Abstract

Long-range lidar systems have been used operationally at the Hong Kong International Airport for wind shear alerting. They are used for monitoring the headwinds over the last 3 n mi of all of the runway corridors of the Hong Kong International Airport (HKIA). This paper discusses the results of a trial of using short-range lidar (SRL) in the alerting of wind shear over a particular runway corridor by performing more frequently updated wind measurements over a specific section of this corridor in which many wind shear reports are received. The radial resolution of the lidar is 75 m and the data are updated every 20 s. Three different ways of wind shear alerting based on SRL’s data are studied, namely, the deviations of the measured radial velocities from the uniform background flow (the “velocity fluctuation”), eddy dissipation rate (EDR), and autocorrelation of radial velocity. The performance of these methods is studied by comparing with the pilot wind shear reports. The velocity fluctuation has the best skill in capturing the wind shear reports. By combining the wind shear alerts from SRL with those from the Wind Shear and Turbulence Warning System (WTWS), it is possible to achieve a probability of detection (POD) of pilot wind shear reports of about 90%, with a percentage of time on alert (PTA) of about 10% only. This even outperforms the existing overall wind shear alerting service (WTWS plus subjective wind shear warnings issued by aviation weather forecasters) by significantly reducing PTA. As such, the present study shows that it is possible to combine wind shear alerts from SRL and WTWS for automatic wind shear alerting without the need of human intervention, at least for a particular runway corridor of HKIA.

1. Introduction

The Hong Kong International Airport (HKIA) is situated in an area of complex terrain. It is surrounded by seas on three sides, whereas on the south there is a mountainous Lantau Island with peaks rising to about 1000 m above mean sea level and valleys as low as 400 m in between. The complex terrain gives rise to low-level wind shear to be encountered by the landing/departing aircraft. The majority of the low-level wind shear (about 70%) is related to the terrain disruption of the prevailing airflow, such as the east-to-southeasterly winds in a stable boundary layer in the spring and the southwest monsoon and strong cross-mountain winds associated with tropical cyclones in the summer. The next major kind of low-level wind shear in terms of the frequency of encounters is sea breeze in fine weather conditions. It could be seen that the major wind shear types occur in nonrainy situations. For their detection and alerting, Doppler lidar systems have been deployed at HKIA since 2002. Some background information about this long-range lidar is given in Table 1. There are currently two units of lidar at the airport, with each serving a specific runway.

Table 1.

Comparing the technical specifications and the cost of the long-range lidar and SRL.

Comparing the technical specifications and the cost of the long-range lidar and SRL.
Comparing the technical specifications and the cost of the long-range lidar and SRL.

There are ongoing developments at HKIA, such as the construction of office blocks, exhibition centers, hotels, and hangars. The pilots have indicated in various arenas that such man-made structures appear to bring about disruptions of the airflow under certain weather conditions, and such turbulence/wind shear is difficult to be differentiated from similar airflow disturbances arising from terrain effect. As such, there are not yet formal statistics about the frequency of building-induced turbulence/wind shear, but works have started to understand the phenomenon in more detail. For instance, Liu et al. (2010) have considered the airflow disruption over a runway as a result of a hypothetical Y-shaped building (similar to a terminal building) in cross-building airflow using a computational fluid dynamics (CFD) model. Similar studies have also been conducted with a row of rectangular blocks (which are similar in appearance to hangars) and an engineering test-run facility in strong northerly winds (Li and Chan 2012). There have also been suggestions that, as a result of the exhibition center at the northeast corner of HKIA, airflow disruptions may occur along the flight path on approach to the north runway of the airport from the east (the so-called 25RA runway corridor) in certain weather conditions, such as the prevalence of the southwest monsoon in the summer.

Apart from the above-mentioned numerical modeling effort for the better understanding of building-induced turbulence/wind shear, the possibility of detecting, and even alerting, such airflow disturbances has also been explored. In this regard, a short-range lidar (SRL) has been tested at the exhibition center at the northeast corner of HKIA for capturing the small-scale airflow disturbances possibly rising from this building. It has much higher spatial and temporal resolution in scanning over a portion of the 25RA runway corridor, compared with the existing lidar systems at the airport (long-range lidar with a maximum measurement range of 10 km).

This paper summarizes the study results of SRL during the field study in Hong Kong in the summer of 2010. The features and the scanning strategies of SRL are described. Three methods have been developed to search for signatures of small-scale airflow disturbances from the SRL’s data, namely, the so-called velocity fluctuation (defined later), the conventional eddy dissipation rate (EDR), and autocorrelation method (Scheffer et al. 2009). The performance of the three methods is studied by using relative operating characteristic (ROC) curves for balancing the hit rate of the pilot wind shear reports and the alert duration (as a percentage of time for the study period). The pilot reports serve as an independent dataset for checking the accuracy of the SRL-based turbulence/wind shear detection algorithm, and they are taken as sky truth, as in the previous development of the lidar-based wind shear alerting system (Shun and Chan 2008). This definition of the ROC diagram is different from the conventional one, in which hit rate is considered against the false alarms. However, false alarms are not addressed in the present paper because of the very small amount of null wind shear reports collected in the study period.

