downscaledl (version 1.0)

ResautoDownscale: ResautoDownscale

Description

This function is the iterative implementation of downscaling with autoencoder based residual network.

Usage

ResautoDownscale(r2, fpredict0, c.grid, ss = 0.2, nepoch = 30,
  cores = 1, thresh = 0.01, ntime = 5)

Arguments

r2

Stack for the covariates for downscaling

fpredict0

Target fine-resolution image to be downscaled, you can just provide a NA image data (this function can fill this NA) or an initial predictions obtained by other methods.

c.grid

Coarsely resolved grid

ss

Sampling propotion for independent test

nepoch

Numder of epoch for residual network training (default: 30)

cores

Number of CPU cores used for computing (default: 1)

thresh

Stopping creation shreshold (default: 0.01)

ntime

Maximum number of iterations (default: 5)

Value

List(performance metrics such as R2, RMSE, and downscaled images)

See Also

AutoEncoderModel for the residual network function used in downscaling.

Examples

# NOT RUN {
  #Load the high-resolution raster of a covariate, 
  # elevation to be used for downscaling  
  eleFile=file.path(system.file(package = "downscaledl"), "extdata", "sample_ele.tif")
  ele=raster::raster(eleFile)

  #Load the coarse-resolution raster of the target variable to be downscaled  
  coarseFile=file.path(system.file(package = "downscaledl"), "extdata", "sample_coarse_res.tif")
  coarseRst=raster::raster(coarseFile)

  #Extract x and y to be used as two predictors in downscaling 
  xRast=ele
  yRast=ele
  pos=raster::rasterToPoints(ele,spatial=TRUE)
  cell=raster::cellFromXY(xRast,pos)
  xyc=sp::coordinates(pos)
  xRast[cell]=xyc[,"x"]
  yRast[cell]=xyc[,"y"]

  #Merge the covariates 
  covStk=raster::stack(xRast,yRast,ele)
  names(covStk)=c("x","y","ele")

  #Use the fine-resolution covariate (elevation) as the target image 
  fineTarget=ele
# }
# NOT RUN {
  #Set the paramneters and start to downscale ...  
  ares=ResautoDownscale(covStk,fineTarget,coarseRst,ss= 0.2, cores= 5, thresh = 0.01,ntime=3)
  
  #Show the iteration results 
  message(paste(capture.output(ares$diogRMSE), collapse = "\n",sep=""))
  
  #Show the optimal results in the final predictions 
  message(paste("test R2:",round(ares$r2,2),",test RMSE:",round(ares$rmse,4),sep="")) 
  
  #Obtain the downscaled
  downscaled_img=ares$raster
  
  #Save the current par setting 
  curpar = par(no.readonly = TRUE)

  #Set the new par setting 
  par(mfrow=c(1,2),mar=c(1,1,1,1))
  
  #Show the final predictions of fine resolution and 
  # original coarse-resolution image for a comparison 
  raster::plot(coarseRst)
  raster::plot(downscaled_img)
  
  #Restore the previous par setting 
  par(curpar) 
# }