The Probabilistic Regressor for Input to the Magnetosphere Estimation (PRIME) is an algorithm capable of predicting the solar wind just upstream of the Earth’s bow shock from solar wind measurements made at the first Earth-Sun Lagrange point (L1) 1,500,000km upstream of the Earth. Since the only continuous solar wind monitors are located at L1, PRIME fills a critical space weather need. Unlike other L1-to-Earth propagation algorithms PRIME is capable of estimating reliable uncertainties for its predictions, which has been shown to be crucial for accurate terrestrial weather prediction. PRIME utilizes a probabilistic recurrent neural network architecture trained to reproduce plasma and magnetic field measurements made by the Magnetospheric Multiscale (MMS) spacecraft constellation given solar wind measurements from the Wind L1 monitor spacecraft to make its predictions. Please see the manuscript for quantitative demonstrations of PRIME’s accuracy and reliability, and follow PRIME on GitHub for the source code, model outputs, and future updates.
Following the success of PRIME, I adapted its architecture to another longstanding problem in space physics: magnetosheath prediction. As the solar wind encounters the Earth’s magnetosphere, it diverts around the obstacle and slows to subsonic speeds. This forms a shock (much like the shock ahead of supersonic aircraft) which heats and compresses the solar wind, forming a structure called the magnetosheath. The magnetosheath is the plasma and magnetic field that actually drives energy transfer processes between the solar wind and Earth’s magnetosphere, however there is no continuous in-situ magnetosheath monitor. Instead, conditions in the magnetosheath must either be measured by one of several spacecraft on orbits that sometimes pass through the magnetosheath or inferred using a magnetosheath model driven by upstream solar wind conditions. My new algorithm PRIME-SH combines both of these approaches by assembling a magnetosheath model directly from spacecraft observations in the magnetosheath and solar wind. Using PRIME’s probabilistic neural network architecture, PRIME-SH is capable of making accurate and reliable predictions of magnetosheath conditions with uncertainties attached. Since it is based on observations, PRIME-SH incorporates physics typically difficult for models to produce (such as temperature anisotropy) while still being computationally efficient enough to run on a laptop. Follow PRIME-SH on GitHub for the source code, model outputs, and future updates.
GEM Climate Survey