Development of a Runway Veer-Off Location Distribution Risk Assessment and Reporting Template (2014)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

16 Modeling Veer-Off Risk General Approach Enhanced lateral runway excursion risk location mod- els that reflect how RSA configuration and the presence of obstacles or unprepared terrain may impact veer-off risk are presented in this chapter. The enhanced location models are integrated to a three-part modeling approach. Event probability, location probability, and veer-off consequence are shown in Figure 8. This is similar to the approaches used under previous ACRP studies in this area. The first component is the Event Probability (Frequency Model). The likelihood of an aircraft veer-off incident depends on the operation conditions, including airport characteris- tics, weather conditions, and aircraft performance. This also includes the interaction between the runway distance required by the aircraft for the given conditions and the runway distance available at the airport. The probability of an accident is not equal for all loca- tions around the runway. The probability of a veer-off close to the edge of the runway is higher than at larger distances from the runway edge. Also, the probability may be different over the length of the runway. This dependence is represented by the Location Probability Model, which is the second main element of the current methodology. Its development was one of the key goals of this study. The last component is the Veer-Off Consequence Model. The basic approach uses the location models to assess the probability that an aircraft strikes an obstacle in the vicinity of the runway or departs the RSA leading to an accident. Each of these three components is discussed in greater detail in the ensuing sections of this chapter. Event Probability The annual probability of an aircraft veer-off accident depends on the probability of an accident per aircraft move- ment and the number of movements (landings and takeoffs) carried out per year. This probability may be different for each operation at the airport because the conditions may change. To estimate the probability of an accident per movement at any specific airport, a sample of historical data for operations, including aircraft, flight, and weather data is applied to the probability model. During the landing, after touchdown, or during the takeoff roll, the pilot may lose directional control. Some common causes and contributing factors include low runway fric- tion, snow accumulation on the runway, mechanical failures, adverse weather conditions, and pilot deviations. The basis of the approach used to model frequency in this study is presented in ACRP Report 50 and ACRP Report 51. The likelihood of an aircraft veer-off incident depends on the operational conditions and human factors. It includes airport characteristics, weather conditions, and the aircraft performance, as well as the relationship between the runway distance required by the aircraft for the given conditions and the runway distance available at the airport. The basic model structure is: _ 1 1 0 1 1 2 2 3 3 P Accident Occurence eb b X b X b X … { } = + + + + + where • P{Accident_Occurrence} is the probability (0–100%) of an accident type occurring given certain operational conditions; • Xi are independent variables (e.g., ceiling, visibility, crosswind, precipitation, aircraft type); and • bi are regression coefficients. One of the parameters is named runway criticality and represents the interaction between the runway distance required by the aircraft and the runway distance avail- able at the airport. The distance required is a function of the aircraft performance under specific conditions. Therefore, C H A P T E R 5

17 every distance required under International Organization for Standardization (ISO) conditions (sea level, 15 degrees centigrade) is converted to actual conditions for opera- tions. Moreover, the distances are adjusted for the runway surface condition (wet, snow, slush, or ice) and for the level of head/tailwind. The adjustment factors for runway surface condition are those recommended by the Flight Safety Foun- dation (FSF, 2009). Table 4 presents the factors applied to the distance required by the aircraft. Parameters Xi are defined in Table 5, which summarizes the model coefficients obtained for each veer-off frequency model. Figure 8. Risk modeling approach (adapted from ACRP Report 50). Three-Part Risk Model Event Probability Location Probability Operating Conditions (Plane Performance, Type of Operation, Runway Distance Available and Elevation, Weather Conditions) RSA Configuration, Available Runway Distances Type, Size and Location of Obstacles Veer-off Consequences Incident Accident Probability Local Factor Unit Reference Adjustment Elevaon (E)i 1000  E = 0  (sea level) FE = 0.07 x E + 1 Temperature (T)i deg C T = 15 deg C FT = 0.01 x (T – (15 – 1.981 E) + 1 Tailwind for Jets(TWLDJ)iii knot TWLDJ = 0 knot FTWJ = (RD + 22 x TWLDJ)/RDii Tailwind for Turboprops(TWLDT) iii knot TWLDT = 0 knot FTWJ = (RD + 30 x TWLDT)/RD Headwind for Jets(HWTOJ)iii knot HWTOJ = 0 knot FTWJ = (RD + 6 x HWTOJ)/RD Headwind (HWTOT) for Turbopropsiii knot HWTOT = 0 knot FTWJ = (RD + 6 x HWTOT)/RD Runway Surface Condion—Wet (W)iv Yes/No Dry FW = 1.4 Runway Surface Condion—Snow (S)iv Yes/No Dry FS = 1.6 Runway Surface Condion—Slush (SL)iv Yes/No Dry FSL = 2.0 Runway Surface Condion—Ice (I)iv Yes/No Dry FI = 3.5 iTemperature and elevaon correcons used for runway design. iiRD is the runway distance required. iiiCorrecon for wind are average values for aircra type (jet or turboprop). ivRunway contaminaon factors are those suggested by FSF. Table 4. Correction factors applied to runway distance required.

