Add information from a raster nc-file to a polygon shapefile

Question

DATA

Download from this website the following ZIP file: NUTS_2006_60M_SH.zip and save all the NUTS_RG_60M_2006 files. In what follows I refer to this as "NUTS3 regions".

Download from this website the following nc-file: cru_ts3.23.1991.2000.pet.dat.nc.gz This is a raster with data all over the world.

Desired output

The nc-file is a raster of the world with values in different colors.

The goal is now to translate these data to a NUTS3 polygon regions and save it back as a nuts3-shapefile with the data added to the polyons.

Software

I can work with R, but since I couldn't make it work in R, I switched to ArcGIS. If the data are added to the NUTS3 shapefile, I can continue working with it in R.

Any solution in R or ArcGIS is welcome, as long as the output is 1 polygon shapefile with all the values added.

Problem

I can open the nc-file in ArcGIS (toolbox, multi-dimension tools, make netCDF Raster Layer). I can also overlap the nuts3-file with the raster. But I can't merge the data together in one shapefile (by taking the average of the underlying rasters that overlap with one polygon) and drop all the data that I don't need.

I found some examples on how to do this on Google, but only one of them explained how to start from an nc-file. As far as I understood, I first have to right click on the data-layer to export the data as a raster (as it is now only an nc-file and not yet a shapefile). However, this is not working since I get the following error: "Output file format with specified pixel type or number of bands or colormap is not supported."

Answer 1

1. Under ArcMap: Toolbox\Multidimension tools\Make netCDF Raster Layer: Raster layer will appear in TOC of map document
2. In ArcMap-document TOC: Right click the Raster layer entry and convert to e.g. TIFF-file. I do not think that you can convert straight away to ESRI-GRID-format
3. Toolbox\Conversion tools\ToRaster\Raster to other format (multiple): choose ESRI GRID
4. Toolbox: Spatial Analyst Tools\Zonal Statistics\Zonal Statistics (as table). Use the NUTS-dataset for input and the ESRI-GRID as the input value raster. Specify the statistic you need.
5. Cheers

Answer 2

I found an answer on the following website on how to do this in R. The solution that is given above also works, but I am having memory problems to run it.

http://article.gmane.org/gmane.comp.lang.r.geo/24010/match=

##########################################################################
library(fields)
library(chron)
library(ncdf4)

setwd("C:/Users/lucp8042/Desktop/CRU")

## we have the files
nutsdsn <- "NUTS_2006_60M_SH/shape/data"
nutslayer <- "NUTS_RG_60M_2006" ## bingo, polygons


library(rgdal)  ## to read shapefile
admin  <- readOGR(".", layer = "NUTS_RG_60M_2006")


##### PET #####
setwd("C:/Users/lucp8042/Desktop/CRU/Pet")
ncfile <- "cru_ts3.23.1981.1990.pet.dat.nc"


library(raster)
## looks fine
plot(raster(ncfile))  ## this gets the first layer (or "band")
plot(admin, add = TRUE)

admin

## get every layer (without actually loading necessarily)
(cru_ts3 <- brick(ncfile))


## extract everything, collapse to mean values

## note that this will automatically transform to the raster
## and warn, you should explore what this does (the plot above does not for example)
ex <- extract(cru_ts3, admin, fun = mean)
dim(ex)
## [1] 1927  120

## tidy up
d <- as.data.frame(ex)
row.names(d) <- row.names(admin)