SEARCH

SEARCH BY CITATION

Keywords:

  • ionosphere;
  • modeling;
  • HF predictions

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Derivation of MUF(3000)F2
  5. 3. Australian Ionosonde and GPS TEC Locations
  6. 4. Discrepancies in GAIM Values of Median foF2, M(3000)F2, and MUF(3000)F2
  7. 5. Comparison of ASAPS, GAIM, and Observed Values of MUF(3000)F2
  8. 6. Discussion
  9. Acknowledgments
  10. References
  11. Supporting Information

[1] The accuracy of the Utah State University Global Assimilation of Ionospheric Measurements (GAIM) monthly median values of the maximum usable frequency for an HF communications circuit has been determined for 3000 km circuits centered on Australian ionosonde locations for March and April 2004. The accuracy is discussed in terms of foF2, M(3000)F2, and the product MUF(3000)F2 = foF2 * M(3000)F2. Ground truth is provided by hourly values of foF2 and M(3000)F2 that have been hand-scaled from ionograms. In general, the accuracy of the GAIM predictions of foF2 exceeds that for M(3000)F2. Given that M(3000)F2 is a function of the height and shape of the F2 subpeak, it follows that these parameters of the GAIM F2 peak need to be improved. The GAIM errors in foF2 are quite large at Learmonth during the day, even though Learmonth is a digital ionospheric sounding systems (DISS) station that provides real-time data to GAIM. We interpret these errors in terms of limitations of the current GAIM model. Comparisons of the GAIM errors in the monthly median MUF(3000)F2 with those given by the propagation prediction program Advanced Stand Alone Prediction System (ASAPS) show that the GAIM errors are generally somewhat greater than the ASAPS errors, but we consider GAIM's performance to be creditable given that most of the data available for assimilation is GPS total electron content observations. Overall, the daytime GAIM errors are 3 MHz or 10%, while the nighttime errors are 3 MHz or 16%. On average, the GAIM errors are about 3/2 the ASAPS errors during the day and twice the ASAPS errors at night.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Derivation of MUF(3000)F2
  5. 3. Australian Ionosonde and GPS TEC Locations
  6. 4. Discrepancies in GAIM Values of Median foF2, M(3000)F2, and MUF(3000)F2
  7. 5. Comparison of ASAPS, GAIM, and Observed Values of MUF(3000)F2
  8. 6. Discussion
  9. Acknowledgments
  10. References
  11. Supporting Information

[2] The Utah State University (USU) Global Assimilation of Ionospheric Measurements (GAIM) model [Scherliess et al., 2004; Schunk et al., 2004], in particular the USU Gauss-Markov Kalman filter (GMKF) model [Thompson et al., 2006], has been tested extensively by both the Air Force Research Laboratory (AFRL) and the USU group for its ability to predict the TOPEX/Jason observations of total electron content (TEC). However, these validations have little to say about the model's ability to provide support for applications that use only that part of the ionosphere below the F2 peak, such as HF communications and surveillance. Multiple validation studies at AFRL in support of the United States Air Force Weather Agency (AFWA) have confirmed the model's capabilities in different scenarios. These (unpublished) studies included comparisons with TOPEX TEC, but also included tests of the validity of the model's plasma frequency profiles at Learmonth (Australia) and Jicamarca (Peru), and of the predicted values of foF2 for Australian locations.

[3] This paper is concerned with the ability of the USU model to provide the ionospheric basis for making high-frequency (HF) radio propagation predictions, in particular the maximum usable frequency (MUF) for the standard 3000 km circuit. Monthly median MUF predictions are provided by propagation prediction programs such as the Australian Advanced Stand-Alone Prediction System [Australian IPS Radio and Space Services, 2005]. Propagation prediction programs also provide statistical distributions for the propagation, as well as signal strengths and signal-to-noise ratios [McNamara et al., 2006]. We do not address either of these options here, but choose to concentrate on the MUF as a first step. We focus on the MUF because it is readily derivable from ionograms, it is a standard parameter used by HF users, and is a useful metric by which to assess the F2 region performance of an ionospheric model.

[4] AFWA uses an assimilative ionospheric model for making real-time HF propagation predictions in its Operational Space Environment Network Display (OpSEND) suite of applications [Bishop et al., 2004]. The USU GAIM model is currently being phased in to replace PRISM [Anderson, 1993] for all OpSEND products. In that context, the primary goal of GAIM is to provide a real-time ionosphere that specifies “today's ionospheric weather.” However, a first test of such a model is to verify that it produces accurate monthly median behavior. This is the capability we focus on in this paper. That is, we study how well GAIM produces the monthly median MUF.