2. Features of SRL

The SRL in the study uses a laser beam with a wavelength of about 1.5 μm and pulse energy of 0.005 mJ. The pulse width is about 0.5 μs and the pulse repetition frequency is 4000 Hz. Some background information about SRL (also compared to that of the long-range lidar) is shown in Table 1. The SRL belongs to class 1M in terms of laser safety, and it is safe to the unaided eyes for use at an airport. The transceiver unit includes the laser diode, a modulator, and a signal amplifier. A laser beam is emitted through a scanner and is backscattered by the suspending particulates in the air. By using the Doppler principle, the SRL tracks the motion of the particulates and provides the line-of-sight (or radial) velocity of the air in the direction of the laser beam. The setup of SRL on the rooftop of the exhibition center is shown in Fig. 1.

Fig. 1.

The appearance of an SRL: the scanner as an outdoor unit, with the signal processor and laser source enclosed in the equipment box.

Fig. 1.

The appearance of an SRL: the scanner as an outdoor unit, with the signal processor and laser source enclosed in the equipment box.

In the field study, the SRL was configured to have a radial resolution of 75 m and a maximum range of 1500 m, that is, 20 range gates. The scanner allows scanning over a limited span only, namely, azimuthal angles of 40°. The elevation angle is fixed mechanically. It was taken to be 5° from the horizon so that the scanning sector of the SRL could be sufficiently close to a portion of the 25RA glide path. The scanner consists of two counterrotating wedges. With that setup, the laser beam is made to scan quasi horizontally (when the elevation angle of the scanner is taken as 0°), with a rotation speed of about 2° s−1 for the laser beam. As such, it takes about 20 s (i.e., an azimuthal span of 40° divided by a rotation speed of 2° s−1) to complete one scan in a direction, and another 20 s for scanning in the opposite direction. As such, for scanning over the particular portion of the 25RA runway corridor, wind data were updated every 20 s with a range resolution of 75 m during the field study. This is compared against the existing long-range lidar for the north runway of HKIA, which has a revisit time of about 2 min (i.e., scanning over the whole 25RA runway corridor every 2 min) and a range resolution of about 100 m. As such, the SRL could provide wind data over a portion of 25RA with higher spatial and temporal resolution, with the objective of capturing the possibly building-induced turbulence/wind shear, which is much smaller in spatial scale and evolves much faster in temporal scale compared to terrain-induced turbulence/wind shear.

The purpose of the SRL is to scan a particular part of the glide path over which turbulence/wind shear often occurs. Its application is different from that of the long-range lidar, which provides the wind data over the whole glide path. As such, from the information in Table 1, the long-range lidar would be too expensive to be deployed for the application purpose of SRL.

3. Scanning strategy of SRL in the field study

The scanning area of the SRL is given in Fig. 2a. The SRL is situated at the rooftop of a small building near the middle of the exhibition center and scans in the direction of the 25RA glide path. The scanning area is shown schematically in Fig. 2a. An overview of the same scanning area from the top is given in Fig. 2b. The azimuthal angle of the scanning area with respect to the north spans from 9° to 49°. This range of the azimuthal angle is chosen in order to see any airflow disturbances downstream of the exhibition center in south-to-southwesterly winds in the summer (when the southwest monsoon prevails over the south China coast). The intersection between the scanning area projected on the horizon and the 25RA glide path projected on the same plane extends from about 1613 m from the eastern threshold of the north runway of HKIA to a distance of about 2390 m from the same threshold. The anemometer R2E, as considered later in the paper, is also located near the eastern threshold of the north runway, at about 350 m to the west-southwest of the threshold.

Fig. 2.

Scanning strategy of SRL: (a) the scanning area of SRL overlaid on a photo showing the exhibition center, and (b) an overview of the scanning area of SRL and the 25RA runway corridor, (c) in three-dimensional perspective.

Fig. 2.

Scanning strategy of SRL: (a) the scanning area of SRL overlaid on a photo showing the exhibition center, and (b) an overview of the scanning area of SRL and the 25RA runway corridor, (c) in three-dimensional perspective.

A 3D picture of the scanning area of the SRL is shown in Fig. 2c. To avoid scanning the landing aircraft, and with the limitation of scanning with a fixed elevation angle only, the laser beam of SRL was configured to scan below a 3° glide path of the aircraft, with a height difference of 7.8 m at the eastern end and 21.7 m at the western end of the scanning area. As such, the laser beam from the SRL was sufficiently close to the glide path of the landing aircraft in order to measure the radial velocity of the winds along this path.

In the present scan strategy of SRL, the laser beam is basically transverse to the glide path. It aims at observing the airflow disturbances in the southerly flow, which is nearly perpendicular to the runway orientation. On the other hand, airflow disturbances in the direction of the runway are expected to be covered by the existing lidar system for the north runway. In the latter part of the present study, it could be seen that the wind shear alerts from the existing lidar system would be combined with the alerts from SRL in order to capture the pilot wind shear reports. It is expected that both instruments are used together to cover the various manifestations of turbulence.

4. An example of the airflow disturbance observed by SRL

With the frequently updated radial velocity data from SRL, some new features of airflow disturbances are captured, which have never been observed before by the long-range lidar systems. An example is shown in Fig. 3 between about 0803 and 0804 local time (LT) (Hong Kong time = UTC + 8 h) 21 September 2010. In that morning, moderate to fresh southerly winds prevailed in the region of the airport. From the SRL’s radial velocity data at 0803:06 LT on that day, there appeared to be an area of strong winds (colored in red in Fig. 3a) at the southwest corner of the scanning area. In the next minute or so (Figs. 3b–d), this area of stronger winds propagated to the north and crossed the glide path of the landing aircraft. There are three yellow lines in Fig. 3; the center one is the extended centerline of the north runway of HKIA, and the other two are located at about 50 m to both sides of the extended centerline to take into account the uncertainty with the flight path of the aircraft.

