Transcript Overview of Space Weather Forecasting

Space Weather Forecasting
2008 Asian-Pacific Region
International Heliophysical Year School
Christopher Balch
NOAA Space Weather Prediction Center
29 October 2008
•
•
•
•
Overview
Solar Flare Forecasting
Geomagnetic Forecasting
Solar Energetic Particle Forecasting
NOAA SPACE WEATHER PREDICTION CENTER
General Points
• Physical Models have not developed to
the point of being useful operationally
• There are efforts to improve this
situation
– Center for Integrated Space weather
Modeling (CISM)
– Coordinated Community Modeling Center
(CCMC)
– Center for Space Environment Modeling
(CSEM at University of Michigan)
Forecasting Today
• Key inputs
– Conceptual models
– Observational data
– Empirical models
• Much depends on the human forecaster to
analyze and synthesize the information
NOAA SPACE WEATHER PREDICTION CENTER
Forecaster Brains
I’m going to need those…
Forecaster
From www.explodingdog.com
NOAA SPACE WEATHER PREDICTION CENTER
The Weather Analysis and
Forecasting Process
What happened?
Diagnosis
Why did it happen?
What is happening?
Nowcasting
Why is it happening?
What is going to happen?
Prognosis Forecasting
Why is it going to happen?
How did/is/will it affect(ing) my customers?
Bosart, 2002
NOAA SPACE WEATHER PREDICTION CENTER
Types of Forecasters
•
•
•
•
•
Intuitive Scientists
Rules-Based Scientists
Procedure-Based Observers
Procedure-Based Mechanics
Disengaged
Pliske et. al., 1997
NOAA SPACE WEATHER PREDICTION CENTER
Good Forecasters…
• maintain situational
awareness
• are organized and can
multitask
• deal well with pressure
• are decisive
• are flexible
• develop good visualization
and conceptualization
skills
Doswell, 2003
• are passionate about [their
work]
• are able to deal with failure
• are continuous learners
• have good “people” skills
• have good communication
skills
• can adapt to shift work
NOAA SPACE WEATHER PREDICTION CENTER
Good Forecasters…
• cultivate increasing technical
proficiency
• synthesize knowledge into
useful wx info
• recognize customers needs,
knowledge level and
expectations
• learn from peers & past
events
• distinguish between
mechanical and diagnostic
prowess
• are interested and passionate
about [their work] –
professionally dedicated
Stuart et. al., 2006
• have good
management/people skills
(delegation, prioritization,
mentoring)
• acknowledge other
perspectives and can tolerate
criticism/disagreement
• are honest & accountable
• maintain a productive rapport
with researchers / modelers
• scrutinize model output
• have stamina for shift work
NOAA SPACE WEATHER PREDICTION CENTER
Challenges
•
•
•
•
•
Time pressure
Too much, too little, bad or conflicting data
Conflicting, or bad model output
Incomplete conceptual models
Human Factors (IT, Rust, Policy, Staffing,
Face-Threat)
NOAA SPACE WEATHER PREDICTION CENTER
Key Data Sources
•Ground Sites
–Magnetometers (NOAA/USGS)
–Thule Riometer and Neutron
monitor (USAF)
–SOON Sites (USAF)
–RSTN (USAF)
–Telescopes and Magnetographs
–Ionosondes (AF, ISES, …)
–GPS (CORS)
•SOHO (ESA/NASA)
–Solar EUV Images
–Solar Corona
(CMEs)
ESA/NASA SOHO
•ACE (NASA)
–Solar wind speed,
density, temperature and
energetic particles
–Vector Magnetic field
•STEREO (NASA)
– Solar EUV Images
– Solar Corona &
Heliosphere (CMEs)
– In-situ plasma & fields
– In-situ energetic particles
– SWAVES
NASA ACE
NOAA GOES
•GOES (NOAA)
–Energetic Particles
–Magnetic Field
–Solar X-ray Flux
–Solar EUV Flux
–Solar X-Ray Images
NOAA POES
•POES (NOAA)
–High Energy Particles
–Total Energy Deposition
–Solar UV Flux
Solar flares
• What is a solar flare ?
