Airport Aviation Activity Forecasting (2007)

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.

18 OVERVIEW OF AVAILABLE METHODS Four general approaches to forecasting airport activity are dis- cussed here. From a statistical point of view these methods range from very simple to very sophisticated; however, it is important to keep in mind that the use of sophisticated statis- tical methods does not always result in better forecasts. As mentioned earlier, most real-world airport forecasts do not use the most sophisticated methods. Nevertheless, the techniques described in this chapter represent current best thinking on how to produce accurate forecasts. The related topics of eval- uating forecasts and assessing the uncertainty associated with forecasts are discussed in chapter four. The four major forecasting methods considered here are: • Market share forecasting • Econometric modeling • Time series modeling • Simulation modeling This list is not exhaustive, but it covers most of the forecast- ing techniques that have been used by airport sponsors or managers in the United States. For an overview of other fore- casting methodologies, see “The Air Transport Industry Since 11 September 2001” (2006). Market share forecasting is a simple top-down approach to forecasting where current activity at an airport is calculated as a share of some other more aggregate external measure for which a forecast has already been produced (typically some regional, state, or national measure of aviation activity). Then an assumption is made about the airport’s projected share of activity in the future. Econometric modeling is a multistep process referring to an approach that posits a causal relationship between a dependent variable (the metric to be forecast) and a set of independent explanatory variables. The explanatory variables are likely to be among those described in chapter two that are thought to influence the demand for or supply of air travel. An equation relating the dependent and independent explanatory variables to each other first is estimated using statistical techniques; the equation then can be tested for a variety of statistical proper- ties and accuracy. Finally, the estimated equation is used to forecast future values of the dependent variable. In principle at least this approach may be more powerful (and potentially more accurate) than simple market share fore- casting because it takes into account factors thought to directly cause changes in the activity metric being forecast, rather than relying only on more aggregate forecasts of other activity mea- sures. However, the data requirements are also far greater. With this approach, one must obtain historical data for both the dependent and explanatory variables to statistically estimate the relationship. Additionally, to then use the estimated rela- tionship to make forecasts of the independent variable, one must have access to forecasts of the independent variables. Time series analysis is a third approach that essentially involves extrapolation of existing historical activity data with- out utilizing independent explanatory variables. A variety of different statistical techniques can be used in time series analy- sis, including simple trend projection, moving average, expo- nential smoothing, and Box–Jenkins analysis. More sophis- ticated “multivariate” time series techniques have also been developed that can incorporate explanatory variables in a re- stricted way. In addition to these three methods, the use of simulation for forecasting is also relevant for airport planners and decision makers. However, simulation methods serve a very different purpose than the other techniques. Simulation models can be used when one needs to obtain high-fidelity estimates of the particular itinerary that passengers or aircraft may take across an airline network and the associated delays they may face, or how passengers may travel through a particular airport termi- nal, or how aircraft may traverse over an airport tarmac and runways. These models impose precise rules that govern how passengers or aircraft are routed, and then aggregate the results so that planners can assess the infrastructure needs of the net- work or airport to be able to handle the estimated traffic. The typical metrics that are the outputs from the other fore- casting methods discussed here (e.g., total enplanements and operations) are used as inputs to simulation models. “Fore- casts” from simulation models represent snapshots of how a given amount of traffic flows across a network or through an airport, rather than a time series of monthly or annual projec- tions of total traffic. The discussions in this chapter on appro- priate selection of methods and data collection center only on forecasts over time, not the snapshot forecasts that arise from simulation models. CHAPTER THREE AIRPORT ACTIVITY FORECASTING METHODS