Fig. 3.

Radial velocity imageries of an SRL for the wind shear case from 0803:06 to 0804:06 LT 21 Sep 2010; the figures are separated by 20 s. For the three yellow lines: the middle one is the nominal flight path of 25RA, and the two other lines represent uncertainty with the aircraft location, each having a distance of 50 m from the yellow line in the center.

Fig. 3.

Radial velocity imageries of an SRL for the wind shear case from 0803:06 to 0804:06 LT 21 Sep 2010; the figures are separated by 20 s. For the three yellow lines: the middle one is the nominal flight path of 25RA, and the two other lines represent uncertainty with the aircraft location, each having a distance of 50 m from the yellow line in the center.

It could be seen from this case that the airflow disturbances could pass across the glide path of the landing aircraft within about 1 min. Such disturbances are captured successfully with the rapidly updated (viz., every 20 s) radial velocity data from the SRL. This shows the advantage of using a dedicated lidar over a particular portion of the glide path, which may be more vulnerable to the occurrence of airflow disturbances as a result of, for instance, man-made structures. On the other hand, each long-range lidar system at HKIA is deployed to measure the winds over all the four arrival and departure runway corridors of a particular runway. It also spends time in rotating between the eastern and western ends of that runway. As a result, the revisit time at a particular runway corridor is on the order of a couple of minutes. With that lower data update rate, the airflow disturbances that show up in the SRL’s data in the present case may not be captured successfully by the long-range lidar.

With the above success of capturing rapidly evolving airflow disturbances near 25RA glide path, the SRL data are processed by a number of novel algorithms in an attempt to issue automatic turbulence/wind shear alerts for the pilots. Three methods have been considered in the present paper, namely, velocity fluctuation, EDR, and autocorrelation method. They are introduced in the following sections.

5. Velocity fluctuation

The basic idea of velocity fluctuation is to find out the deviation of the measured radial velocity of the SRL from the “background” flow. The more disturbed the airflow would be as a result of the man-made structures and/or natural terrain at the airport area, the larger the deviation from the background flow could be expected. To apply this approach to the SRL’s data, two practical issues have to be resolved, namely, (i) the removal of noisy data from the SRL’s measured radial velocity, and (ii) the method for establishing the background wind velocity.

Noise may appear at SRL’s measurements from time to time because of the relatively weak output power of this instrument, reflection from clutter (e.g., birds), laser beam attenuation in rain, etc. It is particularly apparent at longer ranges of the SRL. As seen from the example in Fig. 3, some potential noise could be identified in the far-range data of the SRL, appearing as dots of blue and green among the strong southerly flow in red. Similar to the data quality control (QC) algorithm for the long-range lidar system (Shun and Chan 2008), the noise is identified by comparing each radial measurement of the SRL with those of the four points around it, namely, the two at the adjacent range gates and the other two at the adjacent azimuthal angles. A statistical distribution of the difference in radial velocity is given in Fig. 4a. This is based on all of the measurement radials of all the scans on a typical day in the study period. It could be seen that the majority of the difference lies within ±3 m s−1. As such, a data QC threshold of 3 m s−1 is adopted. If the difference is larger than 3 m s−1, the data point under consideration would be removed. This 3 m s−1 is fixed and does not change with time. In future studies, it may be necessary to establish the QC thresholds for different seasons with different aerosol loading in the atmosphere.

Fig. 4.

Frequency distributions of the (a) radial velocity difference between the adjacent range gates of the SRL and (b) exponent of the power law.

Fig. 4.

Frequency distributions of the (a) radial velocity difference between the adjacent range gates of the SRL and (b) exponent of the power law.

Concerning the background flow, as a first step the anemometer data of R2E is used. This anemometer has good exposure in all directions. In south-to-southwest directions, the terminal building of HKIA is located at least 700 m away. Wind data are available from R2E every second. The anemometer is located at about 10 m above the ground level, which is in turn about 5.2 m above mean sea level. Because the SRL is located at about 25 m above ground level and the laser beam is located at about 106–133 m above mean sea level, the anemometer’s data have to be properly scaled up to represent the “background” flow at the location of the laser beam below the glide path. As a first attempt, the following power law is adopted for the background wind velocity V:

 
formula

where h is the altitude of the scanning plane of the SRL and V0 is the wind speed measured at R2E, located near a flat, open coast with a height of 15.2 m above the mean sea level.

The exponent 0.16 is considered to represent the surface condition in the measurement area, which is an open sea outside the airport and generally sparsely populated buildings around it within a few kilometers from the exhibition center and the eastern end of the north runway. As a demonstration, the frequency distribution of the exponent of the power law is shown in Fig. 4b based on the wind data of a Doppler sodar located at about 1 km upstream (i.e., south) of the exhibition center for measuring the background wind. The wind data of 57 days in June–September 2010 are considered. The sodar has a measurement range from 5 to 200 m above ground. To obtain a sufficient amount of wind data for curve fitting, the altitude range of 25–100 m above ground is considered. The wind speed at 25 m is at least 2 m s−1 (i.e., not considering the light wind cases). The wind direction at 25 m should be within 150°–210° from north. The wind profiles fulfilling these requirements are fitted with power-law curves, and the exponents of the power-law curves are extracted. As seen in Fig. 4b the peak of the frequency distribution occurs at the exponent between 0.1 and 0.15. In fact, considering the whole distribution, the mean exponent is 0.176 and the median is 0.141. As such, the choice of 0.16 is justified with the field measurements. In future studies, the variation of the exponent, such as seasonal variation, would need to be considered.

