[GRASS-user] sample a strds at specific locations (areas)
Stefan Blumentrath
Stefan.Blumentrath at nina.no
Mon Aug 12 14:11:15 PDT 2019
Ciao Vero,
Currently, areas, lines and points are rasterized by default.
Centroids would come in addition.
That means, small polygons that only cover parts of a pixel would get rasterized never the less. This can of course be useful in some cases but cause issues in others. However, given that GRASS centroids are located inside areas it should be mostly OK.
Cheers
Stefan
From: Veronica Andreo <veroandreo at gmail.com>
Sent: mandag 12. august 2019 16:24
To: Stefan Blumentrath <Stefan.Blumentrath at nina.no>
Cc: Margherita Di Leo <diregola at gmail.com>; grass-user <grass-user at lists.osgeo.org>
Subject: Re: [GRASS-user] sample a strds at specific locations (areas)
Ciao Madi,
is the region properly set?
@Stefan in this case, if the centroid only is rasterized, then stats will also belong to that pixel only? AFAIU, we don't want that, but stats for the area. Or rasterization of centroid would have a different effect?
best,
Vero
El vie., 9 ago. 2019 a las 21:29, Stefan Blumentrath (<Stefan.Blumentrath at nina.no<mailto:Stefan.Blumentrath at nina.no>>) escribió:
Ciao Margherita,
This is strange. The area (I assume it is an area though your image just shows the boundary) should definitely get rasterized and thus analysed in v.strds.stats.
Can you post the output of:
v.category input=big_areas2 at PERMANENT option=report
and
v.info<http://v.info> test -t ?
I double-checked, and the possibility to rasterize also centroids is unfortunately not handed down to v.rast.stats (and moduls build ontop of that, like v.strds.stats). This should be done for GRASS version >= 7.6 (e.g. by adding centroids to the default selection of types to rasterize in v.to.rast) and is probably worth a ticket.
Cheers
Stefan
From: Margherita Di Leo <diregola at gmail.com<mailto:diregola at gmail.com>>
Sent: fredag 9. august 2019 16:51
To: Veronica Andreo <veroandreo at gmail.com<mailto:veroandreo at gmail.com>>
Cc: Stefan Blumentrath <Stefan.Blumentrath at nina.no<mailto:Stefan.Blumentrath at nina.no>>; grass-user <grass-user at lists.osgeo.org<mailto:grass-user at lists.osgeo.org>>
Subject: Re: [GRASS-user] sample a strds at specific locations (areas)
Hi,
excuse me if I return on this. I have again the same problem, of v.strds.stats skipping lots of polygons. Now I'm using a brand new dataset - strds of Sentinel 1 and a brand now vector of polygons - and the skipped polygons are not narrow, I'm sure that there are cells centroids in it. See for example screenshot attached, depicting a polygon that was skipped. I also tried to run v.strds.stats on that polygon alone, like:
v.strds.stats in=big_areas2 at PERMANENT where="cat == '1'" strds=db_cross_pol at scaled out=test method=average
v.info<http://v.info> test
+----------------------------------------------------------------------------+
| Name: test |
| Mapset: stats |
| Location: S1 |
| Database: /media/madi/TOSHIBA EXT/S1/grassdata |
| Title: Output from v.patch |
| Map scale: 1:1 |
| Name of creator: madi |
| Organization: |
| Source date: Fri Aug 9 12:12:49 2019 |
| Timestamp (first layer): none |
|----------------------------------------------------------------------------|
| Map format: native |
|----------------------------------------------------------------------------|
| Type of map: vector (level: 2) |
| |
| Number of points: 0 Number of centroids: 0 |
| Number of lines: 0 Number of boundaries: 0 |
| Number of areas: 0 Number of islands: 0 |
| |
| Map is 3D: No |
| Number of dblinks: 0 |
| |
| Projection: UTM (zone 34) |
| |
| N: 0 S: 0 |
| E: 0 W: 0 |
| |
| Digitization threshold: 0 |
| Comment: |
| |
+----------------------------------------------------------------------------+
I confess I hadn't look much further into it because I thought it was a problem with nodata, but this is not the case and I think it's worth of further investigation. Thanks for any pointers.
