CENTER FOR AIR TRANSPORTATION SYSTEMS RESEARCH. ICRAT 08 Keynote Talk June 3, 2008

Air Travel at the Edge of Chaos George L. Donohue, Ph.D. Professor Systems Engineering and Operations Research Director of the Center for Air Transportation Systems Research Volgenau School of Information Technology and Engineering ICRAT 08 Keynote Talk June 3, 2008 CENTER FOR AIR TRANSPORTATION SYSTEMS RESEARCH George L. Donohue 2008

Air Transportation is a CAS NOT an Aircraft Design! The National Air Transportation System is a Network of Networks National Air Transportation Capacity Growth & Congestion Management is a Complex Adaptive System Stochastic Network Control Problem NYC Metroplex is a Current Example of a Major Problem that will Illustrate a General Solution Approach 2

Key Nodes in these Networks are Predicted to be Saturated Even with New Runways and Technology! 3 2007 New York Predicted Congested Metropolitan Regions with all NEXTGEN Technology and Runway Improvements FAA FACT 2 Report May 2007 2015 New York Los Angeles Philadelphia San Francisco 2025 New York Los Angeles Philadelphia San Francisco Atlanta Las Vegas Phoenix San Diego

Air Transportation System (ATS) is a CAS with 6 Interacting Network Layers The ATS is a Public - Private Partnership with conflicting objective functions: Public Commerce and safety; interest groups Private Profit maximization Passenger/Cargo Layer (Delays, Cancellations) Airline Layer (Routes, Schedules, A/C size) TSA/FAA Layer (ATC Radar, Radios, Ctr s, Unions) Weather Layer (Thunderstorms, Ice Storms) Physical Layer (i.e. Cities, Airports, Demographics) Government Regulatory Control Layer 4

Outline How Bad and widespread is the Problem What Has Changed Since 1947 Passenger QOS NYC Example What are the Underlying Causes Too Many Scheduled Flights into Too Few Runways Why the Airlines cannot fix the Problem Themselves Prisoners Dilemma and Curse of the Commons Safety is the Underlying Capacity Constraint Current Safety Trends Airport Arrival Time Slot Auctions Economic Impact High Payoff Research Topics NextGen ATM system 5

What has Changed since 1947? Transonic vs. Subsonic Aircraft 40,000 ft vs. 20,000 ft Altitude Avionics: Flight Management Systems Required Navigation Perf. 0.1nm Required Time of Arrival Traffic Collision Avoidance System On the Aircraft! AOC Data Links Zero Visibility Landing Systems ATC radar Separation WHAT HAS NOT CHANGED Air Traffic Controllers talking to Pilots using WW II AM Radio Technology 6

Some Little Known Facts Modern Jet Aircraft Gate-to-Gate Travel Time is the Same or Longer than Propeller aircraft (DC-6 circa 1947) for many routes in NE Triangle Typical Jet Aircraft is 70% Faster and fly's 80% Higher Jet Aircraft can fly Over most bad weather Modern Commercial Jet Aircraft can land in very low visibility Airport Congestion Causes Most ATC Delays and Airline Schedule Padding Masks Real Gate-to-Gate Delay 7

Airline Load Factors are Increasing Percent Aircraft Seats Occupied (Average) 90 85 80 75 70 65 60 55 50 Load Factor (Anticipated) Load Factor ATA Historical Data 45 1960 1970 1980 1990 2000 2010 Year 8

Air Transportation System is Designed to Move Passengers and Cargo Passenger Tier Performance = f (Vehicle Tier Performance, Passenger Factors i.e. Aircraft Gauge, Load Factor, Cancellations) 9

Passenger Total Delay Airports 10 of the OEP-35 airports 50% Total EPTD some airports significantly impact Passenger Delay more than others (e.g. ORD, ATL, DFW and MCO) 50% 10 Close Network of OEP35 Airport in 2004

