C $Header: /home/ubuntu/mnt/e9_copy/MITgcm/pkg/opps/opps_calc.F,v 1.14 2016/03/10 20:57:11 jmc Exp $ C $Name: $ #include "OPPS_OPTIONS.h" #undef OPPS_ORGCODE C-- File opps_calc.F: C-- Contents C-- o OPPS_CALC C-- o STATE1 C-- o NLOPPS C---+----1----+----2----+----3----+----4----+----5----+----6----+----7-|--+----| CBOP C !ROUTINE: OPPS_CALC C !INTERFACE: ====================================================== SUBROUTINE OPPS_CALC( U tracerEnv, I wVel, I kMax, nTracer, nTracerInuse, I I, J, bi, bj, myTime, myIter, myThid ) C !DESCRIPTION: \bv C *=====================================================================* C | SUBROUTINE OPPS_CALC | C | o Compute all OPPS fields defined in OPPS.h | C *=====================================================================* C | This subroutine is based on the routine 3dconvection.F | C | by E. Skyllingstad (?) | C | plenty of modifications to make it work: | C | - removed many unused parameters and variables | C | - turned everything (back) into 1D code | C | - pass variables, that are originally in common blocks: | C | maxDepth | C | - pass vertical velocity, set in OPPS_INTERFACE | C | - do not use convadj for now (whatever that is) | C | - changed two .LT. 0 to .LE. 0 statements (because of possible | C | division) | C | - replaced statement function state1 by call to a real function | C | - removed range check, actually moved it up to OPPS_INTERFACE | C | - avoid division by zero: if (Wd.EQ.0) dt = ...1/Wd | C | - cleaned-up debugging | C | - replaced local dz and GridThickness by global drF | C | - replaced 1/dz by 1*recip_drF | C | - replaced 9.81 with gravity (=9.81) | C | - added a lot of comments that relate code to equation in paper | C | (Paluszkiewicz+Romea, 1997, Dynamics of Atmospheres and Oceans, | C | 26, pp. 95-130) | C | - included passive tracer support. This is the main change and may | C | not improve the readability of the code because of the joint | C | treatment of active (theta, salt) and passive tracers. The array | C | tracerEnv(Nr,2+PTRACERS_num) contains | C | theta = tracerEnv(:,1), | C | salt = tracerEnv(:,2), and | C | ptracers = tracerEnv(:,3:PTRACERS_num+2). | C | All related array names have been changed accordingly, so that | C | instead of Sd(Nr) and Td(Nr) (plume salinity and temperature), we | C | have Pd(Nr,nTracer) (tracer in plume), with Sd(:) = Pd(:,2), | C | Td(:) = Pd(:,1), etc. | C | o TODO: | C | clean up the logic of the vertical loops and get rid off the | C | GOTO statements | C *=====================================================================* C \ev IMPLICIT NONE C !USES: ============================================================ #include "SIZE.h" #include "EEPARAMS.h" #include "PARAMS.h" #include "OPPS.h" #include "GRID.h" C !INPUT PARAMETERS: =================================================== C Routine arguments C bi, bj :: array indices on which to apply calculations C myTime :: Current time in simulation INTEGER