Automatic Segmentation of Range-Time-Intensity maps of Equatorial Spread-F

The Equatorial Spread-F (ESF) is a nighttime ionospheric phenomenon that can disturb the radio signals of global navigation satellite systems (GNSS) or communication systems in the equatorial zone. This radio phenomenon is related to plasma density irregularities (bubbles) generated at F-region heights in the ionosphere. In Peru, studies of the ESF have been conducted for many years using the Jicamarca ionospheric radar operating in the JULIA mode. The radar measures the ESF backscattered power registered in Range-Time-Intensity (RTI) maps. These RTI maps show the temporal and spatial (height) occurrence of ESF, allowing us to observe different morphological patterns (bottom-type, bottom-side, radar plumes, and others). In this work, our goal is to automatically segment the ESF patterns in the RTI maps using machine learning and deep learning algorithms. Leveraging the data available in the scientific database Madrigal, different techniques such as Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Neural Networks, and U-Net Convolutional Neural Networks are being tested. A comparison study between the techniques reveals the potential of the U-Net algorithm to segment the ESF. The features used in the segmentation are the RTI maps, geospace physical parameters, and statistical texture information that provide spatial knowledge for the segmentation algorithms.