The wind data of each radial of the SRL are available every couple of seconds. They are compared with the wind data from the R2E anemometer at the nearest second. The anemometer’s measured wind velocity is resolved along the measurement radial of the SRL at that particular second, denoted as . The SRL’s radial velocity data are denoted as . The absolute difference between them is calculated. Only the absolute difference for those SRL’s data points along the part of the glide path covered by the SRL’s scanning area is considered, and such data points have the index k. The velocity fluctuation υr is defined as the maximum value of such absolute differences for all points k along the 25RA glide path

 
formula

In summary, the velocity fluctuation is the maximum deviation of the measured radial velocity by the SRL from the “background” wind obtained by scaling up the anemometer reading at the height of the laser beam. For more turbulent flow, the deviation from the background wind is expected to be larger, as is the velocity fluctuation.

As an example of the performance of velocity fluctuation, the strong southerly winds on 28 June 2010 are considered. On that day, southerly winds prevailed in the region of HKIA. A band of north–south-oriented thunderstorms moved from west to east, passing over the northeastern part of the airport at which the SRL was located. The time series of velocity fluctuation on that day is given in Fig. 5. Before and during the passage of the band of thunderstorm over the northeastern part of the airport, south-to-southwesterly winds picked up and a number of pilot wind shear reports were received. In the period in which significant wind shear was reported over 25RA, the velocity fluctuations reached rather large values. If a threshold is adopted for the velocity fluctuation for issuing turbulence/wind shear alerts, say, 6.5 m s−1 (horizontal red line in Fig. 5; its optimization would be discussed later in this section), these pilot wind shear reports could be captured successfully.

Fig. 5.

Time series of velocity fluctuation (m s−1) for the wind shear case on 28 Jun 2010. Individual velocity fluctuation values (blue dots), the 10-min running average of the velocity fluctuation values (green curve), a tentative threshold for velocity fluctuation (red horizontal line), the velocity fluctuation data points that trigger alerts (yellow dots), and the times of pilot wind shear reports (red crosses) are shown.

Fig. 5.

Time series of velocity fluctuation (m s−1) for the wind shear case on 28 Jun 2010. Individual velocity fluctuation values (blue dots), the 10-min running average of the velocity fluctuation values (green curve), a tentative threshold for velocity fluctuation (red horizontal line), the velocity fluctuation data points that trigger alerts (yellow dots), and the times of pilot wind shear reports (red crosses) are shown.

The performance of SRL and the long-range lidar in this particular case is shown in Fig. 6. Although it was raining at the time of wind shear, reasonable data could still be obtained with the SRL to cover a range of about 800 m from this instrument (Fig. 6a), which is sufficient to cover the portion of the 25RA glide path under consideration. On the other hand, the laser beam from the long-range lidar was attenuated in rain, so that there was basically no signal over the part of 25RA glide path under consideration (Fig. 6b). This illustrates another advantage of using SRL over the conventional long-range lidar for focusing on wind measurements, and thus turbulence/wind shear alerting over a specific portion of the glide path.

Fig. 6.

(a) The radial velocity imagery of an SRL at 1059:23 LT 28 Jun 2010 and (b) the corresponding imagery of the long-range lidar at the south runway at about the same time. The scanning area of the SRL is also shown in (b) (white triangle).

Fig. 6.

(a) The radial velocity imagery of an SRL at 1059:23 LT 28 Jun 2010 and (b) the corresponding imagery of the long-range lidar at the south runway at about the same time. The scanning area of the SRL is also shown in (b) (white triangle).

The Hong Kong Observatory (HKO) has operated the Wind Shear and Turbulence Warning System (WTWS) for issuing automatic wind shear alerts for the pilots. WTWS integrates the wind shear alerts issued by a number of automatic algorithms, including the wind shear alerts from the lidar systems [with the algorithm called GLYGA; see Shun and Chan (2008) for details], Terminal Doppler Weather Radar (TDWR), anemometers along the runway and weather buoys, as well as anemometers at the mountain top of Lantau Island. In the study period of 18 June–25 October 2010 with the deployment of SRL at HKIA, there were 97 pilot wind shear reports. The performance of WTWS and velocity fluctuation based on SRL’s data in capturing these reports is summarized in Table 2. A threshold of 5.8 m s−1 is adopted for velocity fluctuation. Both systems capture 56 reports. On top of that, velocity fluctuation algorithm captures 12 more wind shear reports. By using the wind shear alerts of both systems (i.e., a wind shear alert is issued if WTWS and/or SRL’s velocity fluctuation algorithm gives an alert), about 91% of the wind shear reports could be captured. This is to be compared with the use of the existing WTWS only, which has a hit rate of (56 + 20)/97 = 78% only, and is not good enough for operational wind shear alerting purposes.

Table 2.