I, J, bi, bj, KMax, nTracer, nTracerInUse INTEGER myThid, myIter _RL myTime _RL tracerEnv(Nr,nTracer),wVel(Nr) #ifdef ALLOW_OPPS C !FUNCTIONS: ========================================================== c EXTERNAL DIFFERENT_MULTIPLE c LOGICAL DIFFERENT_MULTIPLE _RL STATE1 C !LOCAL VARIABLES: ==================================================== C Local constants C msgBuf :: Informational/error message buffer CHARACTER*(MAX_LEN_MBUF) msgBuf INTEGER K, K2, K2m1, K2p1, ktr INTEGER ntime,nn,kmx,ic INTEGER maxDepth _RL wsqr,oldflux,newflux,entrainrate _RL pmix _RL D1, D2 _RL dz1,dz2 _RL radius,StartingFlux _RL dtts,dt C Arrays _RL Paa(Nr,nTracer) _RL wda(Nr), mda(Nr), pda(Nr,nTracer) C C Pd, Wd :: tracers, vertical velocity in plume C Md :: plume mass flux (?) C Ad :: fractional area covered by plume C Dd :: density in plume C De :: density of environment C PlumeEntrainment :: _RL Ad(Nr),Wd(Nr),Dd(Nr),Md(Nr) _RL De(Nr) _RL PlumeEntrainment(Nr) _RL Pd(Nr,nTracer) CEOP C-- Check to see if should convect now C IF ( DIFFERENT_MULTIPLE(cAdjFreq,myTime,deltaTClock) ) THEN IF ( .true. ) THEN C local initialization C Copy some arrays dtts = dTtracerLev(1) C start k-loop DO k=1,KMax-1 C initialize the plume T,S,density, and w velocity DO ktr=1,nTracerInUse Pd(k,ktr) = tracerEnv(k,ktr) ENDDO Dd(k)=STATE1(Pd(k,2),Pd(k,1),i,j,k,bi,bj,myThid) De(k)=Dd(k) CML print *, 'ml-opps:', i,j,k,tracerEnv(k,2),tracerEnv(k,1), CML & Dd(k),Pd(k,1),Pd(k,2) CML compute vertical velocity at cell centers from GCM velocity Wd(k)= - .5*(wVel(K)+wVel(K+1)) CML( CML avoid division by zero CML IF (Wd(K) .EQ. 0.D0) Wd(K) = 2.23e-16 CML) C guess at initial top grid cell vertical velocity CML Wd(k) = 0.03 C these estimates of initial plume velocity based on plume size and C top grid cell water mass c Wd(k) = 0.5*drF(k)/(dtts*FRACTIONAL_AREA) c Wd(k) = 0.5*drF(k)/dtts wsqr=Wd(k)*Wd(k) PlumeEntrainment(k) = 0.0 #ifdef ALLOW_OPPS_DEBUG IF ( OPPSdebugLevel.GE.debLevB ) THEN WRITE(msgBuf,'(A,I3)') & 'S/R OPPS_CALC: doing old lowerparcel', k CALL PRINT_MESSAGE( msgBuf, standardMessageUnit, & SQUEEZE_RIGHT , 1) ENDIF #endif /* ALLOW_OPPS_DEBUG */ radius=PlumeRadius StartingFlux=radius*radius*Wd(k)*Dd(k) oldflux=StartingFlux dz2=DrF(k) DO k2=k,KMax-1 D1=STATE1( Pd(k2,2), Pd(k2,1),i,j,k2+1,bi,bj,myThid) D2=STATE1( tracerEnv(k2+1,2), tracerEnv(k2+1,1), & i,j,k2+1,bi,bj,myThid) De(k2+1)=D2 C To start downward, parcel has to initially be heavier than environment C but after it has started moving, we continue plume until plume tke or C flux goes negative CML & _hFacC(i,j,k-1,bi,bj) CML & *_hFacC(i,j,k,bi,bj) .GT. 0. CML & .AND. IF (D2-D1 .LT. STABILITY_THRESHOLD.or.k2.ne.k) THEN dz1=dz2 dz2=DrF(k2+1) C find mass flux according to eq.