18 where Model Parameter Ref/Unit Comment/Descrip on Equipment Class Ref: C Large jet of maximum takeoff weight (MTOW) 41k-255k lb (B737, A320, etc.) Heavy AircraŒ AB Heavy jets of MTOW 255k lb+ Commuter AircraŒ D Large commuter of MTOW 41k-255k lb (small RJs, ATR42, etc.) Medium AircraŒ E Medium aircraŒ of MTOW 12.5k-41k lb (biz jets, Embraer 120 Learjet 35, etc.) Small AircraŒ F Small aircraŒ of MTOW 12.5k or less (small, single or twin engine Beech90, Cessna Caravan, etc.) User Class Ref: C = Commercial, or F = Cargo, or T/C = Taxi/Commuter User Class G G = GA Turboprop Turboprop engine (yes/no) – Ref: Turbojet Ceiling Height feet Visibility statute miles Crosswind knots Tailwind knots Gusts Yes/No – Ref: No Icing Condions Yes/No – Ref: No Snow Yes/No – Ref: No Rain Yes/No – Ref: No Frozen Precipitaon Yes/No – Ref: No Fog Yes/No – Ref: No Air Temperature Deg C Turboprop Aircra  Yes/No – Ref: No Foreign Origin/Desnaon Yes/No – Ref: No Non-hub Airport Yes/No – Ref: Yes for hub airport Log Cricality Factor Cricality Factor (CF) is defined as the rao between the runway distance available and the runway distance required. A lower rao means a lower safety margin and greater operaon cricality. Night Condions Night, Dawn, or Dusk – Ref: Daylight Notes: Ref: indicates the reference category against which the odds raos should be interpreted. Non-hub airport: airport having less than 0.05% of annual passenger boardings. These event probability models require the use of historical information on operations and weather for the specific air- port. The necessary information on operations includes the time of the flight, runway used, type of aircraft, type of flight, and if the operation was an arrival or departure. In addition, it is necessary to collect the weather information for the same period that operational data are available, usually for one year. Weather information for U.S. airports can be acquired directly from the National Oceanic and Atmospheric Admin- istration (NOAA) database for the weather station located at the airport. However, the information on operations, par- ticularly for non-towered airports, may be harder to obtain, particularly the identification of the runway used. For tow- ered airports operational data can be requested from the FAA. Another challenge is to run the analysis because com- putations can be made only with the help of a computer and specific software that incorporates these models. To facilitate the analysis, average veer-off rates in the U.S. presented in ACRP Report 51 may be used to simplify the application of the proposed approach. The rates are presented

