|Back to INAG Homepage||Back to UAG-104 Contents page|
AUTOMATIC IONOGRAM PROCESSING SYSTEMS IN JAPAN
S. Igi, K. Nozaki, M. Nagayama, A. Ohtani, H. Kato, and K. Igarashi
Communications Research Laboratory, Ministry of Posts and Telecommunications, 4-2-1 Nukui-kita, Koganei, Tokyo 184, Japan
A new system has been developed that digitises, archives, and autoscales the ionograms obtained by the four observatories located throughout Japan. This paper describes an automatic ionogram processing system and a new display method that shows both ionospheric phenomena and propagation conditions of HF radio waves. The test of the automatic scaling algorithm, performed using nearly 7000 ionograms, indicates that the algorithm can successfully derive the values of foF2, fEs, and fmin within an error of 1 MHz or less, more than 80%, 90% and 99% of the time, respectively. However, the reliability for foF2 is poorer during the summer season because of the presence of the presence of blanketing-type Es. With regard to obtaining the median values of foF2, fEs, fmin, and h'Es, the automatic scaling algorithm is found to perform well. The automatic scaling codes were converted to a M-750V system (IBM compatible system) in 1991, and the UNIX system in 1993.
Synoptic ionospheric measurements in the form of ionograms have provided valuable information both for the HF communicator and for studies related to the physics of the ionosphere. However, the scaling of ionospheric parameters is a laborious task because of complex structures that can often appear on an ionogram. Furthermore, the appearance of two echo traces, called the ordinary and extraordinary components, adds further complexity to ionogram scaling. For the reason cited above, highly specialised experts at a majority of ionospheric observatories routinely do scaling.
In recent years, automatic scaling systems have been developed by several organisations , primarily motivated by the following reasons: the difficulties involved in training skilled scalers; the requirement for the availability of real-time data, and; the elimination of human errors. The Communication Research Laboratory (CRL) has completed the development of a fully automatic system that can be attached to conventional ionosondes for the purpose of collecting and processing ionospheric data .
This paper describes an automatic scaling algorithm for those ionograms which do not distinguish between ordinary and extraordinary mode components. In order to evaluate the performance of this algorithm, automatically scaled parameters are compared with those derived from manual scaling for some 7000 ionograms acquired over an interval of one year. Also, new plots are presented showing the diurnal variation in ionospheric characteristics for surveying ionospheric phenomena.
Automatic Ionogram Scaling
The CRL has completed the development of automatic scaling software to cope with the difficulty of training reliable scalers. In this section, each of the major steps invoked by the algorithm will be described.
Elimination of Multiple Traces
Multiple echo traces are eliminated from the original ionograms by using the criterion that they must be observed at two or more virtual heights at a given frequency. This process becomes important in determining whether or not the F trace is completely blanketed by Es. The details of this decision will be discussed later.
Selection of Areas Containing foF2, fmin, E-layer Region and F-layer Region
This step selects four regions, that is the foF2, fmin, E layer, and F-layer, for the sake of being efficient and economic use of CPU time. The foF2 region is extracted by detecting those echo traces with a large gradient in frequency. As for the fmin region, the lowest frequency region containing echo traces is removed. The criteria used to remove E and F layers are the E and Es layer echoes observed below and the F layers observed above about 180 km are removed.
After applying a low pass filter to the fmin region extracted in the previous step, the scaling program searches for fmin by detecting the lowest frequency of each echo trace. If fmin cannot be found within the frequency range measured by the ionosonde, the program signals that a value fmin cannot be determined because of the presence of radio wave absorption. In this case, the subsequent procedures are terminated, because it is assumed that no ionospheric echoes are observed within the frequency range measured by the ionosonde.
Extraction of E Region Parameters
The echo traces observed in the E regions exhibit very complex structures on an ionogram, because both E and Es layers can exist at similar heights, a situation further complicated by the inability of the ionosonde to separate ordinary and extraordinary mode waves. As a result, the task of automatically scaling E-layer parameters is rather difficult. Therefore, the CRL method scales the top frequency of the Es (fEs) instead of the highest frequency of the ordinary mode component of the Es (foEs), which is often impossible to determine, even by experienced human scalers. Although fEs coincides with fxEs in general, fEs corresponds to foEs in those cases when the extraordinary mode echo is not present.
