Missing value imputation for > 2 MeV electron fluxes in geostationary orbit based on GA-RF model

Area of Science:

• Space Physics and Aeronomy
• Machine Learning Applications in Geophysics

Background:

• Accurate monitoring of high-energy electron fluxes from Geostationary Operational Environmental Satellites (GOES-E/W) is crucial for space weather studies.
• Large-scale data gaps in satellite measurements pose significant challenges for data analysis and model development.

Purpose of the Study:

• To develop and validate a robust machine learning model for imputing large-scale missing data in 5-minute averaged > 2 MeV electron integral fluxes.
• To compare the performance of the proposed model against various established machine learning algorithms and traditional interpolation techniques.

Main Methods:

• Construction of a genetic algorithm-optimized random forest (GA-RF) model.
• Input variables included satellite-derived parameters (V, Vx, proton density, magnetic field components Bx, By, Bz, AU, AE, SYM/H index) and electron fluxes (> 0.6 MeV and > 2 MeV).
• Performance evaluation using metrics such as PE, LC, RMSE, and MAE, comparing GA-RF with BP, LSTM, RF, ELM, and XGBoost models.

Main Results:

• The GA-RF model demonstrated superior performance in data imputation, achieving the highest PE and LC values and the lowest RMSE and MAE compared to other machine learning models.
• The GA-RF model significantly outperformed cubic spline and linear interpolation methods in capturing electron flux variations.
• Imputed data from the GA-RF model closely aligned with actual satellite-detected values, indicating high accuracy.

Conclusions:

• The GA-RF model is highly effective for imputing large-scale missing data in high-energy electron flux measurements from GOES satellites.
• This advanced imputation technique provides a reliable method for enhancing space weather data continuity and improving the accuracy of space environment models.