Day-to-Day Vertical Drift Estimation Using Deep Learning Based on JULIA-MP mode at the Jicamarca Radar Observations: First results

Ionospheric drifts allow the analysis of ion diffusion velocities at different altitudes. These velocities are fundamental for studying the dynamics of the F layer during both quiet and disturbed conditions, contributing to space weather analysis. Since 2023, the new JULIA-Medium Power (MP) mode has been implemented at the Jicamarca Radio Observatory (JRO), offering continuous measurements of ionospheric drifts at F-region. Moreover, several studies have addressed the estimation of vertical drifts using climatological approaches based on sinusoidal series, as well as more recent Machine Learning–based methods, which have demonstrated improved predictive performance compared to traditional climatological models. Nevertheless, most of these models have been implemented by considering many years of vertical drift data but each year consisting of only a few days (around 20 days per year) of measurements, then these models are great to capture the seasonality but not the day-to-day variability..

In this work, we present the development of a Deep Learning model that combines neural networks with Long Short-Term Memory layers and Attention Mechanisms for the prediction of vertical drift velocities, by using geomagnetic and solar indices and considering days with K values of up to 5.

The dataset used to train the model was collected in JULIA-MP mode. It spans approximately 30 years (1994–2025), with a mean of 48 observation days per year. Notably, the data acquisition rate increased after 2023. The results are compared with other existing models, showing that the our proposed approach achieves superior performance in capturing the day-to-day variability of ionospheric drifts and allowing to better estimate vertical drifts when data is not collected due to maintenance activities or when the radar is set up for another experiment.