Improving the performance of solar flare prediction using active longitudes information

Context. Solar flare prediction models normally depend on properties of active regions, such as sunspot area, McIntosh classifications, Mount Wilson classifications, and various measures of the magnetic field. Nevertheless, the positional information of active regions has not been used. Aims. We define a metric, DARAL (distance between active regions and predicted active longitudes), to depict the positional relationship between active regions and predicted active longitudes and add DARAL to our solar flare prediction model to improve its performance. Methods. Combining DARAL with other solar magnetic field parameters, we build a solar flare prediction model with the instance-based learning method, which is a simple and effective algorithm in machine learning. We extracted 70 078 active region instances from the Solar and Heliospheric Observatory (SOHO)/Michelson Doppler Imager (MDI) magnetograms containing 1055 National Oceanic and Atmospheric Administration (NOAA) active regions within 30 degrees of the solar disk center from 1996 to 2007 and used them to train and test the solar flare prediction model. Results. Using four performance measures (true positive rate, true negative rate, true skill statistic, and Heidke skill score), we compare performances of the solar flare prediction model with and without DARAL. True positive rate, true negative rate, true skill statistic, and Heidke skill score increase by 6.7% +/- 1.3%, 4.2% +/- 0.5%, 10.8% +/- 1.4% and 8.7% +/- 1.0%, respectively. Conclusions. The comparison indicates that the metric DARAL is beneficial to performances of the solar flare prediction model.