Annals 47, 6, 2004def 1783 ANNALS OF GEOPHYSICS, VOL. 47, N. 6, December 2004 Mailing address: Dr. Michael Pezzopane, Istituto Na- zionale di Geofisica e Vulcanologia, Via di Vigna Murata 605, 00143 Roma, Italy; e-mail: pezzopane@ingv.it Key words ionograms – automatic scaling – iono- sonde 1. Introduction Ionospheric observations are performed by a high frequency radar know as an ionosonde. The ionosonde sends short pulses of radio waves verti- cally into the ionosphere. These pulses are refract- ed back towards the ground and the ionosonde records the time delay between transmission and reception of the pulses. By varying the frequency of the pulses from 1 to 20 MHz, a record is ob- tained of the time delay at different frequencies. This record is referred to as an ionogram and is usually presented in the form of a graph. From the ionogram an experienced operator is able to obtain the most important ionospher- ic parameters, in particular the critical frequen- cy of the F2 layer (f0F2) and the Maximum Us- able Frequency on a distance of 3000 km (MUF(3000)F2). Because of the growing interest in real time mapping and short term previsions, in recent years efforts have been made to develop soft- ware to achieve real time scaling of ionograms (e.g., Fox and Blundell, 1989; Igi et al., 1993; Tsai and Berkey, 2000). The ARTIST system de- veloped at the University of Lowell, Center for Atmospheric Research, is an automatic scaling program widely used and tested (Reinisch and Huang, 1983; Gilbert and Smith, 1988). The National Institute of Geophysics and Vulcanology (INGV) developed a low power pulse compressed ionosonde to be used for stan- dard ionospheric soundings. This instrument – called an Advanced Ionospheric Sounder (AIS- INGV) designed for minimally attended, re- motely controlled operation – was installed in the INGV Ionospheric Observatory of Gibil- manna (37.9N-14.0E). A program (called Au- toscala) for the automatic scaling of f0F2 and MUF(3000)F2 was developed to provide scaled Software for the automatic scaling of critical frequency f0F2 and MUF(3000)F2 from ionograms applied at the Ionospheric Observatory of Gibilmanna Michael Pezzopane and Carlo Scotto Istituto Nazionale di Geofisica e Vulcanologia, Roma, Italy Abstract Software for the automatic scaling of critical frequency f0F2 and MUF(3000)F2 from ionograms was developed and applied to the ionograms produced by the new Advanced Ionospheric Sounder installed at the Ionospheric Observatory of Gibilmanna. A test based on 1124 ionograms recorded in normal ionospheric condition was per- formed comparing the automatically scaled parameters with the corresponding ones obtained by the standard manual method. The results of a preliminary test performed during a strong magnetic ionospheric storm are al- so presented. 1784 Michael Pezzopane and Carlo Scotto data within a few minutes the ionogram being produced. To date, these data are available real time on the internet at the site . 2. The INGV software for automatic scaling of ionograms The INGV software is based on an image recognition technique and can operate without polarization information. Hence it can be used both with single antenna systems and crossed an- tenna systems. Using a family of functions with the typical F2 layer shape, and applying a maxi- mum contrast technique, a particular element of this family is then selected and defined as repre- sentative of the F2 layer trace. The vertical asymptote of the selected function corresponds to the critical frequency f0F2; the MUF(3000)F2 is calculated numerically establishing the trans- mission curve tangent to the selected function. Compared to the previous version of the IN- GV software (Scotto and Pezzopane, 2002) for automatic scaling of ionograms some improve- ments have been made. 1) The main routine has been redesigned, with improvements in CPU time required and parametrizing the girofrequency, which means that this version will be able to scale ionograms recorded in any location. 