Skip to content

lapply and SpatialGridDataFrame

2 messages · Irucka Embry, Roger Bivand

Original
Roger Bivand
#
On Tue, 29 Jan 2013, Irucka Embry wrote:

            

      
      
      
    

        
You posted to the second list when nobody replied within two hours - some 
of us sleep at some time of the day. The two replies are equivalent, and 
both indicate that either you didn't read the documentation of lapply, or 
didn't understand what you read.

            

      
      
      
    

        
Wrong, you mean object, not file. You also need to learn about trapping 
errors. Simplify what you are doing as a loop, as most of this is 
confusing because you've written it in a confusing way (don't write a 
single line function, it is pointless). In addition, you have not said 
what you are actually doing.

You are reading 54 ASCII Grid files into SpatialGridDataFrame objects, 
then using Grid2Polygons in the Grid2Polygons package to make a simple 
conversion of each input grid cell (pixel) to a Polygons object. The 
output SpatialPolygonsDataFrame objects are simple copies of the input, so 
you do not use the functionality of Grid2Polygons to aggregate to blocks 
of pixels; you could use coercion in sp instead.

However, in your message to R-help, you showed that your goal was to use 
BayesX. So is your actual aim to stack 54 rasters (possibly 54 time 
periods???), create a single set of SpatialPolygons, output to Shapefile, 
read in through the functions in BayesX (which are not based on sp classes 
- had they been, you wouldn't need to write out and read in again), then 
run some space-time analysis?

Or are the 54 grids all different - in which case get your whole workflow 
functioning with a single raster first, and then iterate that in a loop, 
storing results in a list. If your raster has many cells, you may find 
that the time it takes to run BayesX makes the time taken to convert 
SpatialGrid -> SpatialPolygons very small, but then again, it could be the 
use of Grid2Polygons when its advanced features are not needed:

            

      
      
      
    

        
user  system elapsed
  36.129   0.832  37.013

            

      
      
      
    

        
user  system elapsed
   2.875   0.043   2.945

If the SpatialGridDataFrame objects all share the same GridTopology, you 
should make a single SpatialGridDataFrame in a loop, or a single 
RasterBrick or RasterStack in the raster package, and go from there. But 
your proposed workflow is so opaque that helping you fix lapply isn't 
going to move you forward at all.

You must think through your workflow carefully - it is very possible that 
you can create the BayesX bnd object directly from the GridTopology of the 
input data, without any intermediate SpatialPolygons objects or 
shapefiles. The authors of BayesX would have done everyone a big favour if 
they had supported sp and spacetime classes, for example by coercion - 
spatstat is a good example of how to use both sp and package-specific 
classes, and possibly used facilities in sp and spacetime for 
visualisation.

Roger