I have a geopandas dataframe of a bunch of linestrings that have some data associated with each vertex/point:

Point_x = (Lat, Lon, Time, ID, Data1, Data2, Data3)

The points are converted to linestrings based on ID and ordered by Time.

I want to break the linestrings where at the point where some condition is met. Right now that's when the distance between Points is greater than some value. In the future it could be where a function of the Data fields is some value. For instance, split a linestring when Speed crosses 5 kph.

The current problem is that some of the tracks are formed from points that have duplicate ID's so the linestring jumps back and forth over huge distances and I want a threshold to break these lines.

Any ideas on the correct way to structure this or libraries/methods that might be of use?

The dataframe has over 150k tracks with many points per track in it so efficiency would be nice.

Here's an example of the tracks DF:

ID         geometry                                                  
204235000  LINESTRING (37.62001 -28.99535, 37.62015 -28.9...   
205400000  LINESTRING (3.807816666666666 -18.083181666666...   
207138000  LINESTRING (22.73206 -34.97915833333333, 22.73...   
209016000  LINESTRING (8.447673333333331 -23.522783333333...     

Here's an sample from the points DF. There are 18 columns including Datetime, Point(Lon, Lat), Speed, Size etc etc:

Index           Heading   Latitude  Longitude       ID  
20              92.8 -35.946802  13.089695  210725000               
21              93.5 -35.946912  13.091808  210725000               
22              95.4 -35.965520  13.497698  210725000               
23              94.7 -35.965803  13.501898  210725000               
24              94.9 -35.965987  13.504573  210725000               

EDIT: Tried to be a little clearer.

  • What is the structure of your GeoDataFrame ? A copy of gd.head() would be welcomed.
    – gene
    Commented Dec 12, 2016 at 17:55
  • Edited to show head
    – RedM
    Commented Dec 13, 2016 at 6:44
  • I have used GeoPy (geopy.distance.vincenty) in the past for something similar. I needed to connect points, but didn't want them connected if they were longer than my determined threshold. I sent each pair of coordinates through the function and connected them only if they were less than my threshold. geopy.readthedocs.io/en/1.10.0/#geopy.distance.vincenty
    – JohnR
    Commented Feb 23, 2017 at 20:47
  • What is the primary key/sort condition for the the duplicate ID feature: time vs. ID or ID vs. time?
    – huckfinn
    Commented Sep 5, 2017 at 21:36
  • Not really sure what you mean. The points are grouped by ID then sorted by time and then those ordered positions used to create a line string. The ID is sometimes duplicated between objects. Example: There is a car in city A with ID = '123'. It is transmitting it's position and time. There is also a car in city B with ID = '123' and it is also transmitting it's positions and the times are interleaved. A line built up from these points would jump between A and B
    – RedM
    Commented Sep 6, 2017 at 7:36

1 Answer 1


I haven't used shapely/geopandas yet, so I can only provide pseudocode:

distance_threshold = 50 # Value at which distance to cut off
new_lines = [] # Array to hold the newly created, split lines
new_line_marker = 0 # Let's remember where our new line starts
for linestring in linestrings: # Iterate over all linestrings
  for i, coord in enumerate(linestring.coords[:-1]): # Iterate over all coords of the linestring
    if distance(coord, coords[i+1]) >= distance_threshold: # Check if threshold is met
      # If condition is met, we generate a new linestring,
      # starting from the last split to the current one
      new_lines[] = new LineString(coords[new_line_marker:i])
      new_line_marker = i+1 # remember to reset the marker

The distance function should be something that your libs already offer, or you'll have to implement it yourself (ol' buddy Pythagoras will help you out).

Efficiency can be improved as needed from there, but it should be a good starting point.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.