2 Replies Latest reply on Jun 21, 2012 10:11 AM by erice-esristaff

    Classification Statistics vs. Zonal Statistics

      Hi everyone,

      I am trying to calculate the mean population density of India given a population density layer (raster) and the country's border (shapefile). In order to do this, I ran the zonal statistics as table tool, where the zone layer was the shapefile and the value layer was the population density layer. This worked and I got the statistics. However, I remembered that Arc also calculates classification statistics when prompted. After doing so for the population density layer, the classification statistics look quite different than those given by the zonal statistics table (min and max values are the same but otherwise the rest are different!). I think this might be because I chose to ignore NoData during the zonal statistics step. Is that possible?
      Can anyone explain the difference in results or have any thoughts?  

      Thank you for your help and thoughts!! I appreciate it.

        • Re: Classification Statistics vs. Zonal Statistics
          I´m wondering about the same question during my current work. The same difference occurred, when I ran both tools. Did you do a supervised classification? In my case I classified the maximum likelihood. By classifying with this tool,
          you include a set of values to one class, while otherwise in zonal statistics the value of all pixels below your mask (shapefile) is estimated. That might cause the visual difference in your result. But I can´t answer the question,
          whether a classification or pure zonal statistics is more reliable.
          Maybe someone else knows about the purpose and background, both of these tools are applied and interpreted.

          Best regards,

          • Re: Classification Statistics vs. Zonal Statistics
            The classification statistics box on the Classification dialog (on the classified renderer) are an estimate. Not all of the values are used when the input is floating point which typically a density raster would be.  It would be too slow.  If you want to have more accurate stats in this dialog you can follow the steps outlined in one of our knowledge base documents.

            HowTo:  Compute the histogram with all the unique values, when using the classified renderer

            Johannes-  Zonal Stats does not estimate the value of all pixels below your mask. Maximum Likelihood wouldn't be used on a Density surface.  Please describe the issue you are seeing with your data.