ResautoDownscale: ResautoDownscale

In downscaledl: Downscale of RS Images using Deep Learning

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)

Author(s)

Lianfa Li [e-mail]

See Also

AutoEncoderModel for the residual network function used in downscaling.

Examples

  #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

  #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) 

downscaledl documentation built on Oct. 30, 2019, 11:35 a.m.