An Investigation of Optimal Machine Learning Methods for the Prediction of ROTI

doi:10.11947/j.JGGS.2020.0201

Abstract

Abstract:

The rate of the total electron content (TEC)change index (ROTI)can be regarded as an effective indicator of the level of ionospheric scintillation, in particular in low and high latitude regions. An accurate prediction of the ROTI is essential to reduce the impact of the ionospheric scintillation on earth observation systems, such as the global navigation satellite systems. However, it is difficult to predict the ROTI with high accuracy because of the complexity of the ionosphere. In this study, advanced machine learning methods have been investigated for ROTI prediction over a station at high-latitude in Canada. These methods are used to predict the ROTI in the next 5 minutes using the data derived from the past 15 minutes at the same location. Experimental results show that the method of the bidirectional gated recurrent unit network (BGRU)outperforms the other six approaches tested in the research. It is also confirmed that the RMSEs of the predicted ROTI using the BGRU method in all four seasons of 2017 are less than 0.05 TECU/min. It is demonstrated that the BGRU method exhibits a high level of robustness in dealing with abrupt solar activities.

Key words: machine learning; ROTI prediction; ionospheric scintillation; high-latitude region

Fulong XU,Zishen LI,Kefei ZHANG,Ningbo WANG,Suqin WU,Andong HU,Lucas Holden. An Investigation of Optimal Machine Learning Methods for the Prediction of ROTI[J]. Journal of Geodesy and Geoinformation Science, 2020, 3(2): 1-15.