The first step in obtaining the parameters of the E region is to connect the edges of echo traces that are thought to belong to the same line. The layer (E or Es) and polarisation (O-mode or X-mode) of each line are identified by comparing the shape of the echo traces to patterns stored in the computer memory. The E region parameters are then scaled from each identified echo trace.
Recognition of Blanketing-type Es
To recognise whether or not the F layer is completely blanketed by Es or not is a critical step. Fail here and the program, which searches for F-layer parameters regardless of the appearance of F-layer traces, will be in error. The main reason for adding complexity to this recognition process is that multiple Es and F-layer traces are frequently observed at similar heights. Thus, the multiple traces associated with Es are eliminated in the first step, so that no echoes except for F-layer echoes should be observed at F-layer heights. The presence of completely blanketing-type Es can be determined by detecting whether or not traces in the F region in addition to those associated with the Es layer. In this case, the parameters related to the F-layer are not scaled but symbolised as "blanketed" in case of completely blanketed-type Es.
Only echo traces in the neighbourhood of the F2 cusp are extracted from the foF2 region identified in the second step. Next follows noise reduction with a low pass filter, extraction of the leading edge, and preliminary fitting to the F2 trace with a parabolic curve. The hyperbolic curve described is fitted to the leading edge of the echo traces using an "analysis-by-synthesis" method, which was first used in the pattern recognition of handwritten characters. The frequency of foF2 is derived from the asymptote of the hyperbola determined using this procedure.
Scaling the Height of the F-layer
The parameters related to the F layer are scaled using procedures identical to those applied for the detection of the E-layer parameters.
A typical example of automatic scaling is shown in Fig.1; the arrows indicate the characteristic frequencies and virtual heights that have been automatically scaled. The characteristic values are also shown at the bottom of Fig.1.
A Comparison Between Automatic and Manual Ionogram Scaling Methods
In order to evaluate the CRL method, the automatically scaled parameters have been compared with the manually scaled values using ionograms obtained every hour from April, 1986 to February 1987. Since the details are described by Igi et. al , only the results regarding foF2 are shown here.
Hourly Values and Medians of foF2
A histogram, showing the difference in foF2 between manually and automatically scaled ionograms for April 1986, is presented in .2. It can be seen from this figure that the majority of the differences occur within 1 MHz of the manually scaled value. Figure 3 shows the percentage success for error limits of 1 MHz (solid line) and 0.5 MHz (broken line) obtained using the automatic scaling method extending over nearly one year. The increase in the failure rate during June, July, and August is due to the presence of a blanketing-type Es. This arises because the CRL algorithm frequently concludes that the F layer is completely blanketed in spite of the appearance of higher frequency traces near the F2 cusp. Generally speaking, the 1 MHz error limit test is passed by about 80% of all ionograms in June, July, and August, and by 95% of ionograms recorded at other times of the year. The median values of foF2 obtained by manual and automatic scaling are shown in Fig.4 as a function of local time. As can be seen in the diagram, there is good agreement except at the 21 hour LT.
Close agreement is also obtained in similar comparisons made for July and October 1986 and January 1987, which were selected to be representative of the other seasons.
Hourly Values and Medians of the other parameters
Similar comparisons to those made for foF2 have been made for the other parameters. The automatic scaling algorithm for fmin is nearly perfect throughout the one year period studied here and fEs can be automatically scaled with an error of 1 MHz or less in 90% to 95% of all ionograms studied.
A new method of displaying Ionospheric Characteristics
This section describes a new display method that clearly shows both ionospheric phenomena and propagation conditions of the HF radio wave. The new plots consist of three figures: (1) an fc_h'-t plot showing diurnal variations of ionospheric characteristics, (2) an MUF plot showing diurnal variations of the MUF, and (3) a Median plot showing variations of the median values of the ionospheric characteristics as a function of local time. Only the fc_h'-t plot is described here, because the details are shown in the paper by Igi .
Ionospheric Phenomena - fc_h'-t plot -
A new plot is described showing the diurnal variation in ionospheric characteristics for surveying ionospheric phenomena. This plot indicates the time variations of the characteristic frequencies and virtual heights of each layer, this plot is termed the fc_h'-t plot. The fc_h'-t plot consists of three parts for the F region, the E region, and the virtual height.
