Projects at the Radio Communications Research Unit (RCRU) RUTHERFORD APPLETON LABORATORY

 

SHORT TERM FORECASTING OF THE IONOSPHERIC CHANNEL

Introduction
STIF algorithm
Accuracy of forecast of foF2, MUF(3000)F2 and TEC
Ionospheric storm forecasting
Conclusions
References

1. Introduction [TOP]

Access to real-time information on ionospheric conditions over Europe, as a requirement for high frequency communication, satellite-to-ground links and solar-terrestrial research, has needed:

1. A network of vertical ionosondes involving as wide a European participation as possible to be set up for real-time data access;

2. Short-term forecasting algorithms to be developed for foF2, M(3000)F2, TEC and FOT sufficiently robust to predict events up to 24 hours ahead, and even 72 hours ahead to deal with week-end problems;

3. A mapping algorithm for interpolation between the stations in the network to be implemented by using the kriging instantaneous mapping procedure;

4. Validation of the mapping and forecasting algorithms.

As a response to these needs, an operational Short-Term Ionospheric Forecasting (STIF) tool for the European region based on continuous monitoring of the ionosphere has been developed and is available on the World Wide Web for interactive use (http://www.rcru.rl.ac.uk/iono/STIF.htm) [1, 2]. It provides forecast for up to 24 hours ahead and archive measurement maps of the critical frequency foF2, the Maximum Usable Frequency for a 3000 km range MUF(3000)F2, total electron content (TEC) and FOT (Frequency of Optimum Traffic) for the area of interest at each UT hour. As well as offering a useful on-line resource for the NRPP, this Short-Term Ionospheric Forecasting (STIF) over Europe has been recommended by the COST 251 Management Committee as the official COST 251 product applicable to European project on "Improved quality of service in ionospheric telecommunication system planning and operation". In addition, an artificial neural network method is applied to the development of ionospheric storm forecasting technique for one hour ahead as well as EIFM (European Ionospheric Forecasting and Mapping) algorithm for 1 to 24 hours ahead [3, 4, 5].

2. STIF algorithm [TOP]

A network of 22 currently active ground-based vertical incidence ionosondes (Figure 1 and Table 1) provides the basic inputs for the region of interest (10⁰ W - 90⁰ E, 30⁰ - 70⁰ N). Measurements of foF2 and M(3000)F2 are sent from the ionospheric stations by e-mail, mainly in the form of URSIGRAM messages [6]. Data are updated every 24 hours creating valuable database available at CD.

Figure 1. Network of ionospheric stations (click to enlarge)

Table 1. List of contributing vertical incidence ionospheric stations

An auto-correlation procedure was developed for the short-term forecasting of ionospheric characteristics [7, 8]. It is applied to produce forecast values of foF2 and MUF(3000)F2 at integer hours UT up to 72 hours ahead at each vertical incidence station where sufficient measurements are available. This is necessary to ensure a forecast for up to 24 hours ahead over weekends. Values from the past 60 days are used to construct an auto-regressive filter. The values MUF(3000)F2 are derived from the measured values of foF2 and M(3000)F2.

Forecast and archive maps for all characteristics are drawn using a commercial package with a Kriging option, which is particularly suitable for contouring sparse data [9]. The grid resolution is 2.5 degrees in latitude and 5 degrees in longitude. Following [10], an anisotropy factor of 2.1, which gives greater weight to variations along the longitudinal axis, is employed. Contour maps of forecast values and of the most recently available measurements are produced and updated daily at a fixed time.

Examples of forecast and measurement maps of foF2 are given in Figures 2 and 3 for 28 March 2000, 1200 UT. The crosses in Figure 3 indicate measured values at ionospheric stations. Similar examples for MUF(3000)F2 are given in Figures 4 and 5. Qualitative agreement is rather good for these examples, as expected for a quiet ionosphere.

TEC maps are calculated with the last two characteristics given by STIF and simplified models for foF1 and foE. The TEC profiler is a modified Di Giovanni - Radicella (DGR) model that uses 5 semi-Epstein layers and the ionospheric characteristics foE, foF1, foF2 and MUF3000(F2) as inputs. The profiler has been adapted from the NeQuick model [11]. Examples of forecast and measurement maps of TEC are given in Figures 6 and 7 for 28 March 2000 at 1200 UT. No crosses are indicated in Figure 7 as TEC is not directly measured but calculated by NeQuick with measured input parameters.

The Optimum Working Frequency (OWF or FOT) is the lower decile of the corresponding operational MUF at the quoted UT. In these maps the FOT is taken as the larger of 85% of the operational MUF for F2 mode propagation and 95% of that for E and F1 mode propagation. It is derived from measurements of the ionospheric characteristics foF2 and M(30000)F2 at the stated hour. An example of a FOT ‘footprint’ map is given in Figure 8.

