Spread-F Detection and Forecasting Using CNN Autoencoder
Abstract
Spread F is caused by plasma density fluctuations in the ionosphere. The occurrence of spread F can sometimes further develop into small-scale ionosphere irregularities that cause scintillations in the GNSS signals and impact the propagation of trans-ionospheric radio waves used for Land-Satellite Communication. Forecasting the occurrence of ionospheric irregularities as well as spread F remains a big challenge to the space weather community. In the past, the main method of identifying Spread F is by manually reading through ionograms, which is extremely time-consuming. In this study, machine learning (ML) algorithms are adopted to efficiently automate the identification process on Jicamarca data in order to further establish advanced algorithms for forecasting. An unsupervised learning method - convolutional neural network (CNN) autoencoder, which learns to reconstruct the input by encoding the sample and then decoding it - has been established for automatic identification required for the study. The reconstruction error is computed as the difference between the input ionogram and the output ionogram, comparing pixels elementwise. A threshold is then manually chosen so that the algorithm can automatically classify the ionograms as Spread F when the reconstruction error is above the threshold. In this presentation, we will describe the space weather problem, ML algorithms used for the study, and share the preliminary results. We will also outline the possible methods that will be used for forecasting the Spread-F in the near future.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMNG0040035L
- Keywords:
-
- 1914 Data mining;
- INFORMATICS;
- 7833 Mathematical and numerical techniques;
- SPACE PLASMA PHYSICS;
- 7924 Forecasting;
- SPACE WEATHER;
- 7959 Models;
- SPACE WEATHER