Figures/Tables 14

Fig.1

Fig.2

Tab.1

Fig.3

Fig.4

Fig.5

Fig.6

Fig.7

Fig.8

Tab.2

Fig.9

Fig.10

Fig.11

Tab.3

References 36

[1]	LI Zishen, WANG Ningbo, WANG Liang, et al. Regional ionospheric TEC modeling based on a two-layer spherical harmonic approximation for real-time single-frequency PPP[J]. Journal of Geodesy, 2019,93(9):1659-1671. doi: 10.1007/s00190-019-01275-5
[2]	LI Wang, YUE Jianping, YANG Yang, et al. Ionospheric and thermospheric responses to the recent strong solar flares on 6 September 2017[J]. Journal of Geophysical Research: Space Physics, 2018,123(10):8865-8883. doi: 10.1029/2018JA025700
[3]	YEH K C, LIU C H. Radio wave scintillations in the ionosphere[J]. Proceedings of the IEEE, 1982,70(4):324-360. doi: 10.1109/PROC.1982.12313
[4]	DE OLIVEIRA MORAES A, COSTA E, DE PAULA E R, et al. Extended ionospheric amplitude scintillation model for GPS receivers[J]. Radio Science, 2014,49(5):315-329. doi: 10.1002/2013RS005307
[5]	JIAO Yu, MORTON Y T, TAYLOR S, et al. Characterization of high-latitude ionospheric scintillation of GPS signals[J]. Radio Science, 2013,48(6):698-708. doi: 10.1002/2013RS005259
[6]	MUSHINI S C, JAYACHANDRAN P T, LANGLEY R B, et al. Improved amplitude-and phase-scintillation indices derived from wavelet detrended high-latitude GPS data[J]. GPS Solutions, 2012,16(3):363-373. doi: 10.1007/s10291-011-0238-4
[7]	BASU S, BASU S, VALLADARES C E, et al. Ionospheric effects of major magnetic storms during the international space weather period of September and October 1999: GPS observations, VHF/UHF scintillations, and in situ density structures at middle and equatorial latitudes[J]. Journal of Geophysical Research: Space Physics, 2001,106(A12):30389-30413.
[8]	GROVES K M, BASU S, WEBER E J, et al. Equatorial scintillation and systems support[J]. Radio Science, 1997,32(5):2047-2064. doi: 10.1029/97RS00836
[9]	HUANG C Y, BURKE W J, MACHUZAK J S, et al. Equatorial plasma bubbles observed by DMSP satellites during a full solar cycle: Toward a global climatology[J]. Journal of Geophysical Research: Space Physics, 2002, 107(A12): SIA-7-1-SIA-7-10.
[10]	CHERNIAK I, ZAKHARENKOVA I. New advantages of the combined GPS and GLONASS observations for high-latitude ionospheric irregularities monitoring: case study of June 2015 geomagnetic storm[J]. Earth, Planets and Space, 2017,69(1):66. doi: 10.1186/s40623-017-0652-0
[11]	AARONS J. Global positioning system phase fluctuations at auroral latitudes[J]. Journal of Geophysical Research: Space Physics, 1997,102(A8):17219-17231.
[12]	PRIKRYL P, GHODDOUSI-FARD R, SPOGLI L, et al. GPS phase scintillation at high latitudes during geomagnetic storms of 7-17 March 2012-Part 2: interhemispheric comparison[J]. Annales Geophysicae, 2015,33(6):657-670. doi: 10.5194/angeo-33-657-2015
[13]	ALFONSI L, SPOGLI L, DE FRANCESCHI G, et al. Bipolar climatology of GPS ionospheric scintillation at solar minimum[J]. Radio Science, 2011, 46(3): RS0D05.
[14]	LI Guozhu, NING Baiqi, REN Zhipeng, et al. Statistics of GPS ionospheric scintillation and irregularities over polar regions at solar minimum[J]. GPS Solutions, 2010,14(4):331-341. doi: 10.1007/s10291-009-0156-x
[15]	JACOBSEN K S, DÄHNN M. Statistics of ionospheric disturbances and their correlation with GNSS positioning errors at high latitudes[J]. Journal of Space Weather Space Climate, 2014,4(A8):A27. doi: 10.1051/swsc/2014024
[16]	SIERADZKI R, PAZIEWSKI J. GNSS-based analysis of high latitude ionospheric response on a sequence of geomagnetic storms performed with ROTI and a new relative STEC indicator[J]. Journal of Space Weather and Space Climate, 2019,9(4):A5. doi: 10.1051/swsc/2019001
[17]	PI X, MANNUCCI A J, LINDQWISTER U J, et al. Monitoring of global ionospheric irregularities using the worldwide GPS network[J]. Geophysical Research Letters, 1997,24(18):2283-2286. doi: 10.1029/97GL02273
[18]	BASU S, GROVES K M, QUINN J M, et al. A comparison of TEC fluctuations and scintillations at Ascension Island[J]. Journal of Atmospheric and Solar-Terrestrial Physics, 1999,61(16):1219-1226. doi: 10.1016/S1364-6826(99)00052-8
[19]	BEACH T L, KINTNER P M. Simultaneous global positioning system observations of equatorial scintillations and total electron content fluctuations[J]. Journal of Geophysical Research: Space Physics, 1999,104(A10):22553-22565.
[20]	CHERNIAK I, KRANKOWSKI A, ZAKHARENKOVA I. Observation of the ionospheric irregularities over the northern hemisphere: methodology and service[J]. Radio Science, 2014,49(8):653-662. doi: 10.1002/2014RS005433
[21]	CHERNIAK I, ZAKHARENKOVA I. High-latitude ionospheric irregularities: differences between ground-and space-based GPS measurements during the 2015 St. Patrick’s Day storm[J]. Earth, Planets and Space, 2016,68(1):136. doi: 10.1186/s40623-016-0506-1
[22]	SECAN J A, BUSSEY R M, FREMOUW E J, et al. High-latitude upgrade to the wideband ionospheric scintillation model[J]. Radio Science, 1997,32(4):1567-1574. doi: 10.1029/97RS00453
[23]	MAURITS S A, GHERM V E, ZERNOV N N, et al. Modeling of scintillation effects on high-latitude transionospheric paths using ionospheric model ( UAF EPPIM)for background electron density specifications[J]. Radio Science, 2008, 43(4): RS4001.
[24]	PRIKRYL P, JAYACHANDRAN P T, MUSHINI S C, et al. Toward the probabilistic forecasting of high-latitude GPS phase scintillation[J]. Space Weather, 2012,10(8):S08005.
[25]	CHARTIER A, FORTE B, DESHPANDE K, et al. Three-dimensional modeling of high-latitude scintillation observations[J]. Radio Science, 2016,51(7):1022-1029. doi: 10.1002/rds.v51.7
[26]	MCGRANAGHAN R M, MANNUCCI A J, WILSON B, et al. New capabilities for prediction of high-latitude ionospheric scintillation: a novel approach with machine learning[J]. Space Weather, 2018,16(11):1817-1846. doi: 10.1029/2018SW002018
[27]	HU Andong, ZHANG Kaifei. Using bidirectional long short-term memory method for the height of F2 peak forecasting from ionosonde measurements in the Australian region[J]. Remote Sensing, 2018,10(10):1658. doi: 10.3390/rs10101658
[28]	STANKOV S, KUTIEV I, JAKOWSKI N, et al. GPS TEC forecasting based on auto-correlation analysis[J]. Acta Geodaetica et Geophysica Hungarica, 2004,39(1):1-14. doi: 10.1556/AGeod.39.2004.1.1
[29]	NGWIRA C M, MCKINNELL L A CILLIERS P J. GPS phase scintillation observed over a high-latitude Antarctic station during solar minimum[J]. Journal of Atmospheric and Solar-Terrestrial Physics, 2010,72(9-10):718-725. doi: 10.1016/j.jastp.2010.03.014
[30]	MAYAUD P N. Derivation, meaning, and use of geomagnetic indices[M]. Washington, DC: American Geophysical Union Geophysical Monograph Series, 1980: 22.
[31]	MENVIELLE M, IYEMORI T, MARCHAUDON A, et al. Geomagnetic indices[M] //MANDEA M, KORTE M. Geomagnetic Observations and Models. Dordrecht: Springer, 2011: 183-228.
[32]	GILES C L, KUHN G M, WILLIAMS R J. Dynamic recurrent neural networks: theory and applications[J]. IEEE Transactions on Neural Networks, 1994,5(2):153-156. doi: 10.1109/TNN.72
[33]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997,9(8):1735-1780. doi: 10.1162/neco.1997.9.8.1735 pmid: 9377276
[34]	CHO K, VAN MERRIËNBOER B, GULCEHRE C , et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation [C]//Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP). Doha, Qatar: Association for Computational Linguistics, 2014: 1724-1734.
[35]	SCHUSTER M, PALIWAL K K. Bidirectional recurrent neural networks[J]. IEEE Transactions on Signal Processing, 1997,45(11):2673-2681. doi: 10.1109/78.650093
[36]	JAYACHANDRAN P T, LANGLEY R B, MACDOUGALL J W, et al. Canadian high arctic ionospheric network (CHAIN)[J]. Radio Science, 2009, 44(1): RS0A03.

