310 likes | 397 Views
First Step towards a Well Structured Surface Scheme Source Code Han The. Implementing the ECOCLIMAP database in HIRLAM. LAI*. Parameter fields. ECOCLIMAP: structure. landuse dataset. Equations. parameters per landuse class.
E N D
First Step towards a Well Structured Surface Scheme Source Code Han The Implementing the ECOCLIMAP database in HIRLAM
LAI* Parameter fields ECOCLIMAP: structure landuse dataset Equations parameters per landuse class *LAI implemented as a fixed parameter per land use class (10-days avg.)
ECOCLIMAP Quick Implementation: • Use the ECOCLIMAP software to create the parameter fields to replace the Climate script Disadvantages: • Current HIRLAM code not suitable to handle all parameters • Future implementation of assimilated LAI not possible
New Structure (F77-based; provisional) purpose To cluster single fields into logical multi-dimensional arrays to: • reduce the number of function arguments • simplify the global structure • simplify the introduction of new fields and layers
New Structure: storage • (mostly) constant surface fields:REAL TOPO(NX,NY,NFLD)roughness lengths, tile fractions, LAI etc. • index fields:INTEGER ITOPO(NX,NY,MFLD)soil type and vegetation type index • soil physical quantities stored as 3-D arrays:soil moisture, soil temperature: SWI(NX,NY,NSOIL,NTYPE)
New Structure: access to separate fields Field names are stored in include file TOPO.inc as parameters
TOPO.inc (provisional) integer NFLD c fields in TOPO array integer _LAI, _VEG, _ROOT, _RSMIN, _ALB, _SALB, + _Z0VEG, _Z0H, _SAND, _CLAY, _FIS, _TILE, + _SEA, _ICE, _BAR, _LOW, _HIGH, + _Z0ORO, _LAKE, _LAND, _Z0SEA, _TSCL, _SWCL parameter (_SEA = 1, ! sea fraction + _ICE = 2, ! ice fraction ... + _ALB = 8, ! grid albedo + _Z0H = 15, ! z0 for heat + NFLD = 15) ! total number of climate fields c fields in ITOPO array integer MFLD parameter (MFLD = 1) integer _SOIL parameter (_SOIL = 1) ! soil type integer NSOIL parameter (NSOIL = 2) ! number of soil layers <EOF>
New Structure: rules • Never use the parameter value, e.g.:TOPO(1,1,_ALB) instead ofTOPO(1,1,8) • At lowest level pass the required field and not TOPO as an argument, e.g.: CALL SURCOF_LAND(KSTART,KSTOP,NHOR,NLEV,JTYP,LHIRSN, + TOPO(1,_Z0VEG),TOPO(1,_Z0ORO),ZWS,ZWSMX,ZWFC, + ZITM,ZSNS,ZTS,ZVEG0, … • First five fields in TOPO(:,:,:) correspond to the tile fractions FRACI(:,:,:), i.e. you can pass TOPO as a function argument for the tile fractions (FRACI) instead of TOPO(1,1,1)
New Structure: storage ITOPO replaces constructions like: JQQ=NINT(((PFRACI(JL,JTYP)-RFRLIM)/(ZEPS+ABS(PFRACI(JL,JTYP) & -RFRLIM))+1.)*0.5) JSOL=NINT(PSOILI(JL,JTYP))*JQQ+1-JQQ JSOL=MAX(1,MIN(JSOL,11)) by: JSOL = ITOPO(JL,_SOIL)
New Structure: code • Modifications based on version 6.3.7 • Numbers of files affected: 40
Naming conventions • Currently, there are no strict directions • Many different names for the same variable • Unclear coding and • Reduncancy (unnecessary temporary arrays and variables)
Naming conventions Generally, semantic-related naming conventions do not improve coding except for external/global variables: • where to find the variable declaration/initialisation (e.g. reference to an include file) • type (parameter, index, structure, etc.) • usage (Hirlam code) Generally, there is a need for stricter definitions of variable names e.g.: fixed names for global variables (passed through as function arguments) throughout the code
Naming conventions: examples In SAPP.f: Land fraction is defined twice: FRLAND (function argument; not used) ZFRLAND (local variable; initialised and used instead)
Constant parameter fields: initialisation • TXT.DAT: saturation point, field capacity, wilting point • INIISBA: fractions of sand and clay • INI_VEG: veg parameters + WROUS, WVEG, WLAI • INIPHY: tile fraction, vegetation type • SUR2RAD: albedo
Constant parameter fields: checks • INIPHY: most fields, tile fractions, vegetation type • ISBAH4: soil type and vegetation type, z0veg JSOL=MAX(1,MIN(JSOL,11)) JVEG=MAX(1,MIN(JVEG,18)) • TQUV: roughness length DO I=1,NHOR Z0(I)=MAX(Z0(I),5.E-4) Z0(I)=MIN(Z0(I),3.) ENDDO • SAPP: land fraction • SUR2RAD: vegetation type
