Publication Date:
2024-02-15
Description:
〈title xmlns:mml="http://www.w3.org/1998/Math/MathML"〉Abstract〈/title〉〈p xmlns:mml="http://www.w3.org/1998/Math/MathML" xml:lang="en"〉Machine learning (ML) has been increasingly applied to space weather and ionosphere problems in recent years, with the goal of improving modeling and forecasting capabilities through a data‐driven modeling approach of nonlinear relationships. However, little work has been done to quantify the uncertainty of the results, lacking an indication of how confident and reliable the results of an ML system are. In this paper, we implement and analyze several uncertainty quantification approaches for an ML‐based model to forecast Vertical Total Electron Content (VTEC) 1‐day ahead and corresponding uncertainties with 95% confidence intervals (CI): (a) Super‐Ensemble of ML‐based VTEC models (SE), (b) Gradient Tree Boosting with quantile loss function (Quantile Gradient Boosting, QGB), (c) Bayesian neural network (BNN), and (d) BNN including data uncertainty (BNN + D). Techniques that consider only model parameter uncertainties (a and c) predict narrow CI and over‐optimistic results, whereas accounting for both model parameter and data uncertainties with the BNN + D approach leads to a wider CI and the most realistic uncertainties quantification of VTEC forecast. However, the BNN + D approach suffers from a high computational burden, while the QGB approach is the most computationally efficient solution with slightly less realistic uncertainties. The QGB CI are determined to a large extent from space weather indices, as revealed by the feature analysis. They exhibit variations related to daytime/nightime, solar irradiance, geomagnetic activity, and post‐sunset low‐latitude ionosphere enhancement.〈/p〉
Description:
Plain Language Summary: Space weather describes the varying conditions in the space environment between the Sun and Earth that can affect satellites and technologies on Earth, such as navigation systems, power grids, radio, and satellite communications. The manifestation of space weather in the ionosphere can be characterized using the Vertical Total Electron Content (VTEC) derived from Global Navigation Satellite Systems observations. In this study, the machine learning (ML) approach is applied to approximate the nonlinear relationships of Sun‐Earth processes using data on solar activity, solar wind, magnetic field, and VTEC. However, the measurements and the modeling approaches are subject to errors, increasing the uncertainty of the results when forecasting future instances. For reliable forecasting, it is necessary to quantify the uncertainties. Quantifying the uncertainty is also helpful for understanding the ML‐based model and the problem of VTEC and space weather forecasting. Therefore, in this study, ML‐based models are developed to forecast VTEC within the ionosphere, including the manifestation of space weather, while the degree of reliability is quantified with a target value of 95% confidence.〈/p〉
Description:
Key Points:
〈list list-type="bullet"〉
〈list-item〉
〈p xml:lang="en"〉Machine learning‐based Vertical Total Electron Content models with 95% confidence intervals (CI) are developed for the first time using four approaches to quantify uncertainties〈/p〉〈/list-item〉
〈list-item〉
〈p xml:lang="en"〉Bayesian Neural Network quantifying model and data uncertainties contains ground truth within CIs, but is computationally intensive〈/p〉〈/list-item〉
〈list-item〉
〈p xml:lang="en"〉Quantile Gradient Boosting is fastest with comparable performance in terms of uncertainty; CIs largely determined from space weather indices〈/p〉〈/list-item〉
〈/list〉
〈/p〉
Description:
Deutscher Akademischer Austauschdienst
http://dx.doi.org/10.13039/501100001655
Description:
https://www.tensorflow.org/
Description:
https://doi.org/10.21105/joss.03021
Description:
http://www.aiub.unibe.ch/download/CODE
Description:
https://kauai.ccmc.gsfc.nasa.gov/instantrun/iri
Description:
https://doi.org/10.5281/zenodo.7741342
Description:
https://doi.org/10.5281/zenodo.7858906
Description:
https://doi.org/10.5281/zenodo.7858661
Keywords:
ddc:551.5
;
machine learning
;
uncertainty quantification
;
confidence intervals
;
probabilistic ionosphere forecast
;
space weather
;
ensemble
;
Bayesian neural network
;
quantile gradient boosting
Language:
English
Type:
doc-type:article
Permalink