3. Accuracy of forecast of foF2, MUF(3000)F2 and TEC [TOP]

Using the grid values, several statistical comparisons of maps obtained from forecast and measured values of foF2 and MUF(3000)F2 have been carried out. The forecast values selected were those deduced for one day ahead of the measured values. The principal results concern: (1) Comparisons at 0000, 0600, 1200 and 1800 UT on Tuesdays and Thursdays in June 1998 gave the Root Mean Square Error (RMSE) of 0.65 MHz and 2.15 MHz respectively; (2) Global statistical comparisons for the 5 months of March to September 1998 involving a total of 3672 maps (Table 2).

Figure 8. Operational FOT ‘footprint’ from a location at Davos for 28 March 2000 at 1200 UT (click to enlarge)

These gave RMSE for foF2 and MUF(3000)F2 of 0.63 MHz and 2.10 MHz respectively. In both studies, the individual hourly RMS errors in MUF(3000)F2 are generally slightly more than three times those of foF2. This is to be expected as M(3000)F2 factors are typically about 3 and additional errors are introduced by the measurements of the M(3000)F2 factor.

The effect of using measurement values of the previous day as a prediction (‘persistence’) for the current day at the same UT was examined. For the June 1998 study the foF2 RMSE increased to 0.67 MHz (3%) and that for MUF(3000)F2 to 2.36 MHz (10%). For the 5 month study, the foF2 RMSE increased by 0.01 MHz (1.6%) and that for MUF(3000)F2 by 0.12 (6%). In further comparisons, removing the anisotropy factor lead to increases in the RMSE of less than 2%.

These results suggest that autocorrelation forecasting is a more reliable technique than persistence, even in the relatively quiet conditions that prevailed. Single station tests have already revealed that the ACM model is quite robust in disturbed conditions [8], but further testing is required to assess operational performance.

Table 2. foF2 and MUF(3000)F2 mean and RMSE in MHz

Further access to information on plasmaspheric/ionospheric conditions over Europe in real-time and retrospective modes has been completed by creating on-line 10 min time resolution Total Electron Content database from UK GPS network (York, Chilbolton, Chilton, Sparsholt, Camborne and Bath locations).

To test the NeQuick model, TEC data estimated from new GPS measurements at Chilton are compared with TEC derived from the model. Hourly values of TEC estimated from GPS measurements at Chilton are compared with values of TEC from NeQuick model with foF2 and M(3000)F2 inputs obtained from Chilton vertical-incidence ionosonde. The RMSE deviations of the GPS measurements are given in Table 3. Table 3 represents the months of April, May and June 1999 with 5 geomagnetically quiet days (Q1 - Q8), disturbed days (D3 - D5), and 12 non-qualified days. It can be easily seen that modelled TEC values agree well with measured values.

Table 3. RMSE in TECU (10¹⁶ el/m²) for the selected days at Chilton

4. Ionospheric storm forecasting [TOP]

Ionospheric storms represent an extreme form of ionospheric space weather with important effects on increasingly sophisticated ground-and space-based technological systems. These phenomena are driven by highly variable solar and magnetospheric energy inputs to the Earth’s upper atmosphere. Consequently the ionospheric electron density at a given altitude and location is subject to sudden and profound changes lasting from a few hours to several days [12]. These changes are both positive (10 April 1990 at Figure 9) and negative (11, 12 and 13 April) deviations from the norm (monthly median) following a storm commencement (at noon on 10 April). Realistic simulations of their features, particularly through periods of intense geomagnetic activity, are restricted here to neural networks forecasting 1 to 24 hours ahead by looking at major individual storms [13].

Figure 9. The 1-hour ahead forecasting of the critical frequency of F2 ionospheric layer for Slough (click to enlarge)

A new technique has been developed for storm-time forecasting and instantaneous mapping over Europe, based on analytical presentation of the mapped quantities [5]. In this technique diurnal and seasonal variations of the ionospheric foF2 and M (3000) F2 characteristics are represented by a modified version of the regional model ISIRM [14, 15] adjusted to the past measured data. An auto-regressive extrapolation of the data from the past month enables the 15-day-ahead forecast of the quiet ionospheric distribution to be performed. The short-term variations due to geomagnetic activity are defined as a plane surface superimposed on the quiet distribution in function of the geomagnetic three-hour Kp index. In this way the 24-hour forecast can be obtain during quiet as well as disturbed ionospheric conditions. The EIFM algorithm provides variety of options to perform the short-term forecast depending on availability of the measured ionospheric data and predicted Kp values.