19 in Table 6. The average incident rates are based on the num- ber of accidents and incidents, and the total traffic of relevant operations from 1982 to 2009. From Table 6, LDVOs are approximately 4 times more likely to occur than TOVOs. Location Probability Models There are two location probabilities that are modeled to incorporate in the analysis methodology: • Longitudinal location: The probability that the veer-off occurs within a certain distance from the beginning of the runway, where DExit is the distance from the beginning of the runway to where the plane exited the runway, and DStop is the distance from the beginning of the runway to where the plane stopped or returned to the runway paved area. The “D” distances are measured parallel to the runway centerline; and • Lateral location: The probability that the aircraft may travel beyond a certain distance from the runway edge, where L is a given lateral distance from the runway edge. This “L” distance is measured perpendicular from the runway edge. The product of the previous probabilities provides the probability that the aircraft veers off within a certain subarea between DExit and DStop from the beginning of the runway and travels beyond a certain distance L from the runway edge. Such models will support the analysis and evaluation of RSAs of different widths and help estimate the probability that the aircraft strikes an obstacle located near the runway. Three alternatives were evaluated to normalize the longi- tudinal distances for modeling. The normalized models use normalized distances, or distances transformed to a reference (e.g., the runway length). Whether or not the normalization of longitudinal distances could improve model accuracy was also investigated. The three normalization alternatives evalu- ated were as follows: • Alternative 1—Normalization for the runway distance available (RDA), • Alternative 2—Use of raw distances without normalization, and • Alternative 3—Normalization for the runway distance required. The results achieved from each of these three alternatives were evaluated for accuracy and the most accurate alterna- tive was incorporated into the analysis software developed in this study. As mentioned earlier, rather than solely using the aircraft stopping location, this study attempts to characterize the veer-off path of the aircraft. It was essential to obtain infor- mation on where the aircraft departed the runway and the path followed by the aircraft to help identify the subareas of the RSA affected by the excursion as well as its probability distribution over the runway length. Main Challenges to Develop Location Models The main challenge in developing probabilistic models for runway veer-offs was to find information to character- ize the aircraft veer-off path, as most accident and incident reports lack this information. The alternative was to review the narrative and identify any clues that could be used to infer Variable LDVO TOVO Adjusted Constant -13.088 -15.612 User Class G 1.682 2.094 Aircra Class A/B -0.770 -0.852 Aircra Class D/E/F -0.252 -0.091 Visibility less than 2 SM 2.143 2.042 Visibility from 2 to 4 SM 0.808 Visibility from 4 to 8 SM -1.500 Xwind from 5 to 12 kt 0.653 0.102 Xwind from 2 to 5 kt -0.091 Xwind more than 12 kt 2.192 0.706 Tailwind from 5 to 12 kt 0.066 Tailwind more than 12 kt 0.98 Temp less than 5 C 0.558 0.988 Temp from 5 to 15 C -0.453 -0.42 Temp more than 25 C 0.291 -0.921 Icing CondiŽons 2.67 Rain -0.126 -1.541 Snow 0.548 0.963 Frozen PrecipitaŽon -0.103 Gusts -0.036 Fog 1.74 Turboprop -2.517 1.522 Foreign O/D -0.334 -0.236 Hub/Non-Hub Airport -0.692 Log CriŽcality Factor 4.318 1.707 Night CondiŽons -1.36 Note: LDVO = landing veer-off, TOVO = takeoff veer-off, SM = statute miles, kt = knot, OD = origin/desŽnaŽon. Table 5. Independent variables for veer-off frequency models. Table 6. Average veer-off incident rates (ACRP Report 51) Type of Incident Event Rate per Operaon Operaons per Event LDVO 1.195E-06 837,000 TOVO 2.590E-07 3,861,000

20 the pathway. Clue indications in the narrative included such things as: • Runway lights and signs struck by the aircraft; • Speed when aircraft departed the runway; • Specific airfield components referenced (e.g., crossing of specific taxiways); • Airfield structures and obstacles (e.g., ditches, hangars); and • Phase of flight (e.g., “upon touchdown the right landing gear collapsed and the aircraft swerved to the right”). Another important challenge was to identify an approach that could use the veer-off path instead of using only the final location where the aircraft stopped after the veer-off. This fea- ture was deemed critical as some of veer-off accidents and inci- dents may challenge several subareas of the lateral RSA. The veer-off path was approximated by two linear models and it was necessary to develop a specific code to automatically cal- culate the lateral deviations for each subarea of the lateral RSA. Characterization of the Aircraft Veer-Off Path The aircraft veer-off path is defined as the path of the air- craft from the point where the aircraft departs the edge of the runway to the place the aircraft either comes to a stop or reenters the runway. The veer-off pathway was required to generate data to develop the location models. The path was referenced by the longitudinal distance from the beginning of the runway and the lateral distance from the runway edge. Usually, the path cannot be completely characterized from the information provided in the accident/incident report. Some reports may provide the veer-off path in a diagram or a picture; others do not. Some assumptions and inferences were made based on information contained in the narrative of the report, when possible. Figure 9 shows the references used to measure distances to characterize the veer-off path. For takeoffs, the longitudi- nal distances are measured from the beginning of the take- off runway, unless it is reported that an intersection takeoff occurred. The veer-off distances for landings are measured from the landing threshold (beginning of the runway for landing). The lateral distance is always measured from the runway side edge. The following parameters were defined to characterize the veer-off path and are illustrated in Figure 10: • DExit is the longitudinal distance measured from the begin- ning of the runway to the point where the plane crossed the runway edge and departed the runway; • DStop is the longitudinal distance measured from the beginning of the runway to the point where the plane stopped or returned to the runway; • LStop is the lateral deviation where the plane stopped, or nil, if it returned to the runway surface; • LMax is the maximum lateral deviation from the runway side edge; and • DMax is the longitudinal distance measured from the beginning of the runway to where the plane had the LMax. Figure 10 illustrates a veer-off for which the pilot tries to return the plane to the runway but stops prior to reaching the paved surface. In this situation, LMax is larger than LStop. Implementing these parameters tries to mimic the actual veer-off path with some approximations. Figures 11 through 13 Y D- di st an ce (t ak eo ff ) D- di st an ce (la nd in g) Reference: landing (D=0) L-distance Figure 9. References to distances used to characterize veer-off path. Figure 10. Characterization of veer-off path. DE xi t DM ax LMax • D measured from beginning of runway • L measured from runway edge Assumed PathDS to p Actual Path LStop RSA

