find a clever way of connected component labelling for vector data

[marcosci]

No description provided.

[marcosci]

Started in e3b28d0 with that. @Nowosad - can I cast a multipolygon into a single polygon? Currently I end up with a multipolygon after st_union but I would need a polygon.

get_patches.sf is than also a candidate for landscapetools.

[Nowosad]

I am going to take a look at it now. One suggestion though - I think we should avoid naming an example object "landscape" again.

[marcosci]

I am going to take a look at it now. One suggestion though - I think we should avoid naming an example object "landscape" again.

😆 You are totally right, a bad habitat that caused many headaches. The function is a bit rough, needs definitely a bit of tweaking - was just a proof of concept for us that it could work similar.

I was following this issue here:
r-spatial/sf#234

... and the labeling works quite fine (st_relate is awesome). However, the st_union always returns a MULITPOLYGON and can't be casted in a single polygon - which we would need to have a similiar behaviour than in landscapemetrics.

btw to keep you in the loop: We have a student working on the vector metrics and despite being not that familiar with R (she read geocompation with R - and her level afterwards was incredible 👍 ) she has done an awesome job so far - she finished all the metrics on patch level with vectors.

[Nowosad]

Ok, I think I understand the code now @marcosci. I will try to work on cleaning it today/tomorrow.
I can give you a simple response now to "can I cast a multipolygon into a single polygon?". No, you cannot. I will explain later today.

[marcosci]

👍 Thanks Jakub, no rush!

[Nowosad]

http://postgis.net/docs/using_postgis_dbmanagement.html - search for the "(j-m) cannot be represented as single POLYGONs" phrase.

To expand on this topic:

1. In simple features, a polygon is a simple object, that cannot represent patches connected by a point.
2. They (patches connected by a point) should be represented by multipolygon.
3. In the sf package, each patch can be represented by a row with either POLYGON or MULTIPOLYGON geometry.

[marcosci]

OK, thanks for explaining! simple features are still here and there a little struggle for me.

This basically means that we need to be very carefull what we treat as patch then - if you look at:

https://github.com/marcosci/vectormetrics/blob/master/R/vm_p_area.R

... I thought that the most logical input would either be a sfc with a multipolygon for each class, or single polygons for each patch of a class. Hence, when we would cast the object to a polygon, we would end up with the same structure.

But it makes sense now, that this was a rather naive view ... Need to think about that 👍