Performance of SRL alone, WTWS alone, and both SRL and WTWS in the alerting of low-level wind shear at 25RA in the study period. Note that, in the table, three methods of using the SRL’s data are considered, namely, velocity fluctuation (based on the SRL’s radial velocity data and surface anemometer data; see section 5 of the main text), EDR (based on SRL’s radial velocity data; see section 6), and autocorrelation (considering the autocorrelation of the time series of radial velocity data; see section 7).

Performance of SRL alone, WTWS alone, and both SRL and WTWS in the alerting of low-level wind shear at 25RA in the study period. Note that, in the table, three methods of using the SRL’s data are considered, namely, velocity fluctuation (based on the SRL’s radial velocity data and surface anemometer data; see section 5 of the main text), EDR (based on SRL’s radial velocity data; see section 6), and autocorrelation (considering the autocorrelation of the time series of radial velocity data; see section 7).
Performance of SRL alone, WTWS alone, and both SRL and WTWS in the alerting of low-level wind shear at 25RA in the study period. Note that, in the table, three methods of using the SRL’s data are considered, namely, velocity fluctuation (based on the SRL’s radial velocity data and surface anemometer data; see section 5 of the main text), EDR (based on SRL’s radial velocity data; see section 6), and autocorrelation (considering the autocorrelation of the time series of radial velocity data; see section 7).

The threshold of velocity fluctuation for wind shear alerting is determined by considering the ROC curve. Because there is just a small amount of null wind shear reports from the pilots in the study period over 25RA (about 10 only), the ROC curve is expressed in terms of the probability of detection (POD) versus percentage of time on alert (PTA). This could be slightly different from the conventional meaning of the ROC diagram. POD is determined by the fraction of pilot wind shear reports that are covered by the wind shear alerts. The percentage of time on alert is calculated by the fraction of time over the whole study period (when the SRL was working) in which wind shear alerts are issued. Different situations are considered, namely, combining alerts from WTWS and velocity fluctuation, and combining alerts from GLYGA and velocity fluctuation. The performance of such alerts is compared with the overall wind shear alerting service at present, namely, WTWS alerts plus subjective wind shear warnings issued by the aviation weather forecasters. The ROC curve is formed by varying the threshold of the velocity fluctuation. It is used to obtain the right balance between POD and PTA. The best-performing threshold is determined if the corresponding data point is closest to the upper-left corner of the ROC diagram. The results are summarized in Fig. 7. The following are a number of observations from the ROC diagram:

  • In general, the use of WTWS plus velocity fluctuation has better performance than the use of GLYGA plus the velocity fluctuation.

  • Both WTWS plus the velocity fluctuation and GLYGA plus the velocity fluctuation outperform the overall wind shear alerting service, namely, WTWS plus the forecaster.

  • The WTWS plus the velocity fluctuation performs the best (i.e., closest to the upper-left corner of the ROC diagram) with a threshold of 5.8 m s−1 for the latter. The performance could reach POD of more than 90% and PTA of less than 10%. As such, WTWS plus the velocity fluctuation has the potential for use in operational wind shear alerting without the need of forecaster-provided wind shear warnings.

Please note that false alarms are not considered in the present study because of the small amount of null reports (about 20 compared with 97 reports of significant wind shear in the study period over the 25RA glide path). To collect more null reports for studying the performance of SRL, an intensive observation period may be arranged in future studies so that all aircraft (whether or not it encounters wind shear on arrival) over 25RA would be asked to provide wind shear report. If such data are collected, the false alarm study would be reported in a future paper.

Fig. 7.

ROC curves of various alerting methods based on SRL. Velocity fluctuations (red), with threshold values (from right to left) from 3 to 9 m s−1 with a step of 1 m s−1, except for the optimum value of 5.8 m s−1; EDR1/3s (green) with thresholds (from right to left) from 0.15 to 0.42 m2/3 s−1, with a step of 0.03 m2/3 s−1; and autocorrelations (blue) with threshold from 0.35 to 0.9, with a step of 0.05 are shown. The combination of EDR, velocity fluctuation, and WTWS (purple), with fixed EDR1/3 values (0.3 or 0.36 m2/3 s−1) and a threshold of velocity fluctuations from 3 to 12 m s−1, with steps of 1 m s−1 (from right to left) are also indicated.

Fig. 7.

ROC curves of various alerting methods based on SRL. Velocity fluctuations (red), with threshold values (from right to left) from 3 to 9 m s−1 with a step of 1 m s−1, except for the optimum value of 5.8 m s−1; EDR1/3s (green) with thresholds (from right to left) from 0.15 to 0.42 m2/3 s−1, with a step of 0.03 m2/3 s−1; and autocorrelations (blue) with threshold from 0.35 to 0.9, with a step of 0.05 are shown. The combination of EDR, velocity fluctuation, and WTWS (purple), with fixed EDR1/3 values (0.3 or 0.36 m2/3 s−1) and a threshold of velocity fluctuations from 3 to 12 m s−1, with steps of 1 m s−1 (from right to left) are also indicated.

6. EDR

The cube root of EDR is the internationally adopted metric for turbulence intensity (ICAO 2010). It is thus natural to calculate lidar-based EDR and study its performance in capturing the turbulence/wind shear reports from the pilots.