19 MARKET SHARE FORECASTING Market share analysis involves measuring current activity at an airport as a share of some other aggregate measure (typi- cally at the regional, state, or national level), and then assum- ing that the share will remain constant (or perhaps change in some prespecified way) so that airport activity will grow along with the projected growth in the aggregate activity. A number of variations on this technique also have been used; for exam- ple, using current indicators of activity to assign a predeter- mined share of an aggregate TAF growth rate. For the market share method to produce reasonable pre- dictions, it is important that the presumed relationship between airport activity and the larger aggregate measure to which it is tied be relatively constant. Often that relationship may change over time; because of this, some studies only look at the very recent past and use a small number of his- torical data points to establish the numerical constant that ties the relationship. In general, such an approach is not likely to result in accurate forecasts, because the actual relationship may well change again over the forecast horizon. This is a shortcoming of the market share method that is typically not directly addressed. Its extensive use in aviation forecasting is typically justified as a reasonable way of making forecasts in cases where there are data limitations (e.g., lack of accuracy in operation counts at nontowered airports), where past his- tory does not correlate well with other observable factors, or where other more sophisticated methods may not yield sta- tistically reliable results. ECONOMETRIC MODELING Causal econometric modeling with explanatory variables is sometimes referred to as regression analysis. This involves sta- tistical estimation of a regression equation that posits a causal relationship between a dependent variable and a set of inde- pendent explanatory variables. For example, the demand for air travel (measured, say, in terms of enplanements) at a par- ticular airport may be posited to be a function of some of the airline market factors described in chapter two. In its simplest form, the nature of the relationship between the dependent variable and the independent variables is assumed to be linear. With just a single independent variable, this would be written as: where: Y is the dependent variable, X is the independent variable, α is the constant term of the equation, β is the coefficient describing how a change in X affects Y, and ε is a random error term (with a mean value of zero). Y X= + +α β ε After collecting historical data on Y and X, this classical linear regression (CLR) model can be estimated statistically, resulting in estimates of the coefficients α and β representing a regression line through the observed data. This estimated equation then provides a way to make forecasts of Y based on observed or assumed values of X. One way to assess the accu- racy of the estimated equation is to do an “in-sample” fore- cast by comparing the observed values of Y with the estimated values one would get by substituting in the observed X values and then computing Y from the equation. In addition to in- sample forecasts, one can make “out-of-sample” forecasts, estimates of future values of Y (beyond the historical data) that can be estimated by substituting in projected future val- ues of the independent variable X. For example, if Y is annual enplanements and X is population, then future forecasts of population could be substituted into the estimated equation to yield forecasts of future enplanements. Sidebar on Choice Analysis Rather than directly estimating activity at a specific airport, another option often used in multi-airport regions is to esti- mate overall air transportation demand for the region and then distribute that demand among the various airports based on certain characteristics of the population and the airports. Both parts of such an analysis can still be estimated with econo- metric techniques (Maddala 1983). An example of this type of approach is presented in Ishii et al. (2006), although the authors do not use the model directly for forecasting purposes. First, they measure the impact of airport and airline supply characteristics on air travel choices for both business and leisure passengers departing from one of three airports in the San Francisco area (Oak- land International, San Francisco International, or San Jose International), and arriving at one of four airports in the Los Angeles area [Los Angeles International, Ontario Inter- national, Orange County, or Burbank (Bob Hope)]. The primary explanatory variables in the model include flight fre- quency, ground access time, airport delays, and fares. The principal findings indicate that changes in ground access times affect travel choices more than changes in travel delays, and that airport preference differs between leisure and busi- ness travelers. In some cases, air transportation in a region may be cast as part of a more general model that involves other modes of transportation. The design of such a model would look some- thing like the following: • Trip generation—A model is used to estimate how many trips to and from the region are generated over some defined time period. • Trip distribution—Once a decision has been made to tra- vel, a trip distribution model is used to describe how travelers choose among various available destinations.