[5] The interval considered in this paper is the GAIM example period 1 (EX1), which covers the 26 days from 80 through 105 (20 March to 14 April) in 2004. The highest value of the planetary Ap index was 23, so no major ionospheric perturbations would be expected a priori. Nevertheless, there were several days for which the observed values of foF2 were significantly different from the monthly median. A detailed comparison of the GAIM and observed values has shown that GAIM tends to track the disturbed days, but generally tends to be conservative, not matching the extent of the departures of the disturbed days from the median values [Decker and McNamara, 2007].

[6] Section 2 of this paper discusses the origin and physical interpretation of the term M(3000)F2, which is more familiar to HF communicators than to ionospheric modelers. Section 3 gives the locations of the ionosonde and GPS TEC sites. Australia provides an ideal test bed for ionospheric assimilation studies. In section 4, we compare GAIM and observed monthly median values of foF2, M(3000)F2 and MUF(3000)F2. The observed values of foF2 and M(3000)F2 were derived by manual scaling (by IPS staff on a routine basis) of the ionograms produced by IPS Radio and Space Services ionosondes, while the Learmonth ionograms were autoscaled by ARTIST, with a limited amount of manual scaling. The errors in the GAIM values of foF2 at Learmonth are quite high during the day, which we attribute to the way GAIM weights its various data sources. Section 5 compares the values of MUF(3000)F2 given by GAIM, the values given by the propagation prediction program Advanced Stand Alone Prediction System (ASAPS), and the “observed” values of this parameter derived from the manually scaled data. Equality of the GAIM and ASAPS errors, which has not yet been achieved, would be a significant achievement in the development of GAIM. Section 6 summarizes our findings, notes limitations of the current GAIM methodology, and suggests areas for further study.

2. Derivation of MUF(3000)F2

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Derivation of MUF(3000)F2
  5. 3. Australian Ionosonde and GPS TEC Locations
  6. 4. Discrepancies in GAIM Values of Median foF2, M(3000)F2, and MUF(3000)F2
  7. 5. Comparison of ASAPS, GAIM, and Observed Values of MUF(3000)F2
  8. 6. Discussion
  9. Acknowledgments
  10. References
  11. Supporting Information

[7] Is should first be noted that the term MUF is used in two different senses, the instantaneous MUF and the monthly median MUF. The instantaneous MUF for a 3000 km circuit is simply given by

  • equation image

where foF2 is the critical frequency of the F2 layer (the highest frequency that would be reflected by the ionosphere for ordinary-mode vertical propagation), and M(3000)F2 is the so-called m or MUF factor that will be described in more detail. The simple theory for oblique HF propagation via the ionosphere is given by Davies [1990, chapter 6]. The m factor is basically a geometrical factor that depends on the length of the circuit and the height at which the signals are reflected. For a simple one-hop circuit using a flat Earth and flat ionosphere, the m factor is simply sec Φo, where Φo is the angle of incidence of the ray at the base of the ionosphere. Extending this simple model to a curved Earth leads to a “corrected sec Φ”, which is written as M(D)F2, for a circuit length D.

[8] For manual scaling of ionograms, the m factor is derived by sliding a curve known as the MUF slider [see, e.g., Piggott and Rawer, 1972] parallel to the frequency scale until it is tangential to the F2 ordinary-ray trace. The slider is actually an empirical 3000 km transmission curve, as described by Davies [1990, Figures 6.2 and 6.12] In general, the slider will cut the F2 trace at two points, corresponding to the low and high rays that exist on every oblique circuit. Tangency corresponds to the degenerate case in which the low and high rays are superimposed. The values of M(3000)F2 are dimensionless, and usually lie within the range 2.5 to 4.0. The frequency at which the MUF slider is tangential to the F2 trace is typically ∼0.9 foF2, so a correct value of M(3000)F2 relies on the GAIM profile being a good fit to the actual profile over the range 0.8 to 1.0 times foF2.