(3) from paper by vertical integration newflux=oldflux+e2*radius*Wd(k2)*Dd(k2)* & .5*(dz1+dz2) CML print *, 'ml-opps:', i,j,k,oldflux,newflux,e2,radius, CML & Wd(k2),Dd(k2),Pd(k2,1),Pd(k2,2),dz1,dz2 PlumeEntrainment(k2+1) = newflux/StartingFlux IF(newflux.LE.0.0) then #ifdef ALLOW_OPPS_DEBUG IF ( OPPSdebugLevel.GE.debLevA ) THEN WRITE(msgBuf,'(A,I3)') & 'S/R OPPS_CALC: Plume entrained to zero at level ', k2 CALL PRINT_MESSAGE( msgBuf, standardMessageUnit, & SQUEEZE_RIGHT , 1) ENDIF #endif /* ALLOW_OPPS_DEBUG */ maxdepth = k2 if(maxdepth.eq.k) goto 1000 goto 1 endif C entrainment rate is basically a scaled mass flux dM/M entrainrate = (newflux - oldflux)/newflux oldflux = newflux C mix var(s) are the average environmental values over the two grid levels DO ktr=1,nTracerInUse pmix=(dz1*tracerEnv(k2,ktr)+dz2*tracerEnv(k2+1,ktr)) & /(dz1+dz2) Pd(k2+1,ktr)=Pd(k2,ktr) & - entrainrate*(pmix - Pd(k2,ktr)) ENDDO C compute the density at this level for the buoyancy term in the C vertical k.e. equation Dd(k2+1)=STATE1(Pd(k2+1,2),Pd(k2+1,1),i,j,k2+1,bi,bj,myThid) C next, solve for the vertical velocity k.e. using combined eq. (4) C and eq (5) from the paper #ifdef ALLOW_OPPS_DEBUG IF ( OPPSdebugLevel.GE.debLevA ) THEN WRITE(msgBuf,'(A,3E12.4,I3)') & 'S/R OPPS_CALC: Dd,De,entr,k ',Dd(k2),De(k2),entrainrate,k2 CALL PRINT_MESSAGE( msgBuf, standardMessageUnit, & SQUEEZE_RIGHT , 1) ENDIF #endif /* ALLOW_OPPS_DEBUG */ CML insert Eq. (4) into Eq. (5) to get something like this for wp^2 wsqr = wsqr - wsqr*abs(entrainrate)+ gravity* & (dz1*(Dd(k2)-De(k2))/De(k2) & +dz2*(Dd(k2+1)-De(k2+1))/De(k2+1)) C if negative k.e. then plume has reached max depth, get out of loop IF(wsqr.LE.0.0)then maxdepth = k2 #ifdef ALLOW_OPPS_DEBUG IF ( OPPSdebugLevel.GE.debLevA ) THEN WRITE(msgBuf,'(A,I3)') & 'S/R OPPS_CALC: Plume velocity went to zero at level ', k2 CALL PRINT_MESSAGE( msgBuf, standardMessageUnit, & SQUEEZE_RIGHT , 1) WRITE(msgBuf,'(A,4A14)') & 'S/R OPPS_CALC: ', 'wsqr', 'entrainrate', & '(Dd-De)/De up', '(Dd-De)/De do' CALL PRINT_MESSAGE( msgBuf, standardMessageUnit, & SQUEEZE_RIGHT , 1) WRITE(msgBuf,'(A,4E14.6)') & 'S/R OPPS_CALC: ', wsqr, entrainrate, & (Dd(k2)-De(k2))/De(k2), (Dd(k2+1)-De(k2+1))/De(k2+1) CALL PRINT_MESSAGE( msgBuf, standardMessageUnit, & SQUEEZE_RIGHT , 1) ENDIF #endif /* ALLOW_OPPS_DEBUG */ if(maxdepth.eq.k) goto 1000 goto 1 endif Wd(k2+1)=sqrt(wsqr) C compute a new radius based on the new mass flux at this grid level C from Eq. (4) radius=sqrt(newflux/(Wd(k2)*Dd(k2))) ELSE maxdepth=k2 if(maxdepth.eq.k) goto 1000 GOTO 1 ENDIF ENDDO C plume has reached the bottom MaxDepth=kMax 1 CONTINUE Ad(k)=FRACTIONAL_AREA IC=0 C start iteration on fractional area, not used in OGCM implementation DO IC=1,Max_ABE_Iterations C next compute the mass flux beteen each grid box using the entrainment Md(k)=Wd(k)*Ad(k) DO k2=k+1,maxDepth Md(k2)=Md(k)*PlumeEntrainment(k2) #ifdef ALLOW_OPPS_DEBUG IF ( OPPSdebugLevel.GE.debLevA ) THEN WRITE(msgBuf,'(A,2E12.4,I3)') & 'S/R OPPS_CALC: Md, Wd, and k are ',Md(k2),Wd(k2),k2 CALL PRINT_MESSAGE( msgBuf, standardMessageUnit, & SQUEEZE_RIGHT , 1) ENDIF #endif /* ALLOW_OPPS_DEBUG */ ENDDO C Now move on to calculate new temperature using flux from C Td, Sd, Wd, ta, sa, and