2) In a significant percentage of the record- ed ionograms it is not possible to scale a value for the considered ionospheric characteristic (MUF(3000)F2 and f0F2). This can be due to different reasons, that are well known and clas- sified by the standard URSI. The most common cases are reported below. a) No F2 trace is observed owing to blan- keting by the E sporadic layer. In these cases the standard URSI recommends the use of the descriptive letter A on bulletins substituting the value of ionospheric characteristic f0F2. b) The trace near the critical frequency is not clearly recorded for different reasons. In these cases the standard URSI recommends to scale the most reliable value of f0F2 by extrap- olation, hypothetically extending the traces up to the most probable value of the critical fre- quency. If the extrapolated frequency range is greater than 10%, it is recommended to report the highest frequency of the recorded trace, fol- lowed by the qualifying letter D (greater than) and the descriptive letter associated to the rea- son why the trace is not clearly recorded (S in- terference, R absorption, C equipment). The software for automatic scaling of iono- grams recognizes the trace and outputs both the values of the characteristics. This version has been improved and a method for identifying ionograms without sufficient information has been introduced and tested. If an ionogram is considered to lack sufficient information it is discarded by the program and neither the f0F2 nor the MUF(3000)F2 are given as output. 3. Test: INGV software compared with the manual method during normal ionospheric conditions The test was performed using 1124 iono- grams recorded from December 1 to December 15, 2002 by the AIS-INGV installed in the Ionospheric Observatory of Gibilmanna. The period selected is composed by magnetically quiet days in order to test the performance of the software in normal ionospheric conditions. 3.1. Test of the method to identify and separate ionograms with sufficient or insufficient information The set of 1124 ionograms was divided into two subsets. The subset S, containing the iono- grams scaled by the program, and subset N con- taining the ionograms discarded by the software because of insufficient information. For each subset we considered: a) The number of ionograms for which the operator was able to scale neither the f0F2 nor the MUF(3000)F2. b) The number of ionograms for which the operator was able to scale f0F2 only. c) The number of ionograms for which the operator was able to scale MUF(3000)F2 only. d) The number of ionograms for which the operator was able to scale both f0F2 and MUF(3000)F2. 1785 Software for the automatic scaling of f0F2 and MUF(3000)F2 from ionograms recorded at Gibilmanna The results of this analysis are reported in table I and they confirm the general robustness of the overall system. The ionograms not dis- carded by the software and for which the oper- ator was not able to scale any value constitute the most critical cases. These cases are very few (9 out of 1124, equal to 0.8%) and are re- ported in table II where a comparison between the output of Autoscala and the corresponding manual scaling performed by a very experi- enced operator is shown. This comparison highlights that 8 out of 9 critical cases occurred for ionograms characterized by a truncated trace. On the other hand, it is also important to un- derline that the software does not discard iono- grams that present a trace sufficiently defined for an operator. This can be inferred observing that the cases discarded by the software in which the operator was able to scale both f0F2 and MUF(3000)F2 are very few (7 out of 1124, equal to 0.6%). Table I. The behavior of the method to identify and separate ionograms with sufficient or insufficient informa- tion. Scaled by the INGV software Discarded by the INGV software No. of cases [%] No. of cases [%] The operator scaled neither the f0F2 9 0.9% 46 62.1% nor the MUF(3000)F2 The operator scaled both f0F2 847 80.7% 7 9.5% and MUF(3000)F2 (subset A) The operator scaled MUF(3000)F2 only 182 17.3% 21 28.4% (subset B) The operator