[GIS] Fixing polygon self-intersection in QGIS

polygonqgis-2self-intersection

I am attempting to mask a raster to US Forest Service boundaries in QGIS. It fails due to self – intersections of polygons.

Warning 1: Ring Self-intersection at or near point -109.90112962000001 43.649977879999994
ERROR 1: Cutline polygon is invalid.

I have run the check validity tool and a new shapefile is created but it too finds the same error when attempting to mask the raster to the "valid" shapefile.

I have also attempted to dive into the v.clean tool with unsuccessful results using both bpol and snap separately. I don't fully understand what this tool is doing, but it outputs a number of polygons that match some Forest Service land but excludes the majority of their land.

I also ran v.build.check to find topological errors and it created an output that is seemingly empty. Nothing in the attribute table, nothing visible on the map.

There have been similar questions on here but I haven't been able to make their solutions work.

I downloaded and used GRASS to attempt to fix the self-intersecting polygons. GRASS upon v.in.ogr recommended I snap at 1e-6. I did this and the output said I had 1 area overlapping left when previously I had thousands. So I saved the shapefile and imported to QGIS to being my analysis over there. The check validity tool found 185 self-intersecting polygons.

Still unable to use the clipper to create a Mask, I decided to look at the error output on the Map. The snapping tool isn't fixing these nodes.

Here's a larger picture of the area.

I wonder if it would be easiest just to dissolve these shapes in some way.

Best Answer

When I get errors like this I usually find applying a zero distance buffer can fix the geometries. So use the buffer tool in QGIS and set the buffer distance to 0 units. Then try and use this buffered layer as the mask.

Related Solutions

[GIS] Self intersections for polygons, how to solve

Is it feasible to just manually edit the vertices and remove the self intersections?

I do this a lot (using QGIS because it seems to be less picky about this issue). QGIS also has a 'Check Geometry' tool that will show you where all the self intersections are.

I know you said you're running arc but this might be a helpful approach if nothing else is working.

[GIS] Improving Clip Raster workflow in QGIS

Ok, having learnt a lot of stuff in the months since I asked this question, here are a couple of options:

In R, very similar to Bastien's answer:

library(raster)

rootdir <- 'C:/Users/obrl_soil/Downloads'
setwd(rootdir)
tinyraster <- raster('tinyrast.tif')
tinypolygon <- readOGR('tinypolygon.shp')

# alter your mask polygon to line up with the nearest pixel edges
tpoly_aligned <- alignExtent(tinypolygon, tinyraster, snap='near')

# clip and export in one hit
tinyclip <- crop(tinyraster, tpoly_aligned, filename='tinyclip.tif')

I noticed when checking the outputs in QGIS that writeRaster introduced some spurious data in the output tif, but crop didn't - e.g. input elevation for a pixel was 108.10545, output with crop was the same, but output with writeRaster was 108.10545349121094.

Alternate workflow - get the polygon bounding box coordinates and feed them into gdal_translate using -projwin. Just be careful which version of gdal_translate you use!. 1.11 is fine, and 2.1.2 should be when its released.

You can also do this in R, like

gtrans111 <- 'C:/Program Files/GDAL/gdal_translate.exe'
tinypolygon <- readOGR('tinypolygon.shp')
tpbb <- toString(c(tinypolygon@bbox[1], tinypolygon@bbox[4],
                   tinypolygon@bbox[3], tinypolygon@bbox[2]))
tpbb <- gsub(', ', ' ', tpbb)

system2(gtrans111, args= c('-projwin', tpbb, 'tinyraster.tif', 'tinyclip2.tif' ))

or even bypass creating any R objects by just inputting the polygon bounding coordinates directly

system2(gtrans111, args= c('-projwin', '148.665 -20.88 148.67 -20.884', 
                           'tinyraster.tif', 'tinyclip3.tif'))

The output extent is not identical to the raster package methods, but does align correctly. All of these methods are easy to loop across multiple datasets, which is the real advantage of R. You can do something similar in Python too.

Related Question