NYNJ comparison to Comparable European Airports Total Total Average Delays Airport Movements Passengers Minutes 2005 2000 2005 2000 2006 Frankfurt, Gr (FRA) 490,147 458,731 52,219,412 49,360,630 2.7 London, UK (LHR) 477,884 466,815 67,915,403 64,606,826 3 Newark (EWR) 437,402 450,187 33,999,990 34,188,468 28.8 Amsterdam, NL (AMS) 420,736 432,480 44,163,098 39,606,925 0.7 New York Laguardia (LGA) 404,853 384,554 <29,000,000 <28,000,000 23.4 Munich (MUC) 398,838 - <29,000,000 1.8 New York Kennedy (JFK) <353,000 <384,000 41,885,104 32,856,220 24.3 Madrid, Sp (MAD) 415,677 <384,000 41,940,059 32,893,190 1.8 Data taken from ACI-NA, EC PR2006 and FAA ASPM 11

Domestic Markets served by NYC Major Airport Networks: 104 12

Airspace Analysis 3 Airports as One EWR Arrival Airspace 13 JFK Departure Airspace

DOT/FAA response to NYC Congestion Problem Cap Operations at LGA, JFK & EWR Loss of Market Competition & Efficiency Offer Airports the Option to Charge Congestion Pricing Landing Fees PANYNJ rejecting offer Considering using Slot Auctions to find Market Clearing Prices Congressional Authorization may be Required 14

PANYNJ Response to Problem 90 Yr. Lease of Stewart International Airport $79 million + $17 million Improvements 2 Parallel Runways November 1, 2007 assumed control Skybus Airline will connect to Columbus and N.C. Piedmont Triad International Airports Other Airlines offer service to Atlanta, Florida, Detroit and Philadelphia 15

Some Outstanding Issues with these Solutions What are the Airport/Airline/DOT Property Rights? What is the Best Equity Metric? How should Max. Capacity be Determined? What Fraction of Max. Capacity should be Allocated? How Should these Airport Operations be Coordinated? How should Small and Medium sized Communities be Served? How will Market Allocation affect Service? Desired Market Service Redundancy? Desired Market Service Frequency? Desired Aircraft Gauge Distribution? 16

NYC airports Quarter Hour Over-scheduling Percentage (6:00am-10:00pm) in 2007 summer 17 O = n i= 1 I / Demand > maxcapacity n

Airports Served by Multiple NYC Airports # of NYC airports 1 2 3 # of airports served by 29 37 38 % of airports served by 27.9% 35.6% 36.5% What Property Rights? What Equity Metric? What Frequency? What Aircraft Gauge? What Load Factor? What Competition? 18 What Profitability?

Efficiency vs. Profitability? Frequency vs. Seat Size & Unit Revenue High Unit Revenues For Small A/C & High Frequency 19

Reduction in Average Number of Aircraft Seats by Airport All Departures 200 180 160 140-15.5% 120 100-12.7% -0.1% -6.8% 9.9% -18.9% -10.4% -11.4% -4.2% 80 60 40 20 0 ATL DFW EWR IAD IAH JFK LGA ORD PHL July 2002 July 2007 Dorothy Robyn The Brattle Group 20

Interim Observations Airlines are filling planes Flights with low load capacity are mostly to small profitable airports or BOS and DCA Airlines made money in Summer of 2007 So, what's the problem? - Delays! 21

Outline How Bad and widespread is the Problem What Has Changed Since 1947 Passenger QOS NYC Example What are the Underlying Causes Too Many Scheduled Flights into Too Few Runways Why the Airlines cannot fix the Problem Themselves Prisoners Dilemma and Curse of the Commons Safety is the Underlying Capacity Constraint Current Safety Trends Airport Arrival Time Slot Auctions Economic Impact High Payoff Research Topics NextGen ATM system 22

Jet Blue and Delta AL are Competing for the JFK Market: Passengers Pay the Price in Flight Delays and Cancellations JFK Summer 2007 Departures 25 FAA Announced Departure Rate (weekday AVG +/- 2) Airline's Scheduled Departures Flight Departures per 15 Minure Epoch 20 15 10 5 0 5 AM 10 AM 3 PM 8 PM Midnight 0 20 40 60 80 100 120 24 Hours in 15 min. Epochs 23

JFK Scheduled Gate-In/Gate-Out Demand Distribution (Count - Summer 07 ASPM) Gross Over- Scheduling 24

JFK Scheduled Wheels-On/Wheels-Off Demand Distribution (Count - Summer 07 ASPM) Schedule Padding for Expected Taxi Delays 25