scaled f0F2 only 12 1.1% 0 0.0% (subset C) Total 1050 100.0% 74 100.0% Table II. The few cases (9 out of 1124, equal to 0.8%) in which the program scales values but the operator is not able to read a value. A comparison between the output of Autoscala and the corresponding manual scaling performed by a very experienced operator according to the standard URSI is reported. Day and time MUF(3000)F2 MUF(3000)F2 f0F2 f0F2 (software) (operator) (software) (operator) [MHz] [MHz] [MHz] [MHz] 5 December 2002 22.00 8.3 A 3.1 A 6 December 2002 13.00 26.1 R 8.1 8.4 DR 7 December 2002 07.45 30.4 R 9.1 8.9 DR 7 December 2002 08.45 29.0 S 8.6 8.2 DS 7 December 2002 12.45 18.1 S 4.7 9.7 DS 8 December 2002 09.00 32.1 R 10.6 9.8 DR 8 December 2002 11.00 31.2 S 9.6 9.6 DS 11 December 2002 14.00 30.2 S 9.8 9.4 DS 12 December 2002 03.30 30.3 S 9.8 8.6 DS 1786 Michael Pezzopane and Carlo Scotto 3.2. Test of accuracy and acceptability of the automatically scaled parameters With reference to the ionograms scaled by the INGV software, the following three subsets were considered (see table I): 1) Subset A, containing the ionograms in which the operator was able to scale both the critical frequency f0F2 and the MUF(3000)F2. 2) Subset B, containing the ionograms in which the operator was able to scale the MUF(3000)F2 only. 3) Subset C, containing the ionograms in which the operator was able to scale the f0F2 only. Table IIIa-c. Acceptable and accurate values (with corresponding percentage) for the test carried out on the ionograms recorded in Gibilmanna from Decem- ber 1 to December 15, 2002. In (a) the 847 iono- grams for which the operator was able to scale both f0F2 and MUF(3000)F2 are considered (subset A). In (b) the 182 ionograms for which the operator was able to scale only the MUF(3000)F2 are considered (subset B). In (c) the 12 ionograms for which the op- erator was able to scale only the f0F2 are considered (subset C). f0F2 MUF(3000)F2 No. [%] No. [%] Total 847 100.0 847 100.0 Acceptable 833 98.3 847 100.0 Accurate 638 75.3 827 97.5 MUF(3000)F2 No. [%] Total 182 100.0 Acceptable 179 98.5 Accurate 170 93.5 f0F2 No. [%] Total 12 100.0 Acceptable 9 75.0 Accurate 6 50.0 Fig. 1a,b. a) Differences (δ) between the values of f0F2 scaled by Autoscala and by the standard manu- al method considering ionograms for which the op- erator was able to scale both f0F2 and MUF(3000)F2 (subset A). Out of 847 cases it results: for 638 cases (− 0.1 MHz ≤ δ ≤ 0.1 MHz); for 139 cases (0.1 MHz < δ ≤ 0.3 MHz); for 24 cases (− 0.3 MHz ≤ δ < < − 0.1 MHz); for 26 cases (0.3 MHz < δ ≤ 0.5 MHz); for 6 cases (− 0.5 MHz ≤ δ < −0.3 MHz); for 7 cases (δ > 0.5 MHz); for 7 cases (δ < − 0.5 MHz). b) Dif- ferences between the values of MUF(3000)F2 scaled by Autoscala and by the standard manual method considering ionograms for which the operator was able to scale both f0F2 and MUF(3000)F2 (subset A). Out of 847 cases it results: for 827 cases (− 0.5 MHz ≤ δ ≤ 0.5 MHz); for 11 cases (0.5 MHz < δ ≤ 1.5 MHz); for 3 cases (−1.5 MHz ≤ δ < − 0.5 MHz); for 4 cases (1.5 MHz< δ ≤ 2.5 MHz); for 2 cases (− 2.5 MHz ≤ δ < −1.5 MHz); for 0 cases (δ > 2.5 MHz); for 0 cases (δ < − 2.5 MHz). a b a b c 1787 Software for the automatic scaling of f0F2 and MUF(3000)F2 from ionograms recorded at Gibilmanna A quantitative comparison between the val- ues scaled automatically and manually was per- formed and the results of the comparison are re- ported in table IIIa-c. In this work an accurate value is considered to lie within ± 0.1 MHz of the value obtained by the operator for f0F2 and ± 0.5 MHz for MUF(3000)F2. An acceptable value is considered to lie within ± 0.5 MHz for f0F2 and ± 2.5 MHz for MUF(3000)F2. Limits of acceptability have been adopted in line with the URSI limits of ± 5∆ (∆ is the reading accu- racy). For the subset A and B (constituted by a sufficient number of elements), the results of the comparison are presented in form of a his- togram in figs. 1a,b and 2. This analysis shows that the INGV software for automatic scaling gives acceptable values in more than 98% of cases both