Reduced Order Probabilistic Emulation for Physics‐Based Thermosphere Models

Self Cite

Space weather indices are used commonly to drive forecasts of thermosphere density, which affects objects in low‐Earth orbit (LEO) through atmospheric drag. One commonly used space weather proxy, F10.7cm, correlates well with solar extreme ultra‐violet (EUV) energy deposition into the thermosphere. Currently, the USAF contracts Space Environment Technologies (SET), which uses a linear algorithm to forecast F10.7cm. In this work, we introduce methods using neural network ensembles with multi‐layer perceptrons (MLPs) and long‐short term memory (LSTMs) to improve on the SET predictions. We make predictions only from historical F10.7cm values. We investigate data manipulation methods (backwards averaging and lookback) as well as multi step and dynamic forecasting. This work shows an improvement over the popular persistence and the operational SET model when using ensemble methods. The best models found in this work are ensemble approaches using multi step or a combination of multi step and dynamic predictions. Nearly all approaches offer an improvement, with the best models improving between 48% and 59% on relative MSE with respect to persistence. Other relative error metrics were shown to improve greatly when ensembles methods were used. We were also able to leverage the ensemble approach to provide a distribution of predicted values; allowing an investigation into forecast uncertainty. Our work found models that produced less biased predictions at elevated and high solar activity levels. Uncertainty was also investigated through the use of a calibration error score metric (CES), our best ensemble reached similar CES as other work.

Section: Datamentioning

confidence: 99%

Probabilistic Solar Proxy Forecasting With Neural Network Ensembles

Daniell,

Mehta

2023

Self Cite

“…Firstly, we adopted the dimensionality reduction for the ionospheric outputs as obtained from REPPU simulations, by applying the principal component analysis (PCA) (e.g., Halko et al., 2011) using the Python 3 scikit‐learn/pca (Pedregosa et al., 2011). Very similar method was used by Licata and Mehta (2023) for different purpose (thermosphere model emulator). The time series of each parameter z = {Σ xy , Φ, or J//}, at certain (latitude, longitude) position of the grid indices ( i , j ), can be represented by the time averaged spatial pattern z 0 and the linear combination of time‐dependent PCA variables α and PCA component patterns U as follows:

z(i,j,t)={z}_{0}(i,j)+{z}_{1}(i,j,t),

{z}_{1}(i,j,t)=\sum\limits _{r=1}^{{N}_{r}}{\alpha }_{r}(t){U}_{r}(i,j).

In this study, we independently constructed the emulators for J//, Σ xy , and Φ maps.…”

Section: Methodsmentioning

confidence: 99%

Machine Learning‐Based Emulator for the Physics‐Based Simulation of Auroral Current System

Kataoka,

Nakamizo,

Nakano

et al. 2024

Using a machine learning technique called echo state network (ESN), we have developed an emulator to model the physics‐based global magnetohydrodynamic simulation results of REPPU (REProduce Plasma Universe) code. The inputs are the solar wind time series with date and time, and the outputs are the time series of the ionospheric auroral current system in the form of two‐dimensional (2D) patterns of field‐aligned current, potential, and conductivity. We mediated a principal component analysis for a dimensionality reduction of the 2D map time series. In this study, we report the latest upgraded Surrogate Model for REPPU Auroral Ionosphere version 2 (SMRAI2) with significantly improved resolutions in time and space (5 min in time, ∼1° in latitude, and 4.5° in longitude), where the dipole tilt angle is also newly added as one of the input parameters to reproduce the seasonal dependence. The fundamental dependencies of the steady‐state potential and field‐aligned current patterns on the interplanetary magnetic field directions are consistent with those obtained from empirical models. Further, we show that the ESN‐based emulator can output the AE index so that we can evaluate the performance of the dynamically changing results, comparing with the observed AE index. Since the ESN‐based emulator runs a million times faster than the REPPU simulation, it is promising that we can utilize the emulator for the real‐time space weather forecast of the auroral current system as well as to obtain large‐number ensembles to achieve future data assimilation‐based forecast.

“…We chose the same parameters used during the discrete sampling in Section 2.1 (SYM-H index, AL index, IMF B z , and SW V x ) as input drivers with the addition of the SW density. The LSTM input structures are built following the process outlined in Section 2.2 of Licata and Mehta (2023).…”

Section: Dynamic Modelingmentioning

confidence: 99%

“…10.1029/2023SW003706 In certain scenarios, true state outputs may not be accessible. In such cases, when conducting forecasts, predicted state outputs are utilized for forecasting future timesteps, following the approach outlined in Figure 3 of Licata and Mehta (2023). After predicting the current timestep t, the lookbacks are advanced for the subsequent timestep t + 1, with the corresponding lookback for t being updated with the predicted output.…”

Section: Space Weathermentioning

confidence: 99%

See 1 more Smart Citation

Reduced‐Order Probabilistic Emulation of Physics‐Based Ring Current Models: Application to RAM‐SCB Particle Flux

Cruz,

Siddalingappa,

Mehta

et al. 2024

Self Cite

In this work, we address the computational challenge of large‐scale physics‐based simulation models for the ring current. Reduced computational cost allows for significantly faster than real‐time forecasting, enhancing our ability to predict and respond to dynamic changes in the ring current, valuable for space weather monitoring and mitigation efforts. Additionally, it can also be used for a comprehensive investigation of the system. Thus, we aim to create an emulator for the Ring current‐Atmosphere interactions Model with Self‐Consistent magnetic field (RAM‐SCB) particle flux that not only improves efficiency but also facilitates forecasting with reliable estimates of prediction uncertainties. The probabilistic emulator is built upon the methodology developed by Licata and Mehta (2023), https://doi.org/10.1029/2022sw003345. A novel discrete sampling is used to identify 30 simulation periods over 20 years of solar and geomagnetic activity. Focusing on a subset of particle flux, we use Principal Component Analysis for dimensionality reduction and Long Short‐Term Memory (LSTM) neural networks to perform dynamic modeling. Hyperparameter space was explored extensively resulting in about 5% median symmetric accuracy across all data sets for one‐step dynamic prediction. Using a hierarchical ensemble of LSTMs, we have developed a reduced‐order probabilistic emulator (ROPE) tailored for time‐series forecasting of particle flux in the ring current. This ROPE offers accurate predictions of omnidirectional flux at a single energy with no pitch angle information, providing robust predictions on the test set with an error score below 11% and calibration scores under 8% with bias under 2% providing a significant speed up as compared to the full RAM‐SCB run.