## Question about clustering algorithm(projection)

Topics about the trajectory clustering program for HYSPLIT.
mc2128
Posts: 7
Joined: March 4th, 2014, 9:06 am
Registered HYSPLIT User: Yes

### Question about clustering algorithm(projection)

I was looking for the algorithm for the clustering and found out that the clustering is done after projection.
I'm just curious that why clustering process should be done after projection process. Is there any reason for that not using the great circle distance or something?
Because it would be just hypothetic situation but if trajectories are starting from equator to near polar then clustering would be different between each mercator, polar and lambert projection.

I used this make_traj.sh file to test them and trajectories from this sample file showed different clustering for different projection.

*Added, I thought file wass uploaded but it wasn't. I upload the make_traj.sh file.
make_traj.zip
make_traj
Last edited by mc2128 on July 21st, 2015, 9:55 pm, edited 1 time in total.
barbara.stunder
Posts: 451
Joined: November 9th, 2012, 4:23 pm
Registered HYSPLIT User: Yes

### Re: Question about clustering algorithm(projection)

Using a projection avoided longitude issues using the great circle distance for a set of south polar trajectories so the program was changed to a projection.

The 'auto' projection sets the projection based on the trajectory origin latitude - polar with absolute values of latitude greater than 55, mercator less than 25, otherwise, lambert. There probably can be some differences in the cluster results near the borders. How large are the differences you are finding?
mc2128
Posts: 7
Joined: March 4th, 2014, 9:06 am
Registered HYSPLIT User: Yes

### Re: Question about clustering algorithm(projection)

The difference was not that large. It was just curiosity.
If I using trajectories from the file that I uploaded on original message,
they all have 6 cluster, but as you can see in the attached file, for the lower number of cluster the clustering is little different. (polar has 3, but lambert or mercator has 4 cluster instead)

*Here I attach clustering file and image for each projection method.
all the trajectories used were same. (other specific images are on the "iamge.zip" file)
Polar
Mercator
Attachments
image.zip
other image files