Accuracy Estimation of Probabilistic Wake Vortex Prediction Considering Weather Information Errors

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

3rd AIAA Atmospheric Space Environments Conference AIAA 2011-3195

27 - 30 June 2011, Honolulu, Hawaii

Accuracy Estimation of Probabilistic Wake Vortex


Prediction Considering Weather Information Errors

Masahiko Sugiura1 and Naoki Matayoshi2


Japan Aerospace Exploration Agency, Mitaka, Tokyo, 181-0015, Japan

Current wake vortex separation minima are a major impediment to increasing traffic
capacity since they require greater separations than radar separation minima. The concept
of dynamic wake vortex separation minima, which would allow reduced separations in
favorable meteorological conditions when wake durations on flight paths become shorter,
would allow an increase in capacity. This paper proposes a methodology that uses a
probabilistic wake vortex prediction model to probabilistically assure that wake vortex
encounter risks at reduced separations are equal or lower than the risks at current
separations.

I. Introduction conditions by using a probabilistic wake vortex


The demand for air travel continues to grow, and prediction model which considers more complicated
there is a strong demand to reduce aircraft wake decay and advection processes than crosswind
separations to increase traffic capacity. The wake advection. First, we introduce the DLR stochastic
vortex separation minima are a major impediment to two-phase (S2P) wake vortex prediction model 7-9
this since they require 4-6 nm separations, which are which is used in this paper. Second, we describe a
greater than radar separation minima of 2.5-3 nm method to control the hazard probability that wake
(table 1). However, these wake vortex separation vortex encounter risk at reduced separation exceeds
minima were established in the 1970s when the risk at current separations, and clarify the level of
knowledge of wake vortices was limited and might accuracy required for the prediction model for this
therefore be overly conservative, assuming the worst purpose. Finally, we present how to calculate the
case. During the past few decades, knowledge of required information about the prediction model
wake vortices has greatly increased thanks to accuracy.
advances in lidar measurement and CFD analysis
techniques 1, 2. Based on this, the concept of dynamic II. Probabilistic Wake Vortex Prediction
wake vortex separation minima, which would allow Model and Its Error Sources
reduced separations in favorable meteorological There are several parametric wake vortex
conditions when wake durations on flight paths prediction models such as D2P/P2P/S2P developed
become shorter, has been studied intensively 3-6. In by DLR7-9, DVM/PVM developed by UCL10, and the
this concept, the current wake vortex separation AVOSS model developed by NASA/NWRA11. These
minima are considered to be adequately safe, and the models output a wake vortex's circulation strength
wake vortex encounter risks at reduced separations
Table 1. ICAO wake vortex separation minima.
must be equal or lower than the risks at current
Leading Following Non-radar
separations. However, there are still only a limited Radar
aircraft aircraft Departure Arrival
number of the researches that discuss how to control
Heavy 4 nm 2 min. 2 min.
the encounter risk at reduced separations, apart from Heavy
several studies that mainly discuss wake vortex Medium 5 nm 2 min. 2 min.
(136ton)
advection due to crosswinds and so are applicable Light 6 nm 2 min. 3 min.
only to crosswind conditions 5, 6. Heavy
Medium
This paper discusses methods to probabilistically (7ton)
Medium
assure that wake vortex encounter risks at reduced Light 5 nm 3 min.
separations are equal or lower than the risks at all
Light
current separations in a wide range of meteorological categories
______________________________________________________
1
Researcher; Aviation Program Group; 6-13-1 Osawa, Mitaka, Tokyo, 181-0015, Japan; AIAA member.
2
Section Leader; Aviation Program Group; 6-13-1 Osawa, Mitaka, Tokyo, 181-0015, Japan; AIAA member.
1
American Institute of Aeronautics and Astronautics