• H-alpha classification system
• X-ray classification system
June 6, 2000 at 1616 UTC
Holloman Solar Observatory
June 6, 2000 at 1715 UTC SXT
GOES XRS 5-7 June 2000
H-alpha flare classification system
BLOCK
I –H-alpha
Flares
e • Two letter classification
based on
e actual
coverage areabased
of the flare
on theofsolar
surface
• Importance:
on area
brightening
ification
es report H imagery and report their brightness and area
ess
• Brightness:
Line
of intensity
a solar
flare is classified
aswidth
faint (F),
normal (N),increase
or
e
• Example: 1B means area of 100-250, brightness such that ≥20 millionths
is at ≥ 50 % above bkgnd± 1.0 Å of H-alpha center
X-ray flare classification system
• Based on total (spatially integrated) x-ray flux from Sun in 1-8 Å band
• Continuous observations provided by GOES satellites
• Letter, number system: letter for ‘decade’, number for level in decade
Peak
Flux
≥10-8
≥10-7
≥10-6
≥10-5
≥10-4
Class
A
B
C
M
X
Example:
if peak flux is 2.3 x 10-5
then x-ray class is M2.3
Solar Active Regions and Flare Prediction
• C, M, X, Proton probabilities 1-3 days
• Images used: white light, surface
magnetic fields, H-alpha, X-ray
• Focused on active regions and magnetic
field structure
• Baseline – climatology
• Persistence
June 5, 2000 at 1714 UTC
Big Bear Solar Observatory
June 7, 2000 at 1430 UTC
Mt Wilson Solar Observatory
Analysis
• Why is a given region flaring ?
• Evaluate complexity, dynamics,
rate of growth/decay, ‘hot spots’
• Looking for shear, proper motion,
differential rotation effects
• E/W vs N/S inversion lines
Sunspot Classification System: Optical
NOAA SPACE WEATHER PREDICTION CENTER
Solar magnetic field
•
•
•
•
•
Sunspots and magnetic fields
Observations
Magnetic classification system
Vector magnetic fields
Conceptual model for magnetic loops
– ‘Potential fields’
– ‘Non-potential’ fields
– Relationship to flare probability
• Role of growth, decay, differential rotation, proper
motions
NOAA SPACE WEATHER PREDICTION CENTER
Sunspot Classification Systems: Magnetic
NOAA SPACE WEATHER PREDICTION CENTER
Extension of magnetic
fields into the corona
Storing energy in the
coronal magnetic field
NOAA SPACE WEATHER PREDICTION CENTER
Flare Prediction Tool
Forecaster- entered values
Forecast these values
using the formula
RTOTAL = 1-[(1-R1)
*(1-R2)*… *(1-Rn )]
where
R1, R2 ,… ,Rn are C,
M or X flare
probabilities for the
Statistical probabilities
individual regions on
the disk.
Observed flares
NOAA SPACE WEATHER PREDICTION CENTER
Limitations
• Usually do not have vector field observations
• No one really knows what triggers a flare
– Concept of self-organized criticality
• Do not know when new flux will emerge or old
flux will decay
• No physical model guidance
NOAA SPACE WEATHER PREDICTION CENTER
Developmental Activities
that may improve flare forecasting
• Helicity as a driver of eruptive events
• Parameters derived from vector magnetograms
– Also exist studies deriving parameters from lineof-sight magnetograms
• Bayesian methods to combine multiple inputs
(e.g. persistence and sunspot classification)
• Solar dynamics observatory
• Helioseismology (to detect emerging flux…)
NOAA SPACE WEATHER PREDICTION CENTER
Geomagnetic Forecasting
• Physical Drivers of Geomagnetic Activity
– Transients from CME’s
– Recurrent High Speed Streams from Coronal Holes
• Additional Factors to consider
– Seasonal effects (climatology)
– Continuation of current levels (persistence)
Geomagnetic Specification Using Indices
• A summary of activity:
– The indices are intended to provide a summary of variations of the
Earth’s magnetic fields
• An interpretation:
– Helps to extract a particular type of magnetic variation or an ensemble
of variations (as related to one or more magnetospheric or ionospheric
current systems)
• Help users and non-specialists
– Enable users to distinguish between times of high risk and low risk
– Non-specialists can know the the ‘level’ of activity without having to
interpret magnetograms or satellite data
• Facilitate comparative studies
– Compare activity level with related phenomena
– Study cause-effect relationships