[9] Monthly median worldwide maps of M(3000)F2, like those for foF2, are provided by URSI or CCIR/ITU. IPS Radio and Space Services uses its own maps [Fox and McNamara, 1988] of these parameters, which are based on observations that extend back over 50 years. The values of MUF(3000)F2 provided by HF propagation prediction programs are monthly median values. It is these monthly median values, and not the instantaneous values, that we discuss in this paper.

[10] The GAIM ionosphere is specified in terms of the electron density profile at 83 altitudes for each 15 min at an array of 1056 grid points (44 latitudes, 24 longitudes). The vertical incidence ionogram for each profile has been derived by direct integration over height up to the reflection point of (μ′ − 1), where μ′ is the group refractive index, and the value of M(3000)F2 derived from that ionogram. Ionogram autoscaling systems have replaced the manual MUF slider procedure described above by a numerical procedure that uses a fit [Paul, 1982] to the 3000 km transmission curve:

  • equation image

[11] The transmission curve given in Figure 6.12 of Davies [1990] gives h′ versus M, and the above equation represents the inverse curve. The value of M(3000)F2 is found by searching for the maximum in the value of f(i) * M(h′(i)) for points (f, h′) along the F2 ordinary ray trace, and then dividing that maximum value by foF2.

3. Australian Ionosonde and GPS TEC Locations

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Derivation of MUF(3000)F2
  5. 3. Australian Ionosonde and GPS TEC Locations
  6. 4. Discrepancies in GAIM Values of Median foF2, M(3000)F2, and MUF(3000)F2
  7. 5. Comparison of ASAPS, GAIM, and Observed Values of MUF(3000)F2
  8. 6. Discussion
  9. Acknowledgments
  10. References
  11. Supporting Information

[12] The USU GAIM model runs used in this study assimilated TEC observations from a number of GPS observatories in the Australian region, as well as the plasma frequency profile from the ionosonde at Learmonth. These locations, as well as the locations of the IPS ionosondes, are given in Figure 1 and Table 1.

image

Figure 1. Locations of Australian ionosondes (uppercase and lowercase letters) and GPS sites (uppercase letters).

Download figure to PowerPoint

Table 1. Locations of the Australian Ionosondes
SiteLatitude, degLongitude, degGAIM Latitude, degGAIM Longitude, deg
Vanimo, PNG−2.70141.30−2.30142.5
Port Moresby, PNG−9.4147.1−11.67142.5
Darwin−12.45130.95−11.67127.5
Townsville−19.63146.85−21.0142.5
Learmonth (DISS)−22.25114.08−21.0112.5
Brisbane−27.53152.92−25.64157.5
Norfolk Island−29.03167.97−30.33172.5
Mundaring−31.98116.22−30.33112.5
Canberra−35.32149.0−35.0142.5
Hobart−42.92147.32−44.33142.5
Macquarie Island−54.5159.0−53.67157.5

[13] The fourth and fifth columns give the location of the nearest GAIM latitude/longitude grid point to the ionosonde. Comparisons between GAIM and ionosonde parameters are made at these grid points. With the current state of accuracy of the GAIM predictions, this is not a significant source of error. The Hobart and Macquarie ionosonde observations were too limited to be useful for the present study. Learmonth is a digital ionospheric sounding systems (DISS) site that operates a Digisonde whose plasma frequency profiles are assimilated by GAIM. (The DISS network consists of twenty digital ionospheric sounding systems, i.e., the Digisonde 256 [Reinisch, 1996], and is operated by the United States Air Force Weather Agency.) The other ionosondes were manufactured by IPS Radio and Space Services, the main instrument being the IPS 5D Ionosonde.

4. Discrepancies in GAIM Values of Median foF2, M(3000)F2, and MUF(3000)F2

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Derivation of MUF(3000)F2
  5. 3. Australian Ionosonde and GPS TEC Locations
  6. 4. Discrepancies in GAIM Values of Median foF2, M(3000)F2, and MUF(3000)F2
  7. 5. Comparison of ASAPS, GAIM, and Observed Values of MUF(3000)F2
  8. 6. Discussion
  9. Acknowledgments
  10. References
  11. Supporting Information

[14] We have compared the GAIM and observed values of foF2, M(3000)F2 and MUF(3000)F2 for eight IPS ionosonde locations, and for Learmonth, which is a DISS site. However, we restrict our discussions here to three sites that demonstrate the most interesting comparisons. Canberra is a midlatitude IPS site, Vanimo is a low-latitude example, and Learmonth is our single example of a DISS site whose data was assimilated by the model.

