[GRASS-dev] r.resamp.stats error

[Glynn Clements]

Martin Wegmann wrote:

> I found a weird behaviour in r.resamp.stats which causes a shift of the 
> resulting image but only on a large scale (subcontinental scale, shift to 
> south-west).
> 
> I tried to reproduce it with the spearfish dataset but failed.
> 
> Doing it on Africa srtm DEM (res based on gtopo30 DEM):
> 
> g.region -g
> ...
> res=0.00083333
> ..
> 
> g.region res=0.0083333
> r.resamp.stats input=dem output=dem_resampstats method=average
> 
> produces this output of a small region with islands shown in the jpg (overlay 
> with original raster)
> 
> any idea how to fix this? regards, Martin

--- raster/r.resamp.stats/main.c	15 Nov 2006 03:57:24 -0000	1.8
+++ raster/r.resamp.stats/main.c	7 Dec 2006 05:59:11 -0000
@@ -80,13 +80,13 @@
 	for (col = 0; col <= dst_w.cols; col++)
 	{
 		double x = G_col_to_easting(col, &dst_w);
-		col_map[col] = (int) floor(G_easting_to_col(x + 0.5, &src_w));
+		col_map[col] = (int) floor(G_easting_to_col(x, &src_w) + 0.5);
 	}
 
 	for (row = 0; row <= dst_w.rows; row++)
 	{
 		double y = G_row_to_northing(row, &dst_w);
-		row_map[row] = (int) floor(G_northing_to_row(y + 0.5, &src_w));
+		row_map[row] = (int) floor(G_northing_to_row(y, &src_w) + 0.5);
 	}
 
Instead of adding half a cell (to locate the midpoint rather than the
north-west corner), I was adding half a map unit.

For a projected location, the offset was half a metre (or foot), which
is negligible. For a lat-lon location, the offset was half a degree,
which is rather more significant.

Fixed in CVS.

-- 
Glynn Clements <glynn at gclements.plus.com>