[R-sig-Geo] fix color scale

Mon Nov 8 09:34:30 CET 2010

To add to Edzer's answer, I had written this earlier - I just don't
know anything about the way grid/lattice can handle rasters.

The image/grid handling with sp/raster can be fast for two main reasons:

- image.SpatialGridDataFrame now can use rasterImage() to plot grids,
so the graphics-memory footprint is minimized, though the full grid is
still in R memory

- the raster package adds memory-management for grids, reading only
what is required on demand for plotting - I think raster still uses
image() for plotting, but it's only ever reading the required data for
a given level of detail, saving the graphics engine from the full
overhead demanded by image()

Those are good improvements that both provide different implications
for grid/lattice and thereby ggplot images.

Cheers, Mike.

On Mon, Nov 8, 2010 at 5:53 PM, Edzer Pebesma
<edzer.pebesma at uni-muenster.de> wrote:
> Dylan, I believe that lattice (grid) recently improved the plotting of
> rasters quite a bit, by using panel.levelplot.raster instead of
> panel.levelplot - some function in grid now renders rasters faster on
> devices for which this is optimed.
>
> spplot also has to transform the data to long format, by calling
> sp:::spmap.to.lev, which in turn uses a stack method for
> SpatialPointsDataFrame objects; this in turn uses stack.data.frame
>
> Bests,
>
>
> On 11/08/2010 05:00 AM, Dylan Beaudette wrote:
>> Interesting demonstration. However, it seems like sp objects and
>> spplot are still more efficient at storing / plotting large grids. The
>> example listed below required over 2 minutes to complete on a 2.4 Ghz
>> Intel processor. Also, storing the coordinates + data in long format
>> could easily fill available memory with large rasters. I have found
>> that storing multiple raster attributes as columns in the @data slot
>> of a SpatialGridDataFrame object give me flexibility to plot images
>> with a consistent color scheme, without needing to store the data in
>> long format. Also with lattice, the plots are generated in a couple of
>> seconds (500x700 px grid). I would be very interested in knowing how
>> to make ggplot() any faster.
>>
>> Cheers!
>> Dylan
>>
>> On Sun, Nov 7, 2010 at 11:55 AM, Edzer Pebesma
>> <edzer.pebesma at uni-muenster.de> wrote:
>>> Paul, this nicely illustrates the power of ggplot2.
>>>
>>> In the resulting plot, however, it seems to me that the
>>>
>>> + opts(aspect.ratio = 1)
>>>
>>> does not result in the desired effect that 1 m in the x direction equals
>>> 1 m in the y direction. Standard plot (asp = 1), and lattice plots (asp
>>> = "iso") have this; what does ggplot2 need?
>>>
>>> On 10/31/2010 11:31 AM, Paul Hiemstra wrote:
>>>> Hi Peter,
>>>>
>>>> When creating such a large amount of illustrations with the same
>>>> colorscale, I automatically think of lattice graphics. Under the hood
>>>> spplot also uses lattice graphics. Take a look at the levelplot()
>>>> function from lattice which produces the grid plots for spplot (if I'm
>>>> correct). Alternatively, I've been using ggplot now for quite a while to
>>>> make plots of a lot of grids. A small example says more than a thousand
>>>> words:
>>>>
>>>> library(ggplot2)
>>>> library(sp)
>>>>
>>>> data(meuse.grid)
>>>> summary(meuse.grid)
>>>>
>>>> # Note that I do not transform meuse.grid to SpatialPixelsDataFrame
>>>> # Let's make a simple grid plot
>>>> dum = meuse.grid[c("x","y","dist")]
>>>> ggplot(aes(x = x, y = y, fill = dist), data = dum) + geom_tile()
>>>>
>>>> # Let's make a few more attributes to the grid
>>>> # could be measurements on other dates for example
>>>> new_atts = do.call("cbind", lapply(1:100, function(num) dum$dist +
>>>> runif(dum$dist)))
>>>> summary(new_atts)
>>>> dum = data.frame(cbind(dum, new_atts))
>>>>
>>>> # Important step now is to
>>>> # restructure the data
>>>> dum_ggplot = melt(dum, id.vars = c("x","y"))
>>>>
>>>> # Now make a plot using dum_ggplot
>>>> # of 'x' and 'y' using value as a 'fill'
>>>> # with a plot per 'variable', can take a minute to plot
>>>> ggplot(aes(x = x, y = y, fill = value), data = dum_ggplot) + geom_tile()
>>>> + facet_wrap(~variable) +
>>>>       scale_x_continuous('', labels = NA, breaks = NA) +
>>>>       scale_y_continuous('', labels = NA, breaks = NA) +
>>>> opts(aspect.ratio = 1)
>>>> # These last two lines get rid of the labels on the axes and set aspect
>>>> ratio to 1
>>>>
>>>> Now you have a plot with 101 maps with the same colorscale, with ggplot
>>>> doing all the hard work. It takes some time to get the hang of ggplot,
>>>> but I think it is worth the investment, also for spatial plots.
>>>>
>>>> cheers and hope this helps,
>>>> Paul
>>>>
>>>> On 10/28/2010 09:12 PM, Peter Larson wrote:
>>>>> Hello!
>>>>>
>>>>> I have a problem.
>>>>>
>>>>> I am using IDW to interpolate a daily series of geospatial
>>>>> observations. Thus, I want to produce a large number of sequential
>>>>> maps.
>>>>>
>>>>> I want them to all represent the same color scale. Is there any way to
>>>>> fix the color scale so that it is the same for all the plots?
>>>>>
>>>>> Thanks,
>>>>>
>>>>> Pete
>>>>>
>>>>> _______________________________________________
>>>>> R-sig-Geo mailing list
>>>>> R-sig-Geo at stat.math.ethz.ch
>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>>>>
>>>>
>>>>
>>>
>>> --
>>> Edzer Pebesma
>>> Institute for Geoinformatics (ifgi), University of Münster
>>> Weseler Straße 253, 48151 Münster, Germany. Phone: +49 251
>>> 8333081, Fax: +49 251 8339763  http://ifgi.uni-muenster.de
>>> http://www.52north.org/geostatistics      e.pebesma at wwu.de
>>>
>>> _______________________________________________
>>> R-sig-Geo mailing list
>>> R-sig-Geo at stat.math.ethz.ch
>>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>>
>
> --
> Edzer Pebesma
> Institute for Geoinformatics (ifgi), University of Münster
> Weseler Straße 253, 48151 Münster, Germany. Phone: +49 251
> 8333081, Fax: +49 251 8339763  http://ifgi.uni-muenster.de
> http://www.52north.org/geostatistics      e.pebesma at wwu.de
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>

-- 
Michael Sumner
Institute for Marine and Antarctic Studies, University of Tasmania
Hobart, Australia
e-mail: mdsumner at gmail.com