we. Values for these variables are at C center of grid cell, use weighted average to get boundary values C use a timestep limited by the GCM model timestep and the maximum plume C velocity (CFL criteria) C calculate the weighted wd, td, and sd dt = dtts do k2=k,maxDepth-1 IF ( Wd(K2) .NE. 0. _d 0 ) dt = min(dt,drF(k2)/Wd(k2)) C time integration will be integer number of steps to get one C gcm time step ntime = nint(0.5*int(dtts/dt)) if(ntime.eq.0) then ntime = 1 endif C make sure area weighted vertical velocities match; in other words C make sure mass in equals mass out at the intersection of each grid C cell. Eq. (20) mda(k2) = (md(k2)*drF(k2)+md(k2+1)*drF(k2+1))/ * (drF(k2)+drF(k2+1)) wda(k2) = (wd(k2)*drF(k2)+wd(k2+1)*drF(k2+1))/ * (drF(k2)+drF(k2+1)) DO ktr = 1, nTracerInUse Pda(k2,ktr) = Pd(k2,ktr) Paa(k2,ktr) = tracerEnv(k2+1,ktr) ENDDO enddo dt = min(dt,dtts) #ifdef ALLOW_OPPS_DEBUG IF ( OPPSdebugLevel.GE.debLevA ) THEN WRITE(msgBuf,'(A,F14.4)') & 'S/R OPPS_CALC: time step = ', dt CALL PRINT_MESSAGE( msgBuf, standardMessageUnit, & SQUEEZE_RIGHT , 1) ENDIF #endif /* ALLOW_OPPS_DEBUG */ DO ktr=1,nTracerInUse Pda(maxdepth,ktr) = Pd(maxdepth,ktr) ENDDO kmx = maxdepth-1 do nn=1,ntime C top point DO ktr = 1,nTracerInUse tracerEnv(k,ktr) = tracerEnv(k,ktr)- & (mda(k)*(Pda(k,ktr)-Paa(k,ktr)))*dt*recip_drF(k) ENDDO C now do inner points if there are any CML if(Maxdepth-k.gt.1) then CML This if statement is superfluous CML IF ( k .LT. Maxdepth-1 ) THEN CML DO k2=k+1,Maxdepth-1 CML mda(maxDepth) = 0. DO k2=k+1,kmx k2m1 = max(k,k2-1) k2p1 = max(k2+1,maxDepth) DO ktr = 1,nTracerInUse tracerEnv(k2,ktr) = tracerEnv(k2,ktr) + & (mda(k2m1)*(Pda(k2m1,ktr)-Paa(k2m1,ktr)) & -mda(k2) *(Pda(k2,ktr) -Paa(k2,ktr)) ) & *dt*recip_drF(k2) ENDDO ENDDO CML This if statement is superfluous CML ENDIF C bottom point DO ktr=1,nTracerInUse tracerEnv(kmx+1,ktr) = tracerEnv(kmx+1,ktr)+ & mda(kmx)*(Pda(kmx,ktr)-Paa(kmx,ktr))*dt*recip_drF(kmx+1) ENDDO C set the environmental temp and salinity to equal new fields DO ktr=1,nTracerInUse DO k2=1,kmx paa(k2,ktr) = tracerEnv(k2+1,ktr) ENDDO ENDDO C end loop on number of time integration steps enddo ENDDO C count convection event in this grid cell OPPSconvectCount(I,J,K,bi,bj) = & OPPSconvectCount(I,J,K,bi,bj) + 1. _d 0 C jump here if k = maxdepth or if level not unstable, go to next C profile point 1000 continue C end of k-loop ENDDO C-- End IF (DIFFERENT_MULTIPLE) ENDIF RETURN END C---+----1----+----2----+----3----+----4----+----5----+----6----+----7-|--+----| _RL FUNCTION STATE1( sLoc, tLoc, i, j, kRef, bi, bj, myThid ) C !DESCRIPTION: \bv C *===============================================================* C | o SUBROUTINE STATE1 C | Calculates rho(S,T,p) C | It is absolutely necessary to compute C | the full rho and not sigma=rho-rhoConst, because C | density is used as a scale factor for fluxes and velocities C *===============================================================* C \ev C !USES: IMPLICIT NONE C == Global variables == #include "SIZE.h" #include "EEPARAMS.h" #include "PARAMS.h" #include "GRID.h" #include "DYNVARS.h" #ifdef ALLOW_NONHYDROSTATIC # include "NH_VARS.h" #endif /* ALLOW_NONHYDROSTATIC */ C !INPUT/OUTPUT PARAMETERS: C == Routine arguments == INTEGER i, j, kRef, bi, bj, myThid _RL tLoc, sLoc C !LOCAL VARIABLES: C == Local variables == _RL rhoLoc _RL pLoc CMLC estimate pressure from depth at cell centers CML mtoSI = gravity*rhoConst CML pLoc = ABS(rC(kRef))*mtoSI IF ( usingZCoords ) THEN C in Z coordinates the pressure is rho0 * (hydrostatic) Potential #ifdef ALLOW_NONHYDROSTATIC IF ( selectP_inEOS_Zc.EQ.3 ) THEN pLoc = rhoConst*( totPhiHyd(i,j,kRef,bi,bj) & + phi_nh(i,j,kRef,bi,bj) & + phiRef(2*kRef) & )*maskC(i,j,kRef,bi,bj) ELSEIF ( selectP_inEOS_Zc.EQ.2 ) THEN #else /* ALLOW_NONHYDROSTATIC */ IF ( selectP_inEOS_Zc.EQ.2 ) THEN #endif /* ALLOW_NONHYDROSTATIC */ C---------- C NOTE: For now, totPhiHyd only contains the Potential anomaly C since PhiRef has not (yet) been added in S/R DIAGS_PHI_HYD C---------- pLoc = rhoConst*( totPhiHyd(i,j,kRef,bi,bj) & + phiRef(2*kRef) & )*maskC(i,j,kRef,bi,bj) c ELSEIF ( selectP_inEOS_Zc.EQ.1 ) THEN C note: for the case selectP_inEOS_Zc=0, also use pRef4EOS (set to C rhoConst*phiRef(2*k) ) to reproduce same previous machine truncation ELSEIF ( selectP_inEOS_Zc.LE.1 ) THEN pLoc = pRef4EOS(kRef)*maskC(i,j,kRef,bi,bj) ELSE pLoc = rhoConst*phiRef(2*kRef)*maskC(i,j,kRef,bi,bj) ENDIF ELSEIF ( usingPCoords ) THEN C in P coordinates the pressure is just the coordinate of the tracer point pLoc = rC(kRef)* maskC(i,j,kRef,bi,bj) ENDIF CALL FIND_RHO_SCALAR( tLoc, sLoc, pLoc, rhoLoc, myThid ) STATE1 = rhoLoc #endif /* ALLOW_OPPS */ RETURN END #ifdef OPPS_ORGCODE C---+----1----+----2----+----3----+----4----+----5----+----6----+----7-|--+----| C Listed below is the subroutine for use in parallel 3-d circulation code. C It has been used in the parallel semtner-chervin code and is now being used C In the POP code. The subroutine is called nlopps (long story to explain why). C I've attached the version of lopps that we've been using in the simulations. C There is one common block that is different from the standard model commons C (countc) and it is not needed if the array convadj is not used. The routine C does need "kmp" which is why the boundc common is included. For massively C parallel codes (like POP) we think this will work well when converted from a C "slab" (i=is,ie) to a column, which just means removing the "do i=is,ie" loop. C There are differences between this C code and the 1-d code and the earlier scheme implemented in 3-d models. These C differences are described below. subroutine nlopps(j,is,ie,ta,sa,gcmdz) parameter (imt = 361 , jmt = 301 , km = 30 ) C Nlopps: E. Skyllingstad and T. Paluszkiewicz C Version: December 11, 1996 C Nlopps: This version of lopps is significantly different from C the original code developed by R. Romea and T. Paluskiewicz. The C code uses a flux