[GRASSLIST:1096] Re: Reproject raster data - weighted average

[Roger Bivand]

On Wed, 3 Sep 2003, David Orme wrote:

> Hi,
> 
> Providing more information... I need to transform some global data sets 
> in geographic resolutions, for example quarter degree, into a 
> cylindrical equal area grid. The resolution of the CEA location has 
> been chosen so that it matches the geographic resolution of the data 
> set at the latitude of true scale - as a result the longitudinal 
> divisions of the two grids coincide but the latitudinal lines divisions 
> differ (except at the equator - the origin of both grids). What I would 
> like to be able to do is to calculate the values of the new grid as an 
> average of the original cells underlying each new cell weighted by 
> their contribution to the area of the new cell. As a toy example, say I 
> have three cells going north from the equator with values 5, 7, 8 and 
> the new grid happens to form two cells with a division half way through 
> the second cell. The first cell in the new grid should contain the mean 
> of 5 and 7 weighted by 1 and 0.5 (i.e. 5.6667) - the second the 
> weighted mean of 7 and 8 weighted by 0.5 and 1 (7.6667).
> 
> Paul and Rado (thanks for the amazingly fast replies!) suggested:
> 
> 1) Interpolation from a point surface - I think this would give the 
> correct answer if the interpolation neighbourhood was set up correctly. 
> Unfortunately because I'm going to an equal area grid, the 
> neighbourhood for cells in the new grid varies with latitude.
> 
> 2) r.neighbours - again I think this would give the correct answer 
> locally but because the CEA and geographic grids do not overlie each 
> other in a regular fashion it won't work beyond that, unless there is a 
> way of feeding a grid with varying North South resolution into 
> r.neighbours, which I seriously doubt.
> 
> I can think of a way of doing it using a transformation matrix in R, 
> for instance, but I wondered if I was reinventing the wheel.

Were it not for your wanting a weighted average, I think gdalwarp would do 
this, but it is restricted to nearest neighbour resampling, as I 
understand.

Roger


> 
> Thanks,
> David
> 
> 

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Breiviksveien 40, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 93 93
e-mail: Roger.Bivand at nhh.no