– Investigate long time series behavoir
– Simplify predictions of activity levels which are based on solar or solar
wind observations
NOAA SPACE WEATHER PREDICTION CENTER
List of Indices
Index
Year
Application/Interpretation
K
1939
3-hourly range of irregular variations (quasi-log)
Kp
1951
Planetary average of K
A/Ap
1951
Based on K/Kp, Equivalent amplitude, daily average
R
1963
Hourly range (auroral/polar variations – substorms)
am
1969
Globally averaged 3-hourly range (linear)
AE
1969
Global Auroral Electrojet (substorms)
Dst
1969
Equatorial variations (‘ring current’)
aa
1975
Antipodal average, 3-hourly, equivalent amplitude
PC
1979*
Polar variations (related to SW merging electric field)
NOAA SPACE WEATHER PREDICTION CENTER
K-index & A-index
• Measure of 3 hourly range of irregular variations:
– Designed for sub-auroral and mid-latitude stations
– Time interval optimized for 45-180 minute timescales:
• magnetic bays – typical signature of an electrojet injection
– Subtraction of SR: daily regular variation from the magnetometer data
– Maximum of range of horizontal components:
• scaled from 0 to 9 using quasi-log scale (e.g. from Boulder):
Range: 0-4 5-9 10-19 20-39 40-69 70-119 120-199 200-329 330-499 500
K:
0 1
2
3
4
5
6
7
8
9
– Normalization by geomagnetic latitude: (e.g. K9 threshold in Boulder
is 500, College is 2500), to get similar K frequency distributions
– K9 occurs about 0.1% of the time (~30 times per solar cycle)
• 24 hour A-index – average of 3-hourly equivalent amplitutes, ak:
K 0 1 2 3 4 5 6 7 8 9
ak 0 3 7 15 27 48 80 140 240 400
NOAA SPACE WEATHER PREDICTION CENTER
Planetary Kp/Ap
• Kp: a type of planetary average of station K-indices
– A specific set of sub-auroral stations is used
– The calculation includes a ‘standardization’ for each station K-index
– Kp is discretized into 28 levels from 0 to 9
• Kp-est: USAF estimate of Kp based on real-time data
– Different observatory network
– Network has a Northern American bias
– Currently updated on one-hour cadence
• Ap/Ap-est: 24 hourly index of activity
– Calculated just like the 24-hour A-index for a single station
– Based on Kp/Kp-est
NOAA SPACE WEATHER PREDICTION CENTER
NOAA SPACE WEATHER PREDICTION CENTER Berthelier, 1993
Activity Categories
• Qualitative activity level
descriptor based on K & A
K=0,1,2
A є [0,7]
Quiet
K=3
A є [8,15]
Unsettled
K=4
A є [16,29]
Active
K=5
A є [30-49]
Minor Storm
K=6
A є [50,99]
Major Storm
K=7,8,9
A > 100
Severe
Storm
NOAA SPACE WEATHER PREDICTION CENTER
Geomagnetic Forecasts
Sources:
• Earthward-directed
CME’s
• Coronal-hole high speed
streams
Prediction inputs – CME’s
•
•
•
•
•
CME properties
Properties of associated x-ray event
Location of associated activity
Radio signatures (Type II/Type IV)
Medium energy particles (ACE)
Prediction inputs – CH’s
• Size & Location or CH
• Polarity of CH
• Comparison with previous rotation
Solar Wind & Coronal Holes
• Open field regions
– Source of high speed solar wind
• High Speed Stream (HSS)
– Evolve slowly from rotation to
rotation
• Plasma characteristics:
– elevated T
– low density
– presence of Alfven waves
NOAA SPACE WEATHER PREDICTION CENTER
Solar Wind & Coronal Holes
• Co-rotating Interaction
Region (CIR)
• The HSS builds up a
leading CIR
• Plasma characteristics:
– Compressed
– Enhanced Density
– Enhanced Magnetic Field
– Possible shock formation
• In some cases the CIR
shock will accelerate
particles
• Solar wind signatures look
similar to a transient
Recent 27 day
plot of solar
wind data,
showing the
high-speed
stream structure
Russell McPherron Effect
Long-known semi-annual variation
of geomagnetic activity
Svalgaard et al, 2002
A spiral field will have components Bx and By
in the solar-equatorial coordinate system
(GSEQ)
The Solar Magnetospheric coordinate system
is rotated about X, so the By component in
GSEQ will have By and Bz components in
GSM (e.g. the Earth’s ‘dipole’ tilts into or
away from the spiral field)
Russell & McPherron 1973
Away sector more geoeffective in fall
Towards sector more geoeffective in spring
Coronal Hole Forecasting
• Recurrence – what happened 27 days ago?
