Volumetric Imaging of Advanced Modular Incoherent Scatter Radars (AMISR) Data Using Neural Network
Incoherent Scatter Radar (ISR) images with higher spatial and temporal resolution would be extremely beneficial to the space science community in addressing a range of important science questions. This project utilizes an artificial neural network model for a three-dimensional volumetric interpolation model of electron density profiles in the atmosphere. Data files from Advanced Modular Incoherent Scatter Radars (AMISR), specifically Poker Flat ISR and Resolute Bay ISR locations were utilized for experimentation of the code's performance. This project involved evaluating the effectiveness of the current algorithm, improving the reliability of the algorithm, and using the algorithm to assess physical parameters against other techniques.
The required inputs for the algorithm include the processed data file from the AMISR website and the time interval of interest of the experiment. The algorithm initially processes the data file and extracts relevant data required for the neural network. In addition to the neural network, a callback function is implemented to stop the neural network training once a certain loss value is achieved. The volumetric neural network prepares the data by normalizing it, filtering outliers, and defining weights for training. The neural network is trained multiple times to interpolate the electron density within the range of beams used in the experiment. The trained network complies the data points into two and three-dimensional grids which are layered to visually represent the peripheral field of the atmosphere.
The AMISR database consists of individual experiments that provide different integration periods, the number of beams for each mode, and other parameters that are varied for each experiment. This variety of data files provided many aspects of improvements and highlights for this algorithm. We would like to ultimately develop an algorithm that accurately interpolates three-dimensional visualizations of the upper layers of the atmosphere.