block_LEnKPF: block-LEnKPF analysis

In robertsy/assimilr: Local Ensemble Kalman Particle Filters

Description

Assimilate observations sequentially in blocks. Smooth discontinuities by conditional resampling using the empirical covariance structure. The method does not work without a taper for P. The specific update of one block is in one_block_update REM: for historical reasons the indices referred to as u and v in the paper are named in the code as x and v.

Usage

block_LEnKPF(xb, y, H, R, l, block_size = l/2,
  get_partition = ring_partition, ndim = nrow(xb), taper = 1,
  gam.fix = NA, d = length(y), K = ncol(xb), q = nrow(xb),
  unif = runif(1), ...)

Arguments

xb	the background ensemble
y	the observations
H	the observation linear operator
R	the observations error covariance
l	the localization radius (not used directly)
block_size	size of domain considered to include observations in a block
get_partition	function to partition the observations in blocks, depending on geometry (typically ring_partition or sweq_partition)
taper	the tapering correlation matrix applied to P
gam.fix	fixed gamma value
d	dimension of observation, ensemble and state space
K	dimension of observation, ensemble and state space
q	dimension of observation, ensemble and state space
unif	used for balanced sampling
...	additional parameters passed to adaptive_gamma, typically e.0 and e.1

robertsy/assimilr documentation built on May 27, 2019, 10:33 a.m.