– Recurrence can help to identify co-rotating disturbances;
the best view is obtained by looking at a 27 day plot of
solar wind data
– Look for changes in coronal hole morphology for this
rotation relative to last rotation – this should be factored
in as a small modification to recurrence
– Remember semi-annual variation of geomag activity
– Positive holes are more geoeffective in the fall, negative
holes are more geoeffective in the spring
– CIR’s lead coronal holes and can look similar to CME’s
(Temperature provides the key clue)
NOAA SPACE WEATHER PREDICTION CENTER
Coronal Hole
Forecasting Challenges
• Timing can be difficult
– Depends on SW speed
– Coronal hole boundaries don’t
provide full information about
expansion in the heliosphere
– Evolution is slow but occurs
– Can miss timing by 12-24
hours easily
Coronal Hole
Forecasting Challenges
• Timing can be difficult
– Depends on SW speed
– Coronal hole boundaries don’t
provide full information about
expansion in the heliosphere
– Evolution is slow but occurs
– Can miss timing by 12-24
hours easily
• Stereo-B can help
– There are subtle effects in
addition to co-rotation that
have to be accounted for
Solar Wind - Transients
• Coronagraph data are critical
• POS speed can be estimated
– lower constraint on transit time
• Interaction of CME with ambient
solar wind is important
• Back-to-back CME’s – second
CME won’t slow down very much
• Size of CME as it moves out is
difficult to know
NOAA SPACE WEATHER PREDICTION CENTER
Solar Wind - Transients
• CME’s in the solar wind
– Shock, sheath, driver
– Transit times for CMEs
– Using EPAM
ENLIL model
HAF
model
NOAA SPACE WEATHER PREDICTION CENTER
Geomagnetic Forecasting: Transients
• CME-driven disturbances
– As the CME moves out it develops three key
components: shock, sheath (swept up solar wind), and
driver (simple conceptual picture)
– What direction is CME going, how fast, what kind of
solar wind is it ‘plowing’ through ?
– Will Earth go through the center, through the side,
maybe we will only go through the sheath, or maybe it
will miss earth altogether
– Have to rely on grey-matter fusion to estimate CME
trajectory: use location on disk and visual appearance
– STEREO –it will help although we don’t have an
objective technique at this point
NOAA SPACE WEATHER PREDICTION CENTER
Example: X3/4B flare with CME
Fast moving CME, estimated
POS velocity 1500 km/s
d/v calculation ~ 28 hours
NOAA SPACE WEATHER PREDICTION CENTER
Response to X3/CME at 13/0240Z - ~35 hour transit time
Sheath
Driver/Cloud
Shock at 14/1356Z
Geomagnetic Forecasting: Transients
• CME-driven disturbances
– EPAM also provides clues (Energetic Particles on ACE)
– Absence of EPAM signatures is a decent indicator that
we won’t get hit, at least not by a strong CME
– Timing: distance/velocity provides a lower limit on the
arrival time (the fast CME’s tend to decelerate).