for f0F2 and MUF(3000)F2. The ionograms belonging to subset B show the ionogram trace abruptly truncated, but the software considers the information sufficient to identify the trace, and a value of f0F2 is given. The test gives us an evaluation of the capability of the software to scale these cases. An example Fig. 2. Differences between the values of MUF(3000)F2 scaled by Autoscala and by the stan- dard manual method considering ionograms for which the operator was able to scale only the MUF(3000)F2 (subset B). Out of 182 cases it re- sults: for 170 cases (− 0.5 MHz ≤ δ ≤ 0.5 MHz); for 1 case (0.5 MHz < δ ≤ 1.5 MHz); for 6 cases (− 1.5 MHz ≤ δ < − 0.5 MHz); for 0 cases (1.5 MHz < δ ≤ 2.5 MHz); for 2 cases (−2.5 MHz ≤ δ < −1.5 MHz); for 0 cases (δ >2.5 MHz); for 3 cases (δ < −2.5 MHz). Fig. 3. An example of ionogram belonging to the subset B. The ionogram trace is abruptly truncated, so the op- erator is not able to scale the critical frequency f0F2. In this case the software identifies the trace correctly and so the MUF(3000)F2 is correctly scaled (33.0 MHz the value automatically scaled, and 33.0 MHz by the oper- ator). The value of f0F2 = 10.3 MHz given by the program is reasonable. 1788 Michael Pezzopane and Carlo Scotto Fig. 4a-d. With reference to the days from 1 to 7 De- cember 2002: a) comparison between MUF(3000)F2 manually scaled (black square) and automatically scaled data (red line); b) comparison between f0F2 manually scaled (black square) and automatically scaled data (red line); c) plot of ∆ = ( f0F2measured + − f0F2median ) / f0F2measured, used as an index of the ionospheric disturbance; d) plot of the geomagnetic activity index Kp. Fig. 5a-d. With reference to the days from 8 to 15 De- cember 2002: a) comparison between MUF(3000)F2 manually scaled (black square) and automatically scaled data (red line); b) comparison between f0F2 manually scaled (black square) and automatically scaled data (red line); c) plot of ∆ = ( f0F2measured + − f0F2median ) / f0F2measured, used as an index of the ionospheric disturbance; d) plot of the geomagnetic activity index Kp. of an ionogram from subset B is shown in fig. 3. The ionogram trace is abruptly truncated, so the operator is not able to scale the critical fre- quency f0F2, but the software identifies the trace correctly. The MUF(3000)F2 is correctly scaled (33.0 MHz is the value automatically scaled, and 33.0 MHz scaled by the operator) and the value of f0F2 = 10.3 MHz given by the program is reasonable. Figures 4a and 5a report the plot of the MUF(3000)F2 values obtained by Autoscala and the plot of corresponding values obtained by a well experienced operator. Figures 4b and 5b report the same plots for f0F2. Figures 4c and 5c report the plot of the parameter ∆ = ( f0F2measured − f0F2median) / f0F2measured (where f0F2measured is the value obtained by an experi- enced operator), that is considered an index of the ionospheric disturbance. Figures 4d and 5d report the plot of the geomagnetic activity in- dex Kp. From figs. 4a-d and 5a-d one can observe qualitatively that the plots of the automatical- ly scaled data are very close to the data ob- tained by an operator. Therefore we can con- clude that the performance of Autoscala is sta- ble for not disturbed magneto-ionospheric con- ditions (Kp < = 4). a b c d a b c d 1789 Software for the automatic scaling of f0F2 and MUF(3000)F2 from ionograms recorded at Gibilmanna 4. Performance of INGV software for disturb- ed ionospheric conditions: a case study On October 29 2003, the sudden com- mencement of a strong magnetic storm was ob- served. Approximately at 6 UT the arrival of a large high-speed solar wind shock front was detected by the solar wind monitoring satellite. The geomagnetic field disturbance was ob- served in some regions until 7 UT. The AIS-IN- GV of Gibilmanna recorded the start of the ini- tial phase of the associated ionospheric storm at 8.45 UT by an increase in the critical fre- quency f0F2. Figure 6a,b reports the plot of the f0F2 val- ues obtained from the ionograms recorded in Gibilmanna; the