Towards GNSS Scintillation source region prediction using a Machine learning and Neural network-based framework

Space weather may have major economical and social impact when it affects the omnipresent and heavily dependent position-navigation-timing (PNT) services from the Global Navigation Satellite Systems (GNSS). At high latitudes, although the ionosphere brightens up with beautiful auroral display, during a geomagnetic storm, precision mining, farming technology, marine navigation and the polar flights may severely get affected. One of the goals in our space weather community is to be able to predict when ionospheric structures or irregularities are prone to cause scintillations or fluctuations in the radio signals affecting the PNT solutions. Working towards this goal, our interest for this study is to identify the relationship between GNSS scintillations in radio signals and the location of scintillation creating irregularities utilizing the huge amounts of scintillation data collected over several years. We are focusing on the two regions: polar cap and auroral oval. In order to classify the scintillations based on their source regions we have developed a machine learning (ML) and neural network (NN) -based framework. For the database generation, we first use an anomaly detection algorithm called isolation forest (iForest) to identify scintillation events in the 50Hz time series phase and amplitude data. For training a neural network (NN) for the classification of the data between the two source regions, we use phase high rate data for the locations identified using polar cap or auroral oval boundaries from DMSP-SSUSI data. The NN classifies the scintillation events as either originating due to the irregularities in the polar cap or auroral oval based on the observed patterns in the signatures. We are using the GNSS data from a chain of stations in Canada deployed by university of New Brunswick called CHAIN. We look at phase and amplitude scintillations 2016-2020 years from 3 CHAIN stations and different geomagnetic conditions and comment on how different geomagnetic activity level affects the solution of our NN model.