However, if the solar wind has been previously swept up
by a preceding CME, deceleration of the successive
CME is usually small
– For earthbound CME’s, speed is probably the best
indicator of storm severity
NOAA SPACE WEATHER PREDICTION CENTER
ACE EPAM shows rise in flux ahead of
the interplanetary shock
Limitations
• Arrival time is the biggest challenge today
• Must somehow account for the swept up
component of the solar wind
• Geometry of the shock, the sheath, and the
driver are all uncertain
• Magnetic field in the driver is unknown
– Not obvious if it can be determined w/o in-situ
measurements
NOAA SPACE WEATHER PREDICTION CENTER
Solar Energetic Particles:
nowcasts, forecasts, products
• GOES real-time particle flux data
– Range of energies 0.6 – 500 MeV
– Event defined as flux of 10 p cm-2 s-1 ster-1 (PFU) at > 10 MeV
• Daily forecast
– Three day probabilities for SEP events
• Short term warnings
– Based primarily on flare observations
– Thresholds: 10 PFU  10 MeV and 1 PFU  100 MeV
– Prediction for onset time, maximum flux, time of maximum flux, and expected
event duration
• Alerts
– Real-time reports of an observed event
– Issued shortly after threshold has been attained
• Additional notices at 100, 1000, and 10000 PFU.
• Event Summaries
– Important for users to have an ‘all-clear’ indicator
NOAA SPACE WEATHER PREDICTION CENTER
Solar Energetic Particle Event Forecasting
Sources:
• Strong CME-driven shocks
• Energetic flares in active regions
Reames, 1999
Available Prediction Inputs
•
•
•
•
X-ray maximum and integral flux
Location of associated activity
Radio signatures (Type II/Type IV)
CME properties (speed, total mass,
direction)
Cane et al, 1988
NOAA SPACE WEATHER PREDICTION CENTER
Measuring SEPs
GOES Energetic Particle
Sensor (EPS)
Monitors the energetic
electron, proton, and
alpha particle fluxes
e: 0.6 to 4.0 MeV
p: 0.7 to 700 MeV
a: 4 to 3400 MeV
SWPC processes the
data to derive integral
proton fluxes:
≥ 10 MeV
≥ 30 MeV
≥ 60 MeV
≥ 100 MeV
Proton Event defined:
≥ 10 PFU ≥ 10 MeV
NOAA SPACE WEATHER PREDICTION CENTER
Energetic Storm Particles (ESP)
• Prior to shock passage:
particle trapping leads to flat
time-intensity profiles at
lower energies due to
streaming limit
• Leads to “Energetic Storm
Particles” (ESP) when shock
passes the Earth
• Leads to a ‘broken-power
law’ energy spectrum
• Even high energy particles
can be trapped in the big
events
NOAA SPACE WEATHER PREDICTION CENTER
SEP forecasts
• Three day prediction
– Starts with the flare prediction
– Factor in location of the region (west is better)
– Factor in age of the region (older is better)
– Include persistence of ongoing events
• Warnings
– Key inputs: Integrated x-ray flux, type IV, type II,
location, CME speed
– Longitude/spectrum dependence
– Statistical guidance available but has its limits
– If a Earthbound CME is associated there is a possibility
for an ESP event
– History of the active region
– Comparison with past, similar events
NOAA SPACE WEATHER PREDICTION CENTER
SEP statistical guidance
• Current operational model inputs
– integrated x-ray flux
– peak x-ray flux
– occurrence of type II and type IV radio sweeps
• Statistical: based on event data
• Example:
Integrated flux [0.085-0.257]
X-ray class  [M3-M8]
Radio sweep type II and type IV:
16 such events historically, 6 of which were associated with
proton events => 37.5 % probability (±12%)
NOAA SPACE WEATHER PREDICTION CENTER
1999 Verification Result
• 86 Proton Flares
– Mean prediction 0.37
• 1334 Control Flares
– Mean prediction 0.04
• Event Criteria:
–  M1 flare
–  0.01 Integrated flux
NOAA SPACE WEATHER PREDICTION CENTER
Categorical quality measures for the current model
FAR falls as the
threshold level is
increased
However, POD also
decreases with
increasing threshold
Skill score (HSS)
optimization is achieved
for range of probabilities
20-30%
At optimal point:
POD = 56% (75/134)
FAR = 55% (91/166)
PC = 96% (3888/4038)
HSS = 0.48
Balch, 2008
NOAA SPACE WEATHER PREDICTION CENTER
Work in progress
• New proton event database 1986-2004
– 165 events with associated solar activity
• Control event database 1986-2004