values obtained by Autoscala are compared with the corresponding ones scaled by a well experienced operator. Figure 6a,b also reports the plot of the f0F2 values ob- tained from the ionograms recorded in Rome; the values obtained by ARTIST system are com- pared with the corresponding ones scaled by a well experienced operator. Comparing the two plots it is observed that during the morning the commencement of the storm causes the deterioration of the ionograms trace for both digisondes. Autoscala is able to detect this condition and does not give data that would be wrong. The system AIS-INGV de- tects the ionospheric storm as soon as the iono- gram trace is sufficiently clear. On the contrary, ARTIST does not detect this condition so the re- al time data of the digisonde of Rome are wrongly given in output. As a result the real time data of Rome between about 9 and 11 UT are highly underestimated and consequently the plot dissemble the ionospheric effect of the storm. 5. Conclusions The results of the test performed can be sum- marized observing that considering the subset A (constitued by good quality ionograms, for which the operator is able to scale both f0F2 and MUF(3000)F2) the percentage of acceptable values automatically scaled is very high (> 98%). For medium quality ionograms (for which the operator is able to scale only MUF(3000)F2) the percentage of acceptable values automatically scaled is also good (> 98%). Hence we can con- clude that for normal ionospheric conditions Au- toscala is able to provide good quality data. For disturbed ionospheric conditions a pre- liminary test has been performed showing good results. A study performed during the strong ionospheric storm of october 2003 showed that Autoscala better demonstrated the evolution of the ionospheric storm than the ARTIST system. 6. Further developments The tests performed are not able to give in- formation on the performance of Autoscala for spread F conditions as this case was not ob- served in the period considered at the ionos- pheric station of Gibilmanna. A separate study on this item would be necessary. Autoscala has definitively been applied at the ionospheric station of Gibilmanna since April 2003 and to date more than 20 000 ionograms Fig. 6a,b. a) Plot of f0F2 values, from the iono- grams recorded in Gibilmanna on 29 October 2003, obtained by a well experienced operator (black square) compared to the corresponding plot obtained by Autoscala (red line); b) plot of f0F2 values, from the ionograms recorded in Rome on 29 October 2003, obtained by a well experienced operator (black square) compared to the corresponding plot obtained by ARTIST (red line). a b 1790 Michael Pezzopane and Carlo Scotto have been produced. Tests based on larger datasets comprehensive of different seasons and solar conditions are necessary. Extensive com- parative studies of Autoscala with other automat- ic scaling software (with particular reference to ARTIST system) would be also interesting. REFERENCES GILBERT, J.D. and R.W. SMITH (1988): A comparison be- tween automatic ionogram scaling system ARTIST and the standard manual method, Radio Sci., 23, 968-974. FOX, M.W. and C. BLUNDELL (1989): Automatic scaling of digital ionograms, Radio Sci., 24, 747-761. IGI, S., K. NOZAKI, M. NAGAYAMA, A. OHTANI, H. KATO and K. IGARASHI (1993): Automatic Ionogram Processing Systems in Japan, in Proceedings of Session G4 at the XXIVth General Assembly of the International Union of Radio Science (URSI), August 25-September 2, Kyoto, Japan. REINISCH, B.W. and X. HUANG (1983): Automatic calcula- tion of electron density profiles from digital ionograms 3. Processing of bottomside ionograms, Radio Sci., 18, 477-492. SCOTTO, C. and M. PEZZOPANE (2002): A software for auto- matic scaling of f0F2 and MUF(3000)F2 from iono- grams, in Proceedings of URSI 2002, 17-24 August, Maastricht, The Netherlands. TSAI, L.C. and F.T. BERKEY (2000): Ionogram analysis us- ing fuzzy segmentation and connectedness techniques, Radio Sci., 35, 1173-1186. (received January 28, 2004; accepted May 12, 2004)