JFK Actual Wheels-On/Wheels-Off Demand Distribution (Count - Summer 07 ASPM) FAA ATC Effect DOT CAP 26

Result of this Schedule on Network Delay: AVG Wheels-Off Delays At JFK (ASPM) 95 Minutes! 27

Effect of LGA Slot Control Program: Still Unacceptably High Network Delays! 60 Minutes! 28

Delay Incurred at Major Airports Propagate Network Wide (Summer 2005) Total Delay Ordered by Arrival Delay at Outbound Destination. (minute) 1450000 1250000 1050000 850000 650000 450000 250000 50000-150000 -350000-550000 ATL ORD DFW EWR PHL MSP DTW DEN BOS JFK LGA IAH IAD CLT MCO DCA PHX MIA LAX LAS BWI SEA FLL CVG CLE MDW SFO TPA PIT STL SLC SAN MEM Total Delays (minute) PDX 20,000 Flight Hours Airport Delay Early-arrival Gap Inbound Delay Airborne Delay Arrival Delay at Outbound Dest. 29 34 OEP Airport [Ning Xu GMU 2007]

GMU Model Projects Passenger Delays to Greatly Exceed 2000 delays by 2010 Total Passenger Delays Hours (millions) 160 140 120 100 80 60 Delayed Flights Cancelled Flights Poly. (Delayed Flights) Poly. (Cancelled Flights) 40 20 30-1998 2000 2002 2004 2006 2008 2010 2012 Year D. Wang 2007

Outline How Bad and widespread is the Problem What Has Changed Since 1947 Passenger QOS Economic Impact What are the Underlying Causes Too Many Scheduled Flights into Too Few Runways Why the Airlines cannot fix the Problem Themselves Prisoners Dilemma and Curse of the Commons Safety is the Underlying Capacity Constraint Current Safety Trends Airport Arrival Time Slot Auctions High Payoff Research Topics NextGen ATM system 31

Why do the Airlines Schedule beyond the Maximum Safe RW Capacity with Flights that Loose Revenue? There is no government regulation to limit schedules for safety or compensate passengers for delays and cancellations These were errors in the 1978 Deregulation Act Passenger surveys indicate that frequency and price are the most desirable characteristics of a flight Passengers are not told of consequences of schedule to travel predictability If any one airline decided to offer rational schedules, their competition will offer more frequency to capture market share Thus, still producing delays and cancellations for all In Game Theory, this is called the Prisoner s Dilemma 32

Outline How Bad and widespread is the Problem What Has Changed Since 1947 Passenger QOS NYC Example What are the Underlying Causes Too Many Scheduled Flights into Too Few Runways Why the Airlines cannot fix the Problem Themselves Prisoners Dilemma and Curse of the Commons Safety is the Underlying Capacity Constraint Current Safety Trends Airport Arrival Time Slot Auctions Economic Impact High Payoff Research Topics NextGen ATM system 33

Part 121 (Scheduled Commercial) Accident Rates are Increasing My filtered part-121 accidents Analysis from Zohreh Nazeri, PhD GMU 2007 34 c ount per million operations 2.5 2 1.5 1 0.5 0 y = 0.0533x + 1.0647 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Trends for Incidents Associated with Accidents Trends of the factors in incident databases Pilot factors decreasing Aircraft factors slowly decreasing ATC factors increasing Analysis from Zohreh Nazeri, PhD GMU 2007 35 count per million operations count per million operations Pilot primary factors in ASRS reports 160 140 120 100 80 60 40 20 0 50 40 30 20 10 0 y = -7.1868x + 123.33 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 ATC primary factors in ASRS reports 1995 1996 y = 0.0824x + 29.716 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 count per million operations Aircraft primary factors in ASRS reports 1000 800 600 400 200 count per million operations 0 1995 1996 y = -3.8007x + 683.79 1997 1998 1999 2000 2001 2002 2003 2004 2005 ATC incidents in FAA/OED data, Terminal 40 35 30 25 20 15 10 5 0 2006 y = 1.4581x + 18.118 95 96 97 98 99 00 ;01 02 03 04

