[Qgis-developer] Unknown CRS problem

Martin Dobias wonder.sk at gmail.com
Mon Jan 10 18:43:26 EST 2011


Hi Marco

On Fri, Jan 7, 2011 at 9:20 PM, Marco Hugentobler
<marco.hugentobler at sourcepole.ch> wrote:
> Hi Martin
>
>> On-th-fly raster reprojection is not (IMHO) very expensive. The
>> complexity depends on the desired output quality (e.g. what type of
>> interpolation is used) and raster size. I would expect that most of
>> the raster rendering time is fetching of source data, the warping
>> itself is simple.
>
> Afaik raster reprojection is very expensive (see e.g.
> http://mapserver.org/input/raster.html#raster-warping). And the 2-4 times is
> only with umn mapservers clever linearisation along scanlines. With a normal
> warp, it will be much more.

Even though reprojection is expensive, it's probably better to
reproject the layer than to render it misaligned :-)
Anyway reprojecting every single point using the coordinate transform
formulas wastes a lot of processing power and should be done more
wisely. We have this problem also with reprojecting of vector layers
(every single points is reprojected using the expensive formula).


>> > THird: Having OTF automatic still won't help users with data that lacks a
>> > CRS.
>>
>> No, but at least they can get a notification that a layer has unknown
>> CRS and thus can be aligned incorrectly.
>
> I'm concerned that such a warning could scare newbie users who don't know
> about spatial reference systems at all (and it will be one of the first things
> they see if the datasource crs is not recognized. This happens frequently if
> the data comes from a commercial GIS).
>
> A very important use case here is people working with governemental data
> provided by the local GIS department. The GIS dep. provides everything in the
> same meter CRS and the normal user does not need to know about it, since
> everything overlays properly.

Ideally if all the data from government is in the same CRS, no
reprojecting will happen - because the project CRS will be the same as
the CRS of individual layers (thus no warning would be issued).

Martin


More information about the Qgis-developer mailing list