The calculation method of EDR using SRL’s data is similar to that adopted for glide path scans of the long-range lidar systems (Chan 2010). Only a summary of the major steps of the calculation is discussed here. First of all, the sectorial scanning area of the SRL is divided into a number of small sectors, called subsubsectors. The schematic diagram is given in Fig. 8. Each subsubsector is taken to have an azimuthal scan of 20° and a range of radial gates of 10. The subsubsector is incremented with each azimuthal gate and each radial gate. Becuase the whole scanning area of the SRL has about 40 azimuthal gates and 20 radial gates, there are altogether 121 subsubsectors in the whole area.

Fig. 8.

Schematic diagram of the subsubsector for EDR1/3 calculation: the measurement sector of the SRL (red), subsector (blue), and subsubsector (light blue). The labels 75 m along the radial and 2° in azimuth are the respective resolutions (radial resolution and azimuthal angle resolution).

Fig. 8.

Schematic diagram of the subsubsector for EDR1/3 calculation: the measurement sector of the SRL (red), subsector (blue), and subsubsector (light blue). The labels 75 m along the radial and 2° in azimuth are the respective resolutions (radial resolution and azimuthal angle resolution).

For each subsubsector at a particular scan k, the radial velocity “surface” (as a function of range R and azimuth angle θ) is fitted with a plane using the singular value decomposition method. This is essentially the removal of the linear trend. The velocity fluctuation at each point in the space (R, θ) is taken to be the difference between the measured radial velocity and the fitted velocity on the plane,

 
formula

The longitudinal structure function is calculated, as given by

 
formula

where the summation is made over all the possible azimuthal angles and scans over 5 min (about 15 scans, with the scan at 25RA runway corridor updated every 20 s), and N refers to total number of entries in the summation. The error term E is calculated using the covariance method on the radial velocity difference (e.g., Frehlich 2001).

EDR1/3 is determined by fitting the longitudinal structure function with the theoretical von Kármán model (Frehlich et al. 2006) to give σ2 the variance of the radial velocity, and L0 the outer scale of turbulence. EDR (also denoted as ɛ) is given by

 
formula

The accuracy of the EDR thus calculated from the SRL’s data is established by comparing with the EDR data collected on board an instrumented, fixed-wing aircraft, which is the result of collaborative effort between HKO and Government Flying Service (GFS) in Hong Kong. This aircraft is equipped with a special meteorological measuring system, called the Aircraft-Integrated Meteorological Measurement System 20 (AIMMS-20), for collecting wind data with an accuracy of about 0.5 m s−1 and a frequency of 20 Hz. Technical details of this system could be found in Chan et al. (2011). A piece of postprocessing software has been developed through research collaboration between HKO and National Aerospace Laboratory (NLR) in the Netherlands for processing the vertical velocity data of AIMMS-20 to obtain EDR, following the approach similar to that for commercial jets (Haverdings and Chan 2010).

The comparison between SRL’s EDR and GFS aircraft’s EDR could be found in Fig. 9. The GFS aircraft data were collected over many routine flights (3 times a week) in different kinds of weather conditions in the study period. In such flights, the aircraft followed the 3° glide path of the 25RA runway corridor. Figure 9a shows the histogram of the difference between the EDR values from the two platforms. The majority of the difference lies within 0.1 m2/3 s−1. The scatterplot of the two sets of data is shown in Fig. 9b. They have high correlation, with the correlation coefficient squared reaching 0.88. The slope is close to 1. The outliners of the data points (e.g., those in the upper part of the plot) are data collected in rain, in which the SRL data may become rather noisy. Apart from that, the EDR obtained from the SRL’s data is considered to have reasonable quality for trials in capturing the pilot wind shear reports. It is noted that the error in the order of 0.1 m2/3 s−1 could be about 25% of the EDR1/3 value. This would seem to introduce quite a bit of uncertainty into the wind shear detection scheme.

Fig. 9.

(a) A histogram of the frequency distribution of the difference between the SLR EDR1/3 and the EDR1/3 calculated from GFS aircraft data. (b) The corresponding scatterplot of EDR1/3 values is shown.

Fig. 9.

(a) A histogram of the frequency distribution of the difference between the SLR EDR1/3 and the EDR1/3 calculated from GFS aircraft data. (b) The corresponding scatterplot of EDR1/3 values is shown.

Figure 10 shows an example of EDR calculated from SRL’s data. The case under consideration is the prevalence of southerly flow in the airport area in the morning of 21 September 2011. As seen in Fig. 10a, at far ranges of the SRL (say, beyond about 1.2 km from the SRL), the radial velocity data are quite noisy, and a data QC procedure is considered essential in order to obtain “clean” data for the calculation of EDR map, following the method described in section 5. The corresponding distribution of EDR is shown in Fig. 10b. With the structure function approach, the EDR1/3 is calculated as a spatially averaged quantity. The EDR1/3 map is obtained by repeating the calculation of structure function over a subsubsector incremented by a range gate or by an azimuthal gate at each time. As such, the EDR1/3 map appears to have the same spatial resolution as the raw radial velocity data but, in fact, each EDR1/3 value is an average over the subsubsector space. Note in Fig. 10b that two streaks of higher EDR1/3 values appear to propagate away from the building and affect the glide path of the aircraft. The distribution of EDR seems to be reasonable, and this again shows that the quality of SRL’s radial velocity in general and the application of the data QC algorithm in particular are both reasonable for the calculation of turbulence intensity.