21 Figure 11. Runway veer-off distances—LStop  LMax. DE xi t DM ax LStop = LMax D measured from beginning of runway Veer-off Path Assumed Path Figure 12. Runway veer-off – LStop  0. DE xi t DM ax LMax D measured from beginning of runway DS to p LStop = 0 Veer-off Path Assumed Path Figure 13. Runway veer-off— LExit  0. DM ax LMax Threshold (D = 0) DS to p Aircraft Path LExit LExit ≠ 0 DExit = DBR LStop DE xi t show the type of approximation introduced for different types of veer-off path. In Figure 11, the lateral deviation increases until where the plane stops. In this case, DStop is equal to DMax, and LStop is equal to LMax. As shown in the figure, the actual path is normally a curve, which is approximated by a straight line. As shown in Figure 12, the plane veers off the runway and returns to the runway paved area. The location at which the plane has the maximum lateral deviation is characterized with LMax and DMax. The final lateral distance LStop is equal to zero because the plane returned to the runway. The likely curved veer-off path is again approximated with straight lines. Another possible veer-off scenario considered in this study is represented in Figure 13. In this case, the lateral deviation occurs prior to the touchdown, which occurs off the runway. In this case, the runway exit distance Xe is assumed to be the touchdown distance Xtd. In most cases, the aircraft has its veer-off path parallel to the runway, as depicted in the figure. As mentioned earlier, in addition to the veer-off path, data on weather conditions affecting aircraft performance and on runway distance required, such as air temperature, runway elevation, runway surface condition, effective slope, wind direction, and speed were also important information. Finally, it was necessary to characterize the physical condition of the runway, particularly the distances available for landing or takeoff, depending on the type of incident. Normalization of Longitudinal Distances As indicated in earlier sections, the location models devel- oped in this task used a D-L coordinate system where the D-origin was set at the beginning of the runway, and the L-origin was set at the runway edge, as shown in Figure 14, where D1 and L1 coordinates represent the aircraft location off the runway. Three alternatives to transform, or, in other words, to nor- malize the longitudinal distances were evaluated in this study. The normalization procedure consisted of the transformation Figure 14. Reference coordinate System for veer-off location. D L D1 L1 Beginning of Runway Direction of Operations

22 of the longitudinal distances to a reference length, as described below. The runway length was divided into 10 subareas and the location of each subarea is a function of the specific nor- malization procedure used, as follows: • Alternative 1: Normalization for RDA:—actual longitu- dinal distances characterizing the veer-off pathway were divided by the runway distance available for each event. In this case, the beginning of the runway is the origin (D = 0) and the runway end is the maximum value (D = 1). • Alternative 2: Raw Distances: Actual longitudinal distances from the beginning of the runway were used and the run- way subareas were divided into 800-ft intervals with the last interval containing any distance greater than 7,200 ft. • Alternative 3: Normalization for Runway Distance Required (RDR): The runway distance required by the aircraft involved in the event was estimated based on the actual air- craft model, runway elevation, air temperature, and effec- tive runway slope. The subareas were composed of sections with 0.1 RDR in length, with the last interval containing any distance greater than 0.9 RDR. As an illustration, the subareas used for Alternative 1, the normalization of longitudinal distances for the runway dis- tance available, are shown in Figure 15. The runway distance available is divided into 10 sections of equal length and each section includes both the right and the left side subareas of the lateral runway area. Each subarea comprises 5% of the total lateral RSA. It is important to note that the lateral distances were not normalized and only the raw distances in feet were used for modeling. The maximum lateral distance from the runway edge for the grid was set to 1,000 ft. The lateral distance for each event was computed for each subarea that includes any part of the veer-off path. The largest value of L in each subarea was selected to represent the lateral deviation at the subarea for the specific veer-off event, as illustrated in Figure 16. In this example, the aircraft departed the runway in subarea 2R and stopped in subarea 6R. Subareas 1R, 7R, 8R, 9R, 10R and none of the subareas on the left of the runway were challenged by the veer-off event. In subarea 2R, the corresponding D2 is the maximum value of the path in the subarea, which is equiva- lent to the deviation value when the aircraft crossed the inter- face between subareas 2R and 3R. An algorithm was developed and implemented in MS Excel to calculate the lateral distances for each event in each subsec- tion, as a function of the normalization procedure used. The algorithm uses the veer-off distances to define the two linear segments representing the veer-off pathway and calculates the maximum lateral distance in each segment challenged by the veer-off. Data generated was used to develop lateral prob- ability models for each subsection of the RSA. It should be noted that the example presented is quite simple because the veer-off path was approximated with one straight line. For other cases, when the aircraft has a LMax that is greater than LStop (the plane stopping location), the path is repre- sented by two straight lines and the same principle of using the maximum veer-off deviation in the subarea is applied. Location Models The development of a modeling approach for veer-off deviations was one of the key tasks in this study. The basic approach consisted of the following steps for the three nor- malization alternatives evaluated: • Define the grid associated with the selected normalization procedure; Figure 15. Normalization for RDA subareas. Direction of Operation Subarea 1L Subarea 1R Subarea 2L Subarea 2R Subarea 3L Subarea 3R Subarea 4L Subarea 4R Subarea 5L Subarea 5R Subarea 6L Subarea 6R Subarea 7L Subarea 7R Subarea 8L Subarea 8R Subarea 9L Subarea 9R Subarea 10L Subarea 10R Runway Distance Available (RDA) 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA

23 • Conduct normalization for longitudinal veer-off distances; • Identify which subareas were challenged by each event; • Estimate the lateral deviation in each subarea challenged by each veer-off event; • Repeat the process for each veer-off event and count the num- ber of times that each subarea was challenged by all events to calculate the percentage of occurrences in each subarea; • Using the lateral deviation values estimated for each sub- area, develop mathematical models to estimate the prob- ability that an aircraft exceeds a certain lateral deviation during the veer-off event in the specific subarea; • Based on the probability that aircraft may challenge each subarea, develop cumulative probability curves for longi- tudinal distances covered during the veer-off event; and • Develop risk contour curves based on the subarea prob- abilities and the lateral deviation models for each subarea. It is important to note that the modeling effort presented in ensuing sections was developed based on the assumption that aircraft has an equal chance to veer off to the right or to the left side of the runway. However, out of 873 records con- taining information on the veer off side, in 518 events the air- craft departed the left side, in 354 cases the aircraft departed the right side, and in 1 case the aircraft departed one side, crossed the runway and departed the other side. A Chi-Square statistical test was conducted and results dem- onstrated a statistically significant trend toward veer-offs to the left side of the runway. Despite this result, the models were still developed considering an equal split to the left and right side, since runways are used in both directions and splitting the data to model both sides would negatively impact model accuracy. Lateral deviation and longitudinal distance models and risk contour curves were developed for three normalization alternatives described earlier: RDA, raw distances, and RDR; however, only the alternative using the RDA was selected to incorporate in the analysis approach because it was assumed to be the most accurate approach based on the stability of the contour lines generated with the models. High variability in the generated risk contour lines was assumed to be an indica- tion that the models using the specific transformation may not be suitable or may lead to larger errors. To a certain degree, the distance available is related to the aircraft performance during operations in the runway, includ- ing the adjustments for elevation, temperature, slope, wind, and surface conditions. The resulting contour lines using RDA for normalization were more stable and the technique was selected for use in the analysis software. Only the models using the normalization for RDA will be presented in the body of this report. Results for the other two normalization alternatives are presented in Appendix G. A set of lateral deviation models for veer-off was developed using the RDA to transform the longitudinal distances of the veer-off path for each event. The transformation is simply the ratio between the veer-off path distance and the RDA; therefore, the path distances are given as percentages of the RDA for land- ing or takeoff, depending on the type of operation. For example during a landing operation, DExit is equal to 0.25, which means that the aircraft exited the runway at 25% of the landing distance available (LDA), measured from the beginning of the runway. Longitudinal Probability Distribution Figure 17 illustrates the longitudinal probability distribution for both landing and takeoff veer-offs when distances are nor- malized with the RDA. Figures 18 and 19 depict the longitudi- nal probability distributions for LDVO and TOVO, respectively. Based on the results presented in Figures 17, 18 and 19, the cumulative probability distributions for normalization with Figure 16. Representative deviation for each subarea—example. Direction of Operations Runway Distance Available (RDA) 0.1RDA Y2 Y1=0 Y3 Y4 Y5 Y6 Y7=0 Y8=0 Y9=0 Y10=0 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA

24 55 99 146 145 146 105 77 59 54 44 0 20 40 60 80 100 120 140 160 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 N um be r o f C ha lle ng es Subarea Figure 17. Longitudinal probability distribution—both LDVOs and TOVOs—distances normalized by RDA. 40 48 35 29 23 18 17 14 10 9 0 10 20 30 40 50 60 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 N um be r o f C ha lle ng es Subarea Figure 19. Longitudinal probability distribution—TOVOs only—distances normalized by RDA. 15 51 111 116 123 87 60 45 44 35 0 20 40 60 80 100 120 140 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 N um be r o f C ha lle ng es Subarea Figure 18. Longitudinal probability distribution—LDVOs only—distances normalized by RDA.