constraint to control the change in T and S at C each grid level. First, a plume profile of T,S, and W are C determined using the standard plume model, but with a detraining C mass instead of entraining. Thus, the T and S plume C characteristics still change, but the plume contracts in size C rather than expanding ala classical entraining plumes. This C is heuristically more in line with large eddy simulation results. C At each grid level, the convergence of plume velocity determines C the flux of T and S, which is conserved by using an upstream C advection. The vertical velocity is balanced so that the area C weighted upward velocity equals the area weighted downdraft C velocity, ensuring mass conservation. The present implementation C adjusts the plume for a time period equal to the time for 1/2 of C the mass of the fastest moving level to move downward. As a C consequence, the model does not completely adjust the profile at C each model time step, but provides a smooth adjustment over time. c include "params.h" c include "plume_fast_inc.h" c include "plume_fast.h" c #include "loppsd.h" real ta(imt,km),sa(imt,km),gcmdz(km),dz(km) real pdensity,wsqr,oldflux,newflux,entrainrate,adtemp REAL Del,D,dza1,dza2,kd,kd1,Smix,Thmix,PlumeS,PlumeT,PlumeD INTEGER i,j,k Clfh integer is,ie,k2 Clfh REAL D1,D2,state1,Density REAL dz1,dz2 REAL radius,StartingFlux real ttemp(km),stemp(km),taa(km),saa(km) real wda(km),tda(km),sda(km),mda(km) real dtts,dt,sumo,sumn integer ntime,nn,kmx,ic LOGICAL debug,done INTEGER MAX_ABE_ITERATIONS PARAMETER(MAX_ABE_ITERATIONS=1) REAL PlumeRadius REAL STABILITY_THRESHOLD REAL FRACTIONAL_AREA REAL MAX_FRACTIONAL_AREA REAL VERTICAL_VELOCITY REAL ENTRAINMENT_RATE REAL e2 PARAMETER ( PlumeRadius = 100.D0 ) PARAMETER ( STABILITY_THRESHOLD = -1.E-4 ) PARAMETER ( FRACTIONAL_AREA = .1E0 ) PARAMETER ( MAX_FRACTIONAL_AREA = .8E0 ) PARAMETER ( VERTICAL_VELOCITY = .02E0 ) PARAMETER ( ENTRAINMENT_RATE = -.05E0 ) PARAMETER ( e2 = 2.E0*ENTRAINMENT_RATE ) ! Arrays. REAL Ad(km),Sd(km),Td(km),Wd(km),Dd(km),Md(km) REAL Se(km),Te(km),We(km),De(km) REAL PlumeEntrainment(km) REAL GridThickness(km) C input kmp through a common block common / boundc / wsx(imt,jmt),wsy(imt,jmt),hfs(imt,jmt), 1 ple(imt,jmt),kmp(imt,jmt),kmq(imt,jmt) cwmseas & ,wsx1(imt,jmt),wsy1(imt,jmt) 1 ,wsx2(imt,jmt),wsy2(imt,jmt) C input the variables through a common logical problem common /countc/ convadj(imt,jmt,km),ics,depth(km),problem C-----may want to setup an option to get this only on first call C otherwise it is repetive C griddz is initialize by call to setupgrid dtts = 2400 do k=1,km dz(k) = 0.01*gcmdz(k) enddo do k=1,km GridThickness(k) = dz(k) enddo C modified to loop over slab DO i=is,ie numgridpoints=kmp(i,j) C go to next column if only 1 grid point or on land if(numgridpoints.le.1) goto 1100 C loop over depth debug = .false. C first save copy of initial profile DO k=1,NumGridPoints stemp(k)=sa(i,k) ttemp(k)=ta(i,k) C do a check of t and s range, if out of bounds set flag if(problem) then write(0,*)"Code in trouble before this nlopps call" return endif if(sa(i,k).gt.40..or.ta(i,k).lt.