5. Conclusions [TOP]

Broadcasters and official users require up-to-date information on the state of the ionosphere. Several approaches have been under investigation, ranging from physically based regional numerical modelling, artificial neural network techniques to operationally oriented methods like real-time channel sounding. An intermediate approach based on empirical interpolation between a network of vertical ionosondes has been developed as a practical tool for short-term ionospheric predictions over Europe. Consequently the forecasts of foF2, MUF (Maximum Usable Frequency)(3000) F2, TEC (Total Electron Content) and FOT (Frequency of Optimum Traffic) up to 24 hours ahead are available on-line. This facility provides a useful tool to HF radio users who need up-to-date information on ionospheric conditions over Europe to meet their operational requirements. Applications of short term ionospheric forecasting include frequency management, retrospective ionospheric and solar-terrestrial studies and input to OHD sensors. Comparisons between measured and forecast values of foF2 and MUF(3000)F2 and tests for TEC accuracy have been conducted. The accuracy of the forecast maps is satisfactory. These facilities have been frequently accessed with, for example, approximately 30 sites requesting predictions of FOT in a few days.

The follow-on COST 271 Action on "Effects of the upper atmosphere on terrestrial and Earth-space communications" has been extended to space weather forecasting and will certainly involve future improvements of the existing STIF tools based on the access and manipulation of real time ionosonde, TEC and satellite data.

References [TOP]

[1] Levi, M.F., Lj. R. Cander, M.I. Dick, P. Muhtarov and I. Kutiev, "Real-time ionospheric forecasting", IRI News, Vol. 6, No. 2, 1-5, 1999.

[2] Dick, M.I., M.F. Levy, Lj. R. Cander, I. Kutiev, and P. Muhtarov: "Short-term ionospheric forecasting over Europe", IEE Publication number 461, pp. 105-107, 1999.

[3] Cander, Lj.R., M.M. Milosavljevi}, S.S. Stankovi} and S. Tomasevi}: "Ionospheric forecasting technique by artificial neural network", Electronics Letters, Vol. 34, No. 16, pp. 1573-1574, 1998.

[4] Wintoft, P., and Lj. R. Cander, "Short-term prediction of foF2 using time-delay neural networks", Phys. Chem. Earth, vol. 24/4, pp. 343-347, 1999.

[5] Muhtarov, P., I. Kutiev, Lj.R. Cander, B. Zolesi, G. de Franceschi, M.F. Levy, M.I. Dick, European ionospheric forecast and mapping, Submetted for publication in Phys. Chem. Earth (C), 2000.

[6] ITU-R , ITU-R Recommendations, Volume 1997, P Series-Part 1, ITU, Geneva, 1997.

[7] Kutiev, I., P. Muhtarov, Lj. R. Cander and M.F. Levy, "Short-term prediction of ionospheric parameters based on autocorrelation analysis", Annali di Geofisica, Vol. 42, N. 1, 121-127, 1999.

[8] Muhtarov, P. and I. Kutiev, "Autocorrelation method for temporal interpolation and short-term prediction of ionospheric data", Radio Science, 34, 2, 459-464, 1999.

[9] Surfer for Windows Ver 6, Golden Software Ltd, 809 14^th St, Golden, CO 80401, USA

[10] Stanislawska I., G. Juchnikowski and Lj.R.Cander, "Kriging method of ionospheric parameter foF2 instantaneous mapping", Annali di Geofisica,Vol. XXXIX, N. 4, pp. 845-852, 1996.

[11] Hochegger, G., B. Nava, S.M. Radicella and R. Leitinger, "A family of ionospheric models for different uses", Phys. Chem. Earth (C). Vol. 25, No. 4, pp. 295-299, 2000.

[12] Wintoft, P., and Lj.R. Cander, "Ionospheric foF2 Storm Forecasting using Neural Networks", Phys. Chem. Earth (C), Vol. 25, No. 4. pp. 267-273, 2000.

[13] Wintoft, P., and Lj.R. Cander, "Twenty-four hour predictions of foF2 using time delay neural networks", Radio Science, Vol. 35, No. 2, pp. 395-408, 2000.

[14] Zolesi, B., Lj.R. Cander and G. De Franceschi: "On the potential applicability of SIRM (Simplified Ionospheric Regional Model) to different mid-latitude areas", Radio Science, Vol. 31, No. 3, pp.547-552, 1996.

[15] Hanbaba, R., Improved Quality of Service in Ionospheric Telecommunication Systems Planning and Operation, COST Action 251 Final Report, Space Research Centre, Warsaw, 1999.