Page 1 of 1

Program code to cluster air masses

Posted: April 23rd, 2021, 11:58 am
by imanberriban
Hello everyone,

I have a question if there is someone has the answer.

I tried to program a code to cluster air masses for a period of 2 months, following the method used by hysplit but I can't find the same results. after, I tried to make an approximation to minimize the computation time, when I calculated the Euclidean distance between the trajectories, I tried not to go through all the endpoints of the trajectories, but I took a step of 6 endpoints and by chance I find the desired results.

The problem now is when I test this code over a period of 2 years with trajectories of different length, I discover that this "step" is no longer valid. I want to know is what I have to do to get the same results.

Thank you.

Re: Program code to cluster air masses

Posted: October 2nd, 2021, 7:13 am
by nouaytiabdel
Hi,
I have this problem too. I'm trying to code a program that can statistically analyze the air masses generated by Hysplit. And among the techniques I work on, it's clustering. I specifically chose the clustering based on TSV (total spatial variance) as Hysplit does.

But when I compare the clustering results obtained by using my code with the Hysplit results. I found that they are similar if I use a small number of trajectories and become different if I use a large number of trajectories.And when I found this difference . I compared the TSV values calculated by my code and by Hysplit for the same number of clusters. I noticed that my code produces the minimum (better) value. This means that Hysplit does some approximation that I don't. I want to know if that is true with some explanation please.

Thank you in advance.

Re: Program code to cluster air masses

Posted: October 5th, 2021, 9:01 am
by alicec
Can you give some more specific information.

What are you using for the inputs to the clustering utility?
hours to cluster, time interval, trajectory skip
https://www.ready.noaa.gov/documents/Tu ... _clus.html