Designing an Optimal GNSS Receiver Network for Space Weather Studies Using Unsupervised Machine Learning Algorithm

Ionospheric maps from an optimal GNSS (global navigation satellite system) receiver network may lead to better understanding and prediction of the equatorial ionospheric gradients. An unsupervised machine learning framework is proposed using the k $k$‐means clustering approach to optimize the availa...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Radio science Ročník 57; číslo 12
Hlavní autoři: Dashora, N., Harsha, P. Babu Sree
Médium: Journal Article
Jazyk:angličtina
Vydáno: Washington Blackwell Publishing Ltd 01.12.2022
Témata:
Algorithms
Annual variations
Clustering
clustering algorithm
Confidence limits
Global navigation satellite system
GNSS
Histograms
ionospheric maps
Machine learning
NavIC
Optimization
Receivers
Satellite configurations
Space weather
total electron content
Unsupervised learning
unsupervised machine learning
ISSN:0048-6604, 1944-799X
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Ionospheric maps from an optimal GNSS (global navigation satellite system) receiver network may lead to better understanding and prediction of the equatorial ionospheric gradients. An unsupervised machine learning framework is proposed using the k $k$‐means clustering approach to optimize the available number of GNSS receivers. Beginning with k $k$ = 15 receivers, the simulations are performed to obtain distribution of the ionospheric pierce points (IPPs) in different grid resolutions up to k $k$ = 1,000. The algorithm uses a reference grid and subsequently optimizes the locations according to the available IPPs in a cluster. The IPP cluster centers are iteratively optimized based on minimizing the “within cluster sum of squares” error. The IPPs from the optimal receiver locations are used to generate a 2‐D vertical total electron content (VTEC) map using NeQuick‐G model. A one‐sigma confidence limit derived from hourly histogram of the IPPs from the k $k$ locations is used as decisive criteria for useable grid cells. The number of empty bins thus obtained in a given scenarios provides a percentage normalized empty grid ratio (NEGratio $\mathrm{N}\mathrm{E}{\mathrm{G}}_{\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}}$). A NEGratio $\mathrm{N}\mathrm{E}{\mathrm{G}}_{\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}}$ below 10% is used as a threshold to make a decision over eligibility of a scenario to produce a desired VTEC map. Several scenarios are thus tested for varying temporal and spatial grid resolutions, geographical boundaries, GNSS and NavIC satellite configurations, considering the diurnal and annual variation in year 2019. The framework is found to be successful in designing a GNSS network for any geographical region to obtain high resolution 2‐D VTEC maps for space weather applications. Key Points New framework is developed using unsupervised machine learning k $k$‐means clustering approach for optimally placing global navigation satellite system receivers in a network The framework automatically selects a design to produce a 2‐D vertical total electron content map over at a desired resolution or vice versa Successful validations are performed under the equatorial ionospheric conditions by using 100, 500, and 1,000 receivers over Indian landmass
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0048-6604
1944-799X
DOI:10.1029/2022RS007548