Zonal Statistics "area-weighted max" or other similar statistic...

Discussion created by on Feb 24, 2014

I�??m looking for a tool or method to evaluate and rank a set of polygons, representing proposed development rights-of-way (ROWs), using a raster with integer values ranging from 1-880.  Since the raster cells are roughly 2-3 orders of magnitude smaller in spatial extent than the polygons themselves, using zonal statistics would not introduce too much error and may be the most straightforward approach, although the rasters could instead be converted to polygons and used in a polygon in a polygon analysis. 

My dilemma is this:  the input raster that would be used to rank the ROWs is a layer that shows general habitat value for multiple species.  In this layer there are areas that fall within the ROW polys where a relatively small number of cells could have high biological importance, and these should not be completely overwhelmed by their more average neighbors; a mean value is probably inappropriate. However, at the same time, using a statistic based simply on the maximum raster value that intersects would likely inflate the biological importance of certain ROWs and lead to misleading results.  The desired �??sweet spot�?� would be some statistic that was sensitive to high value areas that represent a relatively low proportion of the whole but not overwhelmed by them.

The Spatial Analyst Zonal Statistics tool produces estimates of mean, majority, maximum, median, minimum, minority, range, sd, sum, and variety.  None of these seem appropriate on their own, but some combination of them might be used to produce an index for ROW ranking that was biologically appropriate. Some sort of "area-weighted maximum" seems to be what I'm ultimately after, but I'm yet to find any clues on how to pull that off in Arc. Does anybody have any advice or comments on this type of approach?

The second route would be a polygon in polygon analysis like that done by the IsectPolyPoly tool in GME (  This tool produces an area weighted mean as well as an area weighted sum.  Since the ROWs vary in size, though, and sum would favor the larger ROWs, sum cannot be used directly.  Area weighted mean likely has the problems mentioned in the second paragraph above.  However,  area weighted sum could be normalized by ROW length using that tool�??s output, or maybe some combination statistic could be ginned up using one of the two area weighted statistics in combination with the others that the tool produces (min and max).  Alternatively perhaps another polygon in polygon analysis creates a more appropriate statistic.  Again, any advice would be greatly appreciated. 

There may be some other approach besides the two mentioned above that would fit the intended goal.  I�??m open to any and all suggestions.

Thank you all very much in advance for your help.