Estimated stand-off distance between ADS-B equipped aircraft and obstacles

With the recent update to ASTM F3442/F3442M-20, Standard Specification for Detect and Avoid System Performance Requirements, I'm happy to release a dataset on Zenodo of the estimated stand-off distance between ADS-B equipped aircraft and obstacles. The standard states that an alternative well-clear means may be appropriate in proximity to terrain or obstacles when justified. MIT Lincoln Laboratory has released this dataset in good faith to promote open and transparent research of the low altitude airspace; I envision this research could support?defining alternative well-clear means.

For many aviation safety studies, aircraft behavior is represented using encounter models, which are statistical models of how aircraft behave during close encounters. They are used to provide a realistic representation of the range of encounter flight dynamics where an aircraft collision avoidance system would be likely to alert. These models currently and have historically have been limited to interactions between aircraft; they have not represented the specific interactions between obstacles and aircraft equipped transponders. In response, we calculated the standoff distance between obstacles and ADS-B equipped manned aircraft.

Input Data and Workflow

For robustness, this assessment considered two different datasets of manned aircraft tracks and two datasets of obstacles. For robustness, MIT LL calculated the standoff distance using two different datasets of aircraft tracks and two datasets of obstacles. This approach aligned with the?foundational research?used to support the?ASTM F3442/F3442M-20?well clear boundary of 2000 feet laterally and 250 feet AGL vertically.

The two datasets of processed tracks of ADS-B equipped aircraft curated from the OpenSky Network . It is likely that rotorcraft were underrepresented in these datasets. There were also no considerations for aircraft equipped only with Mode C or not equipped with any transponders. The first dataset was used to train the?v1.3 uncorrelated encounter models?and referred to as the “Monday” dataset. The second dataset is referred to as the “aerodrome” dataset and was used to train the v2.0 and v3.x terminal encounter model. The Monday dataset consisted of 104 Mondays across North America. The other dataset was based on observations at least 8 nautical miles within Class B, C, D aerodromes in the United States for the first 14 days of each month from January 2019 through February 2020. Prior to any processing, the datasets required 714 and 847 Gigabytes of storage. For more details on these datasets, please refer to Appendix A in "Correlated Bayesian Model of Aircraft Encounters in the Terminal Area Given a Straight Takeoff or Landing" and Section III from “Benchmarking the Processing of Aircraft Tracks with Triples Mode and Self-Scheduling.”

Two different datasets of obstacles were also considered. First was point obstacles defined by the FAA digital obstacle file (DOF) and consisted of point obstacle structures of antenna, lighthouse, meteorological tower (met), monument, sign, silo, spire (steeple), stack (chimney; industrial smokestack), transmission line tower (t-l tower), tank (water; fuel), tramway, utility pole (telephone pole, or pole of similar height, supporting wires), windmill (wind turbine), and windsock. Each obstacle was represented by a cylinder with the height reported by the DOF and a radius based on the report horizontal accuracy. We did not consider the actual width and height of the structure itself. Additionally, we only considered obstacles at least 50 feet tall and marked as verified in the DOF.

The other obstacle dataset, termed as “bridges,” was based on the identified bridges in the FAA DOF and additional information provided by the National Bridge Inventory. Due to the potential size and extent of bridges, it would not be appropriate to model them as point obstacles; however, the FAA DOF only provides a point location and no information about the size of the bridge. In response, we correlated the FAA DOF with the National Bridge Inventory, which provides information about the length of many bridges. Instead of sizing the simulated bridge based on horizontal accuracy, like with the point obstacles, the bridges were represented as circles with a radius of the longest, nearest bridge from the NBI. A circle representation was required because neither the FAA DOF or NBI provided sufficient information about orientation to represent bridges as rectangular cuboid. Similar to the point obstacles, the height of the obstacle was based on the height reported by the FAA DOF. Accordingly, the analysis using the bridge dataset should be viewed as risk averse and conservative. It is possible that a manned aircraft was hundreds of feet away from an obstacle in actuality but the estimated standoff distance could be significantly less. Additionally, all obstacles are represented with a fixed height, the potentially flat and low level entrances of the bridge are assumed to have the same height as the tall bridge towers. The following figure illustrates an example simulated bridge.

It would had been extremely computational inefficient to calculate the standoff distance for all possible track points. Instead, we define an encounter between an aircraft and obstacle as when an aircraft flying 3069 feet AGL or less comes within 3000 feet laterally of any obstacle in a 60 second time interval. If the criteria were satisfied, then for that 60 second track segment we calculate the standoff distance to all nearby obstacles. Vertical separation was based on the MSL altitude of the track and the maximum MSL height of an obstacle.

Dataset Format

For each combination of aircraft track and obstacle datasets, the results were organized seven different ways. Filtering criteria were based on aircraft type and distance away from runways. Runway data was sourced from the FAA runways of the United States, Puerto Rico, and Virgin Islands?open dataset. Aircraft type was identified as part of the?em-processing-opensky?workflow.

• All: No filter, all observations that satisfied encounter conditions
• nearRunway: Aircraft within or at 2 nautical miles of a runway
• awayRunway: Observations more than 2 nautical miles from a runway
• glider: Observations when aircraft type is a glider
• fwme: Observations when aircraft type is a fixed-wing multi-engine
• fwse: Observations when aircraft type is a fixed-wing single engine
• rotorcraft: Observations when aircraft type is a rotorcraft

Supporting Datasets