Regards,
On Thu, Feb 7, 2019 at 4:10 PM Margherita Di Leo <diregola at gmail.com<mailto:diregola at gmail.com>> wrote:
Thanks for testing, Vero. I assume it's due to a local problem then.
Cheers,
On Thu, Feb 7, 2019 at 3:29 PM Veronica Andreo <veroandreo at gmail.com<mailto:veroandreo at gmail.com>> wrote:
Hi Stefan and Madi,
Thanks Stefan for the explanations :) Indeed I agree that a flag to avoid the alignment to input rasters (and just use region settings) sounds good.
I tested what Madi said, but cannot reproduce in the climate NC location [0]. This is the command I used:
v.strds.stats in=boundary_county where="cat == '261'" strds=tempmean t_where="start_time >= '2012-01-01'" out=test
to make it faster (it feels indeed kinda slow for the whole vector and full time series), I selected only one polygon and a range of dates. I get the table as expected while leaving methods by default.
best,
Vero
[0] http://courses.ncsu.edu/mea592/common/media/02/nc_climate_spm_2000_2012.zip
El jue., 7 feb. 2019 a las 14:39, Margherita Di Leo (<diregola at gmail.com<mailto:diregola at gmail.com>>) escribió:
Thank you Stefan for your help! I figured what happens. v.strds.stats with default method produces in my case a corrupted output, topology is there but there's no table associated to it. If I specify method=average, I do obtain the table, and the mystery is solved: some of the polygons fall into a nodata (due to cloud mask). If anyone else can reproduce the corrupted table issue, I can file a ticket for that.
Thanks for help!
Regards,
On Thu, Feb 7, 2019 at 12:33 PM Stefan Blumentrath <Stefan.Blumentrath at nina.no<mailto:Stefan.Blumentrath at nina.no>> wrote:
Hi Madi,
With this combination (polygon size vs. raster resolution), the shape of the polygons can be an issue (narrow areas that do not cover the center of any pixel).
Debugging should be simple with v.db.select or v.extract.
Areas that did not get rasterized should be NULL in the column with statistics computed with v.rast.stats.
In verbose mode v.rast.stats (or probably even v.to.rast) should probably give a more informative Warning message (e.g. listing categories not rasterized). It also would help if you can rasterize the areas yourself and provide a raster with categories as (optional) input to v.rast.stats…
For high resolution data like yours, the speed improvement of multiple raster input might help quite a bit esp. with many maps in the time series. Will see if I can come up with a patch rather soon…
Cheers
Stefan
From: Margherita Di Leo <diregola at gmail.com<mailto:diregola at gmail.com>>
Sent: torsdag 7. februar 2019 10:37
To: Stefan Blumentrath <Stefan.Blumentrath at nina.no<mailto:Stefan.Blumentrath at nina.no>>
Cc: Veronica Andreo <veroandreo at gmail.com<mailto:veroandreo at gmail.com>>; grass-user <grass-user at lists.osgeo.org<mailto:grass-user at lists.osgeo.org>>
Subject: Re: [GRASS-user] sample a strds at specific locations (areas)
Hi,
thank you for your replies. To give a little more context: I selected my polygon areas to be > 0.5 ha each (this would be 50000 mq if I'm not mistaken) and I'm sampling NDVI maps at 10m resolution (the region being the same as NDVI maps). So I think I need an idea on how to debug the areas that were excluded to check them individually to see what could be the problem...
Regarding the alignment problem, if I understand it correctly: if the polygon doesn't include the *center* of the raster beneath it, can't retrieve the value and the polygon is discarded? But a value exists, so it would be correct that it returned a value in any case. But I admit I don't have a full grasp of the problem.
