Changes between Version 3 and Version 4 of ticket/0663_MLF
 Timestamp:
 20100422T18:48:52+02:00 (11 years ago)
Legend:
 Unmodified
 Added
 Removed
 Modified

ticket/0663_MLF
v3 v4 13 13 === Description === 14 14 15 Introduce the modified LeapFrog RobertAsselin filter to fit the one presented in Leclair and Madec Ocean Modelling (2009).15 Introduce the modified LeapFrog  RobertAsselin filter (LFRA) to fit the one presented in Leclair and Madec Ocean Modelling (2009). The scheme allows a exact conservation of the heat and salt contents when using variable volume thickness (key_vvl). The nonconservation of the standard LFRA due to the time diffusion of the forcing through the filter. In the modified scheme the exact conservation is achieved by removing the forcing from the filter. Furthermore, the scheme allows to use a much smaller Asselin filter parameter (10^4 instead of 10^1) decreasing the damping effect of the LFRA scheme, so that it is actually a quasi second order scheme. 16 16 17 This update is part of the LOCEAN.8 task (Update of the ocean physics) due by June 25th 17 The LFRA flow chart of ''step'' is the one introduce in the trunk after the v3.1, to restore the key_vvl option. Its main characteristic is that the ''ssh'' is timestepped first, followed immediately by the computation of the ''now'' vertical velocity (see figure below). [[BR]] 18 19 [[Image(NEW_LF_RA_FlowCart.jpg, 50%)]][[BR]] 20 Step routine flow chart, form Leclair and Madec (OM 2009).[[BR]] 21 22 23 24 This update is part of the LOCEAN.8 task (Update of the ocean physics) due by June 25th 2010. 18 25 19 26 The main remaining modifications concern : 20 27 21 (1) forcing terms as a mean between time step n1/2 and n+1/2 (sbcmod) 28 (1) forcing terms as a mean between time step n1/2 and n+1/2 (sbcmod.F90, sbc.F90, trasbc.F90, dynzdf....F90) 29 30 [[Image(NEW_LF_RA.jpg, 50%)]][[BR]] 31 Illustration of forcing integration methods. Old formulation (top) and new formulation (bottom). 22 32 23 (2) Asselin filter: remove the forcing from the filter (dynnxt, tranxt, traqsr) 33 At each time a forcing term appears in the 3 time evolution equations solved in NEMO/OPA (ssh, dynamics and tracer). The mean of kt+1/2 and kt+1/2 has to be used in order to remove the largest source of divergence of two consecutive time step. Therefore, the surface module (including the seaice!) must now provide both kt+1/2 and kt1/2 forcing fields, and all the module involving a forcing term have to be modified. 34 Note that TKE equation is solved with a two level time stepping scheme : no need of a before field for its forcing term (i.e. taum) 35 36 sbc_oce.F90 : add "before" forcing fields (denoted by a '_b' sufix) for all ocean forcing components (utau_b, vtau_b, qns_b, qsr_b, emp_b, emps_b) 37 sbcmod.F90 : add the swap of surface fields and the read/write in the restart file. 38 Beginning of sbs routine, introduce the swap, but only for kt >= nit000+1. If no restart, the '_b' fields is known only at the end of the sbc calculation. Therefore we have choosen to regroup the setting of '_b' fields at nit000 at the end of sbc routine. CAUTION : here the swap must be done at each kt, what ever the nn_fsbc value is. Otherwise the ocean forcing will be wrong. 39 40 (2) Asselin filter: remove the forcing from the filter (dynnxt (?), tranxt, traqsr) 24 41 25 42 (3) tranxt.F90 : semiimplicit hydrostatic pressure gradient : change the coefficients to take into account the Asselin filter parameter 26 43 27 time mean T and S for the semi implicit hpg computation is now rbcp*ta+(2rbcp)*tn+rbcp*tb 28 with rbcp=(1+atfp)*(1+atfp*atfp) / 4, and atfp is the Asselin filter parameter. It used to be (ta+2*tn + tb) / 4. 44 time mean T and S for the semi implicit hpg computation was (ta+2*tn + tb) / 4 . It is now replaced by 29 45 30 The effect of the change is to further extend the stability limit of the time stepping scheme when ln_dynhpg_imp = T 46 rbcp*ta+(2rbcp)*tn+rbcp*tb 47 48 with rbcp=(1+atfp)*(1+atfp*atfp) / 4, and atfp is the Asselin filter parameter. 49 50 The change extend further the stability limit of the time stepping scheme when ln_dynhpg_imp = T 31 51 32 52 33 === Current missing points ===53 === Open questions === 34 54 35 ( 1) and (2)55 (2) Do we need to remove the stress component from the filter in dynnxt? 36 56 37 (3) open questionin vvl case : thickness weighted tracer in the time mean, do we need to compute and use a time mean thickness in the hpg computation?57 (3) in vvl case : thickness weighted tracer in the time mean, do we need to compute and use a time mean thickness in the hpg computation? 38 58 39 59 