The Prediction of Day‐to‐Day Occurrence of Low Latitude Ionospheric Strong Scintillation Using Gradient Boosting Algorithm. Issue 12 (21st December 2021)

• Record Type: Journal Article
• Authors: Zhao, Xiukuan; Li, Guozhu; Xie, Haiyong; Hu, Lianhuan; Sun, Wenjie; Yang, Sipeng; Li, Yi; Ning, Baiqi; Takahashi, Hisao
• Is Part Of: Space weather. Volume 19:Issue 12(2021)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12-21
• Subjects: ionospheric scintillation prediction -- Gradient boosting algorithm -- Equatorial F layer height; Space environment -- Periodicals; 551.509992
• DOI: 10.1029/2021SW002884
• Languages: English
• ISSNs: 1542-7390

Abstract

Abstract: Ionospheric scintillations caused by equatorial plasma bubbles (EPBs) can seriously affect various high technology systems based on Global Navigation Satellite System (GNSS) signals at equatorial and low latitudes. A reliable prediction of ionospheric scintillation occurrence is critical to relieve the effect. Using the long‐term ground‐based GNSS receiver and ionosonde data collected in the Brazilian longitude sector during 2012–2020, an ionospheric strong scintillation prediction model based on the gradient boosting algorithms extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and CatBoost is created and tested. It is for the first time that the XGBoost, LightGBM, and CatBoost are utilized to predict the day‐to‐day occurrence of regional ionospheric scintillation during post‐sunset hours. The relative importance of different parameters affecting EPB/scintillation occurrence for building the prediction model is examined. A comparison of daily scintillation occurrence from the modeled and observed results during 2014 (solar maximum) and 2020 (solar minimum) shows that the gradient boosting algorithms are effective for predicting strong scintillations over low latitude, with a prediction accuracy of ∼85%. The results suggest that the trained model with input of total electron content, equatorial F layer peak height and critical frequency before sunset could be well employed to predict the occurrence/nonoccurrence of intense scintillations over low latitude after sunset on a daily basis. Plain Language Summary: Ionospheric scintillation is the rapid fluctuation of radio signals traversing through ionospheric irregularities. Severe scintillation can cause loss of lock for the systems using Global Navigation Satellite System signals. The dependences of scintillation on seasonal, solar and geomagnetic activities have been widely studied, but its day‐to‐day variability and prediction still remain a challenge. The relationship between scintillation occurrences and a variety of factors is complex. The machine learning algorithm could handle nonlinear problems and thus uncover the implicit correlations between multiple factors. The gradient boosting, which is a type of machine learning technique, has been demonstrated to be effective in many fields, such as in the field of intrusion detection. Here, we employ the gradient boosting algorithm, together with long‐term observations in the Brazilian longitude sector to investigate if the day‐to‐day occurrence of low latitude ionospheric scintillation could be predicted. The results show that with limited input parameters, the prediction accuracy for scintillation occurrence on a daily basis reach ∼85%, suggesting that the gradient boosting algorithms are effective for predicting strong scintillations over low latitude. This opens a possibility for scintillation forecasting with acceptable accuracy under the conditions without physical model and powerful computing capability. Key Points: A model for predicting low latitude scintillation occurrence after sunset is created based on gradient boosting algorithm The model with input of total electron content, equatorial hmF2 and foF2 before sunset, can well capture strong scintillation occurrence after sunset The gradient boosting algorithm is suggested to be effective in predicting low latitude strong scintillation occurrence on a daily basis