stfit_landsat: STFIT for Landsat data

In stfit: Spatio-Temporal Functional Imputation Tool

STFIT for Landsat data

Description

This function is used for Landsat data imputation, which includes five steps: mean estimation, outlier detection, temporal effect estimation, spatial effect estimation and imputation. In real application, one can use this as a template to create a five steps imputation procedure depending on the real data structure.

Usage

stfit_landsat(
  year,
  doy,
  mat,
  img.nrow,
  img.ncol,
  doyeval = 1:365,
  h.tcov = 100,
  h.tsigma2 = 300,
  h.scov = 2,
  h.ssigma2 = 2,
  nnr = 10,
  outlier.action = c("keep", "remove"),
  outlier.tol = 0.2,
  intermediate.save = TRUE,
  intermediate.dir = "./intermediate_output/",
  use.intermediate.result = TRUE,
  teff = TRUE,
  seff = TRUE,
  doy.break = NULL,
  cycle = FALSE,
  t.grid = NULL,
  t.grid.num = 50,
  clipRange = c(0, 1800),
  clipMethod = "nnr",
  var.est = FALSE
)

Arguments

year	vecotr of year
doy	vecotr of DOY (day of the year)
mat	a numeric matrix. Each row contains a row stacked image pixel values.
img.nrow	number of rows of the image
img.ncol	number of columns of the image
doyeval	a vector of DOY on which to get the mean and temporal imputation
h.tcov	bandwidth for temporal covariance estimation
h.tsigma2	bandwith for temporal variance estimation
h.scov	bandwidth for spatial covariance estimation
h.ssigma2	bandwidth for spatial variance estimation
nnr	maximum number of nearest neighbor pixels to use for spatial covariance estimation
outlier.action	"keep" to keep outliers; "remove" to replace outliers with imputed values
outlier.tol	The threshold to use to define outlier image. Default is 0.2, i.e. images with more than 20% outlier pixels are treated as outlier image.
intermediate.save	TRUE or FASLE; whether to save the intermediate results including mean, temporal effect and spacial effect imputation resutls. The intermediate results can be useful to avoid duplicating the computation for some imputation steps.
intermediate.dir	directory where to save the intermediate results
use.intermediate.result	whether to use the intermediate results in the 'intermediate.dir' folder. Default is TRUE.
teff	TRUE or FALSE, wheter to calculate the temporal effect. Default is TRUE.
seff	TRUE or FALSE, wheter to calculate the spatial effect. Default is TRUE.
doy.break	a vector of break points for doy where the spatial effect are estimated seperately on each interval. Default is NULL, i.e. the spatial effect is assumed to be the same over doy.
cycle	TRUE or FALSE. When doy.break is specified, whether to combine the first doy.break interval and the last doy.break together for spatial effect estimation.
t.grid	a vector of grid points on which to calculate the temporal covariance function
t.grid.num	number of grid points to use for temporal covariance estimation. Ignored if t.grid is given.
clipRange	passed to meanEst function
clipMethod	passed to meanEst function
var.est	Whether to estimate the variance of the temporal and spatial effects. Default is FALSE.

Value

List of length 4 with entries:

• imat: imputed matrix of mat

• smat: standard error matrix of the same size as mat

• idx: a list of image indexes

  • idx.allmissing: completely missing image indexes,

  • idx.partialmissing: partially observed image indexes,

  • idx.fullyobserved: fully observed image indexes,

  • idx.outlier: outlier image indexes.

• outlier: a list of image outliers information

  • outidx: image index with outlier pixels,

  • outpct: percentage of outlier pixels corresponding to outidx,

  • outlst: a list of the same length as outidx, with each list the missing pixel index.

Examples

library(doParallel)
library(raster)
library(rasterVis)
library(RColorBrewer)
dfB = landsat106[landsat106$year >= 2000,]
matB = as.matrix(dfB[,-c(1:2)])
year = dfB$year
doy = dfB$doy
if(require(doParallel))
  registerDoParallel(1)
res <- stfit_landsat(year, doy, matB, 31, 31, nnr=30,
use.intermediate.result = FALSE, intermediate.save = FALSE, var.est = TRUE)
## visualize the imputed results
idx = c(res$idx$idx.allmissing[150], res$idx$idx.partialmissing[c(30, 60, 90)])
rst_list = list()
for(i in 1:length(idx)){
  rst_list[(i-1)*3+1] = raster(matrix(matB[idx[i],], 31))
  rst_list[(i-1)*3+2] = raster(matrix(res$imat[idx[i],], 31))
  rst_list[(i-1)*3+3] = raster(matrix(res$sdmat[idx[i],], 31))
}
s = stack(rst_list)
levelplot(s, index.cond=list(c(seq(1, 12, 3), seq(2, 12, 3), seq(3, 12, 3))),
          par.setting = rasterTheme(panel.background=list(col="black"),
                                    region = brewer.pal(9, 'YlOrRd')),
          names.attr = c(rbind(paste0("Original ", idx), 
                               paste0("Imputed ", idx),
                               paste0("Std. Error ", idx))),
          layout = c(4,3))

stfit documentation built on Oct. 18, 2022, 5:07 p.m.