the numbers of emitors

Topics about the HYSPLIT dispersion model.
Post Reply
poprostudaria
Posts: 19
Joined: May 26th, 2015, 7:30 am
Registered HYSPLIT User: No

the numbers of emitors

Post by poprostudaria » June 18th, 2019, 3:17 am

Hi,
I would like to ask if it is some limitation in case of the numbers of emitors in the emitimes file? I would like to run HYSPLIT with almost 70 000 point emitors, but the simulation takes so long (sometimes a few days for only one day of simulation, whereas I need to calculate almost three months), that I am wondering if hysplit is designed to carry so much numer of emitors? Ah, probably important information - I use HYSPLIT at linux.

Maybe anyone could look at my files and tell me what I am doing wrong, that this takes so long times. Maybe should I change some settings? ;) I will be very grateful if someone could do that.

Best regards,
Daria

PS. The emitimes and script files are too large to attached it here so I placed it here: https://drive.google.com/open?id=1yqxFV ... iPdS6doy0m
Attachments
conc.txt
(1.16 KiB) Downloaded 6 times

alicec
Posts: 140
Joined: February 8th, 2016, 12:56 pm
Registered HYSPLIT User: Yes

Re: the numbers of emitors

Post by alicec » June 18th, 2019, 8:50 am

HYSPLIT run time depends on several factors including

- number of computational particles which are being simulated
- model time step (which depends on grid size. Finer grids need smaller time steps).
- system configuration

One thing you can do is to break up your run into smaller runs. Instead of doing one run with 70,000 emitters,
try doing 10 runs with 7,000 emitters. You can run them all at the same time if you have multiple processors and then
add the results together.

If you are doing a lengthy run but are only interested in a limited space domain you may want to remove particles which
go outside your domain. You can do this in two ways.
A. One way is to set KHMAX in the SETUP.CFG file so that particles are dropped from the simulation after a certain time.
https://ready.arl.noaa.gov/hysplitusersguide/S623.htm
B. The other way is to crop your meteorological data around the area that you are interested in so that once particles leave the area they are
dropped.

Other options for speeding up a run. These options may or may not be appropriate depending on your application.

- make your concentration grid more coarse
- Use two or more concentration grids (a fine one close to the source and a coarser one farther out)
- use less computational particles
- set the model time step to be larger (DELT option in SETUP.CFG).

poprostudaria
Posts: 19
Joined: May 26th, 2015, 7:30 am
Registered HYSPLIT User: No

Re: the numbers of emitors

Post by poprostudaria » June 18th, 2019, 12:02 pm

Thank you! I did not think about it earlier.
I have two more questions, maybe someone would know answers for them.
1. What is krandr parameter? I have read in user guide that if the number of particles is greater than 5000, krand should be equal to 1. I have numpar greater than 5000 but when I set krandr=1, I do not have any results files. What is the cause?
2. Will be the noticeable difference in results when I run simulation for three months separately for each day, instead of the whole 3-month run (it means ~90 emission cycles in emitimes file)? The first one is definitelly easier to run especially in case of computing power, but I wonder if the differences in results will be significant.

alicec
Posts: 140
Joined: February 8th, 2016, 12:56 pm
Registered HYSPLIT User: Yes

Re: the numbers of emitors

Post by alicec » June 18th, 2019, 5:30 pm

See the documentation below for an explanation of KRAND.
Do you mean you do not have any cdump file output when setting KRAND?
HYSPLIT will automatically use KRAND=1 if numpar > 5000 so you only have to set it if you want KRAND=2.

https://www.ready.noaa.gov/hysplitusersguide/S640.htm

KRAND (2) - if not otherwise explicitly set, krand=1 when NUMPAR is greater than 5000, which precomputes all the random numbers required for the particle turbulence calculations. When NUMPAR is less than or equal to 5000 then krand=2 and the random numbers are computed at each dispersion time step. The dynamic method is slightly more acurate than the precomputed random numbers but it requires about twice as much CPU time. With large particle release rates both methods will provide comparable results. The special case of krand=3 will result in no turbulent mixing being added to the particle position; the particle track will follow the mean trajectory. This option is primarily intended for diagnostic testing.

The results should not depend on whether you run the simulations separately or together. The results may be slightly different because probably the random numbers that are generated will not be exactly the same. However, there is no reason that the results from one long simulation would be considered any better than the results from several smaller simulations added together.

However you may want to think about how you want to split up the runs. Determine how long you need to track particles which are emitted. Will particles need to be simulated only for the first day, three days, one week, one month?

poprostudaria
Posts: 19
Joined: May 26th, 2015, 7:30 am
Registered HYSPLIT User: No

Re: the numbers of emitors

Post by poprostudaria » June 19th, 2019, 2:56 am

Thank you. I mean that when I set krand to 1 I did not receive any txt file with results (when I set krand = 2 I receive txt file for every one hour of simulation, what means 24 files for 24 h run).
According to second question - I think one day (maybe sometimes longer, but max. 3), I do not need longer particle track.

alicec
Posts: 140
Joined: February 8th, 2016, 12:56 pm
Registered HYSPLIT User: Yes

Re: the numbers of emitors

Post by alicec » June 19th, 2019, 9:19 am

The main output of a dispersion run is a binary file which we often refer to as a cdump file.
This file can be converted to ascii output as described here:
https://ready.arl.noaa.gov/hysplitusersguide/S341.htm

Post Reply