Copyright 2011 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
and position considering its generation, decay and
advection processes. The initial vortex strength
depends on aircraft parameters such as lift (weight)
and airspeed. Ambient atmospheric parameters such
as wind speed, turbulence and stratification govern
the decay and advection processes. The ground effect
is also modeled as a function of the wake vortex's
circulation strength, its height above the ground and
the time elapsed since its generation.
In our study, the S2P model is employed since it
is capable of providing a probabilistic distribution of
wake vortex circulation strength and vertical/lateral
position in real-time. Considering wake decay and
advection processes, the S2P model predicts the
uncertainty bounds and the upper and lower limits of
wake vortex parameters, and then applies prescribed
probability density distributions (PDDs) normalized Figure 1. Example of S2P calculation.
by the calculated uncertainty bounds to express wake Wake vortex generated by a heavy aircraft on approach
vortex random behaviors (Fig. 1). Since the to an airport. Measured values are normalized by the
uncertainty bounds account for wake vortex equation: , , , ,
deterministic behaviors influenced by flight where is a normalized value, is a measured
conditions and local meteorological conditions, the value, is a upper value, is a lower value.
PDDs are independent of these. We therefore require
only a single PDD for each wake vortex parameter.
However, we have to consider the following two
major PDD error sources:
1) The accuracy of the S2P input parameters: The
PDDs are derived from a wake vortex observation
database normalized by the S2Ps uncertainty bounds
calculated using the flight and meteorological
conditions (S2P input parameters) that prevailed
when the observations were conducted. Thus, the
accuracy of a PDD depends on the accuracies of its
S2P input parameters in the observation database.
This means that we have to modify the PDDs when
we use an S2P with input parameters having different
accuracies from those in the original observation
database. Figure 2. Definition of PDD's confidence interval.
2) The uncertainty of the PDD: The PDD has an
uncertainty since it is based on a finite number of Here, we show how to satisfy eq. (1) by using the
observations. The reliability of the PDD increases accuracy of the wake vortex prediction model
with the number of observations from which it is expressed by the confidence intervals of the S2Ps
derived. PDD. Let us assume that we have the lower limits of
the PDDs confidence interval (PDDL) which
III. Evaluation of Wake Vortex Encounter satisfies the requirement that the probability that
Risk Considering Wake Vortex Prediction true PDD is higher than PDDL for all ranges is
Accuracy 1-PL. In the same manner, we assume that we have
the upper limits of the PDDs confidence interval
Probabilistic assurance that the risk of wake
(PDDU) which satisfies the requirement that the
vortex encounter at reduced separation (PRdc) is equal
probability that true PDD is lower than PDDU for all
to or lower than the risk at current separation (PCur) is
ranges is 1-PU. (Fig. 2) If we use PDDL to
realized by controlling the hazard probability (PHzr)
calculate the wake vortex encounter risk at current
that the encounter risk at reduced separation exceeds
separation, the probability that the true risk PCur is
the risk at current separation (PRdc > PCur) to be equal
lower than the calculated risk (PCur_Cal) is PL. If we
to or lower than some threshold value (PThr).
use PDDU to calculate the encounter risk at reduced
PThr PHzr (1)
separation, the probability that the true risk PRdc is
2
American Institute of Aeronautics and Astronautics
higher than the calculated risk (PRdc_Cal) is PU. the PDDs as follows.
Therefore, when we reduce wake vortex separation Step 1: Add extra errors to the S2P input parameters
up to the point at which the calculated risk at the Xi (i=1-N) in the database such that they have the
reduced separation becomes equal to that at current same amount of error as the new input parameters:
separation (PRdc_Cal = PCur_Cal), the hazard probability XMi = Xi + RNDi (4)
PHzr is expressed as shown in eq. (2). where XMi (i=1-N) is a modified input parameter and
1-PHzr (1-PL)(1-PU) > 1-PL-PU (2) RNDi (i=1-N) is a random number according to the
In fact, S2P predicts three wake vortex parameters: standard normal distribution whose average is zero
circulation strength and lateral/vertical positions. We and standard deviation is unity.
therefore have to modify eq. (2) accordingly by Step 2: Re-normalize the wake vortex observation
applying the above procedure to all three variables. results in the original database using the new S2Ps
1-PHzr ((1-PL)(1-PU))3 > 1-3(PL+PU) (2a) uncertainty bounds calculated by using the modified
From eqs. (1) and (2a), we obtain the requirement for S2P input parameters XMi.
PL and PU to control the hazard probability PHzr to be Step 3: Reconstruct the PDDs from the re-normalized
equal to or lower than the threshold PThr database data.
PThr/3 PL+PU (3) Figure 3 shows an example of PDD modification
As shown above, if we have the confidence for wake vortex lateral position that changes the
interval of the S2Ps PDD (PDDU, PDDL) defined by standard deviation of the error of ambient crosswind
the probabilities of PU and PL satisfying eq. (3), we information from 1 m/s for the original PDD to 3 m/s
can satisfy eq. (1) by evaluating encounter risks using for the modified PDD. We simulated 3,000 wake
PDDU and PDDL. In this method, we control the vortex observations at a Japanese airport using
hazard probability by underestimating the risk at realistic meteorological data produced by a
current separations and overestimating the risk at meso-scale weather forecasting model. Table 2
reduced separations. The extent of underestimation summarizes the Johnson SU probabilistic distribution
and overestimation is expressed as the ratio of parameters used to simulate the original PDD shape.
integrated values of PDDU and PDDL, and this ratio The parameter values were determined by referring to
can be used as a metric for the expected separation DLRs wake vortex observation results 9. We can see
reduction. A larger ratio yields a smaller reduction in that the modified PDD becomes broader than the
separation, since the calculated encounter risk at original according to the increases in the S2P input
reduced separation would reach the same level of risk errors. Please also note that we can modify a PDD
at current separation with a smaller reduction, and
vice versa. Table 2. Johnson SU probabilistic distribution
parameters used in simulation.
IV. Quantification of Probabilistic Wake 0.50
Vortex Prediction Accuracy 0.16
position (y, z)
0
In this section, we describe how to calculate the 1.60
confidence interval of an S2P PDD (PDDU, PDDL)
0.50
considering the accuracies of the S2P input
0.65
parameters. circulation ()
0
3.26
A. Modification of S2Ps PDD Considering S2P
Input Parameters Accuracies
As explained previously, we have to modify an
S2Ps PDDs when we use an S2P with input
parameters having different accuracies from those in
the original observation database. Here, we assume
that the errors of the S2P input parameters follow a
normal distribution, and define that the standard
deviations of the input parameter errors in the
original database are Oi (i=1-N, where N is the
number of the input parameters). When we use an
S2P with input parameters whose error standard
deviations are Ui (i=1-N, Ui Oi), we can obtain
modified PDDs for these S2P input parameters by
modifying the observation database and recreating
Figure 3. Example of S2P PDD modification.
3
American Institute of Aeronautics and Astronautics
only for input parameters whose accuracies are worse
than the original ones (Ui Oi). This means that we
cannot improve the accuracy of the PDD even if we
use an S2P with more accurate input parameters than
the original ones.