4.1. Canberra

[15] Figure 2 shows the diurnal variation of the GAIM and observed median values of foF2 for Canberra. The agreement between the GAIM and observed median values of foF2 for Canberra is clearly very good (generally within 10%). This is not surprising, given that there is a GPS site (STR1/2, whose data was used by GAIM) very near the ionosonde, with two others relatively nearby at Melbourne (MOBS) and Hobart (HOB2). Decker and McNamara [2007] show that the GAIM predictions of foF2 are more accurate when there is a GPS TEC site nearby.

image

Figure 2. Observed and USU GAIM median values of foF2 for Canberra.

Download figure to PowerPoint

[16] Figure 3 shows the GAIM and observed median values of M(3000)F2 for Canberra. The largest errors in M(3000)F2 occur around sunrise. For example at 1800 UT (∼0400 LT), we see that GAIM underestimates the observed median by 0.25 or 8%. From an HF communications point of view, the most important parameter is the MUF(3000)F2, which is shown for Canberra in Figure 4. Recall that MUF(3000)F2 is simply the product of foF2 and M(3000)F2 (medians), and is expressed in MHz.

image

Figure 3. Observed and USU GAIM median values of M(3000)F2 for Canberra.

Download figure to PowerPoint

image

Figure 4. Observed and USU GAIM median values of MUF(3000)F2 for Canberra.

Download figure to PowerPoint

[17] As with foF2, there is very good agreement between the GAIM and observed values of MUF(3000)F2. That is, GAIM reproduces the basic diurnal shape of the observed median with errors that are small compared to the magnitude of the diurnal variation. There is some cancellation of the over and underestimates of foF2 and M(3000)F2. The percentage errors are largest around dawn (2000 UT is 0600 LT). Separate studies at AFRL found that the largest GAIM profile discrepancies tended to occur at 0530 LT at many locations. The GAIM values of MUF(3000)F2 are a little over 10% too low during the morning rise.

4.2. Vanimo

[18] Vanimo lies under the southern equatorial anomaly crest. Studies at AFRL have shown that calculated GAIM values of TEC for GPS lines of sight to the north of Darwin (looking into the anomaly) are significantly less accurate that those to the south, where the horizontal gradients in the ionosphere are smaller and where there is also more data for assimilation. GPS observations are not available in real time from either Vanimo or Port Moresby, so they are not used by GAIM. Figure 5 shows the median values of foF2 for Vanimo.

image

Figure 5. Observed and USU GAIM median values of foF2 for Vanimo.

Download figure to PowerPoint

[19] The agreement between the USU and observed median values of foF2 for Vanimo is somewhat surprising, given the dearth of real-time observations. However, it is not quite as good as the previous (midlatitude) example, since there are times when the differences exceed 10%. Note the gap in the observations between 0900 and 1200 UT, for which more than half the observations were qualified in some sense and therefore rejected. This is ∼1830 to 2130 LT, when equatorial spread F is often present. Figure 6 shows the median values of M(3000)F2 for Vanimo.

image

Figure 6. Observed and USU GAIM median values of M(3000)F2 for Vanimo.

Download figure to PowerPoint

[20] This plot is the one that shows some of the largest errors for any station or parameter. The GAIM values of M(3000)F2 are systematically low, by a maximum of ∼15%. These large errors translate into larger errors in MUF(3000)F2, as illustrated in Figure 7. This is particularly true where both foF2 and M(3000)F2 are underestimated by GAIM.

image

Figure 7. Observed and USU GAIM median values of MUF(3000)F2 for Vanimo.

Download figure to PowerPoint

[21] The error at 0000 UT (∼0930 LT) is –23%, while at 1500 UT (∼0030 LT) it is about –30%. Given that the values of M(3000)F2 depend mainly on the shape and the height of the F2 layer peak, it is clear that GAIM's ability to model all aspects (magnitude, height, and shape) of the peak of the F2 layer is important in specifying MUF(3000)F2. GAIM does not do a good job at this equatorial anomaly location.

4.3. Learmonth

[22] The Learmonth ionograms are automatically scaled by the University of Massachusetts program Automatic Real Time Ionogram Scaler with True Height (ARTIST) [see Reinisch et al., 2005], and are subject to the vagaries of the autoscaling procedures. The standard archive output (SAO) files have therefore been processed by the program QualScan [McNamara, 2006], to reject unreliable scaled traces. The median values of foF2 and M(3000)F2 that we are concerned with here are in fact basically unaffected by this filtering process. Figure 8 shows the median ARTIST values of foF2 for Learmonth.

image

Figure 8. Observed and USU GAIM median values of foF2 for Learmonth.

