Tool for predicting the SEP occurrence from microwave data using the UMASEP scheme
Final results
During year 1, the SEP prediction performance results by using this tool were calculated for the period December 2011 - December 2013. The obtained SEP start times are the important data for calculating the POD, the FAR and the Average Warning Time. Table 1 presents the forecast performance results in terms of POD, FAR and average warning time using only the Well-Connected forecasting model with microwave (5 and 9 GHz) or soft X-ray data. We note that the use of soft X-ray and microwave data produces the same POD. The most notable di ? erence is that the use of microwave data does not yield any false alarm. Note that the POD and the average warning time using soft X-ray data and microwave data are similar.
Information about the Task 2.4, in which the tool was developed.
The Task 2.4, called "Tool for predicting SEP events from historical microwave data by using the UMASEP scheme" is lead by UMA, with the participation of the Observatory of Paris. The objective of the task 2.4 is to construct a tool that uses input microwave flux density time series within the UMASEP scheme, instead of SXR flux time series, to predict >10 MeV SEP events, as illustrated in Figure 1.
Figure 1. Use of microwave data in the UMASEP scheme
The starting point of this Task was to construct an uninterrupted series of microwave time profiles from the US Air Force RSTN network and the Nobeyama Radio Observatory in Japan over a duration of three months during 2012. These data, as well a proton flux data, was used to calibrate UMASEP's internal parameters to optimize the forecasting performance in terms of POD, FAR, and average warning time.
Data analysis
It was agreed among the WP participants that as a first test of feasibility a 3 months period was chosen containing both intervals with repeated SEP events and quiet periods. The duration was chosen as a compromise between on the one hand a sufficiently long time to have a meaningful test and on the other hand a limited amount of work to learn how to handle the microwave data.
OBSPARIS was responsible for constructing the contiguous series of time histories at two frequencies during a three-month interval, from 1stJanuary to 31st March 2012 using observations from different ground-based radio telescopes. Whole Sun patrol observations at microwave frequencies, i.e. frequencies between 1 GHz and a few tens of GHz, are routinely carried out by the Radio Solar Telescope Network (RSTN), operated by the US Air Force, and by the Nobeyama Radio Observatory in Japan. RSTN has in principle a 24-hour coverage, through four observatories located in Massachusetts, Hawaii, western Australia and Italy. The four observatories produce four independent data sets of about 8 hours duration each day. They overlap partly with each other, thus enabling consistency checks. Where necessary, consistency checks were also made with Nobeyama observations. Since they are not taken at exactly the same frequencies as RSTN observations, the Nobeyama data are not included in the long-duration data set. The RSTN data are publicly available as ASCII files on the NOAA web site, with the observing time and the calibrated solar flux densities at the observed frequencies given once per second. In order to construct a continuous series of microwave flux densities over an extended time period, OBSPARIS proceeded as follows:
· Automated procedures were developed to download ASCII files from the NOAA web server, remove isolated positive outliers, and calculate a typical quiet-Sun background level for each observatory and each day. The background level differs in general between different sets of observations, because the hardware and environment contribute to the background, and because active regions change intrinsically and become more or less visible in the course of the solar rotation.
· A daily observing interval was chosen for each observatory such as to avoid distortions of the flux density profile that occur at the beginning and the end of the daily observations. The observing interval was identified by visual inspection of the time profiles, and was fixed during the entire 3 months duration. The observing intervals of the four observatories were chosen such as to provide a contiguous 24-hour coverage.
· The daily background evaluated for each observatory was subtracted. An average background over the three-months period was evaluated and added to the data. The flux density time histories were integrated during 5 minute.
· The above series of procedures were applied to observations at two frequencies, namely 4.995 GHz and 8.8 GHz, which provided the most reliable observations. The results were checked by visual inspection and provided to UMA as ASCII files.
UMA took the microwave time profiles and proceeded as follows:
· A general analysis was done to the microwave data to find incompatibilities with the EM data requirements to be provided to the UMASEP tool. UMA detected some minor incompatibilities with the temporal resolution, that were quickly fixed by OBSPARIS.
· The microwave time series at each frequency were used to replace the time derivative of SXR. UMASEP's internal parameters were adjusted to predict the observed >10 MeV SEP events, by maximizing the number of successful predictions and reducing the false alarms (taking into account the SWPC thresholds).