NOTICE: The appearance of external hyperlinks does not constitute endorsement by the United States Department of Defense, the United States Department of the Navy, and the Naval Research Laboratory of the linked web sites, or the information, products or services contained therein. For other than authorized activities such as military exchanges and Morale, Welfare, and Recreation (MWR) sites, the United States Department of Defense, the Department of the Navy and the Naval Research Laboratory does not exercise any editorial control over the information you may find at these locations. Such links are provided consistent with the stated purpose of this DoD Web site.

SPACE WEATHER PREDICTION:

Research and Development

``Space weather'' refers to the state of the magnetosphere and ionosphere which is determined by the solar wind. Under disturbed conditions, satellite- and ground-based technological systems, e.g., communications networks, electric power grids, and satellites, can suffer deleterious effects. Such systems are particularly vulnerable during severe geomagnetic storms. Large storms are relatively infrequent, but when they occur, they can stress the susceptible systems for prolonged periods of time over large geographic areas. Secure operation of systems can still be maintained and hazards can be minimized if the occurrence, duration, and severity of impending storms can be accurately predicted in a timely manner. Thus, space weather forecasting is important for protecting national assets in both the commercial and military sectors. This task is being carried out by the Space Environment Center (SEC) of the National Oceanic and Atmospheric Administration (NOAA) and the U. S. Air Force.

Accurate forecasts of large storms are difficult to achieve because the propagation of solar disturbances to the Earth is difficult to predict with high accuracy. A method of real-time prediction of large nonrecurrent geomagnetic storms is under development at the Plasma Physics Division, Naval Research Laboratory. Using the solar wind quantities measured upstream from the Earth as input, the algorithm recognizes the impending arrival of geoeffective solar wind structures and predicts their geoeffectiveness. The initial tests using historical solar wind data show that the current implementation of the method can accurately identify and predict the occurrence of large storms, with accuracy in the range of 70-80%. Warning time of several hours to more than 10-20 hours may be achieved, depending on the solar cycle.

Real-Time Prediction of Large Geomagnetic Storms

A joint product of the Plasma Physics Division, Naval Research Laboratory, the WIND MFI team of the NASA Goddard Space Flight Center, and the World Data Center C2 (WDC-C2) for Geomagnetism, Kyoto University, Japan.

Testing the Method

The real-time prediction method is currently undergoing testing. The algorithm requires solar wind data as input and calculates the probability that the solar wind being encountered presages the imminent arrival of a geoeffective structure resulting in a storm exceeding a specified threshold. The prediction output also includes the estimated duration (tau') and the maximum value (Bzm') of the Bz component of the geoeffective solar wind event.

In a real-time mode, a continuous stream of real-time solar wind data is required. At this time, such data are not available. We are conducting a comprehensive test of this method in its present form (as of March 1997) using the magnetic field data of the WIND MFI experiment according to a set of quantitative test criteria. The severity of storms is expressed in terms of the hourly Dst. The prediction results are compared with the Dst index published by the WDC, Kyoto University. Currently, it is anticipated that the results will be updated weekly. For the most recent results, Dst values may not be available. As soon as the provisional Dst becomes available, the prediction results and the provisional Dst values are compared. The latest results are shown below.

ACE Data

The algorithm to make predictions on a daily basis using the ACE data is being tested. The real-time ACE data are made available by the Space Environment Center (SEC). Some results of interest will be posted here from time to time. Click here.