Download figure to PowerPoint

[23] The curves in Figure 8 show that the GAIM median is a good match to the autoscaled data, except between 0400 and 0700 UT (1140–1440 LT). Manual examination of the 0600 UT ionograms for EX1 using the University of Massachusetts Lowell application SAO Explorer http://ulcar.uml.edu/SAO-X/SAO-X.html) confirms the general validity of the autoscaled (observed) values of foF2. The values of foF2 manually scaled by IPS Radio and Space Services (see http://www.ips.gov.au/World_Data_Centre/1/2) are also consistent with the autoscaled values during the day. Thus the difference between the GAIM and autoscaled medians must arise from incorrect GAIM values of foF2.

[24] This is a curious result because the bottomside profiles from the Learmonth DISS observations were assimilated in these runs of GAIM. Learmonth is one location at which good agreement with the observed foF2 values might be expected. In an attempt to clarify what is happening in this case, we ran GAIM twice. (The GAIM code was supplied to AFRL by the USU group as part of a validation package.) For the first run, the only data provided to GAIM was the GPS-based TEC. For the second run, the only data provided to GAIM was the DISS data from Learmonth (and other locations outside Australia). Comparison of the two outputs showed that it was the TEC data that led to the larger foF2 values seen during the day. The DISS-only runs produced foF2 values that were more consistent with the observations. We also found that if only TEC data was provided to GAIM, the GAIM values of TEC for Learmonth between 0400 and 0700 UT were very similar to the Karratha (KARR, 347 km from Learmonth) observations of the GPS TEC.

[25] Our interpretation of these results is that the GPS TEC data is having a greater influence than the Learmonth ionosonde data, for two reasons: (1) GAIM assigns fairly conservative (large) error bars to the DISS ionogram values of plasma frequency at each GAIM altitude because of the known difficulties encountered in the autoscaling. Thus it may typically give lesser weight to the ionogram data than to the TEC data. (2) At a location near both DISS and GPS stations, there will be more slant TEC measurements than ionosonde profiles. As pointed out by Thompson et al. [2006], when there are several TEC measurements within a single grid cell the current implementation of the Kalman Filter will “trust” the TEC measurements “too much.” Thus the TEC data in this case pushes GAIM to having larger foF2 values than the Learmonth DISS data indicates. It is intended that the program QualScan [McNamara, 2006] be used to postprocess all autoscaled ionosonde data passed to GAIM, thus allowing more confidence to be assigned to the DISS data. However, the issue of giving too much weight to multiple TEC measurements within a grid cell still remains and will need further study.

[26] Figure 9 shows the USU and observed values of M(3000)F2 for Learmonth. The agreement of the GAIM and observed values of M(3000)F2 is quite good, with similar short-term variations. This is expected, given that GAIM assimilates the Learmonth profiles, and the TEC observations provide very little (if any) profile information.

image

Figure 9. Observed and USU GAIM median values of M(3000)F2 for Learmonth.

Download figure to PowerPoint

[27] Figure 10 shows the diurnal variation of the MUF(3000)F2 for Learmonth. The overestimates of the MUF during the day are largely a result of the GAIM values of foF2 being too high. Underestimates of both foF2 and M(3000)F2 contribute to the underestimates of the MUF around 2000 UT (∼0400 LT).

image

Figure 10. Observed and USU GAIM median values of MUF(3000)F2 for Learmonth.

Download figure to PowerPoint

5. Comparison of ASAPS, GAIM, and Observed Values of MUF(3000)F2

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Derivation of MUF(3000)F2
  5. 3. Australian Ionosonde and GPS TEC Locations
  6. 4. Discrepancies in GAIM Values of Median foF2, M(3000)F2, and MUF(3000)F2
  7. 5. Comparison of ASAPS, GAIM, and Observed Values of MUF(3000)F2
  8. 6. Discussion
  9. Acknowledgments
  10. References
  11. Supporting Information