Fig. 10.

(a) An example of the raw radial velocity imagery of the SRL (0748:11 LT 21 Sep 2010), with (b) the corresponding EDR1/3 map.

Fig. 10.

(a) An example of the raw radial velocity imagery of the SRL (0748:11 LT 21 Sep 2010), with (b) the corresponding EDR1/3 map.

The time series of EDR1/3 on that day is shown in Fig. 11. The pilot reports are given at the bottom of the graph as a number of crosses. The pilot reports occur at times when the EDR1/3 value is relatively high, in excess of about 0.3 m2/3 s−1, although the false alarm may also be quite high. From similar case studies, the EDR obtained from the SRL could be another candidate in capturing the pilot reports if a suitable threshold is chosen to balance between the hit rate and alert duration (or false alarm).

Fig. 11.

Time series of the maximum EDR1/3 from the SRL along the flight path 25RA on 21 Sep 2010. The pilot wind shear (ws) reports are given at the bottom of the plot (crosses).

Fig. 11.

Time series of the maximum EDR1/3 from the SRL along the flight path 25RA on 21 Sep 2010. The pilot wind shear (ws) reports are given at the bottom of the plot (crosses).

By optimization using ROC curve, an EDR1/3 threshold of about 0.3 m2/3 s−1 may be adopted for capturing the pilot reports. The result of wind shear alerting is summarized in Fig. 7 and Table 2. The EDR captures fewer pilot wind shear reports than velocity fluctuation. One possible reason is that EDR tends to capture small-scale velocity fluctuation and thus may not be effective in alerting the larger-scale shear line, such as sea-breeze fronts, which occurred in the study period. On the other hand, velocity fluctuation could also provide alerts for this kind of shear line. Nonetheless, if both WTWS and EDR alerts are used, the POD would reach 89.7% for the 97 pilot reports under consideration in the study period. This POD is close to 90% and also considered to be rather good.

The ROC curves for EDR are again given in Fig. 7. Two separate curves are considered, namely, WTWS plus EDR and GLYGA plus EDR. Their performance is not as good as that for the corresponding combination of WTWS–GLYGA with velocity fluctuation. Another possibility is to combine the three alert sources together, namely, WTWS, velocity fluctuation, and EDR. The threshold of EDR1/3 is also varied, and two such thresholds are considered in Fig. 7, namely, 0.3 and 0.36 m2/3 s−1. Such combinations do not make the ROC curves move further to the upper-left corner of the ROC diagram. It is not unexpected because both velocity fluctuation and EDR are calculated based on the same dataset, namely, the radial velocity measurements from the SRL. Without additional data source, the performance of the various combinations (WTWS + velocity fluctuation, WTWS + EDR, or WTWS + velocity fluctuation + EDR) would be very similar.

7. Autocorrelation

Autocorrelation has been demonstrated to be a candidate of the early warning signal on phenomenon arising from many complex systems (Scheffer et al. 2009). When the system is approaching a phase transition, there is an increase in autocorrelation resulting from the “critical slowing down” phenomenon. Hence, autocorrelation exceeding a certain threshold may be a useful indicator for alerting wind shear. In addition to velocity fluctuation and EDR, the capability of autocorrelation on wind shear alerting is investigated in the present study.

To analyze the dataset, a time series of radial velocity measured by SRL is formed by picking the maximum radial velocity along the glide path from each sector scan (updated every 20 s). The time series is detrended by subtracting a Gaussian kernel smoothing function (the Nadaraya–Watson kernel regression estimate) from the time series. Lag-1 autocorrelations within a sliding window are calculated from the detrended time series.

In the calculation of autocorrelation, there are two empirical parameters, namely, the bandwidth of the data selected for applying the detrending, and the window size of the data selected for calculating the lag-1 autocorrelation. Different settings of these two parameters have been tried out for maximizing the performance of the autocorrelation in capturing significant wind shear in the ROC diagram. After experimentation, a bandwidth of 100 and window size of 10 scans (around 200 s) yields the best performance, and the corresponding results are presented herein.

An example of the time series of autocorrelation could be found in Fig. 12. Compared to velocity fluctuation and EDR, the autocorrelation value varies with time rather rapidly. The ROC curves of autocorrelation could be found in Fig. 7. Table 2 shows the results in capturing the pilot wind shear reports. In general, autocorrelation does not perform well compared to velocity fluctuation, but its performance is basically on par with that of the EDR.

Fig. 12.

Time series of autocorrelation from the SRL along the flight path 25RA on 28 Aug 2010. The pilot wind shear (ws) reports are given at the bottom of the plot (crosses).

Fig. 12.

Time series of autocorrelation from the SRL along the flight path 25RA on 28 Aug 2010. The pilot wind shear (ws) reports are given at the bottom of the plot (crosses).

Among the three methods considered in the present paper, velocity fluctuation has the best performance in terms of balancing between the hit rate and alert duration as a percentage of time. On real-time implementation, it is also relatively easier to compute, at least in comparison with EDR. Please note that the above results are obtained based on the field study of one summer only. It is planned to conduct more field studies of the SRL in the capturing of significant wind shear over 25RA runway corridor.

