NEXTOR Impact of FFP1 on NAS Performance Mark Hansen, Geoff Gosling, Chee Chou Ong, Tatjana Bolic, Wenbin Wei November, 2000 1 NEXTOR NEXTOR FFP1 Evaluation Work Overview q q q q q q Part of much broader effort Database Development (TASC) Simulation (Seagull) Normalization (UCB) Safety Impacts (UCB) Valuation (UCB) 2 NEXTOR Normalization q Translate before/after performance comparisons to with/without comparisons q Focus on FFP1 terminal sites where CTAS and/or SMA will be (has been) deployed q NEXTOR focus to date on ATL, LAX, and DFW 3 NEXTOR Normalization Approach q Macroscopic qAnalysis at daily lev el qIncorporate all flight phases q Exploratory q Statistical q Transferable qUse widely available data sources qApplicable to any terminal area 4 NEXTOR Conceptual Framework Demand System Performance ¥En Route Weather ¥Terminal Area ¥ATM Conditions at other Airports ¥Aircraft Time at Origin Airborne Time Total Flight Time Taxi-In Time 5 NEXTOR Daily Flight Time Index (DFTI) q Daily weighted average of flight times to a given airport from a set of origins q Flight time= q Actual Arrival Time - Scheduled Departure Time q Scheduled Flight Time+Departure Delay+Flight Time Delay q Time-at-Origin+Airborne Time+Taxi-In Time q Origins have at least one completed flight in each day of sample q Weights reflect origin share of flights to study airport over study period 6 NEXTOR DFTI Time Series for LAX 220 DFTI (minutes) 200 180 160 140 120 100 J-95 J-96 J-97 J-98 J-99 J-00 J-01 Date q Generally 140-160 minutes q Spikes to over 180 minutes q Seasonal Pattern 7 NEXTOR 30-Day Moving Average 220 DFTI(minutes) 200 180 Winter Spring 160 Summer Fall 140 120 100 Jan-95 Jan-96 Jan-97 Jan-98 Jan-99 Jan-00 Jan-01 Date q Later fall and winter generally worst q Summer 1999 worse than previous ones q Delayed onset of typical winter pattern in 2000 8 NEXTOR 7-Day Moving Average with Components 200 Average Time per Flight (minutes) 180 160 140 120 Taxi-In Airborne 100 At Origin 80 60 40 20 0 Jan-95 Jul-95 Jan-96 Jul-96 Jan-97 Jul-97 Jan-98 Jul-98 Jan-99 Jul-99 Jan-00 Date q Time-at-origin is major source of variation q Correlation betweem time-at-origin and airborne time 9 NEXTOR Weather Normalization q Based on CODAS hourly weather observations for LAX q Factor analysis of weather data qCreate small number of factors that capture variation in large number of variables qFactors are linear combinations of original variables qFactors correspond to principal ax es of Ndimensional data elipse 10 NEXTOR X2 Factor Analysis with Two Variables First Factor X1 11 NEXTOR 9-Factor Representation of LAX Daily Weather Factor Interpretation 1 Warm temperatures throughout day. 2 VFR operations and absence of low cloud ceiling in the morning. 3 VFR operations and absence of low cloud ceiling in the afternoon. 4 High visibility throughout day. 5 Medium cloud ceiling throughout day. 6 High winds throughout day. 7 High ceiling cloud ceiling throughout day; evening precipitation. 8 Precipitation in late morning and afternoon. 9 Precipitation in early morning. 12 NEXTOR Demand Normalization q Based on CODAS OAG data q Capture number, strength, and duration of demand surges above a given baseline rate q Set of 12 metrics using baseline rates from 10 to 120 arrivals per hour q Summarized by two demand factors: one for higher baseline rates and one for higher rates 13 NEXTOR Flight Schedule Comparison 1200 Cumulative Number 1000 800 6/24/97 1/17/99 600 6/29/99 400 200 0 0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:00 Time of Day 14 NEXTOR Normalization for Conditions at other Airports q Consider airports included in DFTI average q For each compute daily average departure delay for flights not bound to LAX region q Average airport departure delays using DFTI weights 15 NEXTOR Origin Airport Delay Time Series 50 45 40 Origin Airport Delay (min) 35 30 25 20 15 10 5 0 Jan-95 Jan-96 Jan-97 Jan-98 Jan-99 Jan-00 Jan-01 Date 16 NEXTOR Performance Models Yt = f (WX t , DMDt , ODELt ) + ε t Where: Yt is DFTI or DFTI component for day t; WXt is vector of weather factors for day t; DMDt is vector of demand factors for day t; ODELt is average origin departure delay for day t; εt is stochastic error term. 17 NEXTOR Functional Forms Considered q Parametric qLinear (with 3, 6, 9, and 12 weather factors) qQuadratic response surface qNon-linear q Non-parametric q9 clusters based on 3 weather factors q12 clusters based on 9 weather factors 18 NEXTOR Linear Model Estimation Results Variable INTERCEPT ODEL WX1 WX2 WX3 WX4 WX5 WX6 WX7 WX8 WX9 DMD1 DMD2 2 ADJUSTED R Description Intercept Origin airport departure delay Warm daily temperatures VFR ops, no low cloud ceiling in the morning VFR ops, no low cloud ceiling in the afternoon High visibility throughout day Medium cloud ceiling throughout day High winds throughout the day High cloud ceiling throughout day Precipitation in late morning and afternoon Precipitation in early morning Peak demand Base demand Estimate T - statistic P - value 138.055 567.065 0.0001 1.128 44.351 0.0001 -1.357 -12.101 0.0001 -0.988 -7.116 0.0001 -1.123 -7.583 0.0001 -0.449 -3.575 0.0004 1.440 10.555 0.0001 0.512 4.531 0.0001 0.911 4.172 0.0001 1.871 8.324 0.0001 -0.379 -2.614 0.0091 0.075 0.725 0.4685 0.440 4.574 0.0001 0.743 19 NEXTOR Predicted Values (min) Predicted vs Actual Values 210.00 200.00 190.00 180.00 170.00 160.00 150.00 140.00 130.00 130.00 140.00 150.00 160.00 170.00 180.00 190.00 200.00 Observed Values (min) 20 NEXTOR Outliers q Used TMU logs to investigate days for which predictions have large errors q Reasons for higher than predicted DFTI qEast flow qRadar outages qAir Force One qOver-stringent ground delay program q No clear explanations for lower-thanaverage DFTI: No news is good news 21 NEXTOR FFP1 Safety Impact q Need to Address Safety Impact of FFP1 Tools q Operational Error Rate Selected as Safety Metric q Significant Variation in Operational Error Rate q Over time at a given facility q Across facilities q Need to Better Understand Causes of Variation q Account for factors other than FFP1 tools q Measure contribution of FFP1 tools 22 NEXTOR NEXTOR Research on Safety Impact q Identify Potential Safety-Related Effects of FFP1 Tools qBetter information for cont rollers qChanges in controller activities qChanges in traffic patt erns qIncrease in traffic handled q Review Previous Research on Factors Associated with High Rates of Operational Errors q Statistical Analysis of Operational Error Data 23 NEXTOR Valuation Research q Translate operational impact to economic value q Avoid traditional delay-centric approach q NEXTOR work on two areas qBuffered vs non-buffered delay qInfering value from increases in observed throughput rates 24