[28] The current benchmarks for predicted monthly median MUFs in the Australian region are provided by ASAPS. MUFs based on the GAIM model should (ideally) be at least as good as those of ASAPS. This section therefore compares the three sets of median MUFs for 3000 km circuits centered on three ionosonde locations: the GAIM, ASAPS and “observed” (from the ionogram) MUFs. Given its location, the longevity of its ionogram data sets, and the locations of the GPS TEC sites, Canberra should be an ideal location for both GAIM and ASAPS. We therefore start with the Canberra MUF(3000)F2 results, which are shown in Figure 11. The ASAPS values of MUF were averaged for March and April 2004. The values of the ionospheric T index used by ASAPS were those based on historical (i.e., after the fact) observations of foF2 at sixteen worldwide stations for those two months, so ASAPS was effectively run with real-time assimilation of foF2 from those stations.

image

Figure 11. ASAPS, GAIM, and “observed” values of median MUF for a Canberra-centered circuit.

Download figure to PowerPoint

[29] As can be seen from Figure 11, both GAIM and ASAPS track the “observed” median MUFs very well from 0200 to 1600 UT (1200 to 0200 LT). At other times, ASAPS tracks the observations very well, whereas the GAIM values of MUF are underestimates. The lowest curve gives the standard deviation of the individual observed values of the MUF, which is typically ∼3 MHz. This provides another criterion by which to assess the quality of the median specifications. The largest GAIM errors slightly exceed this value. Figure 12 shows the results for Darwin, which is less favorable than Canberra for both GAIM and ASAPS.

image

Figure 12. ASAPS, GAIM, and “observed” values of median MUF for a Darwin-centered circuit.

Download figure to PowerPoint

[30] The ASAPS values of the MUF are more accurate than the GAIM values between 0000 and 1400 UT (∼0900 to 2300 LT), but there is not much difference at the other times. Figure 13 shows the MUF results for Learmonth, which is the DISS location that provides input to GAIM.

image

Figure 13. ASAPS, GAIM, and “observed” values of MUF for a Learmonth-centered circuit.

Download figure to PowerPoint

[31] The largest errors occur during the day, which is when the GAIM values of foF2 are most in error, as discussed earlier. Basically, for much of the daytime GAIM overestimates the MUF, while ASAPS underestimates it. GAIM is at least as, or more accurate than, ASAPS at other times. The RMS and percentage errors in MUF(3000)F2 for GAIM and ASAPS for all nine stations are listed in Table 2.

Table 2. RMS Percentage Errors in MUF(3000)F2 for GAIM and ASAPS
StationLatitude, degDay (0600–1800 LT)Night (1800–0600 LT)
GAIMASAPSGAIMASAPS
Vanimo, PNG−2.74.92.54.23.1
Port Moresby, PNG−9.44.21.76.52.4
Darwin−12.43.82.74.62.5
Townsville−19.63.02.01.80.9
Learmonth (DISS)−22.23.83.71.41.6
Brisbane−27.52.81.52.50.9
Norfolk Island−29.02.01.72.30.6
Mundaring−32.02.01.01.11.4
Canberra−35.31.80.81.80.5
Average (9 stations) 3.12.02.91.6
Average percent 10.46.815.68.3

[32] It can be seen that the GAIM errors are all greater than or equal to the ASAPS errors, except for Mundaring and Learmonth at night. The daytime errors tend to be greatest for the three low-latitude stations, for both GAIM and ASAPS. The daytime errors for Learmonth are atypically high for these mid latitude stations. The high GAIM errors arise from the large errors in foF2 noted earlier in Figure 8. The high ASAPS errors presumably arise from errors in the IPS maps of foF2 or M(3000)F2. The GAIM and ASAPS values of MUF(3000)F2 straddle the observed values, as shown earlier in Figure 13. Overall, the daytime GAIM errors are 3 MHz or 10%, while the nighttime errors are 3 MHz or 16%. On average, the GAIM errors are about 3/2 the ASAPS errors during the day, and twice the ASAPS errors at night. Note that there is some uncertainty with the Learmonth comparisons because of limitations of the ionogram autoscaling. Only the 0600 UT ionograms have been manually rescaled.

6. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Derivation of MUF(3000)F2
  5. 3. Australian Ionosonde and GPS TEC Locations
  6. 4. Discrepancies in GAIM Values of Median foF2, M(3000)F2, and MUF(3000)F2
  7. 5. Comparison of ASAPS, GAIM, and Observed Values of MUF(3000)F2
  8. 6. Discussion
  9. Acknowledgments
  10. References
  11. Supporting Information

[33] Our comparisons have shown that the Utah State University Gauss-Markov Kalman filter Global Assimilation of Ionospheric Measurements (GAIM) model performed creditably in predicting monthly median MUFs for Australian circuits. By creditable we mean that the GAIM errors in the median behavior were in the 10% range. This is a creditable performance because the basic diurnal variation of foF2 and MUF(3000)F2 is large compared to 10% (generally the difference between night and day is over a factor of two). A somewhat more stringent criterion is to compare GAIM errors to the standard deviation of the observations themselves. We saw examples of the observed standard deviations in Figures 11–13, (bottom curve). These values can be compared with the errors given in Table 2. We see that for Canberra, GAIM RMS errors are less than the observed standard deviation. This is not the case for Darwin and Vanimo. GAIM still has room for improvement by this standard. Of course, what we would like to see is the GAIM errors for all locations be clearly smaller than the observed standard deviations.

[34] It is encouraging that while GAIM was effectively driven by just TEC data, it produced an F2 region with reasonably accurate HF propagation characteristics. Although there is room for improvement (see below), this speaks well for the assimilation process and for the theoretical physics model that provides the shape of the F2 peak profile. However, GAIM did not perform as well as the ASAPS empirical HF propagation program in predicting monthly median MUFs. It should not be surprising that an empirical model based on decades of Australian ionosonde data and designed specifically to produce monthly median values of the MUF(3000) performs better in its home territory. However, for regions not well defined by extensive historical ionosonde observations, GAIM could easily be more accurate than ASAPS or other similar programs, provided it is supported by adequate real-time observations of TEC.

[35] GAIM reproduces the monthly median values of foF2 better than it reproduces the values of M(3000)F2. The errors in the GAIM values of M(3000)F2 suggest that there is room for improvements in the underlying physics model, especially in terms of the shape and height of the F2 peak. Having more ionosonde profiles available for assimilation, along with a means to use such data more effectively in a TEC-rich environment, should lead to an improved shape of the GAIM F2 peak.

[36] For near vertical incidence circuits, the ordinary ray instantaneous MUF is equal to foF2, and GAIM has been shown to track even the day-to-day variability of foF2 (the “weather” in foF2) at Australian locations quite well [Decker and McNamara, 2007]. For one-hop circuits lengths between zero and 3000 km, the profile shape between the E and F layer peaks (the F1 region) will be important. The current Gauss-Markov model does not yield good profiles at F1 altitudes, as was found by comparing GAIM and ARTIST profiles for Learmonth and Jicamarca (unpublished AFRL study). However, the “full physics” model being developed by the USU group will be able to provide more reliable profiles at these altitudes (R. W. Schunk, personal communication).

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Derivation of MUF(3000)F2
  5. 3. Australian Ionosonde and GPS TEC Locations
  6. 4. Discrepancies in GAIM Values of Median foF2, M(3000)F2, and MUF(3000)F2
  7. 5. Comparison of ASAPS, GAIM, and Observed Values of MUF(3000)F2
  8. 6. Discussion
  9. Acknowledgments
  10. References
  11. Supporting Information

[37] The GAIM profiles and source code were provided to AFRL by USU to support an AFRL validation effort under AFRL contract FA8178-04-C-0055. We greatly appreciate the cooperation of the USU scientists in our validation studies.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Derivation of MUF(3000)F2
  5. 3. Australian Ionosonde and GPS TEC Locations
  6. 4. Discrepancies in GAIM Values of Median foF2, M(3000)F2, and MUF(3000)F2
  7. 5. Comparison of ASAPS, GAIM, and Observed Values of MUF(3000)F2
  8. 6. Discussion
  9. Acknowledgments
  10. References
  11. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Derivation of MUF(3000)F2
  5. 3. Australian Ionosonde and GPS TEC Locations
  6. 4. Discrepancies in GAIM Values of Median foF2, M(3000)F2, and MUF(3000)F2
  7. 5. Comparison of ASAPS, GAIM, and Observed Values of MUF(3000)F2
  8. 6. Discussion
  9. Acknowledgments
  10. References
  11. Supporting Information
FilenameFormatSizeDescription
rds5414-sup-0001-t01.txtplain text document1KTab-delimited Table 1.
rds5414-sup-0002-t02.txtplain text document1KTab-delimited Table 2.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.