8. Conclusions

This paper summarizes the results of a field study of using SRL for the alerting of low-level wind shear over the runway corridor 25RA at HKIA. The SRL provides velocity data of high spatial and temporal resolutions over a particular section of 25RA where many pilot wind shear reports are received. From the radial velocity imageries of the SRL, there are many wind shear features that are not observable in the measurements of the existing instrumentation, including the long-range lidar. For instance, there could be an area of higher wind speed that passes through the 25RA runway corridor in about 1 min, which would be missed by the radial velocity data of the existing long-range lidar with data updates every 1–2 min over a particular runway corridor. The SRL is also found to provide useful data in heavy rain, because the equipment is sufficiently close to the section of the runway corridor of interest, which could not be achieved with the existing long-range lidar.

An algorithm is then developed for automatically capturing wind shear features based on the radial velocity data of the SRL. Three different methods have been tried out, namely, velocity fluctuation, EDR, and autocorrelation. It turns out that, though the method is rather simple and straightforward, the use of velocity fluctuation could be very effective in capturing additional pilot wind shear reports on top of the existing alerts from WTWS. In fact, using pilot wind shear reports as sky truth data, velocity fluctuation has the best skill by striking a good balance between POD and PTA. By combining wind shear alerts from SRL and WTWS, it is possible to achieve a POD of 90% with a PTA of 10% only. This even outperforms the existing overall wind shear alerting service, which combines WTWS alerts with subjective wind shear warnings issued by the aviation weather forecasters. The results of the present study suggest that there is potential for combining SRL with WTWS for fully automatic wind shear alerting for 25RA without human intervention, which would be the eventual goal of wind shear alerting for the airport.

Please note that the present results are based on a field study of one summer only. The SRL should be tried out in more years, and the multiyear results should be presented in papers in the future. Moreover, the idea of using an SRL should be tried out for other runway corridors of HKIA, and in other airports, by strategically positioning the SRL in order to capture some undetected wind shear features by the existing meteorological instrumentation. The combination of several instruments should also be tried out, namely, on top of the existing WTWS, also considering other remote sensing meteorological instruments, such as a microwave radiometer (Chan and Lee 2011) and Doppler acoustic radar.

The SRL is tried out in the present study in summertime when the southwest monsoon prevails over southern China. The maritime airstream in this season has relatively lower aerosol loading, but the SRL could still reach the desirable measurement range for the purpose of monitoring the airflow along the glide path. The range performance of SRL may be studied in other seasons in future studies, although in these seasons it would be much less often to use 25RA glide path.

The velocity fluctuation method is found to work best in the study period, namely, in summer season without many tropical cyclones. The other methods may work better in other weather conditions, for example, EDR may have better performance in strong cross-mountain winds in tropical cyclone situations, when the terrain-disrupted airflow is very turbulent. As such, a longer period of study would be required to see if velocity fluctuation is the most robust method, or maybe at times it would need to be supplemented by other turbulence/wind shear detection quantities, such as EDR.

REFERENCES

REFERENCES
Chan
,
P. W.
,
2010
:
LIDAR-based turbulence intensity calculation using glide-path scans of the Doppler Light Detection And Ranging (LIDAR) systems at the Hong Kong International Airport and comparison with flight data and a turbulence alerting system
.
Meteor. Z.
,
19
,
549
563
.
Chan
,
P. W.
, and
Y. F.
Lee
,
2011
:
Application of a ground-based, multi-channel microwave radiometer to the alerting of low-level windshear at an airport
.
Meteor. Z.
,
20
,
423
429
.
Chan
,
P. W.
,
K. K.
Hon
, and
S.
Foster
,
2011
:
Wind data collected by a fixed-wing aircraft in the vicinity of a tropical cyclone over the south China coastal waters
.
Meteor. Z.
,
20
,
313
321
.
Frehlich
,
R. G.
,
2001
:
Estimation of velocity error for Doppler lidar measurements
.
J. Atmos. Oceanic Technol.
,
18
,
1628
1639
.
Frehlich
,
R. G.
,
Y.
Meillier
,
M. L.
Jensen
,
B.
Balsley
, and
R.
Sharman
,
2006
:
Measurements of boundary layer profiles in an urban environment
.
J. Appl. Meteor. Climatol.
,
45
,
821
837
.
Haverdings
,
H.
, and
P. W.
Chan
,
2010
:
Quick access recorder data analysis software for windshear and turbulence studies
.
J. Aircr.
,
47
,
1443
1447
.
ICAO
,
2010
:
Meteorological service for international air navigation
.
Annex 3 to the Convention on International Civil Aviation, International Civil Aviation Organization, 206 pp
.
Li
,
L.
, and
P. W.
Chan
,
2012
:
Numerical simulation study of the effect of buildings and complex terrain on the low-level winds experienced by landing aircraft at an airport in typhoon situation
.
Meteor. Z.
,
in press
.
Liu
,
C. H.
,
D. Y. C.
Leung
,
A. C. S.
Man
, and
P. W.
Chan
,
2010
:
Computational fluid dynamics simulation of the wind flow over an airport terminal building
.
Zhejiang Univ. Sci. A
,
11
,
389
401
.
Scheffer
,
M.
, and
Coauthors
,
2009
:
Early-warning signals for critical transitions
.
Nature
,
461
,
53
59
.
Shun
,
C. M.
, and
P. W.
Chan
,
2008
:
Applications of an infrared Doppler lidar in detection of wind shear
.
J. Atmos. Oceanic Technol.
,
25
,
637
655
.