25 runway distance available were developed. The model inte- grating both LDVOs and TOVOs is illustrated in Figure 20. A polynomial curve was fit to the cumulative probability points. A high degree polynomial was used to obtain the models rep- resenting the probabilities for each subarea with the highest accuracy possible. The models are represented by the follow- ing equations. An R2 of 99.99% was achieved (R2 is a statistical measure of fit; R2 = 100% signifies a perfect fit). Integrated Model for TOVOs and LDVOs 12.1793 36.7712 38.3658 13.9251 0.4265 0.4225 99.9% 6 5 4 3 2 2 CP D D D D D D R( ) = − + − + + + = Model for LDVO 20.4465 63.2398 69.4061 29.2622 1.8031 0.1538 99.9% 6 5 4 3 2 2 CP D D D D D D R( ) = − + − + − + = Model for TOVO 13.1509 43.3722 54.6310 32.0242 7.4079 1.2068 6 5 4 3 2 CP D D D D D D = − + − + + where: D is the normalized longitudinal distance from the begin- ning of the runway and CP is the cumulative probability that a veer-off will occur within D. Lateral Probability Distribution The lateral deviation models were developed using the fol- lowing form: 1P L L eaL b{ }> = where P{L > L1} is the probability that the lateral deviation L exceeds a given distance L1 and a, b are model coefficients. Mathematical models were developed for each subarea using the lateral deviations generated for each LDVO and TOVO event challenging each subarea. Therefore, ten different models were developed for this normalization alternative with respect to the runway distance available. Table 5 summarizes the model coefficients for each subarea. Figures comparing the model estimates with actual data are presented in Appendix C. The last column in Table 7 shows the models’ R2, which rep- resent the excellent accuracy achieved. Based on these models, risk contour lines were derived to cover the runway distance available, as shown in Figure 21. It should be noted that the contour lines represent both sides of the runway. Aircraft deviations are referenced to the center point of the aircraft between the main gears. The ISO-risk lines can be used to estimate the probability that an aircraft exceeds the lateral distance in a given subarea. For example, there is a 5% chance that the path of an aircraft veering off the runway and challenging subarea 6 will exceed a lateral deviation of approximately 200 ft. It should be noted that the risk contour curves presented in Figure 21 are applied to individual subareas. It is not pos- sible to calculate the risk of an accident for a given scenario in which the safety area may have limits and some obstacles may be present. However, it is possible to combine the lateral deviation models with the probability that an aircraft will challenge specific subareas of the runway. Figure 22 combines the results from Figure 21 and the lateral deviation models presented in Table 5, where the probabilities for a given dis- tance are multiplied by the subarea probability. In this case, the contour lines represent the probabilities that an aircraft will exceed a given lateral distance during a runway excursion. Figure 20. Longitudinal cumulative probability distribution for LDVOs and TOVOs—distances normalized with RDA. 0% 20% 40% 60% 80% 100% 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Cu m ul ati ve P ro ba bi lit y Distance Normalized for Runway Distance Available Subarea L Range a b R2 1 0–0.1 -0.03399 0.8407 97.4% 2 0.1–0.2 -0.00690 1.1339 99.3% 3 0.2–0.3 -0.01306 1.0032 99.4% 4 0.3–0.4 -0.00644 1.1576 99.5% 5 0.4–0.5 -0.01354 0.9881 99.1% 6 0.5–0.6 -0.00906 1.0482 98.3% 7 0.6–0.7 -0.00909 1.0014 99.0% 8 0.7–0.8 -0.01136 0.9206 99.2% 9 0.8–0.9 -0.01037 0.970348 98.9% 10 0.9–1.0 -0.00361 1.18109 99.1% Table 7. Lateral deviation models for normalization using RDA.

26 35% 25% 10% 5% 2.5% 35% 25% 10% 5% 2.5% -500 -400 -300 -200 -100 0 100 200 300 400 500 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 D ev ia ti on D is ta nc e fr om R un w ay E dg e (ft ) Subarea Normalized for Runway Distance Available Direction of Operation Left Right Figure 21. Risk contours—probability of deviations exceeding a given distance L1 for each subarea—distances normalized with RDA. Figure 22. Risk contours—adjusted probability of deviations exceeding a given distance L1—distances normalized with RDA. 5% 2% 1% 0.1% 5% 2% 1% 0.1% -600 -500 -400 -300 -200 -100 0 100 200 300 400 500 600 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 De vi ati on D is ta nc e fr om R un w ay E dg e (ft ) Subarea Normalized for Runway Distance Available Direction of Operation Left Right

