Opened 11 years ago

Last modified 8 years ago

#36 accepted Enhancement

Convection - don't call every time step

Reported by: pesei Owned by: pesei
Priority: minor Milestone: FLEXPART 9.2
Component: FP physics/numerics Version: flexpart 8.2
Keywords: Cc:

Description

(from minutes of FpDev Workshop 2012)

Since the scheme is computationally costly, it should be investigated whether it could be called not every time step, but less frequently.

Change History (5)

comment:1 Changed 11 years ago by pesei

In the header of convmix.f, there is a statement

c     Changes by Caroline Forster, April 2004 - February 2005:
c       convmix called every lsynctime seconds

Before, convection was only called after reading a new set of wind fields. Thus, ask Caroline why she deemed it necessary to call it now every lsynctime.

It might be an option to introduce an intermedate frequency.

comment:2 Changed 10 years ago by sthen

I don't think calling convection only every 3 (or 1) hours is appropriate. We don't store the redistribution matrix for each grid cell and we also only calculate the convection in grid cells were particles are present. In my view these are two reasons why this needs to be done more frequently.
In our COSMO version we have introduced an additional parameter that gives the time step for convection as a multiple of lsynctime. A value of 5 (every fifth time step) seems to work fine without any significant changes. However, I only tested this for a few example runs.

comment:3 Changed 10 years ago by pesei

  • Milestone set to FLEXPART 9.2
  • Owner changed from somebody to ignacio
  • Status changed from new to assigned
  • Type changed from Task to Enhancement

To be in line with other time steps (e.g. output time step), I would suggest to introduce a new time step lconvtime. Default could be equal to lsynctime. Thanks to the namelist approach, we can then let this to future testers to evaluate what is an appropriate value. I would think that such a value would be defined in minutes rather than in multiples of lsynctime, another reaseon to use a variable lconvtime.

This will - for the moment - change the ticket type from Task to Enhancement.

Ignacio, I'm tentatively assigning this to the Milestone V9.2 and to you - if ok, please accept, if not, please remove these assignments with some comment.

comment:4 Changed 8 years ago by pesei

  • Owner changed from ignacio to pesei
  • Status changed from assigned to accepted
  • Version set to flexpart 8.2

I've asked Caroline: too long ago, does not remember why. So I guess we'll have to do experiments to find out what's good. In any case, introducing lconvtime will be necessary and this should indeed go into Fp9.2, so maybe I should take that ticket.

comment:5 Changed 8 years ago by pesei

Anne has done some tests. The amount of precipitation produced in 3 h depends on how often the scheme is called. That might have been a reason for Caroline to call it more frequently. Note that in a meteorological model, a reasonable time step is needed. As we don't change the T and U profiles due to convectional tendencies, our requirements may be less strict.

Note: See TracTickets for help on using tickets.
hosted by ZAMG