Aurora borealis prediction models
Pro Research Analysisby 
Searched over 200M research papers
Empirical and Statistical Aurora Borealis Prediction Models
Empirical models like OVATION Prime and its updated versions are widely used for predicting auroral precipitation and the location of the auroral oval. These models use solar wind parameters and geomagnetic indices (such as Kp) to estimate auroral power and the probability of visible auroras. OVATION Prime 2013, for example, incorporates UV satellite data and is optimized for high geomagnetic disturbance levels, providing reliable predictions of the auroral oval’s equatorward boundary, especially during moderate activity. However, its accuracy decreases for the poleward boundary and during intense geomagnetic storms, particularly on the nightside and dayside sectors, where it tends to underpredict or overpredict auroral occurrence probabilities 36810.
The Feature Tracking of Aurora (FTA) model is another empirical approach that tracks auroral boundaries and energy fluxes across magnetic local time sectors. FTA has shown improved agreement with satellite observations during high activity, offering more confined and accurate spatial patterns compared to other models .
Machine Learning and Neural Network Approaches for Aurora Forecasting
Recent advances include machine learning models that use geomagnetic indices and satellite data to predict auroral oval locations and intensities. Models based on algorithms like XGBoost, Random Forest, and K-Nearest Neighbors have been tested, with XGBoost outperforming others, especially in predicting the equatorward boundary during geomagnetic disturbances. These models highlight the importance of including additional parameters beyond Kp for improved dayside predictions .
Neural network models, such as generalized regression neural networks (GRNN) and conditional generative adversarial networks (CGAN), have also been developed to predict the spatial distribution of auroral intensity. These models use solar wind, interplanetary magnetic field, and geomagnetic indices as inputs, and have demonstrated good similarity to observed auroral images, with performance improving as more training data is used .
Short-Term and Real-Time Aurora Forecast Systems
Interactive computer models, often web-based, provide short-term (30–70 minutes) forecasts of aurora intensity and probability. These systems typically use empirical models like OVATION Prime as their core, integrating real-time solar wind and geomagnetic data. Validation against ground-based camera networks has shown high agreement rates (up to 86%) between predicted and observed auroras, making these tools valuable for both public viewing and operational space weather forecasting .
Deep Learning and Image-Based Aurora Evolution Models
Deep learning models, such as those based on Convolutional Long Short-Term Memory (ConvLSTM) networks, have been applied to predict the evolution of auroral structures during substorm expansion phases. These models use sequences of auroral images to forecast changes in intensity and boundary movement, achieving reasonable accuracy in both spatial and intensity evolution .
Electron Transport and Spectral Intensity Models
Electron transport models, including the GLOW model and others, are used to predict auroral emission intensities based on electron precipitation characteristics. These models are important for interpreting ground-based and satellite auroral images and have shown strong correlation with observed data when updated with recent chemical and radiative rates .
Model Performance and Limitations
While empirical and machine learning models perform well in predicting the equatorward boundary and general auroral activity, challenges remain in accurately forecasting the poleward boundary, especially during high geomagnetic activity and on the dayside. Probabilistic forecasts tend to underpredict auroral occurrence at moderate probabilities and overpredict at very high probabilities. Incorporating more diverse input parameters and expanding training datasets can help improve model accuracy 3468+2 MORE.
Conclusion
Aurora borealis prediction models have evolved from empirical and statistical approaches to advanced machine learning and deep learning techniques. While current models provide reliable forecasts for operational and public use, especially for the equatorward boundary and during moderate geomagnetic activity, further improvements are needed for high-activity periods and dayside predictions. Integrating more comprehensive data and leveraging modern computational methods will continue to enhance the accuracy and utility of auroral forecasts.
Sources and full results
Most relevant research papers on this topic