update observation operator
This commit is contained in:
@ -1012,16 +1012,16 @@
|
||||
! FILENAME_OBS = '/home/kjw/GEOS-Chem/gcadj_std/runs/v8-02-01/' //
|
||||
! & 'ch4/geocape/' // GET_RES_EXT() // '/adjtmp/' //
|
||||
! & 'gctm.obs.YYYYMMDD.hhmm'
|
||||
FILENAME_OBS = 'gctm.obs.YYYYMMDD.hhmm'
|
||||
FILENAME_OBS = 'gctm.obs.YYYYMMDD.hhmm' ! 观测文件, 居然是直接指定的么
|
||||
CALL EXPAND_DATE( FILENAME_OBS, GET_NYMD(), GET_NHMS() )
|
||||
FILENAME_OBS = TRIM( ADJTMP_DIR ) // TRIM( FILENAME_OBS )
|
||||
XTAU = GET_TAU()
|
||||
CALL READ_BPCH2( TRIM(FILENAME_OBS), 'IJ-OBS-$', 1,
|
||||
& XTAU, IIPAR, JJPAR,
|
||||
& LLPAR, DUMMY_TRUE, QUIET=.TRUE.)
|
||||
GC_CH4_TRUE_ARRAY(:,:,:) = DUMMY_TRUE(:,:,:)
|
||||
& LLPAR, DUMMY_TRUE, QUIET=.TRUE.) ! 读取观测数据
|
||||
GC_CH4_TRUE_ARRAY(:,:,:) = DUMMY_TRUE(:,:,:) ! 将观测数据赋值
|
||||
|
||||
! Convert from [kg] --> [v/v]
|
||||
! Convert from [kg] --> [v/v] 逐个网格转换单位
|
||||
DO II=1,IIPAR
|
||||
DO JJ=1,JJPAR
|
||||
DO LG=1,LLPAR
|
||||
@ -1031,23 +1031,23 @@
|
||||
ENDDO
|
||||
ENDDO
|
||||
|
||||
! Read a priori vertical profiles from file
|
||||
! Read a priori vertical profiles from file 读取先验廓线
|
||||
FILENAME = '/home/kjw/new_satellites/geocape/data/' //
|
||||
& 'geocape_prior.' // GET_RES_EXT() // '.bpch'
|
||||
XTAU = GET_TAU0( 1, 1, 1985 )
|
||||
CALL READ_BPCH2( FILENAME, 'IJ-AVG-$', 1,
|
||||
& XTAU, IGLOB, JGLOB,
|
||||
& LLGEOCAPE, DUMMY_PRIOR, QUIET=.TRUE. )
|
||||
CH4_PRIOR(:,:,:) = DUMMY_PRIOR(:,:,:)
|
||||
CH4_PRIOR(:,:,:) = DUMMY_PRIOR(:,:,:) ! 给先验廓线赋值
|
||||
|
||||
|
||||
|
||||
! Select arbitrary II, JJ and NT value
|
||||
! Select arbitrary II, JJ and NT value 为什么这里的数值都直接给定了
|
||||
II=40
|
||||
JJ=JJPAR-10
|
||||
NT=100
|
||||
|
||||
! Initialize variables
|
||||
! Initialize variables 初始化相关的数组
|
||||
GC_PCENTER(:) = 0d0
|
||||
GC_PEDGE(:) = 0d0
|
||||
GC_AD(:) = 0d0
|
||||
@ -1079,12 +1079,12 @@
|
||||
|
||||
ENDDO
|
||||
|
||||
! Number of vertical levels to use in these observations
|
||||
! Number of vertical levels to use in these observations 处理观测和模式垂直层上的问题
|
||||
! Chop off lowermost levels if
|
||||
! GEOS-Chem surface pressure < GEOCAPE pressure levels
|
||||
nlev = count( PRESSURE_EDGE .LT. GC_PEDGE(1) )
|
||||
IF ( nlev .LT. 13 ) nlev = nlev + 1
|
||||
lind = LLGEOCAPE + 1 - nlev ! minimum vertical index on GEOCAPE grid
|
||||
lind = LLGEOCAPE + 1 - nlev ! minimum vertical index on GEOCAPE grid 垂直层相关
|
||||
|
||||
! Get interpolation matrix that maps GEOS-Chem to GEOCAPE grid
|
||||
GRIDMAP(1:LLPAR, 1:LLGEOCAPE) =
|
||||
@ -1092,11 +1092,11 @@
|
||||
|
||||
! Get GEOS-Chem column from "truth" run to make pseudo-observations
|
||||
GC_CH4_NATIVE_OB(:) = 0d0
|
||||
GC_CH4_NATIVE_OB(:) = GC_CH4_TRUE_ARRAY(II,JJ,:)
|
||||
