Trajectory Cluster Analysis |
|||
 Previous |
 Next |
|
|
|
Trajectory cluster analysis is a means of grouping similar trajectories. The set of January, 1996, trajectories created by Run Daily in the last section will be used for illustration. On the PC, move the 31 daily trajectory endpoints files from c:\hysplit4\working to c:\hysplit4\cluster\endpts. The files from Run Daily were named tdump96010112 (tdumpYYMMDDHH) for the January 1 trajectory to tdump96013112 for January 31. Select Trajectory / Special Simulations / Run Cluster Analysis / Run Standard. Setup the program as in Step 1: Inputs (right). The Run_ID is used on plot labels. With long trajectories and a large set of trajectories, you may want to skip endpoints and trajectories, respectively, to save computational time. Clustering more than several years of daily trajectories will take a while and may even exceed the memory limits of your PC. Note that ALL files with “tdump” in their name in the endpoints folder will be clustered. |
 |
|
In Step 2, Make INFILE and Run cluster are run. INFILE is a text file listing the trajectory endpoints files. The cluster program starts with N trajectories (clusters) and keeps pairing similar clusters until all the trajectories are in one cluster. To determine the final number of clusters, you need to decide when “different” clusters are paired, and save the list of trajectories in each cluster just before that happens. A text listing, output by the Run button, suggests the possibilities and Display plot depicts them. In Step 3, enter the final number of clusters, and Text creates a file CLUSLIST_N (N is the number of clusters), which is a text file listing the trajectories in each cluster. Plots showing the cluster-means (top left) and trajectories in each cluster (1, 2, 3, 4, and 5 below) may be created. Additional final number of clusters may be specified. The Archive button moves all files to the Archive folder. The Help button contains more information. |
|  |
 |
 |
 |
 |
Previous |
Next |