Skip to contents

Smooth a stack with a local-linear-trend Kalman smoother.

kalman_smooth.Rd

Fits a per-pixel local-linear-trend Kalman filter and smoother over the time axis and returns the smoothed level (and optionally its standard error). Convenience wrapper: one scan_over() per requested output, sharing one kalman_llt() parameterisation.

Usage

kalman_smooth(
  x,
  sigma_lvl,
  sigma_slp,
  sigma_obs = 1,
  obs_var = NULL,
  boundaries = NULL,
  outputs = c("mean", "sd"),
  dtype = "f32",
  bands = NULL,
  ...
)

Arguments

x

Observation stack (LazyRaster with a t axis), or a LazyDataset (each band smoothed independently).

sigma_lvl, sigma_slp, sigma_obs

Noise standard deviations (level disturbance, slope disturbance, observation).

obs_var

Optional relative observation-variance stack on the same grid (Var(v_t) = sigma_obs^2 * obs_var_t).

boundaries

Optional 0/1 stack on the same grid marking years that start a new regime (see kalman_llt(), section "Regime boundaries"); needs obs_var (pass a stack of ones for none).

outputs

Which outputs to build ("mean", "sd", the forward-filtered "fmean", "fsd", and "innov"; see kalman_llt()).

dtype

Output dtype (default f32).

bands

LazyDataset only: bands to smooth (default: all value bands). A dataset takes no obs_var or boundaries.

...

Passed to kalman_llt().

Value

A named list of lazy objects, one per requested output.

See also

hampel_smooth() for outlier removal, fill_gaps() for simple gap filling, scan_over() for custom scans.