Title: Use of WAAS for LAAS Ionosphere Threat Status Determination
1LAAS Ionosphere Anomaly Prior Probability Model
Version 3.0
Sam Pullen Stanford University spullen_at_stanford.ed
u
14 October 2005
2Proposed Iono. Anomaly Models for LAAS
- Version 1.0 (November 2002 proposed to FAA)
- Fundamentally based on average or ensemble risk
over all approaches - Insufficient data to back up assumed probability
of threatening storm conditions - Version 2.0 (May 2005 internal to SU)
- Uses enlarged database of iono. storm days to
estimate probability of threatening conditions - Considers several options for threshold Kp
above which threat to LAAS exists - Version 3.0 (October 2005) details in this
briefing - Two results one for fast-moving wave-front
anomalies (detectable by LGF) and one for
slow-moving (potentially undetectable) anomalies - Establishes basis for averaging over both
storm-day probabilities and over hazard
interval within a storm day
3Two Cases for this Study
- For fast-moving storms prior probability of
potentially-hazardous fast-moving storm prior to
LGF detection, but including precursor credit - Result sets PMD for relevant LGF monitors
- For slow-moving storms prior probability of
slow-moving (and thus potentially undetectable by
LGF) storm, including precursor credit - Feasible mitigation is included in prior prob.
4Pirreg Prior Prob. Model used in WAAS
- Cited by Bruce used in GIVE verification in
WAAS PHMI document (October 2002) - Pirreg formerly known as Pstorm
- Examines probability of transition from quiet
to irregular conditions in given time interval - Upcoming GIVE algorithm update does not need it
(can assume Pirreg 1) - Uses a pre-existing model of observed Kp
occurrence probabilities from 1932 - 2000 - Each Kp translates into a computed conditional
risk of unacceptable iono. decorrelation for GIVE
algorithm (decorr. ratio gt 1)
5Key Results from Pirreg Study
Kp Occurrence Probs.
Conditional Decorrelation Probs.
WAAS Safety Constraint
Resulting Pirreg for WAAS 9.0 10-6 per 15
min. (calculated) 1.2 10-5 per 15 min. (add
margin)
6Observed Iono. Storm Totals since Oct. 1999
7Severe Kp State Probability Comparison
- Pirreg model has 5x lower probs. than more
recent numbers - Observations since 10/99 are conservative since
they cover the worst half of a solar cycle - Appears reasonable to use actual fraction of days
potentially threatening to CONUS 4 / 2038
0.00196
8Confidence Interval for Probability of
Threatening Storms (1)
- Use binomial(s,n) model to express confidence
interval (CI) for Pr(threatening storm) ? PTS - i.e., observed s threatening storm days over n
total days (x ? n s number of non-threatening
days) - Analog to Poisson continuous-time model
- CI needed since s 0 for slow-moving storms
- More conservative lower tail limit 1 - L(x)
(Martz and Waller, Bayesian Reliability Analysis,
1991) - Where 100 a 100 (1 g/2) lower
percentile of CI
9Confidence Interval for Probability of
Threatening Storms (2)
- For fast-moving storms
- s 4 n 2038 x n s 2034
- ML (point) estimate PTS s / n 0.00196
- 60th percentile estimate 1 - L(x).4 PTS60th
0.00257 - 80th percentile estimate 1 - L(x).2 PTS80th
0.00330 - For slow-moving storms
- s 0 n 2038 x n s 2038
- ML (point) estimate PTS s / n 0
- Point est. bound for s 1 PTS_bnd s / n
4.91 10-4 - 60th percentile estimate 1 - L(x).4 PTS60th
4.50 10-4 - 80th percentile estimate 1 - L(x).2 PTS80th
7.89 10-4
10Time Averaging over Course of One Day
- For all non-stationary events, anomalous
ionosphere gradient affects a given airport for a
finite amount of time - Model each airport as having Nmax 10 satellite
ionosphere pierce points (IPPs) - Satellites below 12o elevation can be ignored, as
max. slant gradient of 150 mm/km is not
threatening - Conservatively (for this purpose) ignore cases of
multiple IPPs being affected simultaneously - For both cases, determine probability over time
(i.e., over one threatening day) that a given
airport has an ionosphere-induced hazardous error
11Time Averaging for Fast-Moving Storms
- Fast-moving storms are detected by LGF during
rapid growth of PR differential error right after
LGF is impacted by ionosphere wave front - SU IMT detects within 30 seconds of being
affected - Thus, for each satellite impacted, only worst
30-second period represents a potential hazard - Assume EXM excludes all corrections once two
different satellites are impacted - Based on two-satellite Case 6 resolution in SU
IMT EXM - Fast motion of front prevents recovery between
impacts - Assume two fast-moving fronts (rise then fall, or