• Mode choice—Once a decision has been made about where to travel, a shares analysis is used to estimate the percentage of trips going by each available travel mode (e.g., air, automobile, or train). • Traffic assignment—Once a decision has been made about which mode to use, another shares analysis is used to estimate the percentage of air trips departing and arriv- ing at each airport in the region. An example of this type of model is the Regional Airport Demand Allocation Model (RADAM) produced by the South- ern California Association of Governments. This model gen- erates air passenger and cargo demand estimates in geographic zones within Southern California (it bypasses the mode choice analysis by focusing directly on air travel demand) and then allocates the demand to airports in the region based on airport characteristics. Further discussion of this model is given in Transportation Research Circular E-C040 (2002). Assumptions and Potential Problems of Econometric Models There are a number of important assumptions built into the linear regression model that may affect whether the esti- mates it produces are statistically reliable. For example, it is assumed that the dependent variable can be calculated as a linear function of a specific set of independent variables. There are several ways that this assumption could be wrong, including a changing or nonlinear relationship between the dependent and independent variables or using the wrong set of explanatory variables. These so-called “specification errors” refer to the form of the estimating equation possibly having been specified incorrectly. Although there are a number of statistical techniques that can be used to assess whether a regression model is likely to be misspecified, it is often difficult to determine exactly the nature of the specification error and therefore to find the “correct” specification. There are a number of other statistical issues that may affect the reliability of the estimates from an econometric regression. Any standard econometrics textbook will discuss how to deal with such issues [see, for example, Stock and Watson (2006)]. A good supplementary book that describes many techniques in a less formal way is A Guide to Econo- metrics (Kennedy 2003). Econometric Model Validation Once the model specification has been established and the model’s coefficients have been estimated with the appropri- ate statistical techniques, there are other steps that can be undertaken to assess the adequacy of the results before mak- ing any forecasts. First, it is important to establish how well the model fits the data using summary statistics such as R2, which in the CLR model measures how much of the propor- 20 tion of the variation in the dependent variable is explained by the independent variables. One should also assess whether the estimated coefficients on the individual independent variables are reasonable. The signs of the estimates should correspond with the researcher’s prior expectation (e.g., a rise in real income should lead to a rise in air travel demand; therefore, the coefficient on income should be positive), and the magnitude of the effect should correspond with expectations based on the analyst’s judgment and/or previous analyses. Assessing the precision of the coef- ficient estimates is also important. The t-statistic is used for this purpose in the CLR model. Generally speaking, t-statistics greater than about 2.0 are considered “significant.” Although econometric modeling is potentially a very sound and powerful method, there are many ways in which the spe- cific model that is estimated can go wrong. As should be clear from this discussion, there are a large number of potential sta- tistical assumptions and data issues that should be checked in an econometric analysis. In some cases, the model may be found not to pass these statistical tests or data problems, and then it is not always obvious how best to proceed. TIME SERIES MODELING Time series modeling is a conceptually simple approach to forecasting that involves some form of extrapolating existing data out into the future. In its simplest form it is based only on values of the variable being forecast and projects the future based on current or past trends. Because one does not need to collect data on other variables, it can be a low-cost method compared with econometric modeling. In addition, it can often meet or even beat other more sophisticated approaches in terms of forecast accuracy over the short run simply be- cause activity measures often exhibit a strong short-run trend component, whereas estimated relationships with other vari- ables may not hold so tightly in the short run. As will be dis- cussed here, although the approach is conceptually simple, specific statistical techniques that can be employed to make it more accurate are quite sophisticated. This method also can be useful when there are unusual conditions that make the relationship between local activity and other external factors unstable. Time series analysis is more likely to be accurate when a long series of historical data are available, when no large changes in airport use or activity are expected, and when fore- casting over a relatively short time period in the future. These factors make the technique appropriate when forecasting for short-term operational planning needs and/or annual budget- ing; it is less useful for longer-term forecasts that are used to assess future infrastructure and capacity needs. In addition, because time series analysis ignores external factors, it can- not be used to compare alternative policies (e.g., how would activity at a congested airport change if a new runway were