-4.0) then problem = .true. write(0,*)"t out of range at j ",j debug = .true. return endif ENDDO if(debug) then write(*,*)"T and S Profile at ",i,j write(*,*)(ta(i,k),sa(i,k),k=1,NumGridPoints) endif DO k=1,NumGridPoints-1 C initialize the plume T,S,density, and w velocity Sd(k)=stemp(k) Td(k)=ttemp(k) Dd(k)=state1(stemp(k),ttemp(k),k) De(k)=Dd(k) c Wd(k)=VERTICAL_VELOCITY C guess at initial top grid cell vertical velocity Wd(k) = 0.03 C these estimates of initial plume velocity based on plume size and C top grid cell water mass c Wd(k) = 0.5*dz(k)/(dtts*FRACTIONAL_AREA) c Wd(k) = 0.5*dz(k)/dtts wsqr=Wd(k)*Wd(k) PlumeEntrainment(k) = 0.0 if(debug) write(0,*) 'Doing old lowerparcel' radius=PlumeRadius StartingFlux=radius*radius*Wd(k)*Dd(k) oldflux=StartingFlux dz2=GridThickness(k) DO k2=k,NumGridPoints-1 D1=state1(Sd(k2),Td(k2),k2+1) D2=state1(stemp(k2+1),ttemp(k2+1),k2+1) De(k2+1)=D2 C To start downward, parcel has to initially be heavier than environment C but after it has started moving, we continue plume until plume tke or C flux goes negative IF (D2-D1 .LT. STABILITY_THRESHOLD.or.k2.ne.k) THEN dz1=dz2 dz2=GridThickness(k2+1) C define mass flux according to eq. 4 from paper newflux=oldflux+e2*radius*Wd(k2)*Dd(k2)*0.50* . (dz1+dz2) PlumeEntrainment(k2+1) = newflux/StartingFlux IF(newflux.LT.0.0) then if(debug) then write(0,*)"Plume entrained to zero at ",k2 endif maxdepth = k2 if(maxdepth.eq.k) goto 1000 goto 1 endif C entrainment rate is basically a scaled mass flux dM/M entrainrate = (newflux - oldflux)/newflux oldflux = newflux C mix var(s) are the average environmental values over the two grid levels smix=(dz1*stemp(k2)+dz2*stemp(k2+1))/(dz1+dz2) thmix=(dz1*ttemp(k2)+dz2*ttemp(k2+1))/(dz1+dz2) C first compute the new salinity and temperature for this level C using equations 3.6 and 3.7 from the paper sd(k2+1)=sd(k2) - entrainrate*(smix - sd(k2)) td(k2+1)=td(k2) - entrainrate*(thmix - td(k2)) C compute the density at this level for the buoyancy term in the C vertical k.e. equation Dd(k2+1)=state1(Sd(k2+1),Td(k2+1),k2+1) C next, solve for the vertical velocity k.e. using combined eq. 4 C and eq 5 from the paper if(debug) then write(0,*)"Dd,De,entr,k ",Dd(k2),De(k2),entrainrate,k2 endif wsqr = wsqr - wsqr*abs(entrainrate)+ 9.81* . (dz1*(Dd(k2)-De(k2))/De(k2) . +dz2*(Dd(k2+1)-De(k2+1))/De(k2+1)) C if negative k.e. then plume has reached max depth, get out of loop IF(wsqr.LT.0.0)then maxdepth = k2 if(debug) then write(0,*)"Plume velocity went to zero at ",k2 endif if(maxdepth.eq.k) goto 1000 goto 1 endif Wd(k2+1)=sqrt(wsqr) C compute a new radius based on the new mass flux at this grid level radius=sqrt(newflux/(Wd(k2)*Dd(k2))) ELSE maxdepth=k2 if(maxdepth.eq.k) goto 1000 GOTO 1 ENDIF ENDDO C plume has reached the bottom MaxDepth=NumGridPoints 1 continue Ad(k)=FRACTIONAL_AREA IC=0 C start