ATC factors Communication Errors Top complexity factors associated with ATC factors: number of aircraft in airspace -- airspace design runway configuration -- controller experience 36 % comm errors Top-10 traffic complexity factors associated w ith communcation errors 40.00% 30.00% 20.00% 10.00% 0.00% other blank #a/c, other #a/c airspace, #a/c, other rwy config, other airspace, other #a/c, experience, other These factors will get worse over time: airspace #a/c, rwy config air transportation is projected to grow for the next 10 years majority of controllers will retire within next few years Analysis from Zohreh Nazeri, PhD GMU 2007

37 Aircraft factors Flight Control System problems growing Other aircraft factors decreasing 0.3 0.2 0.1 0-0.1-0.2-0.3-0.4-0.5-0.6-0.7 Growth of aircraft problems in SDRS data 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 Aircraft flight control system problems Flight Control System Nacelle/Pylon attachment Collision Avoidance System Wing y = 0.4037x - 0.2704 Compressor Assembly Landing Gear trend line slop count per million operations 1995 1996 1997 1998 1999 2000 2001 2002 2003 Analysis from Zohreh Nazeri, PhD GMU 2007

Safety at Principle Network Nodes (i.e. Airports) is the Constraint Aircraft Safety Separation Time over the Runway Threshold sets the ATS capacity limits Critical Technical Parameters that Define Network Capacity: 38 Runway Occupancy Time (ROT) Landing Aircraft Inter-Arrival Time (IAT) Cap max = 90 sec IAT at 10-3 P SRO = 40 Arr/RW/Hr Queuing Delay Onset at ~ 80% = 32 Arr/RW/Hr limit for Predictable Performance

Queuing Delays set the Practical Capacity Limitation set by Safety Separation Standards THEORETICAL QUEUEING DELAY Lack of Schedule Synchronization and 90 second IAT generate Queuing Delays above about 80% of Maximum Runway Capacity DELAY ( Minutes ) 70.00 60.00 50.00 40.00 30.00 20.00 Cancellations begin ~ K[rho/(1-rho)] 10.00 32 Arr/RW/Hr 0.00 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 DEMAND / MAX. CAPACITY 39

Runway Scarcity/Value and Risk Runways are expensive Sometimes impossible to build because of shortage of space, etc Because of high demand for runways and their scarcity, runways are highly valuable Thus, Maximizing runway utilization is vital Increasing utilization implies potential for increased risks: wake vortex hazard and simultaneous runway occupancy 40 Risk is the other side of the Throughput Coin

ILS Approach to Runway 21L Approach Plate: Runway 21L Detroit Airport IAF Airport Diagram FAF 41 Ref: AirNav.com. http://204.108.4.16/d-tpp/0610/00119i21l.pdf, EC1, AL-119 (FAA 2007)

Typical ILS Landing Process Runway 21L, Detroit Airport IAF FAF PDF 42 At the threshold ROT LTI or IAD Ref: AirNav.com. http://204.108.4.16/d-tpp/0610/00119i21l.pdf, EC1, AL-119 (FAA 2007) PDF

Data Analysis Process to Estimate: IAT, IAD and ROT pdf s Airplane i Threshold Airplane i+1 Runway Aircraft Type Threshold Leave Runway Heavy 10:23:14 10:24:04 Large 10:24:28 10:25:13 Large 10:26:16 10:27:12 Small 10:28:32 10:29:28......... 43 Col. Clint Haynie, USA PhD., 2002 Yue Xie, PhD. 2005

ILS * Separation Standards (nmi) Current standards provide separation Minima for aircraft pairs ILS Approach In-Trail Threshold Separation Minima (nmi) Follow\ Lead Small Large B757 Heavy Small 3 4 5 6 Large 3 3 4 5 B757 3 3 4 5 Heavy 3 3 4 4 1) Ref: FAA 7110.65 Separation Rules For Arrivals and departures Standards should be fully respected 44 In practice, this is not always the case * Instrument Landing System

Simultaneous Runway Occupancy 1. Simultaneous Runway Occupancy (SRO) Can be avoided by go-around procedure P{SRO} = P{LTI k,k+1 < ROT k & k lands} PDF ROT LTI B. Jeddi, GMU PhD 08 45 (Throughput, Risk WV, Risk SRO ) = f(lti, ROT, WV strength/position)

