The Prediction of Storm‐Time Thermospheric Mass Density by LSTM‐Based Ensemble Learning

Wang, Peian; Zhou, Chen; Deng, Xiaohua; Wang, Jingsong; Tang, Rongxing; Li, Haimeng; Hong, Sheng; Wu, Zhiping

doi:10.1029/2021sw002950

Cited by 14 publications

(16 citation statements)

References 40 publications

(49 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The μ models could perform dynamic prediction with low error, but the σ models still struggled to provide meaningful uncertainty estimates. Another approach to UQ, popular in terrestrial and space weather applications, is ensemble modeling (Wang et al., 2022; Xu et al., 2020). We therefore develop many individual LSTMs and combine the results in such a way to obtain accurate predictions and reliable uncertainty estimates.…”

Section: Methodsmentioning

confidence: 99%

Reduced Order Probabilistic Emulation for Physics‐Based Thermosphere Models

Licata

Mehta

2023

Space Weather

View full text Add to dashboard Cite

The geospace environment is volatile and highly driven. Space weather has effects on Earth's magnetosphere that cause a dynamic and enigmatic response in the thermosphere, particularly on the evolution of neutral mass density. Many models exist that use space weather drivers to produce a density response, but these models are typically computationally expensive or inaccurate for certain space weather conditions. In response, this work aims to employ a probabilistic machine learning (ML) method to create an efficient surrogate for the Thermosphere Ionosphere Electrodynamics General Circulation Model (TIE‐GCM), a physics‐based thermosphere model. Our method leverages principal component analysis to reduce the dimensionality of TIE‐GCM and recurrent neural networks to model the dynamic behavior of the thermosphere much quicker than the numerical model. The newly developed reduced order probabilistic emulator (ROPE) uses Long‐Short Term Memory neural networks to perform time‐series forecasting in the reduced state and provide distributions for future density. We show that across the available data, TIE‐GCM ROPE has similar error to previous linear approaches while improving storm‐time modeling. We also conduct a satellite propagation study for the significant November 2003 storm which shows that TIE‐GCM ROPE can capture the position resulting from TIE‐GCM density with <5 km bias. Simultaneously, linear approaches provide point estimates that can result in biases of 7–18 km.

show abstract

Section: Methodsmentioning

confidence: 99%

Reduced Order Probabilistic Emulation for Physics‐Based Thermosphere Models

Licata

Mehta

2023

Space Weather

View full text Add to dashboard Cite

show abstract

“…The MLP and ResNet models share a similar architecture, but the ResNet model replaces the dense layers in the MLP with residual blocks. The MLP model has 7 hidden layers, while the ResNet model has 35 hidden layers, which is deeper than most previous network structures (e.g., Chen et al., 2014; Licata, Mehta, Weimer, et al., 2022; Wang et al., 2022; Weng et al., 2020). In the ResNet blocks, we added a dropout layer to reduce the possibility of overfitting.…”

Section: Deep Learning Modelmentioning

confidence: 99%

Improving the Extraction Ability of Thermospheric Mass Density Variations From Observational Data by Deep Learning

Liu

Chen

et al. 2023

Space Weather

View full text Add to dashboard Cite

Understanding the variation of the Thermospheric Mass Density (TMD) is important for solar‐terrestrial physics and applications for spacecraft safety. The thermosphere, as an open system, is impacted by various space environment conditions and has complicated temporal and spatial features. Consequently, TMD observations contain a wealth of multi‐scale feature information. How to extract such information from observations is a challenge that requires ongoing research. It is vital to improving our understanding of the TMD features. Deep learning (DL) can learn complex representations directly from raw data, which makes it a compelling feature extraction and modeling tool for providing a novel perspective for TMD modeling. The Residual Network is used in this study to build a DL model with deep network architecture. The observations of CHAllenging Minisatellite Payload are utilized in the training phase, while the Gravity Recovery and Climate Experiment, High Accuracy Satellite Drag Model and Naval Research Laboratory Mass Spectrometer and Incoherent Scatter radar Extended model are used to evaluate the performance of the DL model. The results reveal that, compared with the shallow model of the typical Multi‐Layer Perceptron, the DL model can better extract multi‐scale features in the observations while retaining generalization capabilities. Controlled simulation experiments allow us to extract the effects of different physical processes, which improves the interpretability of the DL model. It is demonstrated that the DL model can discriminate the physical processes corresponding to the different space environment indices by simulating Equatorial Mass density Anomaly and geomagnetic storms.

show abstract

“…Wang et al. (2022) adopted the LSTM‐based ensemble learning algorithm to achieve the robust prediction for thermospheric mass density under different geomagnetic activity conditions. Chen et al.…”

Section: Introductionmentioning

confidence: 99%