iteration on fractional area, not used in OGCM implementation DO IC=1,Max_ABE_Iterations C next compute the mass flux beteen each grid box using the entrainment 92 continue Md(k)=Wd(k)*Ad(k) DO k2=k+1,MaxDepth Md(k2)=Md(k)*PlumeEntrainment(k2) if(debug) then write(0,*)"Md, Wd, and k are ",Md(k2),Wd(k2),k2 endif ENDDO C Now move on to calculate new temperature using flux from C Td, Sd, Wd, ta, sa, and we. Values for these variables are at C center of grid cell, use weighted average to get boundary values C use a timestep limited by the GCM model timestep and the maximum plume C velocity (CFL criteria) C calculate the weighted wd, td, and sd dt = dtts do k2=k,maxdepth-1 dt = min(dt,dz(k2)/wd(k2)) C time integration will be integer number of steps to get one C gcm time step ntime = nint(0.5*int(dtts/dt)) if(ntime.eq.0) then ntime = 1 endif C make sure area weighted vertical velocities match; in other words C make sure mass in equals mass out at the intersection of each grid C cell. mda(k2) = (md(k2)*dz(k2)+md(k2+1)*dz(k2+1))/ * (dz(k2)+dz(k2+1)) wda(k2) = (wd(k2)*dz(k2)+wd(k2+1)*dz(k2+1))/ * (dz(k2)+dz(k2+1)) tda(k2) = td(k2) sda(k2) = sd(k2) taa(k2) = ttemp(k2+1) saa(k2) = stemp(k2+1) enddo dt = min(dt,dtts) if(debug) then write(0,*)"Time step is ", dt endif tda(maxdepth) = td(maxdepth) sda(maxdepth) = sd(maxdepth) C do top and bottom points first kmx = maxdepth-1 do nn=1,ntime ttemp(k) = ttemp(k)- * (mda(k)*(tda(k)-taa(k)))*dt*recip_drF(k) stemp(k) = stemp(k)- * (mda(k)*(sda(k)-saa(k)))*dt*recip_drF(k) C now do inner points if there are any if(Maxdepth-k.gt.1) then do k2=k+1,Maxdepth-1 ttemp(k2) = ttemp(k2) + * (mda(k2-1)*(tda(k2-1)-taa(k2-1))- * mda(k2)*(tda(k2)-taa(k2))) * *dt*recip_drF(k2) stemp(k2) = stemp(k2) + * (mda(k2-1)*(sda(k2-1)-saa(k2-1))- * mda(k2)*(sda(k2)-saa(k2))) * *dt*recip_drF(k2) enddo endif ttemp(kmx+1) = ttemp(kmx+1)+ * (mda(kmx)*(tda(kmx)-taa(kmx)))* * dt*recip_drF(kmx+1) stemp(kmx+1) = stemp(kmx+1)+ * (mda(kmx)*(sda(kmx)-saa(kmx)))* * dt*recip_drF(kmx+1) C set the environmental temp and salinity to equal new fields do k2=1,maxdepth-1 taa(k2) = ttemp(k2+1) saa(k2) = stemp(k2+1) enddo C end loop on number of time integration steps enddo ENDDO 999 continue C assume that it converged, so update the ta and sa with new fields c if(i.gt.180.and.j.gt.200.and.i.lt.240) then c write(*,*)"Converged ",i,j,k,maxdepth,ttemp(k+1),ta(i,k+1) c endif do k2=k,maxdepth convadj(i,j,k2) = convadj(i,j,k2) + (ttemp(k2)- * ta(i,k2)) sa(i,k2) = stemp(k2) ta(i,k2) = ttemp(k2) c if(i.gt.180.and.j.gt.200.and.i.lt.240) then c write(*,*)"convadj ",convadj(i,j,k2) c endif C see if nlopps messed up if(sa(i,k).gt.40..or.ta(i,k).lt.-4.0) then problem = .true. write(0,*)"t out of range at j after adjust",j debug = .true. endif enddo C jump here if k = maxdepth or if level not unstable, go to next C profile point 1000 continue C end loop on k, move on to next possible plume ENDDO 1100 continue C i loop ENDDO END #endif /* OPPS_ORGCODE */