| Version 13 (modified by , 3 years ago) (diff) |
|---|
Available options for the different filter algorithms
Contents of this page
There are different operations for each of the filter algorithms that need to be specified in the call to pdaf_init. To display the available options in a compiled assimilation program, one can use run with the specification subtype=-1. In this case pdaf_init wil display the available options for the selected filter algorithm and exit with an error status of -1.
Below we list the options as they are displayed using subtype=-1.
SEEK (filtertype=0)
PDAF Available options for SEEK: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: Evolve unit modes with finite difference approx. of TLM PDAF 1: like 0 with modes scaled by eigenvalues, unit U PDAF 2: Fixed basis vectors; variable U matrix PDAF 3: Fixed covariance matrix (V and U kept constant) PDAF 5: Offline mode PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): Interval for re-diagonalization of P (>0); optional: default 1 PDAF param_int(4): 1 for incremental updating, 0 else; optional: default 0 PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF param_real(2): epsilon for finite-difference approx. of TLM, required PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF >=1 for subtypes 0 and 1; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for the SEEK filter ++++++++++
SEIK (filtertype=1)
PDAF Available options for SEIK: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: full ensemble integration; left-sided application of T PDAF 1: full ensemble integration; right-sided application of T PDAF 2: Fixed error space basis PDAF 3: Fixed state covariance matrix PDAF 4: Implementation with explicit ensemble transformation PDAF 5: Offline mode PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): not used PDAF param_int(4): Apply incremental updating; optional PDAF 0: no incremental updating (default) PDAF 1: apply incremental updating PDAF param_int(5): Type of forgetting factor; optional PDAF 0: fixed forgetting factor (default) PDAF 1: adaptive forgetting factor (experimental) PDAF param_int(6): Type of ensemble transformation matrix; optional PDAF 0: deterministic omega (default) PDAF 1: random orthonormal omega orthogonal to (1,...,1)^T PDAF 2: use product of 0 with random orthonomal matrix with eigenvector (1,...,1)^T PDAF (experimental; for random transformations, 1 is recommended) PDAF param_int(7): Type of transformation matrix square root; optional PDAF (Only relevant for subtype/=3) PDAF 0: symmetric square root (default) PDAF 1: Cholesky decomposition PDAF param_int(8): Application of observation operator H, optional PDAF 0: Apply H to ensemble mean to compute residual (default) PDAF 1: Apply H to all ensemble states and then compute residual from mean of these PDAF param_int(8)=1 is the recomended choice for nonlinear H PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF >=1 for subtypes 0 and 1; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF --- Internal parameter (defined inside PDAF) --- PDAF Nm1vsN: Normalization of covariance matrix; default: 1 PDAF 0: normalization with 1/(Ensemble size) PDAF (original SEIK, mainly for compatibility with older studies) PDAF 1: normalization with 1/(Ensemble size - 1) PDAF (sample covariance matrix consistent with other EnKFs) PDAF +++++++++ End of option overview for the SEIK filter ++++++++++
EnKF (filtertype=2)
PDAF Available options for EnKF: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: Full ensemble integration; analysis for 2*dim_obs>dim_ens PDAF 1: Full ensemble integration; analysis for 2*dim_obs<=dim_ens PDAF 5: Offline mode PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): maximum rank for inversion of HPH^T, optional, default=0 PDAF (for =0, HPH is inverted by solving the representer equation) PDAF (if set to >=ensemble size, it is reset to ensemble size - 1) PDAF param_int(4): not used PDAF param_int(5): Size of smoothing lag (>=0), optional PDAF 0: no smoothing (default) PDAF >0: apply smoother up to specified lag PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF (>=1; not larger than total number of processors) PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for the EnKF ++++++++++
LSEIK (filtertype=3)
