analyseUG: ensemble contructor as input to assimilation
In garciapintado/rDAF: An R Data Assimilation Framework

Description Usage Arguments Details Value Author(s) References

Construct the matrices and objects needed as input to assimilate. Structured grid input metadata are given via the G argument. Unstructured grid input metadata are given via gLON and gLAT.

1
2
3

analyseUG(G = NULL, gLON = NULL, gLAT = NULL, fit, prm, X, yls = NULL,
gauDA = NULL, gauTS = NULL, aug = NULL, ana = NULL, dsn = ".", debugmode = FALSE,
retdy = FALSE, mpi = FALSE, analysis_scn = "a0", theta = NULL, lite=TRUE)

`G`	OPT, LIST, grid metadata for structured grids with a format as `gmeta6` objects in package spgrass6
`gLON`	OPT, LIST, geographical coordinates - unstructured grids [nx,ny] 'matrix'
`gLAT`	OPT, LIST, geographical coordinates - unstructured grids [nx,ny] 'matrix'
`fit`	INTEGER, forecast time index for saved output
`prm`	LIST, global assimilation parameters
`X`	LIST, state vector, structured as X[[xKIND]][[timelab]][[im]]
`yls`	OPT, LIST, observation list, structured as yls[[yKIND]][[yTYPE]]$s lists.
`gauDA`	OPT, LIST, y <-> HE matches, to include observations by state-vector augmentation. Augments x and y
`gauTS`	OPT, LIST, as gauDA lists, without the $y slot. Augments x, and can be used to obtain a smooth estimate (ETKS) at specific locations/variables in the domain.
`aug`	OPT, LIST, augmentation blocks (e.g.; physical parameters)
`ana`	LIST, analysis parameter for each variable type
`dsn`	CHARACTER, output files are stored in the path $dsn/results
`debugmode`	OPT, LOGICAL, defaults to FALSE. TRUE triggers writting files with assimilation details
`retdy`	OPT, LOGICAL, defaults to FALSE. TRUE for no assimilation, and instead return the innovation vector
`mpi`	OPT, LOGICAL, (forced to FALSE in rDAF) whether to parallelize the assimilation via MPI
`analysis_scn`	CHARACTER, label added to output file for eventual identification of assimilation scenarios
`theta`	OPT, LIST, required for finite difference sensitivity Kalman smoothers [pIKS,pMKS]
`lite`	LOGICAL, only TRUE is allowed in this version

A detailed documentation of the arguments to this function is under preparation as a vignette. Specifically, this involves gauDA, prm, aug and ana lists. Please, ask the author/maintainer for documentation requests in the meantime.

analyseUG conducts the preparation of the matrices needed for the assimilation. The actual assimilation is conducted by the function assimilate, which includes Euclidean based localization and localization based on spatial networks, via the use of the SpatialGraph package, as well as MPI parallelization.

For localization based on SpatialGraph objects, it is assumed that the input SpatialGraph is stored as file.path(dsn,prm$sgf) and precalculated along-network distances for nodes in the SpatialGraph are stored in file.path(dsn,prm$gridSGf).

The function assimilate is planned to be publicly released along with the vignette documentation. In the interim time, the version assimilateLite, which conducts global assimilation via the EnKF or the ETKF, serves for the purpose of package testing. analyseUG also includes the Finite Difference Sensitivity (FDS) smoothers described in Garcia-Pintado and Paul (2018).

A LIST with the components

`E`	Updated ensemble matrix
`xdf`	Data frame with state vector information
`dydf`	Data frame with observation and innovation information

Javier Garcia-Pintado

Garcia-Pintado J. and Paul, A., 2018. Evaluation of iterative Kalman smoother schemes for multi-decadal past climate analysis with comprehensive Earth system models. Geoscientific Model Development.

garciapintado/rDAF documentation built on May 25, 2019, 7:26 p.m.