Real-time space weather forecasts based on neural networks Real ...

Real-time space weather forecasts based on neural networks

Henrik Lundstedt and Peter Wintoft Swedish Institute of Space Physics, Lund, Sweden Yurdanur Tulunay and Ersin Tulunay Middle East Technical University, Ankara/Turkey

Outline

• “Space weather” concept: stp/prediction & sw/user/forecast • Workshops on AI Applications in Solar-Terrestrial Physics • Artificial neural networks - Hybrids • Developed solar-terrestrial physics predictions • Operational forcast space weather service for users

Space weather concept Its origin and use • “Space weather” was first mentioned in NASA Technical Document 62-206, 1962. (i.e. by engineers)

Definition IRF-Lund uses

• US Air Force started to use it in 1970

• Prediction within STP

• Scientists Bob McPherron & John Freeman claim they “coined it” in late 1980!

• Forecasting within space weather operational services for users

• In late 1980 power companies (user) contact us in Sweden on space weather forecasts. • NSF/NSWP definition 1995, Report of the Assessment Committee for the NSWP FCM-R24, 2006: (LWS,Heliophysics RM 2035) (CISM,Transitioning scientific STP models into operational tools ≈10 years ahead) • Space weather activities at ESA starts in 1995 i.e. more that 10 years ago. (SWprogramme, SDAs..), EU/COST724…

Workshops arranged by IRF-Lund

Workshops on ”Artificial Intelligence Applications in Solar-Terrestrial Physics” were held in Lund 1993 and 1997. A third workshop: Solar Activity: Exploration, Understanding and Prediction was held in September 19-21, 2005

Artificial neural networks MLBP Neural networks differ by how the neurons (process units) work, how the neurons are connected (topology) and by which learning algorithm that has been chosen.

Elman recurrent neural network Download Lund Dst model in Java and Matlab

Self Organizing Maps

Synoptic field structures were clasified using SOM Structures in solar wind, related to CMEs and CH were clasified with SOM

Neurofuzzy prediction A. Gholipour et al., 2005

Outline

Solar-Terrestrial Physics Predictions Input parameters

Output

KBNM method

Reference

Daily sunspot number

Daily sunspot number

SOM and MLP

Liszka 93;97

Monthly sunspot number

Date of solar cycle max and amplitude

MLP and Elman, Neurofuzzy

Macpherson et al., 95, Conway et al, 98 Maris&Oncica, 06, Gholipour, 05

Monthly sunspot number and aa Date of solar cycle max and amplitude

Elman

Ashmall and Moore, 98

Yearly sunspot number

Date of solar cycle max and amplitude

MLP

Calvo et al., 95

McIntosh sunspot class & MW magn complex.

X class solar flare

MLP expert system

Bradshaw et al., 89

Flare location, duration X-ray and radio flux;X-ray flux

Proton events

MLP, Neurofuzzy

Xue et al., 97; Gabriel et al., 00, Tulunay et al 07

Photospheric magnetic field expansion factor

Solar wind velocity 1-3 days ahead

RBF, PFM MHD

Wintoft & Lundstedt 97;99

Photospheric magnetic synoptic field

Solar magnetic total flux

MLB

Lundstedt et al., 07

Solar-Terrestrial Physics Predictions Input parameters

Output

KBNM method

Reference

Solar wind n, V, Bz

Relativistic electrons in Earth magnetosphere hour ahead

MLP

Wintoft and Lundstedt, 00

Solar wind n,V, Bz, Dst

Relativistic electrons one hour ahead

MLP, MHD, MSFM

Freeman et al., 93

ΣKp

Relativistic electrons day ahead

MLP

Stringer and McPherron, 93

Solar wind V from photospheric B

Daily geomagnetic Ap index

MLP

Detman et al., 00

Ap index

Ap index

MLP

Thompson, 93

Solar wind n, V, Bz

Kp index 3 hours ahead MLP

Boberg et al., 00

Solar wind n, V, B,Bz

Dst 1-8 hours ahead

MLP, Elman

Lundstedt, 91; Wu and Lundstedt, 97

Solar wind n, V, B,Bz

AE 1 hour ahead

Elman, MLP

Gleisner and Lundstedt,00 ,Gavrishchaka et al.,00, 01

Solar-Terrestrial Physics Prediction Input parametrs

Output

KBNM method

References

Solar wind V2Bs, (nV2)1/2, LT, local geomag Δxe, ΔYw

Local geomagnetic field ΔX, ΔY

MLP and RBF

Gleisner and Lundstedt 00

Solar wind n,V, Bz

None, weak or MLP strong aurora (NAO month ahead)

Lundstedt et al., 00 Boberg and Lundstedt 02,03

foF2

foF2 1 hour ahead

MLP

Wintoft and Lundstedt, 99

AE, local time, seasonal information

foF2 1-24 hours ahead

MLP

Wintoft and Cander, 00

MLP

Wintoft and Cander, 99

foF2, TEC, Ap, F10.7 24 hours ahead cm

Tulunay, E., Ozkaptan, C., 00. Tulunay, E., Senalp, E.T., Radicella, S.M., Tulunay, Y.,06

ΣKp

Satellite anomalies

MLP

Wintoft and Lundstedt 00

Solar wind n, V, Bz

dBx/dt, GIC

Elman, MLP

Kronfeldt et al., 01 and Weigel et al.,02

Transitioning Solar-Terrestrial Predictions into operational Space Weather Forecasts

Real-time space weather forecasts and warnings Scientific missions

Solar input data Ends ≈2011

Solar observations with SOHO make warnings 1-3 days ahead possible.

Solar wind observations with ACE make accurate forecasts 1-3 hours ahead possible.

ESA/Lund Space Weather Forecast Service Package

Latest information on forecasts of Kp, Dst, AE and GIC

ESA GIC Pilot Project Real-time predictions of GIC

Real-time forecasts of GIC November 7, 9 2004 events http://www.lund.irf.se/gicpilot/gicforecast/

NN Forecasts from British Geological Survey

Operation neural network space weather forecasts

•

Flare activity based vector magnetogram and NN (Aviation, polar flights, SDO)

• US Air Force MSM (Freeman) forecasts of magnetospheric conditions for satellites (NN Kp + MHD model) (satellites) • Daily Ap NN forecasts (BGS, UK) (Oil companies, power companies) • 3hr Kp NN forecasts (RWC-Sweden, Lund) (STP) • Dst NN forecasts (RWC-Sweden, Lund; RWC-Japan, Tokyo, NICT Space Environment Information Service) (STP) • GIC NN forecasts (RWC-Sweden, Lund) (Swedish power companies)

Coordinate the forecast service with ISES?

ISES Director: D. Boteler Deputy Director: H. Lundstedt Secr. for World days: H. Coffey Secr. Space Weather: J. Kunches WWW for Satellites: J. King