21 built) or to examine alternative environments (e.g., how would counts change if the local economy were to grow at a faster rate than expected). In many cases, time series modeling can be as simple as making year-over-year or month-over-month trend projec- tions based on past values. This is often sufficient when pro- jecting for short-term budgeting or planning needs. In some cases, however, there may not be a consistent trend pattern of growth over time (e.g., at small general aviation airports). In this case, simply using the average of the observed historical data may be a reasonable alternate way to forecast future activity. One can measure the average change in percentage terms or in levels, and simply extrapolate this change to future time periods. As noted by Armstrong (2001a), it will often make sense to weight the most recent data more heavily when making short-term forecasts. An effective way to do this is through “exponential smoothing,” which results in forecasts that are a weighted average of past values with the weights declining geometrically. (This is in contrast to the so-called “moving average” technique, where groups of past observations are each assigned equal weights.) In exponential smoothing, there are one or more smoothing parameters that can either be assumed or estimated through statistical procedures. Both trend and seasonality issues can be addressed by using expo- nential smoothing methods. An analysis by Grubb and Mason (2001) used an expo- nential smoothing method to project long-term aggregate pas- senger demand forecasts for the United Kingdom. Monthly data on passenger movements from 1949 to 1998 were assem- bled, providing nearly 600 time series observations. Both trend and seasonality were clearly apparent in the data. After the initial parameters were estimated, forecasts were generated out to 2015. The projections for 2015 were significantly higher than other forecasts that had been prepared by the United Kingdom Department of the Environment, Transport, and the Regions (2000); therefore, the authors then modified their model until the resulting projections were much closer to the agency estimates. The analysis shows how analysts may make ad hoc adjustments to statistical models to be more consistent with expert judgment or other existing forecasts. Box–Jenkins Analysis In the field of economics, the initial motivation for using time series techniques grew out of the concern that econometric modeling with explanatory variables ignored a fundamental property of time series data—namely, that they tend to grow over time and so do not have a fixed “stationary” mean value. In other words, traditional econometric modeling is typically based on some economic theory about how changes in certain variables may cause changes in other variables (e.g., a rise in income should lead to a rise in the demand for air travel), but the theory typically will not give much guidance on dynam- ics; that is, effects over time, which may be the single most important influence on the variable of interest. The Box–Jenkins approach to time series analysis attempts to directly address this issue of time effects, and can also address issues of seasonality. The resulting general model is called an ARIMA (Autoregressive Integrated Moving Aver- age) model. It is important to understand the very different nature of ARIMA time series modeling as compared with tra- ditional econometric modeling with explanatory variables. Box–Jenkins modeling is essentially a sophisticated extra- polation method; it does not provide any information on spe- cific factors that may explain why airport activity measures go up or down. Rather, it is a completely data-driven tech- nique that exploits and uncovers time dependencies in the data. One drawback is that it typically requires quite a long data series to generate reasonable estimates. However, it can often outperform modeling with explanatory variables in terms of prediction and forecasting of the activity variable of interest. If one is only interested in forecasting accuracy (espe- cially over short time horizons) it may be appropriate to con- sider time series techniques. However, if one wants to consider how alternative policies or economic environments may be ex- pected to affect observed airport activity measures, pure time series techniques cannot help. A study by Pitfield (1993) compared an ARIMA model of air passenger demand in the United Kingdom with a conven- tional regression model with explanatory variables. Weekly data on passenger travel on two airlines operating on specific air routes in the United Kingdom was gathered over a five- year period. Although out-of-sample forecasts were not made, the results indicated that the ARIMA model provided better in-sample predictions of the observed passenger counts. SIMULATION MODELING As mentioned earlier, simulation methods serve a very dif- ferent purpose than the other techniques. Simulation models can be used when one needs to obtain high-fidelity estimates of the particular itinerary that passengers or aircraft may take across an airline network, how passengers may travel through a particular airport terminal, or how aircraft may traverse over an airport tarmac and runways. These models impose precise rules that govern how passengers or aircraft are routed and then aggregate the results so that planners can assess the infra- structure needs of the network or airport to be able to handle the estimated traffic. It is difficult to generalize about these models because each is typically designed for a very specific purpose. One common thread is that the standard metrics that are the outputs from the other forecasting methods discussed here (e.g., total enplane- ments and operations) are typically used as inputs to simula- tion models. “Forecasts” from simulation models represent

