[R-sig-Geo] stars::RasterIO using extent info?

[Howard, Tim G (DEC)]

Mike, 
Thanks for the feedback - I'll check it out. 
Best, 
Tim

From: Michael Sumner [mailto:mdsumner using gmail.com] 
Sent: Wednesday, November 14, 2018 5:55 PM
To: Howard, Tim G (DEC) <tim.howard using dec.ny.gov>
Cc: r-sig-geo using r-project.org
Subject: Re: [R-sig-Geo] stars::RasterIO using extent info?

FWIW I have a raster-delivering-front-end for RasterIO in this dev package:  

https://github.com/hypertidy/lazyraster

It uses the more obvious extent() idioms and will even use an open plot if nothing else is specified. 

(It uses an independent binding to GDAL in the vapour package). 

That might help, or not. It's on my list to find a sensible way for stars to leverage this obvious ease-of-use.  rgdal::readGDAL always had an interface to RasterIO, but only raster ever made use of that, and it's indirect - raster doesn't allow a mix of extent and resolution in its approach. 

Cheers, Mike. 

On Thu, 15 Nov 2018 at 02:27 Howard, Tim G (DEC) via R-sig-Geo <mailto:r-sig-geo using r-project.org> wrote:
Ok, fair enough that there's no magic involved. I've worked through the details with the small example as follows. The result is only a couple of cells different in each direction.

library(stars)
tif <- system.file("tif/L7_ETMs.tif", package = "stars")
x <- read_stars(tif) 
x <- x[,,,1]
# calculate a circular polygon at the center of the raster
pol <- x %>% st_bbox() %>% st_as_sfc() %>% st_centroid() %>% st_buffer(500)

bb_pol <- st_bbox(pol)

xoff <- st_dimensions(x)$x$offset
xdelt <- st_dimensions(x)$x$delta
yoff <- st_dimensions(x)$y$offset
ydelt <- st_dimensions(x)$y$delta
cropXoff <- (bb_pol$xmin - xoff + xdelt)/xdelt
cropXsize <- (bb_pol$xmax - bb_pol$xmin)/xdelt
cropYoff <- (bb_pol$ymin - yoff + ydelt)/ydelt
cropYsize <- (bb_pol$ymax - bb_pol$ymin)/ydelt

# if ydelt is negative, get abs of ysize and move yoffset to the top
if(cropYsize < 0){
  cropYsize <- abs(cropYsize)
  cropYoff <- cropYoff - cropYsize  
}

rasterio <- list(nXOff = cropXoff, nYOff = cropYoff, nXSize = cropXsize, nYSize = cropYsize, bands = c(1))
(z = read_stars(tif, RasterIO = rasterio))
plot(z)

# crop it if desired
plot(z[pol])

## compare to proxy method

x <- read_stars(tif, proxy = TRUE) 
x <- x[,,,1]
y <- st_as_stars(x[pol])
plot(y)

# only a couple of cells off!
z
y




------------------------------
Date: Tue, 13 Nov 2018 17:26:16 +0100
From: Edzer Pebesma <mailto:edzer.pebesma using uni-muenster.de>
To: mailto:r-sig-geo using r-project.org
Subject: Re: [R-sig-Geo] stars::RasterIO using extent info?
Message-ID: <mailto:9d037da7-dc9c-9886-d6fc-5864cf8b4a17 using uni-muenster.de>
Content-Type: text/plain; charset="utf-8"



On 11/13/18 4:10 PM, Howard, Tim G (DEC) via R-sig-Geo wrote:
> Dear list,
>
> I am exploring the different options for reading parts of large imagery object in stars, as discussed here:
>
> https://r-spatial.github.io/stars/articles/proxy.html
>
> My ultimate goal is to read into RAM only a clipped portion of a large raster (well, actually a raster stack, but taking baby steps here).
>
> My immediate question: the `RasterIO` option of read_stars defines cell offsets and cell counts (*Size). Is there a straightforward way to calculate these values given extent information?
>
> Reproducible example (mostly taken from here:  https://www.r-spatial.org/r/2018/03/22/stars2.html):
>
> library(stars)
> tif <- system.file("tif/L7_ETMs.tif", package = "stars")
> x <- read_stars(tif) # read entire tif into ram
> x <- x[,,,1] #get just one layer for now
> # calculate a circular polygon at the center of the raster
> pol <- x %>% st_bbox() %>% st_as_sfc() %>% st_centroid() %>% st_buffer(500)
> plot(x)
> # interestingly, I don't think the circle is in the right place when plotted
> plot(st_geometry(pol), add = TRUE, border = "red")
> # this is what I'd like to be able to restrict to what is read in memory:
> plot(x[pol])
>
> ## read only portion of tif using proxy object
> x <- read_stars(tif, proxy = TRUE)
> x <- x[,,,1]
> y <- st_as_stars(x[pol])
> plot(y) # this is cropped to the extent (but not the circle - let's not worry about that right now)
>
> Question: can I do the equivalent with the RasterIO options in stars?  Said another way, instead of setting up the proxy, can I map my extent object (or bounding box) directly to the cell count values needed for RasterIO?

stars can do the math, and so can you; it is explained here:

https://r-spatial.github.io/stars/articles/data_model.html

stars uses some functions directly from GDAL which it doesn't expose to
the user, but there is no magic going on here.

>
>
> Thanks in advance for any tips.
> Tim
>
> _______________________________________________
> R-sig-Geo mailing list
> mailto:R-sig-Geo using r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>

--
Edzer Pebesma
Institute for Geoinformatics
Heisenbergstrasse 2, 48151 Muenster, Germany
Phone: tel:+49%20251%208333081

_______________________________________________
R-sig-Geo mailing list
mailto:R-sig-Geo using r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
-- 
Dr. Michael Sumner
Software and Database Engineer
Australian Antarctic Division
203 Channel Highway
Kingston Tasmania 7050 Australia