PDAF Available options for LSEIK: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: full ensemble integration; left-sided application of T PDAF 2: Fixed error space basis PDAF 3: Fixed state covariance matrix PDAF 5: Offline mode PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): not used PDAF param_int(4): Apply incremental updating; optional PDAF 0: no incremental updating (default) PDAF 1: apply incremental updating PDAF param_int(5): Type of forgetting factor; optional PDAF 0: fixed forgetting factor (default) PDAF 1: adaptive forgetting factor for full domain (experimental) PDAF 2: locally adaptive forgetting factor (experimental) PDAF param_int(6): Type of ensemble transformation matrix; optional PDAF 0: deterministic omega (default) PDAF 1: random orthonormal omega orthogonal to (1,...,1)^T PDAF 2: use product of 0 with random orthonomal matrix with eigenvector (1,...,1)^T PDAF (experimental; for random transformations, 1 is recommended) PDAF param_int(7): Type of transformation matrix square root; optional PDAF (Only relevant for subtype/=3) PDAF 0: symmetric square root (default) PDAF 1: Cholesky decomposition PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF >=1 for subtypes 0 and 1; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF --- Internal parameter (defined inside PDAF) --- PDAF Nm1vsN: Normalization of covariance matrix; default: 1 PDAF 0: normalization with 1/(Ensemble size) PDAF (original SEIK, mainly for compatibility with older studies) PDAF 1: normalization with 1/(Ensemble size - 1) PDAF (sample covariance matrix consistent with other EnKFs) PDAF +++++++++ End of option overview for the LSEIK filter ++++++++++
ETKF (filtertype=4)
PDAF Available options for ETKF: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: full ensemble integration; apply T-matrix analogously to SEIK PDAF 1: full ensemble integration; formulation without T matrix PDAF 2: Fixed error space basis; analysis with T-matrix PDAF 3: Fixed state covariance matrix; analysis with T-matrix PDAF 5: Offline mode; analysis with T-matrix PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): Size of smoothing lag (>=0), optional PDAF 0: no smoothing (default) PDAF >0: apply smoother up to specified lag PDAF param_int(4): not used PDAF param_int(5): Type of forgetting factor; optional PDAF 0: fixed forgetting factor (default) PDAF 1: adaptive forgetting factor (experimental) PDAF param_int(6): Type of ensemble transformation matrix; optional PDAF 0: deterministic transformation (default) PDAF 2: use product of 0 with random orthonomal matrix with eigenvector (1,...,1)^T PDAF param_int(7): not used PDAF param_int(8): Application of observation operator H PDAF 0: Apply H to ensemble mean to compute residual (default) PDAF 1: Apply H to all ensemble states; then compute residual from mean of these PDAF param_int(8)=1 is the recomended choice for nonlinear H PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF >=1 for subtypes 0 and 1; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for the ETKF ++++++++++
LETKF (filtertype=5)
PDAF Available options for LETKF: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: full ensemble integration; apply T-matrix analogously to SEIK PDAF 2: Fixed error space basis; analysis with T-matrix PDAF 3: Fixed state covariance matrix; analysis with T-matrix PDAF 5: Offline mode; analysis with T-matrix PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): Size of smoothing lag (>=0), optional PDAF 0: no smoothing (default) PDAF >0: apply smoother up to specified lag PDAF param_int(4): not used PDAF param_int(5): Type of forgetting factor; optional PDAF 0: fixed forgetting factor (default) PDAF 1: adaptive forgetting factor for full domain (experimental) PDAF 2: locally adaptive forgetting factor (experimental) PDAF param_int(6): Type of ensemble transformation matrix; optional PDAF 0: deterministic transformation (default) PDAF 2: use product of 0 with random orthonomal matrix with eigenvector (1,...,1)^T PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF >=1 for subtypes 0 and 1; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for the LETKF ++++++++++
ESTKF (filtertype=6)
