5:15 PM - 6:45 PM
[PEM15-P06] Spatiotemporal Sequence Prediction of Global Ionospheric Total Electron Content Map Based on Machine Learning
Keywords:Total Electron Content Map, Spatiotemporal Sequence Prediction, Machine Learning
Global ionospheric Total Electron Content (TEC) map that indicates the total number of electrons, is an important physical quantity for the Earth ionosphere. Since 1995, 132,960 global TEC maps have been provided by the Centre for Orbit Determination in Europe (CODE) based on the signal delay between the globally distributed ground receivers and satellites.
Machine learning technologies that develop rapidly nowadays are leveraged to predict upcoming frames of temporal and spatiotemporal sequences. Recurrent Neural Network (RNN) can restore current output and state of the network as the input of prediction on the next timestamp, thus RNN and its improved version (for example, bi-directional multilayer RNN) are used for temporal sequence prediction including global TEC map prediction. However, these models only learn the temporal trend of sequential input data without considering the spatial association. To solve this problem, the first spatiotemporal sequence prediction model, ConvLSTM, was proposed in 2015 which used convolution operation to learn additional spatial distribution features. Moreover, recent researches show that Transformer and Convolution Neural Networks (CNN) have better performance. After several years of development, new advanced spatiotemporal sequence prediction models such as MIM, E3D-LSTM, SimVP and PredRNN were proposed but have not been applied on global TEC map prediction yet.
To put the auxiliary data into the consideration, previous multimodal fusion methods of neural network can be classified into arithmetic operation (adding/multiplying) or concatenating operation for convolutional layer output tensor of different channels. For example, the action conditioned PredRNN model fuses different channels by multiplying after convolution. This multimodal fusion method cannot improve the predictive ability because it is obvious that the accuracy is lower as the auxiliary channels are more. To make the network have reasoning ability that can increase the prediction accuracy by inputting more auxiliary data, this research proposes a new multimodal fusion framework shown as the following figure. In previous fusion-after-convolution approach, there are distribution gaps between different convolutional kernels that learns from different data channels, but in the fusion-before-convolution approach of this research, the data of the same position of the different channels are rearranged into the same channel by a new fusion layer. So that data distribution gaps between different channels can be suppressed. The experiment result shows that proposed method in this research can improve the performance of existing models.
Machine learning technologies that develop rapidly nowadays are leveraged to predict upcoming frames of temporal and spatiotemporal sequences. Recurrent Neural Network (RNN) can restore current output and state of the network as the input of prediction on the next timestamp, thus RNN and its improved version (for example, bi-directional multilayer RNN) are used for temporal sequence prediction including global TEC map prediction. However, these models only learn the temporal trend of sequential input data without considering the spatial association. To solve this problem, the first spatiotemporal sequence prediction model, ConvLSTM, was proposed in 2015 which used convolution operation to learn additional spatial distribution features. Moreover, recent researches show that Transformer and Convolution Neural Networks (CNN) have better performance. After several years of development, new advanced spatiotemporal sequence prediction models such as MIM, E3D-LSTM, SimVP and PredRNN were proposed but have not been applied on global TEC map prediction yet.
To put the auxiliary data into the consideration, previous multimodal fusion methods of neural network can be classified into arithmetic operation (adding/multiplying) or concatenating operation for convolutional layer output tensor of different channels. For example, the action conditioned PredRNN model fuses different channels by multiplying after convolution. This multimodal fusion method cannot improve the predictive ability because it is obvious that the accuracy is lower as the auxiliary channels are more. To make the network have reasoning ability that can increase the prediction accuracy by inputting more auxiliary data, this research proposes a new multimodal fusion framework shown as the following figure. In previous fusion-after-convolution approach, there are distribution gaps between different convolutional kernels that learns from different data channels, but in the fusion-before-convolution approach of this research, the data of the same position of the different channels are rearranged into the same channel by a new fusion layer. So that data distribution gaps between different channels can be suppressed. The experiment result shows that proposed method in this research can improve the performance of existing models.