Real-Time Prediction and Forecasting of Geoelectric Fields Using Machine Learning

TERRAVOLT Phase I will serve as initial proof of concept for a real-time geoelectric field prediction capability. Specifically, the proposed work will leverage historical magnetometer databases and the USGS/Earthscope EMTFs to develop a capability to calculate and predict geoelectric fields in real time. In order to accomplish this goal, we will leverage recent advances in data analytics and machine learning that have not been widely exploited in space weather prediction:The construction of a sampling-based comprehensive database of historical GMD time series suitable for scalable computation, using historical magnetometer data sets and densely sampled EMTFs.New local and global machine learning methods for identifying nearest-neighbor surrogate GMD time series.New techniques using Analogue Ensemble (AnEn) modeling for the time-evolved machine learning prediction of geomagnetic fields and associated quantitative forecast error boundsScalable search methods based on advances in data indexing and retrieval algorithms such as Locality Preserving Hashing and optimized tree-based search and retrieval algorithms for real-time geoelectric field estimation.Our proposed effort will leverage all of these advances to provide a new capability for providing real-time forecasts of geoelectric fields that can be applied at any location where an operational time series is available and a suitable EMTF is known.nbsp;nbsp;