Analysis of surveillance position error for airfield detection

Overview of air transportation simulation and surveillance position error

Air transportation simulation is the process of reproducing and calculating the movement of aircraft in airspace and airports. Factors that can be measured during an actual aircraft operation, such as travel time, delay time, and fuel consumption, are generally the analysis targets. Aircraft transportation routes, including air routes, airport locations and airfield locations, are reproduced. General air transportation simulations use a modeled pop-up time (i.e. schedule time) and aircraft performance to perform quick calculations on a computer. The physical forms of the airspace and the airport, as well as the stochastic characteristics of flight trajectories that occur under conditions of uncertainty, such as aircraft performance, atmosphere and temperature, are required for a realistic and accurate air transportation simulation.

A radar installed near the airport is generally used to monitor the position and the movement of aircraft in the airspace and airports. The primary radars can monitor aircraft by analyzing the time required for radio waves transmitted from a rotating radar antenna to return after being reflected by an aircraft. Aircraft surveillance can only be performed in intervals of the order of a few seconds because of the operational limitations of current radars. The detected positions may occasionally appear to be scattered even if the target aircraft is stationary because of the variation in the relative positions of the reflected waves. The magnitude of these errors varies depending on whether the ground or the airspace is being monitored and increases with the increase in the distance between the aircraft and the radar. The automatic dependent surveillance–broadcast (ADS-B) technology is considered the next-generation of air surveillance technology because it enables an aircraft to transmit its position information to nearby aircraft or control towers (ICAO, 2004 and Ku and Baik, 2015). The position information measured by the global navigation satellite system (GNSS), such as global positioning system data, is mainly used to realize this function. The GNSS provides relatively accurate position information compared with radar technology. The precision can be further enhanced using additional correction measures. However, the ADS-B technology is completely dependent on the aircraft for the position information, and ATC facilities cannot directly manage the stage of position information generation, which results in an inconsistent reliability of the obtained information (Schuster and Ochieng, 2011; Ju et al., 2014).

Air transportation simulation and surveillance for safety analysis

A high level of safety management is required in the aviation industry because even a single accident has a great impact. Therefore, the measurement of aviation safety, including planning, operation, and post-evaluation, is an important concern in the field of air transportation. Many studies are currently investigating the use of simulation as a substitute for actual operation experiments in safety assessment, which necessitate a substantial amount of time and resources.

Appropriate surveillance functions are essential for achieving a safe and efficient aircraft operation in airspace and airports. The US Federal Aviation Administration, which developed ASDE-X, is researching on improving the surveillance accuracy to ensure a safe and efficient airport operation (Sandberg, 2012). The A-SMGCS, which is an integrated support system for safe and efficient airport management, has been used in Europe, and surveillance functions with high precision are planned under this system. However, actual operational data is required to verify the effectiveness of such systems. Simulation is an effective alternative that can overcome this drawback. Scholars at MIT have identified the following steps and conditions for realizing a fast air transportation simulation. Data on the surveillance error, which varies with the airport and equipment, are included as a condition to verify the safety level (Campbell et al., 2014).

Airfield surveillance trajectory data

ASDE is a type of radar that provides target information to the air traffic controller by detecting aircraft and vehicle movements in the airfield. ASDE has high precision, but its detection range is smaller than that of the radar that monitors the airspace around the airport. The surveillance target accuracy of ASDE varies with various factors, such as the distance between the aircraft and antenna, area where the radio wave is reflected off the aircraft, and weather conditions. Therefore, position errors can occur in ASDE even when the target object is completely stationary, such as a stationary aircraft waiting to enter a runway. These errors can cause a corresponding error in runway holding positions (i.e. the minimum separation between aircraft approaching a runway). Consequently, a violation alarm may be triggered even when the aircraft is safe and maintaining the regulated separation or an actual violation may not be registered by the system.