Characteristic indices	ANN	RNN	LSTM	GRU	BRNN	BLSTM	BGRU
Long-term memory	No	No	Yes	Yes	No	Yes	Yes
Direction of recurrent layer	No	Unidirectional	Unidirectional	Unidirectional	Bidirectional	Bidirectional	Bidirectional
Connection between the hidden layer and output layer	All neurons	The last time step	The last time step	The last time step	All time- steps	All time- steps	All time- steps
Number of gate structures	0	0	3	2	0	3	2

The set which contains data for 250th day
Test set	ANN	0.0476	0.0606	0.0643
	RNN	0.0482	0.0566	0.0569
	LSTM	0.0471	0.0548	0.0506
	GRU	0.0472	0.0551	0.0493
	BRNN	0.0487	0.0569	0.0561
	BLSM	0.0468	0.0550	0.0506
	BGRU	0.0466	0.0539	0.0491
Validation set	ANN	0.0442	0.0519	0.0940
	RNN	0.0441	0.0520	0.0520
	LSTM	0.0440	0.0486	0.0498
	GRU	0.0440	0.0474	0.0490
	BRNN	0.0450	0.0530	0.0521
	BLSM	0.0447	0.0515	0.0496
	BGRU	0.0440	0.0474	0.0484
Training set	ANN	0.0486	0.0766	0.0524
	RNN	0.0486	0.0645	0.0436
	LSTM	0.0485	0.0629	0.0383
	GRU	0.0486	0.0618	0.0380
	BRNN	0.0503	0.0657	0.0420
	BLSM	0.0483	0.0627	0.0388
	BGRU	0.0483	0.0625	0.0379

Seasons	Methods	RMSE of training set	RMSE of validation set	RMSE of test set
Spring	ANN	0.0503	0.0446	0.0552
	RNN	0.0501	0.0442	0.0383
	LSTM	0.0501	0.0413	0.0338
	GRU	0.0492	0.0412	0.0329
	BRNN	0.0493	0.0426	0.0357
	BLSM	0.0491	0.0408	0.0325
	BGRU	0.0490	0.0406	0.0322
Summer	ANN	0.0426	0.0570	0.0477
	RNN	0.0434	0.0539	0.0447
	LSTM	0.0429	0.0514	0.0433
	GRU	0.0427	0.0513	0.0433
	BRNN	0.0428	0.0537	0.0449
	BLSM	0.0424	0.0510	0.0432
	BGRU	0.0432	0.0517	0.0432
Autumn	ANN	0.0455	0.0568	0.0499
	RNN	0.0466	0.0558	0.0479
	LSTM	0.0446	0.0527	0.0466
	GRU	0.0448	0.0524	0.0464
	BRNN	0.0470	0.0590	0.0496
	BLSM	0.0446	0.0525	0.0463
	BGRU	0.0450	0.0522	0.0460
Winter	ANN	0.0574	0.0511	0.0497
	RNN	0.0568	0.0498	0.0485
	LSTM	0.0565	0.0494	0.0463
	GRU	0.0577	0.0489	0.0459
	BRNN	0.0574	0.0511	0.0496
	BLSM	0.0577	0.0497	0.0459
	BGRU	0.0564	0.0485	0.0451