B. Calculation of PDD Confidence Interval


Considering the Limitation of Observation
Number
The confidence interval of an S2Ps PDD (PDDU,
PDDL) can be quantified by evaluating the variance
of the PDDs obtained by repeating the process of
PDD construction. We can simulate this PDD
construction process as follows: 1) generate random Figure 4. Example of confidence interval calculation
numbers according to a PDD created from the of S2P PDD.
observation database; 2) create simulated PDDs using
the generated random numbers which simulate We proposed the following two methods.
normalized observation results. Using these 1) A method of evaluating wake vortex encounter
simulated PDDs, we can calculate the confidence risk: We proposed that the encounter risk at reduced
interval of the PDD as follows. Here, we consider the separation should be evaluated using the upper
calculation of PDDU. PDDL can be calculated in the confidence interval of an S2P PDD, and the risk at
same manner. current separation evaluated by using the lower
PDDU is defined as the upper envelope of a confidence interval. By doing this, we can control the
simulated PDDi (i=1-NT), where NT is the number of hazard probability that the encounter risk at reduced
the simulated PDDs. To satisfy the requirement for separation exceeds the risk at current separation to be
PDDU that the probability of true PDD is lower than equal to or lower than a threshold value.
PDDU for all ranges is 1-PU, NT is given by the 2) A method of calculating an S2P PDD confidence
following equation: interval: First, a PDD modification method was
NTlog(1-PCF) / log(1-PU) (5) presented that enables the use of S2P with arbitrary
where PCF is the confidence level. input parameter accuracies. Second, we presented a
Equation (5) is obtained by using the following method of calculating an S2P PDD confidence
conditions: 1) The probability distribution of the interval as a one-sided confidence interval of the
events PDD exceeds PDDU and PDD does not binomial distribution. By combining these two
exceed PDDU is binomial; and 2) The upper value of methods, an S2P PDD confidence interval can be
the one-sided confidence interval of the probability of calculated with arbitrary input parameter accuracies.
a PDD exceeds PDDU event is PU with the Using the proposed methodology, we expect to
confidence level PCF when the number of PDD satisfy the safety requirements for dynamic wake
exceeds PDDU events is zero under NT trials. We use vortex separation reduction according to
the normal distribution as an approximation to the meteorological conditions. However, the separation
binomial distribution to derive eq. (5). Please refer to reduction that can be expected from the proposed
refs.12 and 13 for further details. methodology is still unknown. To quantify this, we
Figure 4 shows an example of the confidence plan to conduct an airport terminal traffic simulation
interval calculation of an S2Ps PDD. We calculate applying reduced separations calculated by the
PDDU and PDDL for the PDDs shown in Fig. 3. proposed methodology.
Values of PU=PL=10-3/6 and PCF = 0.99 are used. The
number of the simulated PDDs used for the References
confidence interval calculation was NT = 27,629 for 1
Gerz, T., Holzpfel, F. and Darracq, D., Aircraft
these PU, PL and PCF values. Wake Vortices: A Position Paper, Wakenet, 2001.
2
WakeNet2-Europe, Wake Vortex Research Needs for
V. Conclusion Improved Wake Vortex Separation Ruling and Reduced
This paper presented a methodology that uses the Wake Signatures, Final Report of the Thematic
Network WakeNet2-Europe, 6th Framework Program,
S2P probabilistic wake vortex prediction model
National Aerospace Laboratory, NLR-CR-2006-171,
developed by DLR to probabilistically assure that the Amsterdam, 2006.
risk of wake vortex encounter at reduced separation 3
Holzpfel, F., Gerz, T., Frech, M., Tafferner, A.,
is equal or lower than the risk at current separation. Kopp, F., Smalikho, I., Rahm, S., Hahn, K.-U., and