In this study, the ASDE trajectories of a stationary aircraft at the holding position of runway 07 at Jeju International Airport, which is an operating airport in South Korea (Figure 1), were analyzed to characterize the surveillance position errors. A total of 10,380 trajectories from 103 aircraft were analyzed.

Trajectory analysis

The lateral distribution of the aircraft position was recorded as the left and right separation distance of the trajectories based on the taxiway centerline. The trajectories located on the left were set as negative numbers (−), while those on the right were set as positive numbers (+). The longitudinal distribution was recorded as the trajectory position based on the runway holding position, with negative numbers (−) corresponding to the trajectories that did not pass through the runway holding position and positive numbers (+) corresponding to those that did. It is possible to make the lateral and longitudinal errors when pilots do not follow exactly the taxiway centerline or the holding position. But, surveillance errors already include these errors.

The trajectory lateral mean value was −12.18 m; the standard deviation was 2.9 m; and the maximum and minimum values were 7.16 m and −21.81 m, respectively. The aircraft trajectory position was biased to the left of the taxiway centerline. The longitudinal mean value was −38.15 m; the standard deviation was 10.13 m; and the maximum and minimum values were 11.84 m and −67.79 m, respectively. Most trajectory positions did not pass through the runway holding position. However, a few surveillance trajectories exceeded the runway holding position in the stop condition.

Airfield surveillance trajectory data simulation with position error

The lateral and longitudinal surveillance errors must be simultaneously applied to simulate the surveillance error of the stationary aircraft. Two-dimensional (2D) position errors can be complexly applied using multivariate normal distribution. The drawback of this method, however, is the difficulty of implementing tail values, which are present in the actual data. Therefore, principal component analysis (PCA) and kernel density estimation were used to generate the tail values (Figure 2). First, the observation data was subject to PCA, following which the angle of the principal component axis was measured. Second, the 2D x-axis and y-axis data were generated through kernel density estimation, following the data that rotated as much as the measured PCA angle. Finally, the lateral and longitudinal 2D position data were re-rotated to the original angle. MATLAB was used for the simulation and the data analysis.

Figure 3 presents the ASDE-observed data (left), the simulated position data that only follows multivariate normal distribution (center) and the simulated position data subjected to complexly applied PCA and kernel density estimation (right). The ASDE-observed data are the actual observation value, including the surveillance position error, of each stopped aircraft. A simulation that followed multivariate normal distribution based on this data and a simulation that complexly followed the normal distribution and kernel density estimation were performed.

Compared with the simulation performed using multivariate normal distribution, the ASDE observation values generated in the simulation data that complexly used the PCA and kernel density estimation were closer in form to the observed data. The simulation obtained from multivariate normal distribution with PCA and kernel density estimation was more similar to that obtained from the ASDE surveillance data than that obtained from multivariate normal distribution.

Goodness of fit test for the airport surface detection equipment-observed and simulated data

The chi-square goodness-of-fit test was conducted to verify whether the data generated by the simulation could properly implement the small tail value characteristics observed in the actual data. The absolute distance from the center point (0.0) of each data location was set as the main measurement variable. In all, 5 m of distance was set for each rank, and a tail-part distance of 24.1 m or longer was set as one rank to deduct the minimum value of the nominal scale variable for the chi-square test to be 5 or higher (Table I). The mean of 30 iterations of the simulated data was rounded to the nearest tenth.

Table I shows the chi-square test results for the ASDE-observed and simulated data. Here, the null hypothesis is as follows:

The alternative hypothesis is as follows:

The chi-square value of the simulation data obtained considering multivariate normal distribution was 774.82, which was significant (P = 0.05 and Df = 5). Therefore, this simulation data cannot be used because H₀ was rejected. In contrast, the chi-square value of the simulation data through kernel density estimation and PCA was 5.52, which was non-significant (P = 0.05 and Df = 5). Thus, H₀ holds, indicating the appropriate implementation of the ASDE observation values with the simulation data.