Conflicting Goals (B. Jeddi, GMU PhD 08) 2. Wake Vortex (WV) hazard Depends on follow-lead aircraft pair type Meteorological condition Strength and position of the WV and position of the following aircraft WV hazard pdf LTI pdf 46 Distance

ROT vs. IAT to find Simultaneous Runway Occupancy (SRO) Probability: est to be ~ 2 / 1000 Runway Occupancy Time (sec) SRO Region Inter-Arrival Time (sec) 47 Detroit Metropolitan Airport (DTW) Freq (IAT < ROT) ~= 0.0016 in peak periods and 0.0007 overall (including non-peak periods - 1870 total samples) IMC: 1 / 669= 0.0015 in peak periods Correlation coefficient = 0.15 [Babak, Shortle and Sherry, 2006]

It does Not Have to Be this Way Runway Occupancy Time (sec) New Avionics & Procedures Existing Avionics and Slot Controls Inter-Arrival Time (sec) 48 Changes in FAA Procedures, Airport Slot Controls and New Avionics Will Improve BOTH Safety and Capacity

Variance Reduction of LTI (B. Jeddi, GMU PhD 08) PDF 0.05 0.04 0.03 0.02 0.01 ROT, Small and Large aircraft %50 %70 %100 LTI ~ LogNormal(40; 4.1, 0.45) Mean= 104.2, Std=49, mode= 87 ROT ~ 0.59 Beta(11.8, 27.9) + 0.41Beta ( 9.0, 16.6) LTI of 3 nm pairs Mean= 104 sec Std= % of 49 sec Std=% of 30 sec 49 0 0 20 40 60 80 100 120 140 160 180 200 Time ( sec ) Current P{SRO} 0.007 30% Std reduction: P{SRO}=0.0014 50% Std reduction: P{SRO}=0.0002

Risk vs. Throughput (B. Jeddi, GMU PhD 08) 0.07 Runway Related Risk vs. Throughput 0.06 RISK = P ( LTI < ROT ) 0.05 0.04 0.03 0.02 0.01 Worse than Linear Linear 0 7 7.5 8 8.5 9 9.5 10 10.5 THROUGHPUT (Arr per quarter hour) 50 Risk is the other side of the throughput coin!

Variance Reduction Effect (B. Jeddi, GMU PhD 08) 44 42 Effect of LTI Variance Reduction on Throughput Curve Percentages are of the Original LTI Standard Deviation %50 Successful Landing ( land / h ) 40 38 36 34 32 3 nm pairs %70 %50 %100 %70 3.6 4 nm pairs 30 %100 51 28 28 30 32 34 36 38 40 42 44 46 48 50 w ( attempt / h )

Weaknesses of Current Standard 1. Many landings fall below given minimum For example, for 3 nm separation pairs, at DFW 26% under 3.0 nm and 7.5% under 2.5 nm (Ballin 1996) Why are they violated so frequently? Well designed?! Are standards too conservative? 2. Minima are not directly operational Controllers add some buffer spacing to assure the minimum (Lebron 1987, Vandavenne 1992, ) These standards do not directly incorporate uncertainty 52

Summary on Capacity 40 Arrival per Runway per Hour is current Safety Maximum 32 Arrivals per Runway per Hour is ONSET of Queuing Delays Using Current (OLD) Technology Using Current (OUTDATED) ATC Procedures FAA has Refused to Mandate New Technology and Procedures to Reduce the Variability in IAT to Increase BOTH Safety and Capacity 53

A Natural DoT Congressional Question? Is There an Optimal Allocation of Scarce Runway Resources? What would happen if schedules at major airports were Capped at Safe, Predictable Runway Capacity and allocated by a Market mechanism? What markets would be served? How would airline schedules change? Frequency Equipment (#seats per aircraft) How would passenger demand change? At airport On routes How would airfares change? What would happen to airline profit margins? 54 How would airport and network delays be altered?