4
American Institute of Aeronautics and Astronautics
Schwarz, C., The Wake Vortex Prediction and Monitoring Assessment, Journal of Aircraft, Vol. 43, No. 3, 2006.
9
System WSVBS Part I: Design, 1st CEAS European Air Holzpfel, F., Steen, M., Aircraft WakeVortex
and Space Conference, Berlin, Germany, 2007. Evolution in Ground Proximity: Analysis and
4
Gerz, T., Holzpfel, F., Gerling, W., Scharnweber, A., Parameterization, AIAA Journal, Vol. 45, No. 1, 2007.
10
Frech, M., Wiegele, A., Kober, K., Dengler, K. and Rahm, Jackson. W., Yaras M., Harvey J., Winckelmans G.,
S., The Wake Vortex Prediction and Monitoring System Fournier G., and Belotserkovsky A., Wake Vortex
WSVBS Part II: Performance and ATC Integration at Prediction - An Overview, Phase 6 and Project Final
Frankfurt Airport, 1st CEAS European Air and Space Report, Transport Canada and its Transportation
Conference, Berlin, Germany, 2007. Development Center, TP 13629e.pdf, 2001,
5
Lang, S., Wake Turbulence Mitigation for www.tc.gc.ca/tdc/summary/13600/13629e.htm.
Departures (WTMD) and Wake Turbulence Mitigation for 11
Robins, R. E., and Delisi, D. P., NWRA AVOSS
Arrivals (WTMA), Global wake conference, Brussels, Wake Vortex Prediction Algorithm Version 3.1.1, NASA/
Belgium, 2009. CR-2002-211746, 2002.
6 12
Treve, V., Crosswind dependent separations Shakarian, A., Application of Monte-Carlo
(CROPS), Wakenet 3 Specific Workshop WV Concepts Techniques to the 757/767 Autoland Dispersion Analysis
& Capacity, London, UK, 2011. by Simulation, AIAA Paper 83-2193, 1983.
7 13
Holzpfel, F., Probabilistic Two-Phase Wake Vortex Conover, W. J., Practical Nonparametric Statics,
Decay and Transport Model, Journal of Aircraft, Vol. 40, 3rd ed., John Wiley & Sons, Inc., New York, 1999, pp.
No. 2, 2003. 130-133.
8
Holzpfel, F., Probabilistic Two-Phase Aircraft
Wake-Vortex Model: Further Development and

5
American Institute of Aeronautics and Astronautics

You might also like