27 Veer-Off Consequences Approach Using both the lateral deviation models for individual sub- areas in combination with the cumulative probability model for the longitudinal distance, it is possible to evaluate the risk that an aircraft strikes an obstacle during the veer-off. However, the risk of accidents during veer-offs is not always associated with the aircraft collision with an obstacle. For example, in many events the landing gear collapsed during the touchdown resulting in major damage to the aircraft, even before the aircraft departed the runway. In other situ- ations, uneven terrain, sometimes resulting from transitions between paved and unpaved areas, caused the landing gear to collapse or wing/engine to collide with the terrain. Another common occurrence is the collapse of the landing gear dur- ing the runway excursion due to high stresses when tires sink in soft terrain. In many cases, minor damage was caused by aircraft striking runway/taxiway lights and signs. Probability of Accidents Figure 23 summarizes different causes of damage to air- craft during veer-off events with associated frequencies. The illustration contains three groups involving both accidents and incidents, accidents only, and incidents only. The follow- ing categories of aircraft damage cause were identified: • Touchdown Hard—aircraft suffers damage as a result of high stresses or striking the wingtip on the ground. In many cases, damage was a result of the collapse of land- ing gears. • Rough Terrain—aircraft departed the prepared surface of the safety area, crossed the transitions between paved and unpaved surfaces (e.g., crossing taxiways), unprepared terrain, or areas with varying bearing capacity, in many cases off the RSA. • Soft Terrain—aircraft wheels sinking in soft terrain causing high stresses to landing gear that lead to collapse. • Struck Light/Sign—although frangible, these structures may still cause damage to aircraft during runway excur- sions and increase severity of veer-offs. • Mechanical Collapse of Landing Gear—this category does not include cases in which gear collapse occurred due to hard touchdown and is only related to the collapse of the gear during normal touchdowns. • Struck Obstacles—aircraft striking obstacles other than runway/taxiway lights and signs. It may include hangars, ditches, other aircraft, etc. • Other damage causes may include foreign object debris (FOD) ingestion, blown tires, gear-up landings, wildlife strikes, etc. The frequency observed for each of the seven categories of damage causes are represented in Figure 23. Based on Figure 23, the main causes of damage to aircraft during veer-offs were rough terrain and the striking of lights and signs. For accidents, the main causes of damage to air- craft were rough terrain, soft terrain, and striking of obsta- cles. Striking lights and signs were the main cause of damage to aircraft during veer-off incidents. It is important to note that the damage cause may or may not be the cause of the veer-off. For instance, if the landing gear collapses due to high stresses during touchdown, it may be the cause of the veer-off and the cause of damage. However, if an aircraft strikes an obstacle off the runway, it is normally the result of the veer-off rather than the cause of the event. Figure 23. Damage causes during aircraft veer-offs (Mech  mechanical, Acc  accident, Inc  incident).

28 Figure 24 summarizes the data for accidents/incidents for which records contained a veer-off path. This figure indicates aircraft damage frequencies and if the damage occurred on or off the runway. In some cases, aircraft was damaged both on the runway and off. These results are very important to support the modeling approach for consequences because accidents occurring dur- ing veer-offs are not always related to aircraft striking obsta- cles in the vicinity of the runway. Since the damage cause for many veer-off events is not associated with the presence of obstacles in the safety area or its vicinity, it was necessary to combine the probability of striking an obstacle with the probability of substantial damage to the aircraft from other causes based on evidence from veer-off accidents and inci- dents. Historically, approximately 25% of reported veer-off events result in substantial damage to aircraft. Out of those 25%, approximately 3% resulted from aircraft colliding with obstacles. Therefore, the probability of an accident from causes not related with obstacles was approximately 22%. Probability of Aircraft Striking Obstacles Modeling the probability of an aircraft striking an obstacle will require evaluating the probability that the aircraft path passes within the obstacle area. Each veer-off event has a wreck- age path associated with it and Figures 25 and 26 illustrate the average longitudinal distances for each subarea covered by the aircraft path during the runway excursion for LDVOs and TOVOs, respectively. Based on the results presented in Figure 25 for LDVOs, the average distance covered is fairly constant for all the subareas. For TOVOs, the distance is small for subareas near the start of the takeoff and becomes constant for subareas beyond the runway midpoint, as shown in Figure 26. The average distances for each subarea will be used to define an area of influence associated with the position of the obstacle along the runway, as shown in Figure 27. Two areas are characterized in the figure. The first area is called Area of Influence 1 and its length is associated with the average distance X1 covered during veer-offs in the subarea where the obstacle is located, as presented in Figures 25 and 26. X1 depends on the type of operation (landing or takeoff) and the subarea in which the beginning of the obstacle is located. It is assumed that veer-offs initiated in this region will impact the obstacle. The end of this region is located at a distance equivalent to half of the wingspan (WS) of the aircraft consid- ered in the analysis. In this case, it is assumed that the aircraft may collide with the obstacle if located at the farthest point of this region if it deviates enough from the runway edge. The second region is defined as Area of Influence 2. This area has a length X2 that can be calculated with the following formula: X2 L WS 2 WS 2 L WSobs obs= + + = + where X2 is the length of Area of Influence 2, Lobs is the length of the obstacle, and WS is the wingspan of the aircraft considered. Figure 24. Location at which damage was caused to aircraft (Rwy  runway). Figure 25. Average longitudinal distance covered during landing veer-off path— fraction of RDA. 0 0.04 0.08 0.12 0.16 0.2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Av er ag e Lo ng itu di na l Di st an ce o f L an di ng V ee r- off Pa th (R DA ) Subarea Figure 26. Average longitudinal distance covered during takeoff veer-off path— fraction of RDA. 0 0.04 0.08 0.12 0.16 0.2 0.24 0.28 0.32 0.36 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Av er ag e Lo ng itu di na l Di st an ce o f T ak eo ff V ee r- off Pa th (R DA ) Subarea