Figure 5 shows the process for producing a fc_h'-t plot. Noise and multiple echoes are eliminated from the original digital ionogram. Eliminating multiple echoes help to discriminate F-layer traces from multiple Es-layer traces, because the F-layer trace and the echo traces reflected twice between the Es-layer and the ground are observed at similar heights in the ionogram.
Next, the method of representing the F region and E region is described. The echoes reflected from the F region are plotted by summing up the pixels above 180 km along the virtual height axis at each frequency step from 1 MHz to 25 MHz (see (a) in Fig.6). The echoes for the E region are drawn by summing up the pixels below 180 km (see (b) in Fig.5). The F and E regions of the fc_h'-t plot are made by recording such histograms every fifteen minutes.
The top and middle panels of Fig.6 indicate the diurnal variation of the frequency range of the echoes reflected from the F and E regions, respectively. The two solid curves are the predicted values of fxE and foE calculated using the model recommended by CCIR (shown as fxE(P) and foE(P)).
The bottom panel of Fig.6 shows the diurnal variation of virtual height. First, a histogram of the virtual height is calculated by summing up the pixels along the frequency axis at each virtual height sample from 0 km to 719 km (see (c) in Fig.5). Since virtual heights are scaled from the lowest height of each layer, the differences between successive bars in the histogram (c) are calculated to detect the edge of each layer (see (d) of Fig.5). As positive values of the histogram (d) corresponds to the lowest height of each layer, the virtual height figure is made by sequentially recording positive values of the histogram (d) every 15 minutes.
The fc_h'-t plot aides in surveying ionospheric phenomena.
An automatic ionogram scaling algorithm incorporating a pattern recognition method has been developed at CRL. The CRL method is applicable to the data recorded by conventional type ionosondes, i.e., those that do not distinguish between ordinary and extraordinary mode echoes. Moreover, with a high degree of reliability, the CRL method is able to recognise Es layer blanketing, so that it is useful for scaling ionograms recorded at those latitudes where Es occurrence in dominant. Data obtained using the automatic ionogram scaling system has been routinely published since June 1988 for all ionospheric observatories in Japan (Wakkanai, Kokubunji, Yamagawa, and Okinawa).
Ionospheric phenomena can be easily identified from the automatically scaled parameters and also from sequential records of ionograms called Ionospheric Summary Plots. According to the URSI standard, most ionospheric observatories scale fourteen ionospheric parameters: fmin, foE, h'E, foEs, h'Es, type of Es, fbEs, foF1, M(3000)F1, h'F, h'F2, foF2, fxI, and M(3000)F2. The fc_h'-t plot displays parameters corresponding to fmin, foE, h'E, foEs, h'Es, type of Es, fbEs, h'F, h'F2, foF2, and fxI. The fc_h'-t plot provides an easy method to monitor ionospheric conditions. The examples of the fc_h'-t plot showing some ionospheric phenomena will be useful for users of fc_h'-t plot, which are published in the monthly bulletin from the CRL as "Ionospheric Data in Japan". However, users should check the original ionograms to confirm the phenomena displayed in the plot, since the fc_h'-t plot has slightly less information than the original ionogram.
 Reinisch, B. W. and X. Huang, "Automatic calculation of electron density profiles from digital ionograms, 3.Processing of bottomside ionograms", Radio Sci., 18(3), 477-492, 1983.
 Fox, M. W. and C. A. Blundell, "Automatic scaling of digital ionogram", Radio Sci., 24(6), 747-761, 1989.
 Nozaki, K., M. Nagayama, and H. Kato, "Automatic ionogram processing system 1. Data reduction and transmission of ionograms", J. Commun. Res. Lab., 39(2), 357-365, 1992.
 Igi, S., H. Minakoshi, and M. Yoshida, "Automatic ionogram processing system 2. Automatic scaling of ionospheric parameters", J.Commun.Res.Lab., 39(2), 367-379, 1992.
 Igi, S., "Automatic ionogram processing system 3. A new method of displaying ionospheric characteristics", J. Commun. Res. Lab., 39(2), 381-402, 1992.