R-sig-Geo Digest, Vol 153, Issue 4

Hi,
I think there is a more fundamental problem that you should consider. The
use of p values and set significant levels are seriously being questioned
as good statistical modelling practises. You might want to look at the
recent policy statement by the American Statistical Association on this
topic. There is even a strong movement to convince editors not to review
papers which base the results on the above methods or at least advise the
use of different procedures.
If you are interested in reading this policy statement I would be happy to
send a pdf copy.

Cheers,
John Lewis

----------------------------------------------------------------------

Message: 1
Date: Tue, 3 May 2016 23:11:29 +0000 (UTC)
From: "Thiago V. dos Santos" <thi_veloso at yahoo.com.br>
To: R-sig-geo Mailing List <r-sig-geo at r-project.org>
Subject: [R-sig-Geo] Mask a map using statistical significance
Message-ID:
       <1350390896.5790739.1462317089174.JavaMail.yahoo at mail.yahoo.com>
Content-Type: text/plain; charset=UTF-8

Dear all,

In climate studies, it is a common practice to perform some statistical
test between two fields (maps), and then plot the resulting map using a
significance mask. This masking is usually done by adding some kind of
pattern (shading, crosshatching etc) on top of the actual color palette.

Examples can be seen here in this image https://www3.epa.gov/climatechange
/images/science/GlobalPrecipMap-large.png and in the left images of this
panel:
http://www.nature.com/nclimate/journal/vaop/ncurrent
/images_article/nclimate2996-f3.jpg
In my case, I ran a statistical test for detecting trend on a time-series
raster and I now have one raster with the trend for rainfall (in degree C
per year) and one with the p-values associated to the test.

My data looks roughly like this:

require(raster)

## scratch raster objects and fill them with some random values
r.trend <- raster(nrows=50, ncols=50, xmn=-58, xmx=-48, ymn=-33, ymx=-22)
r.trend[] <- round(runif(50 * 50, min=0, max=3), digits=2)

r.pvalue <- raster(nrows=50, ncols=50, xmn=-58, xmx=-48, ymn=-33, ymx=-22)
r.pvalue[] <- round(runif(50 * 50, min=0.0001, max=0.1), digits=5)


What I would like to do is to plot r.trend, and on top of it plot r.pvalue
(as a pattern) where r.pvalues < 0.01 (corresponding to a significance
level of 99%).

Has anyone here ever tried to do a similar plot and can point me to any
direction?

I usually use rasterVis and ggplot2 to plot maps, but I would be open to
try some other package and even other SIG software other than R.

Many thanks in advance, -- Thiago V. dos Santos

PhD student
Land and Atmospheric Science
University of Minnesota



------------------------------

Subject: Digest Footer