29 The next step in the modeling approach is illustrated in Fig- ure 28. In this figure (not to scale), the obstacle is located on the left side of the runway at a distance from the runway edge. To use a simple example, the obstacle is parallel to the runway and both the beginning and end of the obstacle are located at the same distance from the runway edge (L1 = L2) (L1 is the lateral distance to the beginning of the obstacle measured from the edge of the runway. L2 is the lateral distance to the end of the obstacle measured from the edge of the runway. The beginning and end of the obstacle are defined based on the direction of operation). For a given aircraft WS, both the length of the obstacle parallel to the runway and the distance from the runway edge are adjusted to include half of the WS (WS/2) as shown in the illustration. The adjustment is to con- sider the difference between the center of the aircraft, which is the reference for the distances (D – L) used in the probability Figure 27. Areas of influence (WS  wingspan, Lobs  length of the obstacle, X1  average distance covered during veer-offs in the subarea where the obstacle is located, X2  length of Area of Influence 2). Direction of Operations Runway Distance Available (RDA) 0.1RDA Obstacle Area of Influence 1 Area of Influence 2 WS/2WS/2 X1 X2 Lobs 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA 0.1RDA Figure 28. Total area of influence and calculation of probabilities (D1 is the longitudinal distance from the runway approach end to the beginning of the obstacle. D2 is the longitudinal distance to the end of the obstacle. Beginning and end of obstacle are defined according to the direction of operation.). Obstacle 0% 20% 40% 60% 80% 100% 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 P1 P2 D1 D2 L1 = L2 Direction of Operations L1 – WS/2

30 models and the tip of the wing. A collision is assumed when the aircraft wingtip or any part of the aircraft strikes the obstacle. With D1, probability (P1) is calculated from the cumulative probability model developed in this study. Using the same model, probability (P2) is estimated based on a distance (D2), as shown in the illustration. The probability that the aircraft will veer off in the longitudinal region of the obstacle (PD) is estimated by PD = P2 – P1. Next, the probability that the lateral deviation from the runway edge exceeds L1 – WS/2 (PL) is estimated using the lateral deviation model for the subarea(s) and the total prob- ability that the aircraft may have struck the obstacle is calculated by the product PD*PL. One or more obstacles may be considered using the approach. In some cases, where the obstacle is at the ground level (e.g., ditches), the center of the aircraft or the width of the main landing gear is considered instead of the wingspan. In addi- tion, the approach may also be applied to obstacles with vari- able distances to the runway edge, by splitting the obstacle in two or more elements. Theoretically, the lateral deviation models could be used to evaluate an obstacle with limited depth, in case the aircraft veers off the runway and has its path behind the obstacle; however, the approach was conservative and the models incorporated in the analysis software cannot evaluate this scenario; instead, it is assumed that obstacles extend to the limits of the RSA. The ultimate goal of modeling the consequences is to esti- mate the probability of accidents resulting from the presence of obstacles. If desired, an adjustment factor can be applied to the probability of veer-offs to estimate the probability of accidents resulting from collision with obstacles. Therefore, the probability of an accident can be calculated using the fol- lowing equation: P P P 0.22acc vo cobs( )= ∗ + where • Pacc is the probability of an accident in the event of a veer-off, • Pvo is the probability of a veer-off (calculated from the frequency model), • Pcobs is the probability of a collision with an obstacle resulting from the veer-off, and • 0.22 is a factor used to add the probability of accidents not related to collision with obstacles.