On Wed, Feb 6, 2019 at 10:52 PM Stefan Blumentrath <Stefan.Blumentrath at nina.no<mailto:Stefan.Blumentrath at nina.no>> wrote:
Hi Vero,
I think there is a little misunderstanding.
v.rast.stats did not change it behaviour with regards to the computational region (at least not if only one raster map is used). The alignment to the input raster (resolution) has been around since the module got ported to Python (like 10 years ago):
https://trac.osgeo.org/grass/browser/grass/trunk/scripts/v.rast.stats/v.rast.stats.py?rev=33522#L148
So, adding a flag for skipping the alignment was more an idea for an enhancement that allows the behaviour you seem to prefer (too).
Cheers
Stefan
From: Veronica Andreo <veroandreo at gmail.com<mailto:veroandreo at gmail.com>>
Sent: onsdag 6. februar 2019 21:38
To: Stefan Blumentrath <Stefan.Blumentrath at nina.no<mailto:Stefan.Blumentrath at nina.no>>
Cc: Margherita Di Leo <diregola at gmail.com<mailto:diregola at gmail.com>>; grass-user <grass-user at lists.osgeo.org<mailto:grass-user at lists.osgeo.org>>
Subject: Re: [GRASS-user] sample a strds at specific locations (areas)
I had a similar problem some time ago, just that it was not raster resolution, but region resolution that I changed to solve my problem (see this thread and MM's answer: http://osgeo-org.1560.x6.nabble.com/v-to-rast-for-polygons-not-overlapping-center-of-raster-cell-td5355686.html#a5355729)
IIUC, MM's proposed solution to my case then does not work anymore because v.to.rast call inside v.rast.stats is affected by the region alignment to the raster to be queried. So, the solution is indeed now, to change raster resolution... ? Then the region would be aligned to it (them)?
If one has large areas or long time series and has to resample all rasters to get smallish polygons rasterized, I do not see the advantage of this new behavior... but maybe I'm missing something
Cheers,
Vero
El mié., 6 feb. 2019 16:54, Stefan Blumentrath <Stefan.Blumentrath at nina.no<mailto:Stefan.Blumentrath at nina.no>> escribió:
Ciao Madi, Vero,
Starting with GRASS 7.6, also centroids are used to get the raster representation of your area vector map. That increases the likelihood of smaller areas to be rasterized.
Increasing the resolution of the current region alone does not help, because v.rast.stats temporarily changes the computational region to align with the input raster map(s) (see also: https://trac.osgeo.org/grass/ticket/3523 and https://trac.osgeo.org/grass/ticket/3598 for discussion) Even if the first ticket is closed, comments are welcome.
It might make sense to add a flag to v.rast.stats like in r.slope.aspect to not align the computational region.
Furthermore, with regards to efficiency, v.strds.stats could gain some speed if multi-raster support in v.rast.stats - added in G 7.6 - would be handed down to the addon. Might almost double the speed for larger STRDS…
Cheers
Stefan
From: grass-user <grass-user-bounces at lists.osgeo.org<mailto:grass-user-bounces at lists.osgeo.org>> On Behalf Of Veronica Andreo
Sent: onsdag 6. februar 2019 17:20
To: Margherita Di Leo <diregola at gmail.com<mailto:diregola at gmail.com>>
Cc: GRASS user list <grass-user at lists.osgeo.org<mailto:grass-user at lists.osgeo.org>>
Subject: Re: [GRASS-user] sample a strds at specific locations (areas)
Hi Madi
El mié., 6 feb. 2019 a las 16:31, Margherita Di Leo (<diregola at gmail.com<mailto:diregola at gmail.com>>) escribió:
I have a question regarding v.strds.stats. I get the following warning message:
WARNING: Not all vector categories converted to raster. Converted 120 of 265.
What could be the reason for that?
Some vector areas might not be converted because they are too small with respect to the pixel size that you try to query. Others will tell better but I think the polygon must overlap the center of the pixel in order to be converted into raster. One solution could be to resample your rasters to a higher resolution.
HTH,
Vero
--
Margherita Di Leo
--
Margherita Di Leo
--
Margherita Di Leo
--
Margherita Di Leo
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/grass-user/attachments/20190812/3fb8ad85/attachment-0001.html>
More information about the grass-user
mailing list