snapshots of how a given amount of traffic flows across a net- work or through an airport, rather than a time series of monthly or annual projections of total traffic. Unlike econometric or time series methods, simulation models do not follow any standard framework that can be used as a guide; each model is essentially built from the ground up using its own rules and methods for distributing and forecasting activity. Such models trace the movement of individual aircraft at airports and in the national airspace route system. The primary inputs to the models include air- line schedules and fleets, route structures, runway configura- tions, separation rules and control procedures, aircraft per- formance characteristics, and weather conditions. Typical outputs from these models include measures of aircraft movements over time, passenger travel time, and fuel con- sumption. By running multiple simulations, it is possible to investigate how and to what extent a particular capacity expansion project (e.g., adding a runway) may be able to accommodate additional aviation activity. Other simulation models have been developed to project queuing and service at landside facilities such as security checkpoints and termi- nal curbsides. SELECTION OF APPROPRIATE METHOD The selection of an appropriate forecasting method may depend on both technical and budgetary factors. From a tech- nical standpoint, two primary drivers in determining which may be the most appropriate method are the purpose for which a forecast is being made and how the metrics being forecast 22 relate to available historical data. Table 2 presents basic rec- ommendations relating forecast methods to these factors. Although the table can be used to help forecasters determine the most suitable method, it should not be interpreted as a com- plete reference tool that applies to every situation. Each case will be different, and forecasters should consider other factors specific to their own situation that may be important in deter- mining the best approach. As noted in the table, for short-term operational planning or budgeting purposes, a simple time series trend analysis can be a low-cost method of obtaining reasonable forecasts if one is confident that future changes in activity are likely to be simi- lar to the historical record. If there are concerns regarding sea- sonal, daily, or hourly peaking effects, more sophisticated time series techniques could be considered. For medium- or long- term forecasts that will be used to assess landside or airside capacity needs, or for financial planning purposes related to already-planned capacity expansions, market share forecasting or econometric modeling would be more appropriate. In practical terms, budgetary constraints may also affect the particular method selected. Small airports may have very small budgets that limit their efforts to a basic review of already existing forecasts and development of a few derivative fore- cast elements. Beyond these considerations, there are a large number of criteria that planners could consider in selecting an appropri- ate forecast method. Yokum and Armstrong (1995) reported findings from a collection of surveys that indicated that TABLE 2 RECOMMENDED FORECASTING METHODS Historical Data Availability Stable Relationship With: Purpose of Activity Forecast Stable Trend External Forecasts Causal Variables Short-Term Operational Planning; Annual Budgeting Time series trend extrapolation, or smoothing/Box–Jenkins if complex time dependencies Market share forecasting Econometric modeling Identify Long-Term Capacity Needs; Financial Planning to Support Facility Expansion Market share forecasting or econometric modeling Market share forecasting Econometric modeling Examine Alternative Environments; Compare Alternative Policies Econometric modeling Obtain High-Fidelity Estimates of Travel Time and Delays (aircraft or passengers) Simulation modeling INCREASING DATA REQUIREMENTS

23 many different criteria may be important in the selection process including accuracy, timeliness in providing fore- casts, costs, ease of interpretation, flexibility, ease in using available data, ease of use, reliability of confidence inter- vals, and ability to compare alternative policies or examine alternative environments. Once a decision about the general type of forecasting method is made, statistical criteria may be useful in selecting the particular model. For example, there are a variety of “goodness-of-fit” measures that are often used to summarize how well a particular method fits the data. However, the use of formal statistical criteria to help choose the appropriate method has limitations. As noted by Armstrong (2001b), an overreliance on methods that are statistically significant can lead one to overlook other criteria; statistical significance is not the same as practical significance. It may well be more impor- tant to implement a forecasting method that is understandable by the intended audience. This is generally understood by air- port planners and their experts who develop aviation forecasts; it is relatively unusual to see forecasts that rely on highly com- plex statistical techniques, although