vice-versa) can occur in one day
12Modeling Precursor Event Probabilities
- Ionosphere anomalies are typically accompanied by
amplitude fading, phase variations, etc. that
make reliable signal tracking difficult - CORS data usually shows L1 and (particularly) L2
losses of lock during time frame of ionosphere
anomalies - This fact makes searching CORS data for
verifiable ionosphere anomalies quite difficult - LGF receivers and MQM should be more sensitive to
these transients than CORS receivers - Multiple gaps in data render over 80 of CORS
station pairs unusable for gradient/speed
estimation during iono. storms - Therefore, pending further quantification,
conservatively assume that 80 of threatening
ionosphere fronts are preceded by precursor
events that make the affected satellites unusable - Actual probability is likely above 90
13Probability Model for Fast-Moving Storms
gt Resulting fast-moving-storm prior prob. for a
single airport is 7.14 10-7 per approach
14Triangle Distribution for Slow-Speed Gradients
- For slow-moving storms, both point estimate bound
and 60th-pct bound seem too conservative - no gradients large enough to be threatening
(i.e., gt 200 mm/km) have been observed at all - To address expected rarity of slow-moving and
threatening gradients, a triangle distribution is
proposed - Linearly decreasing PDF as slant gradient
increases - Assume practical maximum of 250 mm/km
PDF
q
btot 2/150 to give Atot 0.5 atot btot 1
atot 150
aexc 50
100
200
150
250
Slant Gradient (mm/km)
? Aexc threatening fraction of PDF 0.5 aexc
bexc 1/9 0.1111
15Time Averaging for Slow-Moving Storms
- Slow-moving storms may not be detected by LGF
during worst-case approach, but would be detected
soon afterward - Thus, for each satellite impacted, one 150-second
approach duration represents the hazard interval - Slow-moving (linear-front) storms can only affect
one satellite at a time - Very wide front might affect multiple satellites,
but gradient would not be hazardous - Slow motion of front prevents recovery between
impacts - Assume only one slow-moving front event can occur
in one day
16Possibility of Truly Stationary Storms
- Time averaging for slow-moving storms assumes a
minimum practical speed of roughly 20 m/s - Below this speed, a hazardous gradient could
persist for more than one approach (indefinitely
for zero speed) - We have seen no suggestion of storms with zero
velocity (relative to LGF) in CORS data - Even if an event were stationary relative to the
solar-ionosphere frame, it would be moving
relative to LGF due to IPP motion - In other words, stationary relative to LGF
implies motion in iono. frame cancelled out by
IPP motion - Recommendation is to presume some risk of truly
stationary that is a fraction of slow-speed risk
and can be allocated separately within H2 (see
slide 18)
17Probability Model for Slow-Moving Storms
gt Resulting slow-moving-storm prior prob. for a
single airport is 1.74 10-8 per approach
18Observations from these Results
- Feasible CAT I (GSL C) sub-allocation from H2
integrity allocation is as follows - Total Pr(H2) ? 1.5 10-7 per approach (from
MASPS) - Allocate 20 (3.0 10-8) to all hazardous iono.
anomalies - 58 of this (1.74 10-8) must be allocated to
slow-moving iono. anomalies - Reserve an additional 5 of this (7.5 10-9) for
the possibility of truly stationary iono.
anomalies - Then, 37 of allocation (1.11 10-8) remains for
fast-moving ionosphere anomalies - Implied PMD for fast-moving anomalies is 0.111 /
7.14 0.01555 (KMD 2.42) - Given a threatening iono. event, implied
probability that threat is from slow-moving storm
is roughly 0.174 / 7.14 0.024 - This makes sense given apparent rarity of
(non-threatening) slow-moving storms in CORS data
sets
19Summary
- A feasible prior probability model has been
developed to support CAT I (GSL C) LAAS - The key probability averaging steps are
- Averaging over probability of threatening
iono-storm days (used by WAAS for Pirreg) - Time averaging based on fraction of time that a
given airport would face a potential hazard - Triangle distribution for probability of
slow-speed iono. gradients large enough to be
threatening - Some probabilities used here depend on magnitude
of hazardous gradient - Need to iterate between prior model and
mitigation analysis - For extension to CAT III (GSL D), additional
(airborne?) monitoring is needed against
slow-speed events
20Appendix
21User Differential Error vs. Front Speed
LGF impact times
22Differential Error vs Airplane Approach Direction
Iono front hits LGF