– These are events that meet the necessary solar
conditions for an event (from the 165 events), but
did not result in an SEP event
NOAA SPACE WEATHER PREDICTION CENTER
Event Parameters
Proton Events parameters
XRS/CME event parameters
Onset
Time when 10 MeV flux begins to rise
Onset/Max/End
XRS event times (End at ½ power point)
Threshold
Time when 10 MeV flux reaches 10 PFU
Peak Flux
Maximum 1-8 Å x-ray flux (from XRS)
Maxtime
Time of 10 MeV maximum flux
OptClass
H-alpha optical class of associated flare
Endtime
Time when 10 MeV flux drops below 10 PFU
OptLocation
H-alpha location of associated flare
EventType
Characteristics of the event
Type II
Identify association of type II radio sweep (Yes or No)
P10Max
Maximum flux at 10 MeV
Type IV
Identify association of type IV radio sweep (Yes of No)
P30Max
Maximum flux at 30 MeV
CME onset
Time of onset of associated CME
P60Max
Maximum flux at 60 MeV
CME speed
Leading edge CME speed (linear fit)
P100Max
Maxmum flux at 100 MeV
Integrated XRS flux
Integral of XRS flux from onset to end
P10Fluence
Integral of 10 MeV flux from threshold to end
P30Fluence
Integral of 30 MeV flux from threshold to end
Bkgd Subtracted
integrated XRS flux
Integral of background subtracted XRS flux from onset to
end (background is taken to be the pre-event level)
Temperature1
Derived Temperature using ratio of two XRS channels
P60Fluence
Integral of 60 MeV flux from threshold to end
Emission Measure1
P100Fluence
Integral of 100 MeV flux from threshold to end
Derived Emission Measure using ratio of two XRS
channels
Associated XRS
event
Identifies associated XRS event
Associated SEP
event
Identifies associated proton event (or set to none if there
is no association)
Associated
CME event
Identifies associated CME event
1 Temperature
and Emission measure were derived using the SolarSoft library routines
GOES_CHIANTI_TEM.PRO and GOES_MEWE_TEM.PRO
Single Variable Density Estimation
• Density Estimation (Silverman, 1998) is a method of deducing
continuous probability density functions from observed data
• We consider the distribution of one of the parameters in our
data set, for example integrated x-ray flux
• We expect to have different probability distributions for this
parameter, depending on whether it is from the set of proton
associated events, or from the set of events not associated with
proton events
• Once smooth distributions are found for these two cases, we
can deduce a continuous function for the probability for an
event as a function of the parameter
Description of the Method
• Let {Xi} be the set of n observed values of the parameter, i=1, 2, …, n
• The probability density function fˆ is defined on the continuous
domain of values x for this parameter, such that:
n
1
 x  Xi 
ˆf x  
K


nh i 1  h 
Where we use the normal probability density function for the kernel
z2 2
function K:
e
K ( z) 
2
and h is a smoothing parameter
• Geometrically, each observed value contributes a ‘bump’ to the
overall density estimate, so that the resulting function is a smooth,
continuous probability density for the parameter Xi
Example
Density estimate for log of background
subtracted integrated x-ray flux for proton
events using the normal probability
density with a window width of 0.15
Density estimate for log of background
subtracted integrated x-ray flux for events
not associated with proton events (control
events)
Probability model for proton events, using the density functions
Probability model for proton events,
using the density functions
Probability = np*f1/(np*f1 +nc*f2),
np - number of proton events
f1 - density estimate - proton associated events
nc - number of control events
f2 - density estimate for the control events
The probability is set to constant
once maximum probability is reached
Density Model metrics:
• Accuracy = 0.0235
• Rms error = 0.153
• Skill = 0.269
• Reliability=2.59 x 10-4
• Resolution=8.94 x 10-3
For comparison - metrics
for operational model:
• Accuracy = 0.0246
• Rms error = 0.157
• Skill = 0.234
Density analysis for one parameter (background
subtracted integrated x-ray flux) gives better
accuracy, skill, reliability, and resolution !