such approaches are more common in academic studies of aviation demand. Although this discussion has focused on selecting the most preferred method, in some cases it will not be clear which method is the most appropriate. In such a situation, it may be advisable to implement and evaluate multiple methods to reveal the likely range of activity levels as assumptions and inputs are changed. SELECTED EXAMPLES OF FORECASTING METHODS IN PRACTICE This section reviews a small, but representative sample of air- port master plans and regional system plans, focusing on the typical methods, data sources, and variables used to produce airport aviation activity forecasts. For confidentiality pur- poses, the discussion does not identify individual facilities. These examples are meant to show how different forecasting methods may be appropriate under different conditions. Other forecasting studies from professional and academic sources are referenced in other sections of this report. Small Towered Airport with Commercial Service This small airport provides commercial service primarily in support of a nearby, large higher education campus. A sepa- rate general aviation terminal and FBO caters to private busi- ness and recreational users. The commercial terminal cur- rently houses four commuter carriers that provide connecting service to many destinations through their parent carriers’ hubs. An airport-wide Master Plan Update was completed in 2003, and a Terminal Area Master Plan (TAMP), focusing only on commercial service was completed in 2005; these plans were subsequently approved by FAA. The forecasting process in the TAMP analysis involved a varied mix of data sources and methods. Three different methodologies were considered, including trend extrapola- tion, market share projections, and econometric modeling, to predict future scheduled passenger enplanements. As noted in the TAMP study, the use of trend projections can be signifi- cantly influenced by abrupt changes in available service. Using 2003 as the baseline year, the trend projections esti- mated a large one-year increase in 2004 as a result of the introduction of new service by one of the commuter carriers, and then used the 10-year average growth rate at the airport from 1994 to 2003 as the basis for extrapolating enplanements out to 2023, the end of the forecast horizon. This resulted in an average annual growth rate of 3.1% over the entire forecast period. The market share projection analysis found that the air- port’s share of national enplanements grew over the 1994– 2003 period; this growth relative to national activity was assumed to continue for the first five years of the projection period and then to remain at a constant share of national enplanements for the rest of the forecast horizon. The FAA Aerospace Forecasts were used as the basis for projecting future enplanement activity. This resulted in an average annual growth rate of 3.5% over the entire forecast period. The econometric modeling analysis considered three differ- ent local measures—population, employment, and income— as potential drivers of enplanement activity at the airport. Each indicator was regressed separately against historical enplane- ments, and it was found that income had the best fit to the data; therefore, future enplanements were projected using the estimated income equation along with forecasted local income levels from a third-party provider. This resulted in an average annual growth rate of 2.6% over the entire forecast period. All of these forecasts were then compared with the prior projections from the 2003 Master Plan Update. In addition, a passenger demand analysis from the 2003 study was reviewed to assess the effect on enplanements from local passenger diversion to a competing airport, the volume of traffic travel- ing to specific destinations, and the potential for service improvements. This analysis supported the use of the TAMP forecasts. Additionally, a high-growth scenario was developed that assumed significant roadway improvements in the surround- ing region, as well as loss of commercial air service at certain neighboring airports. Under this scenario, average annual growth was estimated at 4.0% over the forecast period. The various forecasts were also compared with FAA’s TAF, which predicted a 3.5% annual growth rate for the air- port. Based on this comparison and the other factors consid- ered, the projections from the market share method (also showing a 3.5% growth rate) were identified as the recom-

mended enplanement forecasts for long-term planning at the airport. The enplanement forecasts were then also used as the basis for projecting air carrier operations and fleet mix by supplementing them with historical and expected trends in load factors, types of aircraft, and average seats per depar- ture. Peak activity at the airport was also estimated from the baseline enplanement forecasts by using the peak month- average day approach described earlier. Peak-hour forecasts were derived based on an analysis of airline schedules and carrier-specific load factors provided by the scheduled car- riers at the airport. As is the practice with most airport master plans, forecast uncertainty was addressed only in an indirect way by includ- ing the high-growth scenario projections that relied on more optimistic assumptions about future enplanement growth. General Aviation Reliever Airport This general aviation facility serves as a reliever airport to a large commercial airport approximately 15 miles away. Although a draft master plan was completed in 