Why are FlightAware data points snapped to an underlying lat/long grid?


#1

First time post for me. I volunteer for a non-profit, providing them with mapping and data analysis. The client has asked me to analyze some FlightAware data. They provided me with a .csv file containing flights over northern Alaska for 2014 and 2015. This is my first experience with FA data, so I want to familiarize myself with it before conducting any analysis.

The data contain about 3 million records. Each record represents a point location containing latitude and longitude fields, along with a variety of descriptive fields (flight id, aircraft type, altitude, timestamp, etc).

When I plot a single flight from the data, those points are displayed in a line stretching from the flight’s beginning to end. This is as expected.

However, I notice that about 98% of the points align perfectly with a 1x1 arc-minute (about 1.0 nm N-S by 0.4 nm E-W spacing) lat/long grid. Thus, those flight lines micro-zigzag from one lat/long intersection to another. Subsequent flight lines passing nearby will also be snapped to the same intersection; thus, one intersection may contain multiple points, each from a different flight.

I’m a commercial pilot, and I know that flights do not follow such patterns. I’m assuming, therefore, that someone (FA?) has moved (“snapped”) the actual reported locations to the nearest lat/long intersection.

But wait! The remaining 2% of the points are not snapped to a lat/long grid; they follow a normal flight route, as one would expect. Here’s a screenshot showing the snapped (blue) and non-snapped (red) points:

Additional observations of the non-snapped (red) points:
a) they only occur in 2015
b) they are clustered around Deadhorse and Kuparuk
c) on average they involve aircraft that are lower, slower and climbing; what I would consider to be in a departure configuration. This pattern holds even when comparing the same aircraft type between snapped and non-snapped.

I am confused why only a small minority of the flight points follow a normal routing, while the vast majority are snapped to a lat/long grid. If it is possible to represent the flight points in their correct locations, why have so many been snapped???

The data contain a field named “Estimated position”, which I thought might be relevant. However, all records contain the value NULL, so this does not appear to be a snapping factor.

I’ve searched this forum, along with the FA website and a general internet search for an answer with no success. I welcome your advice.

Also, a tip of the hat to anyone who can direct me to FA metadata, which may contain my answer.


#2

There are a few different possible datasources here, but I’d expect that for Alaska in 2014-2015 it’ll be primarily FAA-sourced radar data. The snapping may be a limitation of that data. For ADS-B sourced positions, the resolution is much finer (on the order of meters, not miles); that may be your red dots.

If you have an example of data showing the snapping I can check the underlying datasource.


#3

Thanks for the quick reply. I have selected the points from three flights for your review. Each flight’s data is in a separate .csv file. One file contains a flight where all points are snapped to a lat/long grid. The second file contains the reverse; all points are not snapped. The third file is a flight with mixed snapping (denoted by an additional field “snapped”). The three files are zipped, available for download at https://www.dropbox.com/s/2ydoecq7w1igblu/snap_latlong.zip?dl=0

I look forward to your assistance!


#4

Thanks for the data.

In “snapped_to_latlon.csv” all the datapoints are from Anchorage Center - i.e. FAA radar data.
In “not_snapped_to_latlon.csv” all the (non-estimated) datapoints are from ADS-B data.
In the mixed case, all the snapped datapoints are from Anchorage Center, and the other non-snapped datapoints are estimated positions. Estimated positions are essentially a computed great circle based on the expected route.

The datasource is shown the “facility_name” / “facility_description” columns.

Looking at current data I see much the same pattern for radar data from Canadian & Alaskan radar sites - positions are only to arcminute-level precision.

So it looks like this is just a limitation of the radar datasources in these areas - we’re only receiving arcminute-level precision in the positions they provide. Where available, we prefer ADS-B data over radar, but to get ADS-B data requires both that we have ADS-B receivers installed in the area, and that the aircraft itself is ADS-B equipped.


#5

Thank you! Your response has been of great assistance.