GC_CH4_NATIVE_OB(:) = GC_CH4_TRUE_ARRAY(II,JJ,:) ! 给观测数据赋值
|
||||
|
||||
|
||||
! Interpolate GEOS-Chem CH4 column and observation to GEOCAPE grid
|
||||
! Column in [v/v]
|
||||
! Column in [v/v] 在垂直上进行循环, 进行插值
|
||||
DO LN = lind, LLGEOCAPE
|
||||
GC_CH4_onGEOCAPE(LN) = 0d0
|
||||
GC_CH4_onGEOCAPE_OB(LN) = 0d0
|
||||
@ -1122,8 +1122,8 @@
|
||||
! A = GEOCAPE averaging kernel
|
||||
!--------------------------------------------------------------
|
||||
|
||||
! x_m - x_a for model and "observation"
|
||||
! [v/v] --> ln( v/v ) happens here
|
||||
! x_m - x_a for model and "observation" 以下部分似乎是在计算模式柱浓度
|
||||
! [v/v] --> ln( v/v ) happens here 这部分是在观测垂直层上计算的
|
||||
DO LN = lind, LLGEOCAPE
|
||||
GC_CH4_onGEOCAPE(LN) =MAX(GC_CH4_onGEOCAPE(LN), 1d-10)
|
||||
GC_CH4_onGEOCAPE_OB(LN)=MAX(GC_CH4_onGEOCAPE_OB(LN),1d-10)
|
||||
@ -1153,7 +1153,7 @@
|
||||
|
||||
! For safety, initialize these up to LLGEOCAPE
|
||||
|
||||
! Add random error to this observation
|
||||
! Add random error to this observation 给观测增加随机误差
|
||||
CH4_HAT_werr(:) = CH4_HAT(:)
|
||||
DO LN = lind, LLGEOCAPE
|
||||
|
||||
@ -1171,59 +1171,59 @@
|
||||
! J = [ model - obs ]^T S_{obs}^{-1} [ model - obs ]
|
||||
!--------------------------------------------------------------
|
||||
|
||||
! Calculate difference between modeled and observed profile
|
||||
! Calculate difference between modeled and observed profile 计算差分部分
|
||||
DO LN = lind, LLGEOCAPE
|
||||
DIFF(LN) = CH4_HAT_werr(LN) - CH4_HAT_OB(LN)
|
||||
ENDDO
|
||||
|
||||
! Calculate adjoint forcing: 2 * DIFF^T * S_{obs}^{-1}
|
||||
! and cost function: DIFF^T * S_{obs}^{-1} * DIFF
|
||||
DO LN = lind, LLGEOCAPE
|
||||
DO LLN = lind, LLGEOCAPE
|
||||
FORCE(LN) = FORCE(LN) +
|
||||
! 因为卫星垂直上各层的误差存在相关性, 所以最终每层都要处理与其他各层之间的关联
|
||||
! Calculate adjoint forcing: 2 * DIFF^T * S_{obs}^{-1} 计算伴随强迫
|
||||
! and cost function: DIFF^T * S_{obs}^{-1} * DIFF 代价函数就是距离
|
||||
DO LN = lind, LLGEOCAPE ! 在垂直上进行循环
|
||||
DO LLN = lind, LLGEOCAPE ! 继续在垂直上进行循环?
|
||||
FORCE(LN) = FORCE(LN) + ! 伴随强迫的每一层都加上了该层与其他层之间的协方差
|
||||
& 2d0 * OBSERROR_INV(LN,LLN) * DIFF(LLN)
|
||||
ENDDO
|
||||
NEW_COST(LN) = NEW_COST(LN) + 0.5*DIFF(LN)*FORCE(LN)
|
||||
ENDDO
|
||||
ENDDO ! 也就是说, 这部分是将有限差分作为切线性算子的替代, 计算相关参数
|
||||
|
||||
|
||||
!--------------------------------------------------------------
|
||||
! Begin adjoint calculations
|
||||
! Begin adjoint calculations 开始计算伴随部分
|
||||
!--------------------------------------------------------------
|
||||
|
||||
! 整理以下这部分就是: FORCE = DIFF = CH4_HAT_werr_ADJ = CH4_HAT_ADJ
|
||||
|
||||
! The adjoint forcing is 2 * S_{obs}^{-1} * DIFF = FORCE
|
||||
DIFF_ADJ(:) = FORCE(:)
|
||||
DIFF_ADJ(:) = FORCE(:) ! 差分的伴随就是伴随强迫
|
||||
|
||||
! Adjoint of GEOS-Chem - Observation difference