Latest Prediction Results

INPUT: By and Bz of the solar wind (panels a and b) (WIND MFI data) ,(ACE SEC).
PREDICTION OUTPUT: tau', Bzm', and P1 (panels c, d, and e)
COMPARISON: Dst index (panel f) (WDC data)

Latest ACE Prediction

==== 19 Nov 2009 ====

New prediction including solar wind speed:

19 Nov

Latest WIND Prediction

==== 08 Nov - 14 Nov 2009 (DOY 312 - 318) ====

Weekly ACE Prediction

==== 08 Nov - 14 Nov 2009 (DOY 312 - 318) ====

What the Figure Shows

The magnetic field vector components are given in the GSM (Geocentric Solar Magnetosphere) coordinate system, with x pointing toward the Sun from the Earth, y in the east-west direction, and z pointing north. The solar wind (SW) is thought of as consisting of a series of events. Each event is defined to be the SW between successive times when Bz changes sign. Each event is characterized by two quantities: (1) tau=duration of the event and (2) Bzm=maximum Bz value (positive or negative) during the event. In the present version of the algorithm, we use only Bz (panel b) as the primary input supplemented by By (panel a). The input is the 5-minute average of the WIND MFI data. For each event being encountered, the duration and maximum Bz value are predicted at each time step. These are given in panels (c) and (d), where the primed symbols (tau' and Bzm') are quantities estimated by using the SW data available up to the time of prediction. tau' is measured from the beginning of the event (Tz), i.e., the last time when Bz = 0. The output of the pediction algorithm at each time step is the probability P1, given in panel (e), that a the event being encountered is geoeffective (threshold: Dst < -80 nT for two hours or longer) with the estimated duration (tau') and maximum Bz (Bzm'). For the purpose of the test, if P1 (solid line) exceeds 0.5 for 2 hours or more, it is regarded as a positive prediction. If a storm with Dst below the -80 nT threshold occurs, the prediction is correct. Presently, the duration of the storm is not predicted. Even if no geoeffective SW event is encountered, P1 can exhibit spikes but should not exceed 0.5 for 2 hours or more. The prediction output (P1) is compared with the actual, provisional, or quick-look Dst index provided by the WDC-C2 (Kyoto University). In addition, the predicted tau' (panel c) and Bzm' (panel d) can be compared with the actual SW input data (panels a and b). Hits, misses, and false alarms are tabulated below.

NOTE: Occasionally, the WIND spacecraft is moved out of the solar wind into the magnetosphere. When they occur, such periods are designated "Non-IMF" and are bracketed by vertical lines (and shaded in panels a and b). The prediction results during such periods are not meaningful.

DISCUSSION OF THE LATEST RESULTS

HOW TO INTERPRET PANEL (e) -- In showing the results (P1) of the Bayesian classification algorithm, several different situations are distinguished by line style.

Solid line:
  Probability that the event being encountered is geoeffective 
  (Dst < -80 nT) with the estimated values of tau' and Bzm'.  
  (Typically, P1 = 0 unless Bzm' < 0.)

Dashed line:
  Probability that the event, having Bzm' > 0, is not geoeffective but 
  will be immediately followed by a geoeffective event of comparable tau' 
  and Bzm' < 0. The time until the predicted southward turning of Bz is 
  equal to [tau' - (present time - Tz)], where Tz is the beginning of the 
  event.  Dashed lines occur only when Bzm' > 0 with long predicted 
  duration and may be interpreted as "storm watches."  
  See DOY 185 below for an example.

Dotted line:
  Probability that the current event is a long-duration nongeoeffective 
  (Bzm' > 0) event and will not be immediately followed by a geoeffective 
  event of comparable tau' and Bzm' < 0.

Previous Results

Earlier results are archived and can be accessed by clicking on the desired time periods listed below.

Last 5 Days (ACE):

Last 5 Weeks:

01 Nov - 07 Nov 2009 (DOY 305 - 311)

25 Oct - 31 Oct 2009 (DOY 298 - 304)

18 Oct - 24 Oct 2009 (DOY 291 - 297)

11 Oct - 17 Oct 2009 (DOY 284 - 290)

04 Oct - 10 Oct 2009 (DOY 277 - 283)

Monthly (WIND):

Performance Characteristics

The performance characteristics of the current version (March 1997) are tabulated below.

