Engineer School - University of São Paulo
VTEC prediction using a recursive artificial
neural networks approach in Brazil: initial
results
Wagner Carrupt Machado
Edvaldo Simões da Fonseca Junior
MImOSA workshop – february 26th 2013 – INPE - São José dos Campos - Brazil
Presentation outline
• IBGE interest, infrastructure and needs;
• Artificial Neural Networks approach;
• Experiments and Results:
• Solar activity and geomagnetic field status;
• Data processing and results;
• Conclusion and future work.
2
IBGE and Ionosphere
GNSS positioning.
(X,Y,Z)
Ionospheric delay
• First-order delay
pseudorange
- More than 99%
- Proportional to TEC
carrier-phase
RBMC
• Since 1996;
• Actually 88 stations;
• Needs densification.
RBMC-IP
• Real time GNSS data
stream on internet (NTRIP –
Networked Transport of
RTCM via Internet
Protocol);
• Since 2009;
• Actually 28 stations.
On-line PPP service
• Double or single
frequency data
processing;
• Global Ionospheric
Maps (IONEX)
applied to single
frequency solutions;
IGS Global Ionospheric Maps
• Combination of four different solutions:
• CODE (Center for Orbit Determination in Europe);
• ESOC (European Space Operations Centre ESA);
• UPC (Polytechnical University of Catalonia);
• JPL (Jet Propulsion Laboratory).
IBGE collaborations
• Providing GNSS data free of charge to
ionosphere monitoring projects:
•
•
•
•
Unesp - Presidente Prudente (Brazil);
INPE/EMBRACE (Brazil);
La Plata University (Argentina);
IGS – currently 9 stations (international).
9
ANN approach
• Architecture:
• Multilayer Perceptrons;
• 1 hidden layer with 16
neurons.
• Recursive training:
• Updated daily.
• samples taken from 3 previous
days IGS GIM, resulting in 39
grids with 276 points (10,764
samples)
• Output:
• 72 hours ahead of regional ionospheric Maps (IONEX).
Experiments
• 30 ANNs trained;
• IGS GIM
high: March 21 to April 04 2001 (Day 80 to 94)
low: June 16 to june 30 2009 (Day 167 to 181)
• Comparison between VTECGIM and
VTECANN in four cases:
1) high solar activity;
2) day of the geomagnetic storm;
3) 3 days after the day of the geomagnetic storm;
4) low solar activity.
Solar activity status
Solar Flux 10.7 cm (NOAA - Pentiction station)
• High solar activity:
• from 139.8 sfu
to 273.5 sfu
• Low solar activity:
• from 66.5 sfu
to 68.5 sfu
Geomagnetic field status
Dst index (Kyoto)
• High solar activity:
• Day 90 => -400nT.
• Low solar activity:
• less than -50 nT.
Differences
Increases as approaching to
daily VTEC maximum;
High solar activity
• < 15 TECU:
– 86% - geomagnetic storm;
– 88% - not disturbed days.
Low solar activity:
• < 5 TECU:
– 99%.
Relative differences ()
• from 0% to 20% during most
of the time in both periods;
• Day 90 (case 2) between 2 h
and 5 h (Local Time)
VTECGIM were pushed down
due to the geomagnetic
storm;
Conclusion
• 70% to 85% of VTECGIM was correctly mapped
by the ANN;
• Vertical ionospheric delay from 0.24 m to 1.79 m
can be expected in L1 observables;
• Insufficient for high precision applications (ambiguity
resolution);
• The proposed approach:
• auto-adaptive to seasonal and longer period variations;
• real-time GNSS positioning;
Future work
• Mod_Ion regional ionospheric maps with spacial
resolution of 2° x 4° and 1 hour frequency;
• Extend the model coverage to South America;
• Use data from the actual solar cycle maximum;
• Include solar activity and geomagnetic indices in
the model.
Acknowledgments