Modeling Approach and Assumptions Port Authority of NY&NJ has the ability to Determine and Set an Optimum Schedule to: Operate at Competitive Profit Margins Maximize Passenger Throughput Ensure an Airline Operating Profit (Max, 90%,80%) All Current Origin and Destination Markets are Considered 67 Scheduled Daily Serviced Markets Current Market Price Elasticity Remains Constant 55

NY LGA Has 67 Daily Markets 56

Airline Competitive Scheduling: Modeling Framework ASPM, BTS databases Demand-Price Elasticity $ Auction IMC Rate: 32 Slots/Hr S 1 S2 D Network Flow Optimization Problem # Flight schedules Fleet mix Average fare Flight delays Delay Network Simulation 57 (Le, 2006)

Model Estimate of Airline Response to an All Weather Predictable Schedule Restriction Estimate of Aircraft Up-Gauging 300 250 Current Fleet Allocation - 1010 Flts 90% Optimum Fleet Allocation - 806 Flts 20 % Fewer Scheduled Flights Number of Daily Flights 200 150 100 50 Increased Passenger Throughput Same Airfares Loss of 3 Unprofitable Markets 0 70% Less Delay 19 to 37 44 to 50 69 to 77 88 to 110 117 to 133 138 to 158 166 to 181 194 to 225 Aircraft Seating Capacity 58

Annual Passenger Enplanements Predicted to be Lost: FAA Forecast to 2025 Annual Projected Enplanements Foregone Because of Airport Capacity Constraints 30 Annual Enplanements Lost (Millions) 25 20 15 10 5 All available landing slots fully utilized regardless of congestion EWR JFK ORD LGA Optimistic: All Planned Airport Improvements Occur 2025 2015 2005 0 ATL BOS BWI CLE CLT CVG DCA DEN DFW DTW EWR FLL HNL IAD IAH JFK LAS LAX LGA MCO MDW MEM MIA MSP ORD PDX PHL PHX PIT SAN SEA SFO SLC STL TPA 59 Airports FAA 2005 TAF & 2004 Benchmark

Economic Optimum Slot Allocation is at 80-90% Max Capacity $40 $35 Some Costs Approximated Other Costs Not Included Slot Reductions Imposed at Top 35 Airports $ B/Yr $30 $25 $20 Sum of Both Costs ($B) Costs Resulting From Reducing Airport Landing Slots $15 $10 $5 Data Scaled to Reflect CY2025 Costs Resulting from Passenge Delays and Flight Cancellations $0 0.0% 2.5% 5.0% 7.5% 10.0% 12.5% 15.0% 17.5% 20.0% Percent of Slots Withheld at Top 35 Airport Donohue CENTER FOR AIR and TRANSPORTATION Shaver, forthcoming SYSTEMS RESEARCH spring 2008

Outline How Bad and widespread is the Problem What Has Changed Since 1947 Passenger QOS What are the Underlying Causes Too Many Scheduled Flights into Too Few Runways Why the Airlines cannot fix the Problem Themselves Prisoners Dilemma and Curse of the Commons Safety is the Underlying Capacity Constraint Current Safety Trends Airport Arrival Time Slot Auctions Economic Impact High Payoff Research Topics NextGen ATM system 61

IS NextGen addressing the Problem? ADS-B (out), 4-D trajectories, RNP-0.1 Good but NOT ENOUGH Aircraft Gauge, Schedule Synchronization and Network Load Balancing will Be Required Annual Combinatorial Clock Slot Auctions? Aircraft Separation in Terminal Airspace and on the Runways MUST be REDUCED by X3! Closely Spaced, Fully-coupled Autopilot Formation Landings with 2 Lane Runways? Closely Spaced Airports need to be Cross-linked with Runway Independent Air Transport New Generation of Heavy Lift Helicopters? 62

The Predicted Growth in Aviation Demand is based on Passenger Demand NOT Aircraft Operations Larger Aircraft will be required to meet X2 or X3 demand Business Jet and VLJ Air Taxi Service will emerge to compete with Commercial aviation due to current System Failure May not be able to put the Geni back in the Bottle Environmental Implications? New Aircraft (e.g. B 787) should be Environmentally Friendly (Emissions/passenger/mi.?) US airlines are not currently ordering them due to poor financial position New Public Policy will be needed to Deal with these Complex Adaptive System Problems NextGen System not addressing these issues 63

Center for Air Transportation System Research Publications and Information http://catsr.ite.gmu.edu Other Useful Web Sites http://mytravelrights.com http://gao.gov http://www.airconsumer.ost.dot.gov 64