Modeling low-latitude ionospheric vertical drifts using the random forest technique

Abstract:

The vertical component of the E x B plasma drifts at low magnetic latitudes plays an important role in the dynamics of the geospace environment with implications for space weather. The vertical drifts are one of the main drivers of ionospheric plasma transport at low and mid-latitude regions. Additionally, the behavior of the drifts can lead to conditions favorable for the growth of plasma instabilities and the development of ionospheric irregularities, which affect radio signals used for communication, navigation and remote sensing.

The geospace science community has recognized the usefulness of machine learning techniques for fundamental and applied studies. Here we present the results of our efforts to model quiet-time vertical plasma drifts in the low-latitude F-region ionosphere using the random forest machine learning technique.

To develop the model, we analyzed vertical plasma drift measurements made by the incoherent scatter radar of the Jicamarca Radio Observatory between 1996 and 2018. The measurements allowed us to model the drifts as a function of day-of-year, local time, solar flux, and height.

The results show that our machine learning model can describe the expected diurnal variation of the drifts, including the pre-reversal enhancement, a phenomenon that is strongly associated with the occurrence of plasma irregularities. Using an independent test data set, we compared our model predictions with those from the Scherliess and Fejer (1999) empirical model of the drifts, and we find that our model has a slightly smaller root mean squared error. Additionally, our model is also able to accurately predict the expected local time variation of the mean height gradients of the vertical drifts including gradient enhancements found near dawn and dusk.

The model is computationally inexpensive to train and optimize, and can be easily improved as new observations and data sources become available.