• Reliability=6.0 x 10-4
• Resolution=8.1 x 10-3
White – POD
Red – FAR
Green – PC
Yellow – PCH
Blue – GSS
Cyan - HSS
Performance of
categorical forecasts
using probability
thresholds
one parameter density
model
Parameter is background
subtracted integrated xray flux
Skill score optimization
is at probability
threshold of 0.18
At optimal point:
POD = 59% (79/134)
FAR = 59% (113/192)
PC = 96% (3773/4020)
HSS = 0.46
Two Parameter Density Estimation
• The method can be applied using more than one variable as a prediction inputs
• In this case we consider a set of n observation vectors Xi which consist two
components
• Each component represents one of the observed variables (e.g. 1st component
could be integrated x-ray flux, 2nd component could be CME speed)
• The density estimate in this two-dimensional parameter space is defined to be:
n
 x  Xi 
ˆf x  1
K

2  
nh i 1  h 
where we use the standard, multivariate normal density function:

exp  12 zT z
K (z ) 
2

• In order to use a scalar smoothing parameter, h, it is necessary to
normalize the observation vectors – we rescale all of the parameters so
that their range is restricted to the interval [0,1]
NOAA SPACE WEATHER PREDICTION CENTER
2D Density Maps for Background
Subtracted Integrated X-ray flux
combined with H-alpha flare longitude
Red – proton event associated
Green – control event associated
Corresponding Probability
Map for Proton Events
Performance Statistics for this 2D probability model
(Log Background Subtracted Integrated X-ray Flux & Longitude)
QR = 0.0263
RMS = 0.162
Skill = 0.308
REL = 7.60 x 10-4
RES = 1.34 x 10-2
Performance for Categorical Forecasts – 2D density model
White – POD
Red – FAR
Green – PC
Yellow – PCH
Blue – GSS
Cyan - HSS
At optimal point, threshold probability = 21%,
POD = 53.7%, FAR = 41.5%, PC = 96.6%, HSS = 0.542
2D Density Maps for CME speed
combined with Emission Measure
Red – proton event associated
Green – control event associated
Corresponding Probability
Map for Proton Events
Performance Statistics for this 2D probability model
(CME speed and Emission Measure)
QR = 0.0423
RMS = 0.206
Skill = 0.316
REL = 1.86 x 10-3
RES = 2.14 x 10-2
Performance for Categorical Forecasts – 2D density model
White – POD
Red – FAR
Green – PC
Yellow – PCH
Blue – GSS
Cyan - HSS
At optimal point, threshold probability = 24%,
POD = 51.6%, FAR = 41.8%, PC = 94.3%, HSS = 0.517
Top Performers with respect to HSS
2D probability density models
Parameter1
Parameter2
QR
RMS
BgSub Int Xray Flux
longitude
0.0263
0.162
Int Xray Flux
longitude
0.0269
EM (Mewe)
CME speed
CME speed
SS
REL
RES
HSS
Thresh
POD
0.325
7.60E-04
1.34E-02
0.543
0.21
53.7
41.5
96.6
0.164
0.308
1.19E-03
1.32E-02
0.524
0.18
53.7
45.0
96.3
0.0423
0.206
0.316
1.86E-03
2.14E-02
0.517
0.24
51.6
41.8
94.3
longitude
0.0544
0.233
0.313
3.87E-03
2.86E-02
0.495
0.25
43.6
32.5
93.3
EM (Chianti)
CME speed
0.0442
0.210
0.286
1.37E-03
1.91E-02
0.490
0.28
45.2
39.1
94.4
Log Max Xray Flux
CME speed
0.0443
0.210
0.285
1.45E-03
1.91E-02
0.484
0.27
45.2
40.4
94.3
All of these pairs of parameters result in a prediction model
the HSS better than the existing operational model
NOAA SPACE WEATHER PREDICTION CENTER
FAR
PC
Summary
• Forecasting today depends on
– Observing data
– Empirical/statistical approaches
– An experienced person
• Research and Development of practical operational physical
models is in progress
– Importance of establishing and using common set of verification
measures to compare performance & track progress
• Flare forecasting:
– Climatology, persistence, imagery are key inputs
– Eruption of magnetic flux is dominant uncertainty
• Geomagnetic forecasting:
– CME’s and CH’s are key inputs
– Arrival time of disturbed solar wind is dominant uncertainty
• Solar Energetic Particle forecasting
– Currently depends on solar observables and statistics of past events to
deduce mostly likely outcomes
– Physical modeling is particularly difficult
NOAA SPACE WEATHER PREDICTION CENTER
End of Part II