(Update) My final goal is to take some logged point data (1 point about every 20 ft or so there are over 15,000 points) and interpolate that point data. Then I would like to convert the interpolated raster into distinct polygon zones based on the number of classes the data is in using the Natural Breaks classifications. So if I want it in 5 classes, there will be several polygons that make up these 5 classes, but within the tabular data I can see which of the 5 classes each polygon belongs to. In addition to this, I would like to eliminate the smaller polygons created by the interpolation, like say anything under 1 acre in size. Maybe there is a GP process I can run to do this? Let me know if I need to explain this better, and thanks for the help so far.

I was thinking the "output cell size" setting would be the trick here but it defaults to a very small number (0.00003) and if I make it any bigger like even say "0.003" it results in the interpolated raster having virtually no data where both high and low values is a really high number and very low number.

If I use the default settings I get a decent kriged raster but I would like to adjust to exclude smaller areas, like say less than 1 acre in size. Basically I want the kriging interpolation to ignore areas that are smaller than an acre or even a 1/2 acre. Currently it is showing areas that are easily less then 0.001 acres in size of which I would like it to ignore those. Also if I zoom into it looks like the cell size grids are 8ft by 8ft. How do I get that larger to say 100 ft by 100 ft? Like I said I have tired putting larger number in the output cell size but get back basically nothing but extreme values. I must be doing something wrong. Any help would be appreciated.

  • 2
    I'm sorry, I find most of this question incomprehensible--kriging isn't about finding areas or excluding them--but one thing is clear: you are using unprojected data; the values 0.00003 etc. are in decimal degrees; you need to use a projected coordinate system for any kind of kriging. To make more progress, consider explaining what you're trying to accomplish rather than what you are failing to accomplish.
    – whuber
    Sep 14 '11 at 2:42
  • 1
    1) what are you trying to predict 2) how many datapoints do you have
    – johanvdw
    Sep 14 '11 at 11:19
  • whuber, thanks for the tip, I didn't realize I couldn't use WGS_84 with kriging. So I converted it to Web Mercator and it seems to work better. I'll update my post with what I'm trying to accomplish. Sorry for the confusion
    – wilbev
    Sep 14 '11 at 20:12
  • johanvdw, Please read my update above. There is over 15,000 data points. Not sure where you're going with what I'm trying to predict. Basically try to interpolate the logged point data and convert those into distinct polygons.
    – wilbev
    Sep 14 '11 at 20:26
  • Wouldn't that be a post-processing step rather than something directly related to Kriging?
    – underdark
    Sep 14 '11 at 20:56

The details about kriging, five classes, and natural breaks are irrelevant: the question ultimately asks about how to "eliminate" small polygons in an ordinally classified grid. One of the simplest ways uses the generalization operators, especially nibble. To apply these, you first have to identify the small-polygon patches. Do this by regiongrouping the grid. Convert that to the nibble mask by setting the small-count features to NoData via SetNull.

This process is fairly crude: any cell in a small polygon is assigned the value of its nearest large neighbor, regardless of how close that neighbor's value might be to the cell's original value. Expand provides more control. It allows you to specify which values a large polygon can "absorb." For instance, you might use it to allow large polygons to expand into any neighboring small polygons having an adjacent value. This would require cycling through several expand operations, one per value. The result might depend upon the sequence of expansions and could still leave a few small "orphan" features. Polishing the procedure with a final nibble would take care of those.

Another option is to return to the continuous grid on which the classification was based. Smooth it, such as by means of a focal mean or focal median. Start with a circular neighborhood approximately as large as the polygons to be eliminated and iteratively adjust its radius to achieve the desired effect. The smoothing will tend to reduce the fluctuations, which when classified will create fewer small polygons. Use con to paste the new classifications only into the small polygons (identified, as before, with regiongroup). Experimentation and iteration will be needed.

  • whuber, thanks for comments and suggestions. I had no idea it was going to be this involved for a concept that seems simply. So with the 1st option, I ran into a problems first off with the many of the GP tools producing outputs as doubles when many of the GP tools require integers. So I kept having to use the Int tool. I wish there was a way to force integer values in the rasters for the outputs?? So the first option you gave, has many steps, and as I got thru them, the end map is definitely smoothed but looks too different from the original. The second option is easier but still....
    – wilbev
    Sep 21 '11 at 0:50
  • .....leaves some small areas. Neither is exactly what I'm looking for. I think the focal mean is the closest but I don't like how it makes some of regions smaller while others it makes bigger.
    – wilbev
    Sep 21 '11 at 0:51
  • whuber, one last question for you. I don't understand what you mean by [con] to paste the new classifications only in the small polygons. Can you elaborate what you mean by that? Is that something I would add during the focal statistics gp?
    – wilbev
    Sep 21 '11 at 3:43
  • 1
    @wilbev I added the missing link to the help page for con. I don't understand the issues about forcing integer output, because all the operations (except focal stats) described here have integer output when applied to integer input, which is what you are starting with (the classified grid). Finally, if there's something you don't like, that means you are applying additional unstated criteria. If you could characterize what is "likable" and what is not, we might be able to point you in the right direction.
    – whuber
    Sep 21 '11 at 14:52
  • I'm still not understanding the use of the con tool with the focal stat. I don't think the regiongroup is working well on my data, since it only reduces the number of rows from 18K to 11K. Like it's not grouping well enough. On the integer problem, it starts with Kriging, since it outputs floating pt, & so does SetNull. I wish I could prevent that. So what I'm looking for(likable), is basically deleting areas/regions/polygons smaller than a specific size (<1000 sq meters). Maybe I should do after RasterToPolygon but unsure how I could delete polygons of certain size?
    – wilbev
    Sep 21 '11 at 18:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.