|
||||
CH4_HAT_werr_ADJ(:) = DIFF_ADJ(:)
|
||||
CH4_HAT_werr_ADJ(:) = DIFF_ADJ(:) ! 观测模拟差的伴随
|
||||
|
||||
! Adjoint of adding random error to observation
|
||||
! Adjoint of adding random error to observation 随机观测误差的伴随
|
||||
DO LN=lind,LLGEOCAPE
|
||||
CH4_HAT_ADJ(LN) = 0d0
|
||||
CH4_HAT_ADJ(LN) = 0d0 ! 初始化为 0
|
||||
|
||||
DO LLN=lind,LLGEOCAPE
|
||||
DO LLN=lind,LLGEOCAPE ! 考虑协方差, 计算最终误差
|
||||
CH4_HAT_ADJ(LN) = CH4_HAT_ADJ(LN) +
|
||||
& CH4_HAT_werr_ADJ(LLN) * RANDNUM(NT) * OBSERROR(LLN,LN)
|
||||
ENDDO
|
||||
ENDDO
|
||||
CH4_HAT_ADJ(:) = CH4_HAT_werr_ADJ(:)
|
||||
CH4_HAT_ADJ(:) = CH4_HAT_werr_ADJ(:) ! 那你这上面计算的有什么意义
|
||||
|
||||
! Adjoint of GEOCAPE observation operator
|
||||
! Adjoint of GEOCAPE observation operator 观测算子的伴随
|
||||
CH4_PERT_ADJ(:) = CH4_HAT_ADJ(:)
|
||||
DO LN=lind,LLGEOCAPE
|
||||
CH4_PERT_ADJ(LN) = 0D0
|
||||
CH4_PERT_ADJ(LN) = 0D0 ! 又给初始化了, 那上面的赋值有啥意义
|
||||
|
||||
DO LLN=lind,LLGEOCAPE
|
||||
DO LLN=lind,LLGEOCAPE ! 反正和协方差计算相关, 累积
|
||||
CH4_PERT_ADJ(LN) = CH4_PERT_ADJ(LN) +
|
||||
& AVGKERNEL(LLN,LN) * CH4_HAT_ADJ(LLN)
|
||||
ENDDO
|
||||
ENDDO
|
||||
|
||||
! Adjoint of x_m - x_a
|
||||
DO LN = lind, LLGEOCAPE
|
||||
! fwd code:
|
||||
!GC_CH4(LN) = MAX(GC_CH4(LN), 1d-10)
|
||||
! Adjoint of x_m - x_a 状态变量的伴随?
|
||||
DO LN = lind, LLGEOCAPE ! 对于观测垂直层进行循环
|
||||
! fwd code: 看上去这边是针对正向模拟写的, 因为正向模拟这边存在范围限制
|
||||
!GC_CH4(LN) = MAX(GC_CH4(LN), 1d-10) 限制了最小值为 1e-10
|
||||
!CH4_PERT(LN) = LOG(GC_CH4(LN)) - LOG(PRIOR(LN))
|
||||
! adj code:
|
||||
IF ( GC_CH4_onGEOCAPE(LN) > 1d-10 ) THEN
|
||||
@ -1235,28 +1235,28 @@
|
||||
ENDDO
|
||||
|
||||
|
||||
! Adjoint of interpolation
|
||||
DO LN=lind,LLGEOCAPE
|
||||
DO LG=1,LLPAR
|
||||
! Adjoint of interpolation 插值还需要伴随的么
|
||||
DO LN=lind,LLGEOCAPE ! LN 是观测的垂直层
|
||||
DO LG=1,LLPAR ! LG 是模式的垂直层
|
||||
GC_CH4_NATIVE_ADJ(LG) = GC_CH4_NATIVE_ADJ(LG) +
|
||||
& GRIDMAP(LG,LN) * GC_CH4_onGEOCAPE_ADJ(LN)
|
||||
& GRIDMAP(LG,LN) * GC_CH4_onGEOCAPE_ADJ(LN) ! 从卫星垂直层插值到模式垂直层
|
||||
ENDDO
|
||||
ENDDO
|
||||
|
||||
|
||||
! Adjoint of unit conversion
|
||||
! Adjoint of unit conversion 单位转换的伴随, 大概就是从原始值到体积比的换算
|
||||
DO LG=1,LLPAR
|
||||
GC_CH4_NATIVE_ADJ(LG) = GC_CH4_NATIVE_ADJ(LG)
|
||||
& * XNUMOL(1) / ( XNUMOLAIR * GC_AD(LG) )
|
||||
ENDDO
|
||||
|
||||
|
||||
! Pass adjoing forcing back to adjoint tracer array
|
||||
DO LG=1,LLPAR
|
||||
! Pass adjoing forcing back to adjoint tracer array 将伴随强迫返回到大数组上
|
||||
DO LG=1,LLPAR ! 最终的伴随强迫是在模式的垂直层上
|
||||
ADJ(LG) = GC_CH4_NATIVE_ADJ(LG) * CHK_STT(II,JJ,LG,1)
|
||||
ENDDO
|
||||
|
||||
! Update cost function
|
||||
! Update cost function 更新代价函数
|
||||
COST_FUNC_A = COST_FUNC_A + SUM(NEW_COST(:))
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user