Identify regions in common on two rasters

227
5
07-08-2020 07:02 PM
RachelBadlan
New Contributor

Hi,

Am new here so excuse me if my question is naive! I have a terrain-filter raster in pink and another raster filter in purple (shown overlying the first). I am only interested in the areas where the pink and purple are connected.  In other words, I want to remove all the purple areas that are not associated with (adjacent to or overlayed) the pink regions, leaving just the areas where the two filters touch.  I then want to merge them together to make a filter combining both (I do not need to separate the pink from the purple - I just want to identify regions where the two filters are connected. 

How can I achieve this?Pink filterPink filter overlain by purple filter

Tags (1)
0 Kudos
5 Replies
DanPatterson
MVP Esteemed Contributor

several ways, but Combine will give you some flexibility

Combine—Help | Documentation 

It will produce the unique combinations of the inputs, and you can query the output accordingly

purple  pink

1           1

1            nodata

nodata   1

nodata   nodata

so you are interested in the first two classes from the resultant


... sort of retired...
0 Kudos
RachelBadlan
New Contributor

Thanks! Have used the combine tool (below), but how do I remove the values of '3' (which represent areas of light purple) that are isolated? i.e. are not next to a '2' or a '4'? (See pic below for example of areas to remove and keep)

Thanks in advance!

0 Kudos
DanPatterson
MVP Esteemed Contributor

Rachel‌ ... when you mean 'adjacent' to you mean within 1 cell? 2 cells? adjecent-ish?

In that case I would use the Expand tool on the 'pink' and effectively buffer them, then use combine as before except exclude those cases that are denoted as  1, nodata from above.

I think you will have corner cases in any event.

Addendum

I would also have a look at the Generalization toolset to prepare input rasters first.

Some combination of the tools in there and a combine (perhaps) might get you close

How Aggregate works—Help | Documentation 

Boundary Clean—Help | Documentation  did you look at this?

Majority Filter—Help | Documentation  did you try it?

How about a focal variety? Focal Statistics—Help | Documentation 

The list goes on, some things working in some areas and not in others 


... sort of retired...
0 Kudos
DanPatterson
MVP Esteemed Contributor

Brought up an old thread perhaps closely related

https://community.esri.com/thread/219411-how-to-count-the-adjacent-cells-that-have-a-different-value... 


... sort of retired...
0 Kudos
curtvprice
MVP Esteemed Contributor

Not a dumb question, this is a tricky raster processing puzzle. Fortunately, the Raster Calculator  makes it fairly easy to combine tools to get your answer. Map Algebra is cool.

A key part of this is the use of the Con (conditional) tool to make choices about what cell values you want.  Note I'm assuming the cells you don't want are zero (not NoData), this is based on your second screen shot.

1. Make a mask of combined grids where purple==1 and pink==2.  Where pink and purple cells overlap, you get a value of 1, if neither are >0, NoData.

If using raster calculator tool, make the output cgroup and start the expression AFTER the = sign.

cgroup = Con("pink.tif" == 1, 1, Con("blue.tif" == 1, 2))‍‍‍

2. Convert all data cells to 1 (cgroup > 0 creates a grid that has values of 1 and nodata).

Make zones (unique cell value) for each continuous group of "data" cells

mgroup = RegionGroup(cgroup > 0)‍‍‍‍‍‍‍

3. Select all cells that are in a grouped region (zone) with more than one of the two cgroup values, extract cgroup values from those zones.

Pick one of the following depending on what you want the output to look like.

mgroupv = Con(ZonalStatistics(mgroup, "VALUE", cgroup, "VARIETY") > 1, 1)‍‍‍‍‍‍‍

The above returns cells with values of 1 and NoData. You can use Con one last time to convert your NoData to zero if you want (like your input rasters)

mgroupv0 = Con(IsNull(mgroupv), 0, mgroupv)‍‍‍
0 Kudos