PDAF Available options for ESTKF: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: Standard implementation with ensemble integration PDAF 2: Fixed error space basis PDAF 3: Fixed state covariance matrix PDAF 5: Offline mode PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): Size of smoothing lag (>=0), optional PDAF 0: no smoothing (default) PDAF >0: apply smoother up to specified lag PDAF param_int(4): not used PDAF param_int(5): Type of forgetting factor; optional PDAF 0: fixed forgetting factor (default) PDAF 1: adaptive forgetting factor (experimental) PDAF param_int(6): Type of ensemble transformation matrix; optional PDAF 0: deterministic omega (default) PDAF 1: random orthonormal omega orthogonal to (1,...,1)^T PDAF 2: use product of 0 with random orthonomal matrix with eigenvector (1,...,1)^T PDAF (experimental; for random transformations, 0 or 1 are recommended) PDAF param_int(7): Type of transformation matrix square root; optional PDAF 0: symmetric square root (default) PDAF 1: Cholesky decomposition PDAF param_int(8): Application of observation operator H PDAF 0: Apply H to ensemble mean to compute residual (default) PDAF 1: Apply H to all ensemble states; then compute residual from mean of these PDAF param_int(8)=1 is the recomended choice for nonlinear H PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF >=1 for subtypes 0 and 1; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for the ESTKF ++++++++++
LESTKF (filtertype=7)
PDAF Available options for LESTKF: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: Standard implementation with ensemble integration PDAF 2: Fixed error space basis PDAF 3: Fixed state covariance matrix PDAF 5: Offline mode PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): Size of smoothing lag (>=0), optional PDAF 0: no smoothing (default) PDAF >0: apply smoother up to specified lag PDAF param_int(4): not used PDAF param_int(5): Type of forgetting factor; optional PDAF 0: fixed forgetting factor (default) PDAF 1: adaptive forgetting factor for full domain (experimental) PDAF 2: locally adaptive forgetting factor (experimental) PDAF param_int(6): Type of ensemble transformation matrix; optional PDAF 0: deterministic omega (default) PDAF 1: random orthonormal omega orthogonal to (1,...,1)^T PDAF 2: use product of 0 with random orthonomal matrix with eigenvector (1,...,1)^T PDAF (experimental; for random transformations, 1 is recommended) PDAF param_int(7): Type of transformation matrix square root; optional PDAF 0: symmetric square root (default) PDAF 1: Cholesky decomposition PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF >=1 for subtypes 0 and 1; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for the LESTKF ++++++++++
LEnKF (filtertype=8)
PDAF Available options for LEnKF: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: Full ensemble integration; analysis with covariance localization PDAF 5: Offline mode PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): maximum rank for inversion of HPH^T, optional, default=0 PDAF (for =0, HPH is inverted by solving the representer equation) PDAF (if set to >=ensemble size, it is reset to ensemble size - 1) PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF (>=1; not larger than total number of processors) PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for the LEnKF ++++++++++
NETF (filtertype=9)
PDAF Available options for NETF: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: Standard implementation with ensemble integration PDAF 5: Offline mode PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): Size of smoothing lag (>=0), optional PDAF 0: no smoothing (default) PDAF >0: apply smoother up to specified lag PDAF param_int(4): not used PDAF param_int(5): Type of forgetting factor; optional PDAF 0: forgetting factor on forecast ensemble (default) PDAF 2: forgetting factor on analysis ensemble PDAF param_int(6): Type of ensemble transformation matrix; optional PDAF 0: random orthonormal matrix orthogonal to (1,...,1)^T (default) PDAF 1: deterministic transformation PDAF param_int(7): Type of weights inflation; optional PDAF 0: no weights inflation (default) PDAF 1: inflate so that N_eff/N > param_real(2) PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF param_real(2): Limit for weigts inflation N_eff/N > param_real(2), optional, default=0.0 PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF >=1 for subtypes 0 and 1; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for the NETF ++++++++++
LNETF (filtertype=10)
