10:45 AM - 12:15 PM
[PEM09-P12] Probability forecast of relativistic electron fluence levels on geostationary orbit by the multioutput classification with multi-layer perceptron
Keywords:Space weather, Deep learning, Radiation belt electrons
Relativistic electrons that constitute Earth’s radiation belts are considered as a cause of anomalies on artificial satellite. As space weather monitoring and forecast in NICT, we have monitored the variation of the radiation belt electron flux on geostationary orbit observed by GOES-16 and 18 satellites. The temporal variation of 24-hour electron fluence have been monitored (https://swc.nict.go.jp/trend/electron.html) and been forecast with 72 hours lead time (https://radi.nict.go.jp//).
We developed the probability forecast model for radiation belt electrons by using a deep learning (DL) method to improve the accuracy of the prediction and the ease of utilization. The method implements multi-layer perceptron which has the input layer, three hidden layers, and the output layer. The softmax function is applied on the output layer which generates probabilities of four fluence levels of radiation belt electrons with 24 hours lead time. The fluence levels are set to be F < 3.8×107 as the low level, 3.8×107 < F < 3.8×108 as the slightly high level, 3.8×108 < F < 3.8×109 as the high level, and F > 3.8×109 as the very high level. Here F is the 24-hour electron fluence with the unit of “/cm2/str”. The input layer receives the solar wind and electron fluence data past 72 hours with time resolution of 1 hour. Here the solar wind data are the velocity, density, magnetic field amplitude, east-west magnetic field, southward magnetic field, and the minimum values of the field in one hour. The two-dimensional data set (72 x 6) are applied to the forecast model, and then it generates the probabilities of four fluence levels on geostationary orbit 24 hours ahead.
In this presentation, we will show the performance of the DL model for real time solar wind data observed by DSCOVR and evaluate the accuracy of the forecast.
We developed the probability forecast model for radiation belt electrons by using a deep learning (DL) method to improve the accuracy of the prediction and the ease of utilization. The method implements multi-layer perceptron which has the input layer, three hidden layers, and the output layer. The softmax function is applied on the output layer which generates probabilities of four fluence levels of radiation belt electrons with 24 hours lead time. The fluence levels are set to be F < 3.8×107 as the low level, 3.8×107 < F < 3.8×108 as the slightly high level, 3.8×108 < F < 3.8×109 as the high level, and F > 3.8×109 as the very high level. Here F is the 24-hour electron fluence with the unit of “/cm2/str”. The input layer receives the solar wind and electron fluence data past 72 hours with time resolution of 1 hour. Here the solar wind data are the velocity, density, magnetic field amplitude, east-west magnetic field, southward magnetic field, and the minimum values of the field in one hour. The two-dimensional data set (72 x 6) are applied to the forecast model, and then it generates the probabilities of four fluence levels on geostationary orbit 24 hours ahead.
In this presentation, we will show the performance of the DL model for real time solar wind data observed by DSCOVR and evaluate the accuracy of the forecast.