Problem with cluster analysis

Topics about the trajectory clustering program for HYSPLIT.
Post Reply
markS_MT
Posts: 5
Joined: March 29th, 2015, 1:03 pm
Registered HYSPLIT User: No

Problem with cluster analysis

Post by markS_MT » March 29th, 2015, 1:12 pm

I am performing a cluster analysis of back-trajectories from 2008 to 2013 (length 120hr at 6hr intervals) for 6 years and I keep getting the following error: "Error: can not find channel named "0"". I do not get the error when I run the cluster analysis for 2008 to 2012 or from 2009 to 2013 or for any number of years less than 6. Can anybody help?

barbara.stunder
Posts: 260
Joined: November 9th, 2012, 4:23 pm
Registered HYSPLIT User: Yes

Re: Problem with cluster analysis

Post by barbara.stunder » March 30th, 2015, 8:15 am

You may get a more meaningful error message if you run the cluster program from the Windows Command Prompt.

Make the INFILE. In the Command Prompt window (do a Windows search for "Command Prompt"), change directory (cd) to the cluster working directory. This will have the file, INFILE.

Open Windows (File) Explorer and find the file cluster.exe, which I am assuming is causing the problem (in the GUI, "Run Cluster analysis").

From the Command Prompt window, run cluster.exe. From the working directory c:\hysplit4\cluster\working, this would be ..\..\exec\cluster.exe. If the file "CCONTROL" is in the working directory, cluster.exe will read the inputs from that file. Otherwise, you will be prompted for the inputs Hours for clustering, etc.

markS_MT
Posts: 5
Joined: March 29th, 2015, 1:03 pm
Registered HYSPLIT User: No

Re: Problem with cluster analysis

Post by markS_MT » April 15th, 2015, 2:08 am

I have tried this too, the calculation starts, It says Number of trajectories in INFILE:
Pass 1 out of ....

But then it stays like this for a very long time, to the extent that I left the pc on, checked after 4 hours and nothing changed, i left the pc on overnight, and I found that it had shut down! When I re-switched, I got the message that windows had not shut down properly.

barbara.stunder
Posts: 260
Joined: November 9th, 2012, 4:23 pm
Registered HYSPLIT User: Yes

Re: Problem with cluster analysis

Post by barbara.stunder » April 15th, 2015, 7:29 am

It is probably a memory issue. Try fewer trajectories or endpoints, such as every 9 hours. With one trajectory a day it would be ~2190 trajectories. Has anyone done this many trajectories before?

markS_MT
Posts: 5
Joined: March 29th, 2015, 1:03 pm
Registered HYSPLIT User: No

Re: Problem with cluster analysis

Post by markS_MT » April 15th, 2015, 3:46 pm

In research articles yes, not using the clustering function in Hysplit though. I am using a 2008 Sony vaio with 4 GB Ram. How much memory would be enough to carry out this type of calculation?

markS_MT
Posts: 5
Joined: March 29th, 2015, 1:03 pm
Registered HYSPLIT User: No

Re: Problem with cluster analysis

Post by markS_MT » May 24th, 2015, 11:43 am

How I can extract the trajectories apportioned to each of the clusters? I need to perform a cluster analysis however I need the number of trajectories assigned to each month to, so somehow I have to extract all the data for the trajectory together with the cluster number.

Thanks

barbara.stunder
Posts: 260
Joined: November 9th, 2012, 4:23 pm
Registered HYSPLIT User: Yes

Re: Problem with cluster analysis

Post by barbara.stunder » May 26th, 2015, 7:51 am

The HYSPLIT cluster output file "CLUSLIST_N", where N is the number of clusters, lists which trajectories are in each cluster. This is a text file. Through the cluster GUI, under Step 3, "Assign trajectories to clusters", you can check "View", and then "Run" to see the file.

markS_MT
Posts: 5
Joined: March 29th, 2015, 1:03 pm
Registered HYSPLIT User: No

Re: Problem with cluster analysis

Post by markS_MT » May 26th, 2015, 11:18 am

Ok, so if I want to find out the number of trajectories in cluster N in January for example, all i have to do is convert this file to .csv and look for the required info?

barbara.stunder
Posts: 260
Joined: November 9th, 2012, 4:23 pm
Registered HYSPLIT User: Yes

Re: Problem with cluster analysis

Post by barbara.stunder » May 28th, 2015, 7:19 am

To find the number of trajectories in a cluster, you could convert the CLUSLIST_N file to .csv, and look at it, or open CLUSLIST_N in Notepad, but these don't give column headings.

In the GUI if you do Step 3, Assign trajectories to clusters, View, Run, the "Simulation Log" window opens showing the file CLUSLIST_N, with headings - cluster number, number of trajectories in a cluster, trajectory start year,month,date, hour, file number (in INFILE), and filename.

Also, if you plot the trajectories in each cluster, Step 3, "Display Clusters", one of the labels at the top of the plot gives the number of trajectories in each cluster.

Post Reply