PDAF Available options for LNETF: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: Standard implementation with ensemble integration PDAF 5: Offline mode PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): Size of smoothing lag (>=0), optional PDAF 0: no smoothing (default) PDAF >0: apply smoother up to specified lag PDAF param_int(4): not used PDAF param_int(5): Type of forgetting factor; optional PDAF 0: forgetting factor on forecast ensemble (default) PDAF 1: forgetting factor on forecast ensemble only observed domains PDAF 2: forgetting factor on analysis ensemble PDAF 3: forgetting factor on analysis ensemble only observed domains PDAF param_int(6): Type of ensemble transformation matrix; optional PDAF 0: random orthonormal matrix orthogonal to (1,...,1)^T (default) PDAF 1: deterministic transformation PDAF param_int(7): Type of weights inflation; optional PDAF 0: no weights inflation (default) PDAF 1: inflate so that N_eff/N > param_real(2) PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF param_real(2): Limit for weigts inflation N_eff/N > param_real(2), optional, default=0.0 PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF >=1 for subtypes 0 and 1; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for the LNETF ++++++++++
PF (filtertype=12)
PDAF Available options for PF: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: Standard implementation with ensemble integration PDAF 5: Offline mode PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): Resampling type, optional PDAF 1: probabilistic resamping (default) PDAF 2: stochastic universal resampling PDAF 3: residual resampling PDAF param_int(4): Type of ensemble perturbations, optional PDAF 0: no perturbations (default) PDAF 1: constant standard deviation PDAF 2: relative to ensemble standard deviation PDAF param_int(5): Type of forgetting factor; optional PDAF 0: forgetting factor on forecast ensemble (default) PDAF 2: forgetting factor on analysis ensemble PDAF param_int(6): Type of weights inflation; optional PDAF 0: no weights inflation (default) PDAF 1: inflate so that N_eff/N > param_real(2) PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Ensemble pert. level (>0), required, only used if param_int(4)>0 PDAF param_real(2): Forgetting factor (usually >0 and <=1), optional, default=1.0 PDAF param_real(3): Limit for weigts inflation N_eff/N > param_real(2), optional, default=0.0 PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF >=1 for subtypes 0 and 1; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for the PF ++++++++++
GENOBS (filtertype=100)
Note: GENOBS used filtertype=11 before PDAF V2.0.
PDAF Available options for GENOBS: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: Standard implementation with ensemble integration PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required, but not used PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF =1 for GENOBS; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for GENOBS ++++++++++
3DVAR (filtertype=200)
Note: 3D-Var methods were added in PDAF V2.0.
PDAF Available options for 3D-Var: PDAF --- Sub-types (Parameter subtype) --- PDAF 0: incremental 3D-Var with parameterized covariance matrix PDAF 1: 3D ensemble Var using LESTKF for ensemble transformation PDAF 4: 3D ensemble Var using ESTKF for ensemble transformation PDAF 5: Offline mode; analysis chosen by PDAF_put_state/PDAF_assimilate PDAF 6: hybrid 3D-Var using LESTKF for ensemble transformation PDAF 7: hybrid 3D-Var using ESTKF for ensemble transformation PDAF --- Integer parameters (Array param_int) --- PDAF param_int(1): Dimension of state vector (>0), required PDAF param_int(2): Ensemble size (>0), required PDAF param_int(3): Select optimization method (solver), required PDAF 0: LBFGS (default) PDAF 1: CG+ PDAF 2: direct implementation of CG PDAF 3: direct implementation of CG with decomposed control vector PDAF param_int(4): size of parameterized control vector (for parameterized and hybrid 3D-Var), required PDAF param_int(5): size of ensemble control vector (required for ensemble and hybrid 3D-Var), PDAF param_int(4): Dimension of parameterized control vector PDAF --- Floating point parameters (Array param_real) --- PDAF param_real(1): Forgetting factor (usually >0 and <=1), required PDAF param_real(2): hybrid weight beta, optional PDAF >=0.0 and <=1.0 (default = 0.5) PDAF --- Further parameters --- PDAF n_modeltasks: Number of parallel model integration tasks PDAF >=1 for subtypes 0 and 1; not larger than total number of processors PDAF =1 required for subtypes 2 and 3 PDAF screen: Control verbosity of PDAF PDAF 0: no outputs PDAF 1: basic output (default) PDAF 2: 1 plus timing output PDAF 3: 2 plus debug output PDAF +++++++++ End of option overview for 3DVAR ++++++++++}}}