The probability distribution functions (PDFs) are constructed using the OMNI data set from 1973 to 1981. The storms in the last 5 months of 1978 are excluded because they were used to test the algorithm [Chen et al., 1997].
The input: By and Bz components of the interplanetary magnetic field (IMF).
The algorithm seeks to identify long-duration southward Bz structures but does not look for high-speed solar wind streams. It is expected that there will be more misses near solar minimum.
Test criteria :
Geoeffective: produces storm with Dst < -80 nT for 2 hours or longer
Positive prediction: P1 > 0.5 for 2 hours or longer

HITS, MISSES, STORM WATCHES, AND FALSE ALARMS

 
   1 January 1998 - 31 Mar 1998 
       Hits:          2 (DOY: 48, 69) 
       Misses:        0
       Storm Watches: 0
       False Alarms:  0
 
   1 January 1997 - 31 December 1997
       Hits:          6 (DOY: 10, 135, 159, 246, 327, 364)
       Misses:        3 (DOY: 111,284,311)
       Storm Watches: 2 (DOY: 185, 265)
       False Alarms:  1 (DOY: 344/345)
      *Storm on DOY 274 occurred while WIND was out of the IMF

   1 January 1996 - 31 December 1996
       Hits*:         0
       Misses:        1 (DOY: 297)
       Storm Watches: 0
       False Alarms:  0
      *Storm on DOY 13 occurred while WIND was out of the IMF

   1 January 1995 - 31 December 1995 -- coming soon

   1978 DOY 225-365 (Click here to see the 1978 results.)
       Hits:          5  (DOY: 240, 272, 300, 303, 329)
       Misses:        1  (DOY: 316)
       Storm Watches: 0
       False Alarms:  1  (DOY: 268)

   Cumulative:
       Hits:          13
       Misses:         5
       Storm Watches:  2
       False Alarms:   2
       No. Large Storms (hits + misses) = 18

Note: A storm watch, shown by dashed lines for P1 [panel (e) above], 
      is included only if no storm follows - a misclassification 
      that does not lead to an "alarm" (solid lines).

FAILURE MODES - Several failure modes have been identified to date. The following situations tend to cause misses: (1) long-duration southward Bz periods interrupted by short Bz > 0 periods; (2) large storms caused by weak magnetic clouds following periods of moderate (e.g., -80 nT < Dst < -40 nT); (3) storms primarily caused by high-speed solar wind. So far, during this solar cycle minimum period, incorrect predictions have been mainly for "marginal" storms, those with minimum Dst values in the vicinity of the threshold of Dst = -80 nT.
EXAMPLES - Some examples of hits, misses, and false alarms are given and discussed. Click here .
For a list of periods since 1995 when the Dst was less than -80 nT for at list 2 hours, click here .

Future Work

We expect to analyze misses and false alarms in detail to better understand the performance characteristics of the methods. A number of improvements are being developed to deal with the known failure modes, and as they are incorporated, we will update this web page to include descriptions and discussions. We expect to extend the method to analyze solar wind particle data in the near future. A real-time version of the algorithm is under development. When it is completed, it will be tested using the real-time solar wind data obtained by the Advanced Composition Explorer (ACE) satellite.

Please Note ...

The content of this web site is the product of the combined work of the three groups. The use of the WIND MFI data and the Dst index is for demonstration. The results of the prediction algorithm have not been generated as predictions in real-time. We welcome comments, suggestions, and criticisms. If you are interested in publishing any part of this web page, please contact the appropriate team.

SPACE WEATHER PREDICTION:

Research and Development

Real-Time Prediction of Large Geomagnetic Storms

Testing the Method

ACE Data

Latest Prediction Results

Latest ACE Prediction

New prediction including solar wind speed:

Latest WIND Prediction

Weekly ACE Prediction

What the Figure Shows

DISCUSSION OF THE LATEST RESULTS

Previous Results

Last 5 Days (ACE):

Last 5 Weeks:

Monthly (WIND):

2009

2008

2007

2006

2005

2004

2003

2002

2001

2000

1999

1998

1997

1996

1978

Performance Characteristics

Future Work

Please Note ...