errors in the empirical formula for the FLUF.

5.4. Comparison With Irirangi-Sydney Observations

[42] Irirangi is located in New Zealand at 39.32°S, 175.40°E. The receive site in Sydney is at 33.9°S, 151.2°E, so the circuit length is ∼2250 km. The expected dominant propagation mode would be the 1F mode. GAIM was run for days 271 to 295 (28 September through 22 October) for other purposes. The closest Digisonde was at Learmonth (114°E), which is too far away to be useful for either assimilation or ground truth. Thus the main data assimilated was GPS TEC data.

[43] Figure 10 shows the observed 1F MOFs for the Irirangi-Sydney circuit for October 2006. Part of the scatter in the observations is probably due to vagaries of the automatic scaling procedure. The highest observed values occur at 0000 to 0400 UT (∼1100 to 1500 LT), when they are 15–17.5 MHz. Figure 11 shows the HFNowcast 1F MUFs on the same scale. (Figure 11 shows the MUFs for all days 271 to 295. In fact, there were observations for only days 274–276 and 291–293. However, the following conclusions also apply to the restricted HFNowcast data set.)

Figure 10.

Observed values of the 1F Maximum Observed Frequency (MOF), Irirangi-Sydney, October 2006.

Figure 11.

HFNowcast values of the 1F MOF, Irirangi-Sydney, October 2006.

[44] Comparison of Figure 11 with Figure 10 shows that the HFNowcast values are ∼2 MHz too low during the day. This is consistent with the underestimation of the MOF seen with the other circuits during the day. The MOFs are perhaps a little high during the night, but the observations are rather spread and make it hard to derive a reliable median. The MOFs show spikes at 1600 and 2000 UT, which are 0500 and 0900 LT at the circuit midpoint. The London-Rome circuit showed a discontinuity at 0700 UT or ∼0800 LT, which was attributed to issues with the GAIM plasma frequency profiles.

[45] The closest ionosonde to the Irirangi-Sydney circuit is at Canberra (35.32°S, 149.0°E), which is ∼200 km southeast of Sydney. Thus the GAIM errors in foF2 and M(3000)F2 at Canberra can be only indicative of the errors at the circuit midpoint, although the latitudes are reasonably close. Figures 12 and 13 show the observed and GAIM median values of foF2 and M(3000)F2 at Canberra for October 2006.

Figure 12.

Median observed and GAIM values of foF2, Canberra, October 2006.

Figure 13.

Median observed and GAIM values of M(3000)F2, Canberra, October 2006.

[46] The major source of error in the HFNowcast MOF would be GAIM's underestimate of foF2, which is ∼10% during the middle of the day. However, GAIM also underestimates the value of M(3000)F2 by ∼3% during the middle of the day. The combined errors result in an underestimate of foF2 × M(3000)F2 by ∼13%. This is consistent with the difference between the HFNowcast values of the MOF (∼15 MHz, Figure 11) and observed values of the MOF (∼17 MHz, Figure 10) near midday. The accuracy of the GAIM values of foF2 is commensurate with the small number of widely separated GPC TEC stations used in the assimilation runs (from Townsville, Canberra, and Darwin).

6. Conclusions

[47] The utility of programs such as HFNowcast will clearly depend on the accuracy of assimilative models such as USU-GAIM that provide a real-time global ionospheric specification. In regions that provide a large amount of observations that can be assimilated, GAIM/HFNowcast should already provide more accurate specifications of the frequency support for HF communications than are provided by monthly median propagation prediction programs. McNamara et al. [2007] showed that this happy stage has not yet been reached with the widely dispersed GPS TEC stations in Australia.

[48] As the accuracy of GAIM specifications increases, circuit planning should still be based on the standard prediction programs, but the real-time specification of the range of frequency support should benefit from the use of GAIM-type models of the ionosphere. It can be expected that GAIM/HFNowcast will start to win out over the monthly median programs as the new solar cycle develops, when a single global value of R12 will become less reliable and the GAIM models continue to be improved.


[49] The development of HFNowcast was funded by AFWA under contract FA8718-08-C-0012. AFRL also runs a program of USU-GAIM validation on behalf of AFWA under the same contract. The Digisonde ionograms for Chilton and Rome were obtained from DIDBase ( and rescaled using SAO Explorer ( The HIDIVE observations were obtained from∼rice/hidive/index.html. The Inskip-Rome observations were made available by QinetiQ, United Kingdom, in a continuing program of cooperation. The Irirangi-Sydney observations were provided by IPS Radio and Space Services.