Machine Learning Models as an Alternative to Standard Interpolation Techniques for Estimating OMNI Data Gaps

NASAs OMNI data provide information about the solar wind plasma parameters and interplanetary magnetic field in the near-Earth environment. OMNI data is widely used to drive numerical and machine learning models. However, especially during geomagnetic storms, there can be significant data gaps in the OMNI data, which could significantly affect the model performance and results. In this study, we have tested traditionally used interpolation techniques and different machine learning models to predict data gaps in plasma parameters. We have created artificial data gaps in the OMNI data to evaluate the performance of different methods. We found that among different interpolation methods linear interpolation resulted in the lowest root mean square error for data gaps between 30 to 60 minutes. In addition, we found that the Random Forest regression model outperformed all other methods in predicting plasma parameters, especially for data gaps over 2 hours. The results suggest that machine learning models can serve as a better alternative to standard interpolation methods in filling data gaps in plasma parameters.