Constant parameter fields: tile fractions In HDF-file: sum of fraction equals 1 (within single precision) With conversion to GRIB precision is lost. Internal check is always required before usage Currently, this happens in INIPHY In the surface analysis routines there is no check. As a result the land masks in AN en FC differ
Constant parameter fields: tile fractions in ANALYSIS: FIELD PARAMETER INDEX LONG. LAT. AVERAGE STD. DEV. FRACTION OF LAND (1=LAND; 1 -41.61 11.95 0.461639 0.475845 in FORECAST: FIELD PARAMETER INDEX LONG. LAT. AVERAGE STD. DEV. FRACTION OF LAND (1=LAND; 179 -13.63 26.51 0.449246 0.479794
Constant parameter fields: I/O There must be a single API routine to read/write for both Hirlam and Hirvda: GETDAT/PUTDAT • All consistancy checks must do done within GETDAT • The user must have the guarantee that the fields throughout the code is obtained from GETDAT
Function declarations and argument lists: examples Not readable (representative): C----------------------------------------------------------------------- SUBROUTINE ACVEG ( KIDIA,KFDIA,KLON,KLEV, C----------------------------------------------------------------------- C - INPUT 2D . * PAPRSF, clrb PFRSO, * SSWNET, PQ,PQSAT,PT, clre C - INPUT 1D . * PD2,PITM,PIVEG, clr PALB, * PLAI,PNEIJ,PVEG,PRSMIN, * PCHROV,PGWDCS,PWFC,PWSAT,PWL,PWWILT,PWP,PQSATS, C - OUTPUT 1D . * PHQ,PHTR,PHU,PHV,PWLMX, C - TEMPORAIRE 1D . * PZDELTA,PZRSTO )
Function declarations and calls: examples Not useful (I: integer; R real): SUBROUTINE LNDTQ2( KMONTH, KDAY, I KLON , KLAT , KLEV , KTYP R , PAHYB , PBHYB ... Random: SUBROUTINE POSTPP(MYPE $ ,NLON_GLOBAL ,NLAT_GLOBAL X ,NLEVML, LTYPML,ALEVML,NWMOML,IWMOML X ,NSL , LTYPSL, IWMOSL, ALEVSL X ,LPHYS , LOMEGA $ ,NLEV , NLON , NLAT , NTYP ...
Function declarations and calls: proposed format SUBROUTINE ACVEG ( r KIDIA, KFDIA, KLON, KLEV, PAPRSF, r SSWNET, PQ, PQSAT, PT, PD2, PITM, PIVEG, r PLAI, PNEIJ, PVEG, PRSMIN, r PCHROV, PGWDCS, PWFC, PWSAT, PWL, PWWILT, PWP, PQSATS, w PHQ, PHTR, PHU, PHV, PWLMX, m PZDELTA, PZRSTO) rinput (constant/readable): must be initialised before call; no local copy needed; woutput (writable): no initialisation needed; making a local copy in calling function depends on usage; minput/output (modified): initialised; local copy depending on usage;
Example of variable life cycle: Original code SWI: surface layer soil water index DSWI: surface layer soil water tendency CALL PHYS(...,SWI,...,DSWI,...) WSWI = SWI ! WSWI: local variable DSWI = 0 CALL ISBAH4(...,SWI [PSWI],...,DSWI [PDSWI],...) PSWI = MIN(PSWI, 0.5) ZWS = PSWI*RD1*GCONV ! unit conv PSWI; ZWS: local variable ZDWS = 0 ! local variable CALL SURTEND_LAND(...,ZWS [PWS],...,ZDWS [PDWS],...) ZSWP = PWS + calculated change ! local variable ZDSW = ZSWP - PWS ! local variable PDWS = ZDSW*ZRDT/ZCONS10 ! unit conversion of tendency END SURTEND_LAND PDSWI = ZDWS END ISBAH4 WSWI = WSWI + DTVDIF*DSWI DSWI = (WSWI - SWI)/ZDTIME END PHYS
Improved code m/c SWI: surface layer soil water index w DTSWI: surface layer soil water tendency CALL PHYS(...,SWI,...,DTSWI,...) CALL ISBAH4(...,SWI,...,DSWI,...) SWI = MIN(SWI, 0.5) CALL SURTEND_LAND(...,SWI,...,DWSI,...) DSWI = calculated change ! in correct units END SURTEND_LAND END ISBAH4 DTSWI = DSWI/ZDTIME END PHYS
Gain: • removal of 12 temporary fields: WSWI(NHOR,NTYP), DSWI(NHOR,NTYP),ZWS(NHOR), ZDWS(NHOR) • similarly for second soil water layer • simplified code
Life cycle of variables • where are the variables initialised • where are the variables modified • what are the modifications • pathways of the variables
Summary • 2D fields are clustered into 3-dimensional arrays to improve structure and enhance future flexibility • Documented naming conventions based on functionality rather than semantic meaning of the variable • Well structured I/O of climate fields is required (GETDAT) • Description of the life cycle of the time-dependent variables must be included in the documentation