Interpolating Gridded 3D Data to a finer scale

Question

I have a NetCDF file of a probability surface. It's a 30x30 grid of 0.25 degree lat/lon intervals with a probability surface described in the z dimension. I can easily import this into Panoply, a NetCDF viewer:

And it's then a breeze (checking one box) to interpolate/smooth the raw data to a finer grid size:

However, I don't just want to visualize the data, I want to plot it in R along with bathymetry and point data. That all is no problem, but I have not found a straightforward way to interpolate the gridded data in R. Here's the code I use to import and plot the data:

library(RNetCDF)

nc <- open.nc("132235-1.nc")
print.nc(nc)
tmp <- read.nc(nc)
probs<-tmp$likelihoods

xran <- range(tmp$longitude)
yran <- range(tmp$latitude)
zran <- range(probs,na.rm=T)
lon <- tmp$longitude
lat <- tmp$latitude[30:1]

z <- array(probs, dim=dim(probs))

z <- z[,rev(seq(ncol(z)))]
z <- z[,seq(ncol(z))]



prob.pal<-colorRampPalette(
  c("#C1FFC1","#8FBC8F","#2F4F4F")
)

zbreaks <- seq(0.0001, 0.063, by=0.001)

cols<- c(prob.pal(length(zbreaks)-1))

png("ProbTest.png", width=7.5, height=6, units="in", res=200)
layout(matrix(1:2, 1,2), widths=c(6,1.5), heights=c(6))

par(mar=c(2,2,1,1), ps=10)
image(lon, lat, z=z, col=cols, breaks=zbreaks, useRaster=TRUE, ylim=c(13,28), xlim=c(-115,-100))

dev.off()

And I end up with this, which is the same as using Panoply but with a different color scheme:

Is there a straightforward way to interpolate/smooth this data? I know how to create kernel utilization densities etc. using point data, but not using gridded data.

Many thanks for your assistance!

Wieche

I'd use the raster package to first read in the NetCDF file using raster() or brick() and to then smooth the data to a finer-scale resolution using resample(). – Andyane

Answer 1

This is the solution I think you're looking for, which uses bilinear resampling. However this is not the only way to do such interpolation and you'd likely need to justify not using a more sophisticated approach (e.g. geostatistical, splines, etc.):

library(raster)
set.seed(2002)

##  Create an extent boundary:
ex <- extent(c(0, 20, 0, 20))

##  Simulate a coarse raster:
r.small <- raster(ex, vals=rnorm(10*10, mean=5, sd=1), nrow=10, ncol=10)

##  Simulate the grid of a finer-scale raster:
r.big <- raster(ncol=200, nrow=200, ext=ex)

##  Resample the coarser raster to match finer grid:
r.res <- resample(x=r.small, y=r.big, method="bilinear")

plot(r.small)
plot(r.res)

Coarse:

Fine:

Klotz

Hi Forrest—this most certainly did the trick, and solved a lot of other problems I was having, as I didn't realize raster could read NetCDF files. Many thanks for the solution! – Wieche

You're very welcome, but I should say thank you to Robert Hijmans and the rest of the contributors to the raster package as their hard work makes this kind of thing very easy in R. – Klotz