Zonal Statistics overestimates area

[PamFroemke]

I'm having considerable difficulty getting a correct value for area from the zonal statistics tool (64-bit Win XP, ArcMap 10.1, SP1). I have Excel county ag data tables that I want to summarize by watershed.

My procedure is:
1. Calculate zonal stats for ag area by county FIPS (zone data = county raster, value data = NLCD ag area raster)
2. Join stats table to county polygon data
3. Add and calculate a new field to be equal to the stats AREA field (I do this so I have a meaningful field name at the end of my analysis)
4. Intersect joined county ag polygons with watershed polygons
5. Calculate the proportion of county ag area in each new polygon so I can sum the county ag area in each watershed
([area of county ag within intersected polygon] / [total ag area in county])

When I use this procedure and then summarize the ag area proportions for each watershed I get many values greater than 1.0. This leads to overestimates of ag area for a lot of the watersheds (most, but not all, of which are along the coastlines).

I'm using the exact same procedure to estimate the proportion of each county in the watersheds. When I sum all the watershed proportions, I get total values close to 1.0, which is what they should be.

A rather convoluted question, hope this all makes sense.
Thanks!

[curtvprice]

When doing zonal statistics, I recommend making sure you are minimizing resampling errors. The polygon area and zone area (after rasterizing) are not the same - and how close they are is controlled by the cell size. It's best practice to process using the cell size and snap raster set to your value raster, so you minimize resampling. You can set the cell size smaller if you find you need more accuracy than that.

As for the results adding up, I would suggest getting percentages at step 2 by creating a table of summary area from your zonastats output table by county using the Summary Statistics tool and joining to that instead of back to the original polygon areas. An alternative approach is to  run the Pivot Table tool on your zonalstats output table.

Raster is faster, but vector is correcter!