2002, recent developments have necessitated important updates to that plan. In particular, several manufacturers are developing VLJ aircraft, advanced technology twin-jet aircraft weighing less than 10,000 lb that may be certified for single-pilot operations (although air-taxi carriers operating under Part 135 may be required to use two pilots). Some forecasters predict that sev- eral thousand VLJs will enter service over the next ten years, and a leading VLJ manufacturer has already announced plans to locate a manufacturing, testing, training, and maintenance center at this general aviation facility. This has a potentially large effect on operational activity at the airport; as a conse- quence, an update to the airport’s master plan scheduled for 2006 was prepared that provides forecasts of based aircraft, fleet mix, local and itinerant operations, peak activity, opera- tional mix, and instrument approaches. Primary focus was placed on projecting based aircraft, and the remaining activ- ity measures were then projected (with some adjustments for the projected VLJ facility) by means of a relationship with the based aircraft estimates. According to the draft report, historical data on based air- craft at the airport were available only for selected years; therefore, this ruled out forecasting methods such as econo- metric modeling and time series analysis that rely heavily on complete and accurate historical information for the metric being forecast. Instead, a market share analysis of the number of registered general aviation aircraft in the local area was undertaken; available annual data on this metric was com- pared against changes in both local population and United States active general aviation aircraft registrations. Although both relationships exhibited a relatively constant share over the historical period examined (1993–2005), the relationship with 24 national aircraft registrations was selected as the preferred planning tool to forecast local aircraft ownership out to 2025. This resulted in an estimated average annual growth rate of approximately 1.3% over the entire forecast time horizon. Historical based aircraft levels at the airport (for the years available) were then related to local aircraft ownership. Given this relationship, low and high forecasts for based aircraft were developed assuming, respectively, a constant and increasing share of projected local aircraft ownership counts. Further analysis then was undertaken by comparing these forecasts with the high and low forecasts from the 2002 master plan, as well as independent based aircraft projections shown in FAA’s TAF and the department of transportation state airport system plan for the state where the airport is located. In all, seven different based aircraft forecasts were considered and a preferred forecast was developed from the range of based aircraft counts, including a final upward adjustment to account for the effects of the planned VLJ facility. The esti- mated average annual growth rate of based aircraft for the preferred forecast was approximately 3.7% over the entire fore- cast period. As mentioned earlier, once the based aircraft forecast was completed, it provided the primary basis on which to project other measures. The draft report did not contain substantive discussion or numerical estimates of forecast uncertainty; however, such considerations may be forthcoming as the analy- sis is refined and updated in the future. Regional Airport System Plan The agency responsible for aviation systems planning in a large eastern metropolitan area prepared a system plan that was completed in 2001. As a result of 9-11 and other events an update was prepared in 2005, which was revised further in 2006. The update plan identifies capacity needs out to 2030 for 30 public and private aviation facilities throughout the region, including three commercial airports; of these 30 facil- ities, 14 are eligible for federal subsidies, whereas the others must rely on state or private investments. The initial system plan focused primarily on consideration of general aviation activity at the facilities; conservative esti- mates of commercial activity (operations and enplanements) were taken directly from the commercial airports’ own mas- ter plans. General aviation activity for the region as a whole was tied to regional population and employment forecasts, in which the agency had a long history of being able to make reasonable projections. These forecasts were then broken out to individual facilities based on local variations in the popu- lation and employment projections. For the 2005 update, it was found that although increasing numbers of aircraft were based at the region’s general avia- tion and reliever airports, operations per general aviation air- craft had declined since 2001, owing primarily to increased

25 costs associated with general aviation flying and more restric- tive flying rules imposed after 9-11. The 2005 forecast for general aviation operations used the same growth rate devel- oped in the 2001 plan, but applied it to a reduced base of oper- ations for 2005. Additional growth, however, was projected for general aviation jet operations owing to the anticipated intro- duction of VLJs and increases in corporate operations at some of the smaller suburban airports as the primary commercial air- port in the region becomes more congested. This upsurge in demand at suburban facilities would be further induced by sig- nificant projected increases in population and employment away from the primary city center of the metropolitan area.